You are on page 1of 456

VOLUME NINETY FOUR

ADVANCES IN
PARASITOLOGY
Mathematical Models for
Neglected Tropical Diseases:
Essential Tools for Control and
Elimination, Part B
SERIES EDITOR
D. ROLLINSON J. R. STOTHARD
Life Sciences Department Department of Parasitology
The Natural History Museum, Liverpool School of Tropical
London, UK Medicine Liverpool, UK
d.rollinson@nhm.ac.uk russell.stothard@lstmed.ac.uk

EDITORIAL BOARD
T. J. C. ANDERSON R. C. OLIVEIRA
Department of Genetics, Texas Centro de Pesquisas Rene Rachou/
Biomedical Research Institute, CPqRR - A FIOCRUZ em Minas
San Antonio, TX, USA Gerais, Rene Rachou Research
Center/CPqRR - The Oswaldo Cruz
M. G. BASA  NEZ
~ Foundation in the State of Minas
Professor of Neglected Tropical Gerais-Brazil, Brazil
Diseases, Department of Infectious
Disease Epidemiology, Faculty of R. E. SINDEN
Medicine (St Mary’s Campus), Immunology and Infection
Imperial College London, Section, Department of Biological
London, UK Sciences, Sir Alexander Fleming
Building, Imperial College of
S. BROOKER Science, Technology and
Wellcome Trust Research Fellow Medicine, London, UK
and Professor, London School of
Hygiene and Tropical Medicine, D. L. SMITH
Faculty of Infectious and Tropical, Johns Hopkins Malaria Research
Diseases, London, UK Institute & Department of
Epidemiology, Johns Hopkins
Bloomberg School of Public Health,
R. B. GASSER Baltimore, MD, USA
Faculty of Veterinary and
Agricultural Sciences, The R. C. A. THOMPSON
University of Melbourne, Parkville, Head, WHO Collaborating Centre
Victoria, Australia for the Molecular Epidemiology
of Parasitic Infections, Principal
N. HALL Investigator, Environmental
School of Biological Sciences, Biotechnology CRC (EBCRC), School
Biosciences Building, University of of Veterinary and Biomedical
Liverpool, Liverpool, UK Sciences, Murdoch University,
Murdoch, WA, Australia
J. KEISER
Head, Helminth Drug X.-N. ZHOU
Development Unit, Department Professor, Director, National
of Medical Parasitology and Institute of Parasitic Diseases,
Infection Biology, Swiss Tropical Chinese Center for Disease Control
and Public Health Institute, Basel, and Prevention, Shanghai, People’s
Switzerland Republic of China
VOLUME NINETY FOUR

ADVANCES IN
PARASITOLOGY
Mathematical Models
for Neglected Tropical
Diseases: Essential Tools for
Control and Elimination, Part B
Edited by
 NEZ
MARIA GLORIA BASA ~
Department of Infectious Disease Epidemiology,
Faculty of Medicine, Imperial College London,
London, UK

ROY M. ANDERSON
Department of Infectious Disease Epidemiology,
Faculty of Medicine, Imperial College London,
London, UK

AMSTERDAM • BOSTON • HEIDELBERG • LONDON


NEW YORK • OXFORD • PARIS • SAN DIEGO
SAN FRANCISCO • SINGAPORE • SYDNEY • TOKYO
Academic Press is an imprint of Elsevier
Academic Press is an imprint of Elsevier
125 London Wall, London EC2Y 5AS, United Kingdom
The Boulevard, Langford Lane, Kidlington, Oxford OX5 1GB, United Kingdom
50 Hampshire Street, 5th Floor, Cambridge, MA 02139, United States
525 B Street, Suite 1800, San Diego, CA 92101-4495, United States

First edition 2016

Copyright © 2016 Elsevier Ltd. All rights reserved.

No part of this publication may be reproduced or transmitted in any form or by any means, electronic or
mechanical, including photocopying, recording, or any information storage and retrieval system, without
permission in writing from the publisher. Details on how to seek permission, further information about
the Publisher’s permissions policies and our arrangements with organizations such as the Copyright
Clearance Center and the Copyright Licensing Agency, can be found at our website: www.elsevier.com/
permissions.

This book and the individual contributions contained in it are protected under copyright by the Publisher
(other than as may be noted herein).

Notices
Knowledge and best practice in this field are constantly changing. As new research and experience
broaden our understanding, changes in research methods, professional practices, or medical treatment
may become necessary.

Practitioners and researchers must always rely on their own experience and knowledge in evaluating and
using any information, methods, compounds, or experiments described herein. In using such information
or methods they should be mindful of their own safety and the safety of others, including parties for
whom they have a professional responsibility.

To the fullest extent of the law, neither the Publisher nor the authors, contributors, or editors, assume any
liability for any injury and/or damage to persons or property as a matter of products liability, negligence
or otherwise, or from any use or operation of any methods, products, instructions, or ideas contained in
the material herein.

ISBN: 978-0-12-809971-1
ISSN: 0065-308X

For information on all Academic Press publications visit our


website at https://www.elsevier.com

Publisher: Zoe Kruze


Acquisition Editor: Alex White
Editorial Project Manager: Helene Kabes
Production Project Manager: Magesh Kumar Mahalingam
Cover Designer: Greg Harris

Typeset by TNQ Books and Journals


CONTRIBUTORS

R.M. Anderson
London Centre for Neglected Tropical Disease Research, London, United Kingdom;
School of Public Health, Imperial College London, London, United Kingdom
R.F. Baggaley
London School of Hygiene and Tropical Medicine, London, United Kingdom
M.G. Basan~ ez
Imperial College London, London, United Kingdom
I.M. Blake
Imperial College London, London, United Kingdom
L.E. Coffeng
University Medical Center, Rotterdam, Rotterdam, The Netherlands
O. Courtenay
University of Warwick, Coventry, United Kingdom
S.H. Farrell
London Centre for Neglected Tropical Disease Research, London, United Kingdom;
School of Public Health, Imperial College London, London, United Kingdom
M. Gambhir
Monash University, Melbourne, VIC, Australia
C. Holland
Trinity College Dublin, Dublin, Ireland
T.D. Hollingsworth
University of Warwick, Coventry, United Kingdom
R.J. Kastner
Swiss Tropical and Public Health Institute, Basel, Switzerland; University of Basel, Basel,
Switzerland
G.F. Medley
London School of Hygiene and Tropical Medicine, London, United Kingdom
A. Pinsent
Monash University, Melbourne, VIC, Australia
R.J. Quinnell
University of Leeds, Leeds, United Kingdom
K.S. Rock
University of Warwick, Coventry, United Kingdom

ix j
x Contributors

P. Steinmann
Swiss Tropical and Public Health Institute, Basel, Switzerland; University of Basel, Basel,
Switzerland
W.A. Stolk
University Medical Center, Rotterdam, Rotterdam, The Netherlands
C.M. Stone
Swiss Tropical and Public Health Institute, Basel, Switzerland; University of Basel, Basel,
Switzerland
M. Tanner
Swiss Tropical and Public Health Institute, Basel, Switzerland; University of Basel, Basel,
Switzerland
F. Tediosi
Swiss Tropical and Public Health Institute, Basel, Switzerland; University of Basel, Basel,
Switzerland
J.E. Truscott
London Centre for Neglected Tropical Disease Research, London, United Kingdom; School
of Public Health, Imperial College London, London, United Kingdom
H.C. Turner
London Centre for Neglected Tropical Disease Research, London, United Kingdom; School
of Public Health, Imperial College London, London, United Kingdom
S.J. de Vlas
University Medical Center, Rotterdam, Rotterdam, The Netherlands
M. Walker
Imperial College London, London, United Kingdom
PREFACE

This is the second (Part B) of two volumes for Advances in Parasitology on the
topic of ‘Mathematical Models for Neglected Tropical Diseases: Essential
Tools for Control and Elimination’. Since the publication of Part A in
2015, considerable progress has been made, in this research field, and under
the fillip of both the NTD Modelling Consortium (http://www.
ntdmodelling.org/) and the London Centre for Neglected Tropical Disease
Research (LCNTDR) (http://www.londonntd.org/), to bring together
diverse modelling groups and approaches: (1) to scrutinize the effectiveness
of current interventions for achieving the goals set in 2012 by the World
Health Organization (WHO) (http://apps.who.int/iris/bitstream/10665/
70809/1/WHO_HTM_NTD_2012.1_eng.pdf) for the control or elimina-
tion of 17 NTDs by 2020, and (2) to investigate how best to deploy alter-
native/complementary strategies should the former be deemed insufficient
in all endemic settings. In January 2013, the WHO published its second
report on NTDs (http://www.who.int/neglected_diseases/2012report/
en/), aiming to sustain the drive fuelled by the original roadmap and in
May that year, the 66th World Health Assembly adopted resolution
WHA66.12, calling on member states to intensify and integrate measures,
as well as to plan investments to improve the health and social well-being
of populations affected by NTDs (http://www.who.int/neglected_
diseases/WHA_66_seventh_day_resolution_adopted/en/). To this end,
mathematical models of NTD transmission and control have a crucial role
to play in guiding programmes and helping policy makers to plan those strat-
egies that have the highest impact with available intervention tools and to
identify those that are the most cost effective. Models also help in designing
monitoring and evaluation protocols to assess progress towards the 2020
goals.
In October 2015, the NTD Modelling Consortium published a thematic
collection of papers on quantitative analyses of the current strategies to
achieve the 2020 goals for NTDs (https://www.biomedcentral.com/
collections/ntdmodels2015), and in January 2016 the LCNTDR launched
a collection of papers to highlight recent advances in scientific research for
NTD control (http://www.parasitesandvectors.com/series/LCNTDR).
This volume complements both these thematic collections by offering
state-of-the-art reviews on modelling bacterial (trachoma), protozoan

xi j
xii Preface

(leishmaniasis), and helminth (intestinal nematode, schistosome and filarial)


infections with the aim of identifying approaches to support the control and
elimination goals proposed by the WHO and endorsed by the London Dec-
laration on NTDs (http://unitingtocombatntds.org/resource/london-
declaration).
Part B starts with a chapter by Amy Pinsent and colleagues reviewing
mathematical models of trachoma (caused by the bacterium Chlamydia tra-
chomatis) in the context of the four-pronged approach adopted by the Global
Elimination of blinding Trachoma (GET 2020) alliance. This approach relies
on the SAFE strategy (S for trichiasis surgery; A for antibiotic treatment; F
for facial cleanliness and E for environmental improvement) for eliminating
trachoma (the foremost cause of infectious blindness worldwide). This chap-
ter highlights that although a large body of work has focussed on modelling
the effectiveness of antibiotic administration (A) in reducing or interrupting
trachoma transmission, much less work has been conducted to model the
effect of nonpharmaceutical interventions (the F and E components of
SAFE). Two further gaps were identified, the former relating to a scarcity
of models linking infection and disease to enable rigorous evaluation of
progress towards GET 2020 (elimination of blinding trachoma) and the latter
on the little consideration that has been given to modelling the cost effec-
tiveness of the SAFE interventions.
The chapter by Kat Rock and coworkers reviews progress in the math-
ematical modelling of the life-threatening form of leishmaniasis, visceral
leishmaniasis (VL). VL (also known as kala-azar and caused by members
of the protozoan parasite species complex Leishmania donovani) has been tar-
geted for elimination as a public health problem on the Indian subcontinent
(where it is deemed to be mainly an anthroponosis). In parallel to issues dis-
cussed in the chapter on trachoma, quantifying progress towards elimination
of an NTD as a public health problem is not a trivial pursuit. It requires a
fundamental understanding of the relationship between infection and disease
as well as of the progression across disease stages. In VL, disease progression
through different stages and the influence of asymptomatic infection and of
post-kala-azar dermal leishmaniasis (PKDL) in diagnosis and infectivity to
vectors are not yet clearly understood. The authors highlight the success
of recent control efforts, which have prompted the original goal of elimina-
tion as a public health problem to be brought forward from 2020 to 2017.
However, they also recommend caution, as modelling and transmission
studies have indicated that asymptomatically infected individuals might
Preface xiii

hinder elimination efforts and/or obscure the true number of leishmaniasis


cases.
Following on with the theme of NTDs whose amelioration as a public
health problem (control rather than elimination) constitutes the 2020 goal,
James Truscott and colleagues review the underlying biology and epidemi-
ology of the three main intestinal nematode (ascariasis, trichuriasis and hook-
worm) infections included under the umbrella term of soil-transmitted
helminthiases (STHs). These authors review efforts to model the transmis-
sion cycle of the (directly transmitted) helminth species (Ascaris lumbricoides,
Trichuris trichiura and Necator americanus/Ancylostoma duodenale) responsible
for STHs and the effects of preventative chemotherapy on their control
and potential elimination. Although considered together under the STH
term, recent modelling work conducted by the authors has shown that
the different epidemiological characteristics of these nematode species means
that a one-size-fits-all mass drug administration (MDA) approach is not
equally suitable for the attainment of control (or elimination) goals. When
linked to intervention cost-effectiveness analyses, these models can helpfully
inform public heath policy. Research gaps identified include a poor under-
standing of the diagnostic performance of the tools currently used to detect
and quantify infection, which consequently hinders efforts to test and
improve the models.
Schistosomiasis – a term encompassing intestinal and urinary schistoso-
miasis caused by a number of (snail-transmitted) schistosome species – is ear-
marked for elimination (of infection) in selected African countries according
to the WHO 2012 roadmap on NTDs. Roy Anderson and coauthors
review modelling frameworks for schistosomiasis transmission and control,
with particular reference to the role of exposure and/or acquired immunity
on age profiles of infection and reinfection by Schistosoma mansoni and Schis-
tosoma haematobium. These authors also discuss the influence of predisposition
to infection and to treatment uptake (adherence/compliance) on model
outcomes of intervention impact and present deterministic and stochastic
approaches. In a similar fashion to the chapter on trachoma, the authors
highlight that key epidemiological measures, such as the basic reproduction
number and the force of infection (in helminths the rate at which new
worms are acquired), are seldom derived from field data and presented in
published modelling studies. Given the ambitions of the WHO to achieve
elimination of schistosomiasis transmission in defined regions, the authors
advise caution, echoing the conclusions of the chapter on VL. Stochastic
model outputs indicate potentially long durations of parasite persistence in
xiv Preface

some communities even after achieving levels of therapeutic coverage above


those required to break transmission. In common with the paper on STH
modelling, this chapter also discusses issues regarding the performance of
available diagnostics, a theme running through the volume and that is the
subject of the chapter by Medley and coauthors.
A review of control programmes and associated models for onchocercia-
sis (river blindness) is presented by Maria-Gloria Basan ~ez and colleagues.
Human onchocerciasis (a Simulium-transmitted, anthroponotic filarial infec-
tion) has been targeted for elimination in selected African countries by 2020
and in 80% of endemic countries by 2025. These authors focus on the com-
parison between the EPIONCHO (deterministic) and ONCHOSIM (sto-
chastic microsimulation) models and discuss how these two distinct
frameworks are addressing the key questions of where can onchocerciasis
be eliminated with current intervention strategies by 2020/2025, and
what alternative strategies could help to accelerate elimination. Currently,
onchocerciasis programmes deliver annual or 6-monthly MDA with iver-
mectin, and the importance of achieving high therapeutic coverage and
minimizing systematic nonadherence is highlighted. However, ivermectin
is deemed to have little efficacy in reducing the survival of the long-lived
(Onchocerca volvulus) adult worms (although it affects the worm’s ability to
produce live progeny). Therefore, the authors discuss a number of alterna-
tive and complementary strategies (including vector control) and their
potential modalities of delivery to infected individuals and endemic
populations.
Graham Medley and colleagues devote the following chapter to intro-
duce the properties of current diagnostic assays for STHs and highlight
the programmatic consequences of their dwindling sensitivity as control
progresses and infection prevalence falls. Using A. lumbricoides as a case study
– due to the availability of high-quality data – the authors investigate how
more sensitive diagnostics would affect important features of MDA
programmes such as: (1) their implementation and duration; (2) the number
of treatment rounds that would be deployed (including the question of
distributing unnecessary treatments); (3) the health impact of the
programme, and (4) the probability of achieving elimination. Although
focussed on the STHs in general, and ascariasis in particular, the issues
addressed in the chapter are relevant to all the NTDs discussed in this
volume.
Finally, Fabrizio Tediosi and coauthors use lymphatic filariasis (LF) as an
example from which to learn lessons when developing an eradication
Preface xv

investment case (EIC). LF is a mosquito-transmitted infection caused by a


number of filarial nematodes, namely Wuchereria bancrofti, Brugia malayi
and Brugia timori (with the former being responsible for >90% of LF cases).
The Global Programme for the Elimination of Lymphatic Filariasis (GPELF)
was established from the outset to aim at elimination of the infection (the
transmission of W. bancrofti involves only humans and mosquitoes) under
the broader Global Alliance for the Elimination of Lymphatic Filariasis
(GAELF). In this chapter, the authors highlight the value and implications
of undertaking a comprehensive analysis of the overall situation prior to
embarking on an elimination or eradication programme and developing
an EIC. Considerations for LF elimination/eradication and its associated
modelling include an assessment of the level and commitment of the neces-
sary investments; the availability and efficacy of treatments (including their
efficacy in killing the long-lived adult worms); the feasibility of scaling up
the coverage of MDA to all endemic communities, the quantification of
the health impact of the programme, its cost effectiveness and the consider-
ation of broader economic benefits.
With the fifth anniversary of the WHO NTD roadmap and the London
Declaration on NTDs approaching in January 2017, the publication of this
volume is timely. The WHO and Uniting to Combat NTDs (http://
unitingtocombatntds.org/) will host an NTD summit in Geneva, Switzer-
land, in April 19–22, 2017. The purpose of this joint summit will be both
to celebrate the progress made thus far and to rally the NTD community
towards the 2020 milestones and beyond, recognizing that increased efforts
are required to push towards 2020 across all NTDs. Thanks to a number of
programmes of unprecedented scale and exemplary public–private partner-
ship, several NTDs are receding, whereas others are well positioned to
achieve the 2020 targets. However, perceived success can be a double-edged
sword, giving the impression that the job is done or leading to donor fatigue.
In fact, some programmes are in the process of winding down or have closed
altogether (e.g., the African Programme for Onchocerciasis Control,
APOC), placing the NTD community at a crucial crossroad. Substantial
implementation, funding and R&D gaps remain and the recently launched
Expanded Special Project for Elimination of NTDs (ESPEN) that will
succeed APOC requires considerable additional investment. As the coordi-
nation of NTD programmes is devolved to endemic countries, the diseases
will have to compete with a myriad of other national health priorities at a
pivotal time in the road to elimination. This time, however, is also
xvi Preface

accompanied by a growing interest in developing and understanding the fea-


tures of novel diagnostics, drugs and vector control strategies for neglected
diseases.
This second volume on the mathematical modelling of NTDs has, there-
fore, been designed to support the NTD community and policy makers, by
describing a set of quantitative tools to enhance scientific understanding of
how best to suppress transmission and reduce morbidity at the community
level. Quantitative analyses, based on mathematical frameworks that capture
the known biology and epidemiology of the aetiological agents responsible
for NTDs, help to identify clearly which are the optimal and cost-effective
strategies for their control. This volume focuses on the strengths and weak-
nesses of current models, on the estimation of key parameters from field epi-
demiological studies and on future research needs. We thank the support of
the editorial team of Advances in Parasitology and hope that our readers will
enjoy this volume as much as we have enjoyed preparing it.

~ ez and Roy M. Anderson


Maria-Gloria Basan
September 2016
CHAPTER ONE

Mathematical Modelling of
Trachoma Transmission, Control
and Elimination
A. Pinsent*, 1, I.M. Blakex, M.G. Bas
an~ ezx, M. Gambhir*
*Monash University, Melbourne, VIC, Australia
x
Imperial College London, London, United Kingdom
1
Corresponding author: E-mail: amy.pinsent@monash.edu

Contents
1. Introduction 2
1.1 Clinical and epidemiological features 2
1.2 Trachoma control and elimination 5
1.3 The role of mechanistic and statistical models 6
1.3.1 Deterministic and stochastic modelling 6
1.3.2 Modelling infection and modelling disease 7
1.4 Aims and objectives of this review 8
2. Methods 8
3. Results 9
3.1 Characteristics of identified studies 9
3.2 Deterministic and stochastic models 24
3.3 Data sets used and model fitting to data 25
3.4 Transmission intensity and the basic reproduction number 29
3.4.1 Estimating the force of infection 29
3.4.2 Estimating the basic reproduction ratio 30
3.5 Acquired immunity: recovery rate and infectivity 32
3.6 Infection and active disease, modelling of disease sequelae 33
3.7 Simulation of interventions, forecasting infection and disease, and analysis of 36
their cost-effectiveness
4. Discussion 40
4.1 Using modelling to determine the feasibility of the GET 2020 goals 40
4.2 Conclusions 41
Acknowledgements 42
References 42

Abstract
The World Health Organization has targeted the elimination of blinding trachoma by
the year 2020. To this end, the Global Elimination of Blinding Trachoma (GET, 2020)
Advances in Parasitology, Volume 94
© 2016 Elsevier Ltd.
j
ISSN 0065-308X
http://dx.doi.org/10.1016/bs.apar.2016.06.002 All rights reserved. 1
2 A. Pinsent et al.

alliance relies on a four-pronged approach, known as the SAFE strategy (S for trichiasis
surgery; A for antibiotic treatment; F for facial cleanliness and E for environmental
improvement). Well-constructed and parameterized mathematical models provide use-
ful tools that can be used in policy making and forecasting in order to help to control
trachoma and understand the feasibility of this large-scale elimination effort. As we
approach this goal, the need to understand the transmission dynamics of infection
within areas of different endemicities, to optimize available resources and to identify
which strategies are the most cost-effective becomes more pressing. In this study,
we conducted a review of the modelling literature for trachoma and identified 23 ar-
ticles that included a mechanistic or statistical model of the transmission, dynamics
and/or control of (ocular) Chlamydia trachomatis. Insights into the dynamics of
trachoma transmission have been generated through both deterministic and stochas-
tic models. A large body of the modelling work conducted to date has shown that, to
varying degrees of effectiveness, antibiotic administration can reduce or interrupt
trachoma transmission. However, very little analysis has been conducted to consider
the effect of nonpharmaceutical interventions (and particularly the F and E compo-
nents of the SAFE strategy) in helping to reduce transmission. Furthermore, very few
of the models identified in the literature review included a structure that permitted
tracking of the prevalence of active disease (in the absence of active infection) and
the subsequent progression to disease sequelae (the morbidity associated with
trachoma and ultimately the target of GET 2020 goals). This represents a critical gap
in the current trachoma modelling literature, which makes it difficult to reliably link
infection and disease. In addition, it hinders the application of modelling to assist
the public health community in understanding whether trachoma programmes are
on track to reach the GET goals by 2020. Another gap identified in this review was
that of the 23 articles examined, only one considered the cost-effectiveness of the in-
terventions implemented. We conclude that although good progress has been made
towards the development of modelling frameworks for trachoma transmission, key
components of disease sequelae representation and economic evaluation of interven-
tions are currently missing from the available literature. We recommend that rapid ad-
vances in these areas should be urgently made to ensure that mathematical models for
trachoma transmission can robustly guide elimination efforts and quantify progress to-
wards GET 2020.

1. INTRODUCTION
1.1 Clinical and epidemiological features
Trachoma is one of 17 neglected tropical diseases (NTDs) prioritized
by the World Health Organization (WHO) for control and elimination
through preventive chemotherapy or intensified disease management strate-
gies (WHO, 2015a). NTDs are mostly responsible for chronic infections/
conditions that can cause severe morbidity in affected individuals, leading
Mathematical Modelling 3

to long-term disability but are deemed to be associated with relatively low


mortality in comparison to acute, epidemic infectious diseases (WHO,
2015a; Hotez et al., 2014). Transmission of NTDs is facilitated by living
conditions that are associated with poverty, such as poor housing and
sanitation, and limited access to clean water and basic health care (Hotez
et al., 2009).
Trachoma is the leading global cause of infectious blindness and is
currently estimated to affect 84 million people across 51 endemic countries
(WHO, 2012). An estimated 1.8 million people are visually impaired as a
result of the disease, of which 0.5 million people are irreversibly blind
(WHO, 2012; WHO, 2015b). Active inflammatory diseasedtrachomatous
follicular and trachomatous inflammatory [TF, TI, according to the WHO
simplified grading scheme (Taylor et al., 2014)]dis caused by infection with
the bacterium Chlamydia trachomatis. Repeated infection with these bacteria
leads to an immunopathological response characterized by scarring of the in-
ner part of the eyelid, and an eventual curling-in of the eyelashes, which
abrades the corneal surface. This can lead to trachomatous trichiasis (TT),
corneal opacity (CO) and blindness. Excess mortality is also reported to
be associated with blinding trachoma (Hotez et al., 2014). Estimates of
the Disability-Adjusted Life Years (DALYs) due to trachoma have been var-
iable. The Global Burden of Disease (GBD) 1990 Study estimated the
burden of trachoma (all ages) to be 144,000 (95% uncertainty interval
[95% UI],104,000e189,000) DALYs, whereas the GBD 2010 Study re-
ported a value of 334,000 (95% UI 243,000e438,000) (Murray et al.,
2012). Other authors have set this figure at least at one million DALYs
(Evans and Ranson, 1995) or as high as 3.6 million, with the highest propor-
tion (72%) contributed by sub-Saharan Africa (Frick et al., 2003). Among
the major causes of blindness in 2010, trachoma represented 5.2% in sub-
Saharan Africa (Naidoo et al., 2014). An accurate quantitative estimate of
the burden of trachoma remains, however, challenging due to several fac-
tors, including scarce data availabilitydwhich limits the ability to estimate
accurately the number of people infecteddand an unresolved issue as to
whether trichiasis should be considered as a disabling disease sequela (Burton
and Mabey, 2009). The economic impact of trachoma in terms of lost pro-
ductivity is estimated to range between US $2.9 and $5.3 billion annually,
rising to US $8 billion if trichiasis is included in the estimate (WHO, 2015b).
Countries reported to have the highest prevalence of infection are
located in East Africa and the Sahel belt; however, trachoma is also prevalent
in Southeast Asia, the Middle East, the Indian subcontinent and Latin
4 A. Pinsent et al.

America (Burton and Mabey, 2009), although the distribution and the
prevalence of infection are far more heterogeneous in these regions in com-
parison to sub-Saharan Africa. While trachoma was previously prevalent in
Europe and North America only 100 years ago, improvements in sanitary
and living conditions resulted in the gradual disappearance of infection
(Burton and Mabey, 2009).
Infection with C. trachomatis is spread through two primary routes. The
first is direct personal contact which could be direct hand contact with an
infected individual or through contact with clothing which has contacted
infectious discharge (Burton and Mabey, 2009). The second route involves
eye-seeking flies (e.g., Musca sorbens) which have contacted the discharge
from an infected person’s eyes or nose (Emerson et al., 2004). For transmis-
sion of infection to be sustained, it must be consistently transmitted from
person to person. The severity of disease experienced by an infected individ-
ual varies with age and hence their duration of exposure to infection (Bailey
et al., 1999; Grassly et al., 2008). Infection with C. trachomatis becomes
shorter in duration and reduced in frequency as individuals age; therefore,
the highest burden of C. trachomatis is observed in young children (Bailey
et al., 1999; Grassly et al., 2008). Repeated infection with age (continuous
exposure) leads to conjunctival scarring, ultimately leading to TT, CO and
blindness as mentioned previously (Burton and Mabey, 2009; West et al.,
1991). Several epidemiological surveys have suggested that severe sequelae
in the form of TT and CO disproportionally affect women in comparison
to men, as a result of women having a higher exposure to the reservoir
source of infection, which is reported to be young children (Courtright
and West, 2004; West et al., 1991).
A number of risk factors for trachoma transmission have been identified,
including (1) secretions from the eye which other individuals may come into
contact with, and which may also attract flies which help to facilitate
transmission (Emerson et al., 2004; Ngondi et al., 2008); (2) overcrowding
within the household, which increases the frequency of contact between in-
dividuals potentially leading to more frequent infection events (Abdou et al.,
2007; Ngondi et al., 2008) and (3) limited supplies of clean water resulting in
infrequent face washing, general poor hygiene practice and lack of easy
access to latrines, which can lead to a buildup of faecal matter in the envi-
ronment which attracts eye-seeking flies (Emerson et al., 2004). Transmis-
sion intensity of trachoma within a community is classified according to
the prevalence of active disease in 1- to 9-year olds. Communities are
considered hyperendemic if the prevalence of active disease in this age group
Mathematical Modelling 5

is >20%, mesoendemic if prevalence is >10% but <20%, and hypoendemic


if prevalence is <10% (Wright and Taylor, 2005).

1.2 Trachoma control and elimination


The WHO has advocated for the elimination of blinding trachoma by 2020,
and the Global Alliance for the Elimination of Trachoma by 2020 (GET
2020) was established to develop criteria to help to achieve this goal. These
criteria are: (1) to reduce prevalence of TF and TI (active disease) to <5% in
1- to 9-year olds across all endemic communities; (2) to reduce the popula-
tion prevalence of TT to <1 per 1000 persons and (3) to enhance the use of
the Facial cleanliness (F) and Environmental improvement (E) components
of the SAFE strategy (Surgery for trichiasis, Antibiotics to combat the infec-
tion, Facial cleanliness, and Environmental improvement) (West, 2003).
The GET 2020 goals were developed to eliminate blinding trachoma by
2020, not to completely eliminate infection in the population. Within the
SAFE framework, as the public health burden of trachoma is reduced, infec-
tion will be controlled, but not completely eliminated (hereafter, therefore,
any reference to elimination refers to the elimination of blinding trachoma,
unless otherwise specified). The WHO endorses the implementation of the
full SAFE strategy in order to treat and control trachoma. As mentioned pre-
viously this comprises four key components, more specifically described as:
(1) surgery in order to correct trichiasis; (2) mass distribution of azithromycin
as the antibiotic of choice used to treat and clear active infection in the com-
munity (topical tetracycline is used in very young children), (3) promotion
of facial cleanliness in order to reduce transmission via eye discharge and (4)
environmental modifications to improve living conditions, ensuring that the
environment no longer helps to facilitate the transmission of infection (this
can be in principle achieved through a number of avenues, such as facili-
tating the provision of clean water, increasing access to well-designed
latrines, helping to reduce the population density of flies, and reducing
overcrowding) (West, 2003).
The London Declaration on Neglected Tropical Diseases (NTDs)
(Uniting to Combat NTDs, 2012) has led to a renewed commitment to
control and/or eliminate these morbidity-causing diseases from some of
the poorest countries in the world. In addition, it is expected that the
recently established Neglected Tropical Diseases Modelling Consortium
(http://www.ntdmodelling.org/) will help to facilitate the exchange and
discussion of ideas across a range of NTDs and will foster and enhance
collaboration between different mathematical modelling groups. This will
6 A. Pinsent et al.

help to address many urgent policy issues concerning the control and elim-
ination of NTDs that can only be answered through the use of quantitative
~ez and Anderson, 2015). Models are ideally suited to answering a
tools (Basan
wide range of questions relating to the possible impact of various interven-
tions within populations affected by NTDs, e.g., which interventions to
deploy, to whom, how often and for how long should they be administered.

1.3 The role of mechanistic and statistical models


The epidemiology and dynamics of trachoma infection arise from a complex
set of contributory factors, including the natural history of infection and dis-
ease, which is governed by an individual’s time-varying immunological
response to and clearance of infection. Transmission of trachoma can be
altered by the behaviour and social contact patterns of people within the
community. At an even greater scale, the climate and local ecology may
affect the transmissibility of the bacterium causative of trachoma, C. tracho-
matis. Integrating these effects across several hierarchical scales is a challenge
for any infectious disease.
However, through careful determination of key features of the epidemi-
ology of infection, models can be formulated as time-dependent mathemat-
ical expressions and solved computationally. These include, but are not
limited to, discrete event, agent-based, or differential equation-based simu-
lations. Equally, a quantitative understanding of data generating processes
can be provided through the development of statistical models. Statistical
models can help to understand the relationship between an outcome of in-
terest and external variables which may be important within the system, and
this can be done through techniques such as linear or logistic regression.
Moreover, epidemiological data rarely capture observations at every point
in the transmission or infection cycle, thus statistical models such as hidden
Markov models can be used to help to provide insights into the dynamic
infection process where certain states of the infection process are not
observed, but the observed epidemiological outcome is dependent on these
states. Mathematical and statistical models, which seek to incorporate key
features of the population biology, demography and ecological covariates
(often referred to as mechanistic models) remain among the only quantita-
tive methods capable of performing this kind of integration.

1.3.1 Deterministic and stochastic modelling


Mechanistic models of disease transmission are commonly divided into
either deterministic or stochastic. Put simply, results from a deterministic
Mathematical Modelling 7

model are fully determined by the parameter values chosen and the initial
conditions and, therefore, model outputs are the same every time the model
is run (for a given parameter set). In contrast, stochastic models contain
inherent randomness (demographic and/or parametric) and the same param-
eter sets (which will have distributions rather than only nominal values) and
initial conditions used will result in a range of output results. Deterministic
models describe the average behaviour of a system and, therefore, the
average of many stochastic realizations tends to approximate the determin-
istic solution.
The deterministic approach may miss some aspects of disease transmission,
especially in the context of small populations and low infection prevalence,
where stochastic fade-out or take-off may play an important role. However,
the simplification gained through the use of deterministic models allows
simpler fitting of models to data to estimate parameters of interest, a more
transparent representation of complex natural histories of infection, along
with realistic population demography. Irrespective of the model structure
used, mathematical models are unlikely to be informative unless they are
fitted to or informed by high-quality baseline and follow-up surveillance
data. The decision to develop a deterministic or a stochastic model should
depend upon the question under investigation. For example, if data from a
small community were analyzed and the possibility of elimination was being
explored, a stochastic model would likely be considered more suitable. How-
ever, if one were analyzing impacts of different interventions at a population
level, a deterministic model may be more appropriate.

1.3.2 Modelling infection and modelling disease


The vast majority of data collected from control programmes within
endemic communities have been obtained through clinical diagnosis of
active disease or the disease sequelae (Jimenez et al., 2015). More recently,
testing for active infection has been performed in clinical field studies using
polymerase chain reaction (PCR) methods to identify active infection from
conjunctival swabs. However, laboratory tests can suffer from cross-
contamination across numerous stages of the DNA amplification process
and the presence of inhibitors of DNA amplification within samples can
also cause amplification problems (Solomon et al., 2003). In addition, the
collection of individual-level PCR data within a community is prohibitively
expensive, limiting its widespread use in surveillance. However, studies that
have been able to collect data on active infection and active disease preva-
lence highlight the complex relationship between active infection and
8 A. Pinsent et al.

disease, whereby individuals can have detectable active disease but do not
have a PCR detectable infection (Solomon et al., 2004a,b). Hence, active
disease can persist much longer than infection (Grassly et al., 2008;
Harding-Esch et al., 2009). This highlights the need for modelling
studies to explicitly account for the period of infection and active disease sepa-
rately in order to ensure the prevalence of active disease is not underestimated.
The first two GET 2020 goals relate to the specific outcomes of active
disease. Therefore, in order to understand if the GET 2020 goals are going
to be achieved in different localities, models of C. trachomatis transmission
must also include progression towards disease sequelae.

1.4 Aims and objectives of this review


In this review we report previously published mechanistic and statistical
mathematical models that have been developed to provide insight into
the transmission, dynamics and control of trachoma. We compare the
different mathematical model structures that have been published, whether
they are stochastic or deterministic and the type of data the models were
fitted to. We consider the availability of data on infection and disease
prevalence, and the estimation of epidemiological parameters such as the
basic reproduction number of the infection, R0. We evaluate how models
have been used to assess the impact of different interventions, and finally
how such models can be used as disease forecasting tools to help to under-
stand how achievable the GET 2020 goals are. We then move to a discussion
about which of these published studies contain results (or useful modelling
frameworks) pertaining to the GET 2020 goals. Finally, we address what
questions require further elucidation before critical trachoma control-related
questions can be answered by mathematical models and provide a perspec-
tive on future modelling directions.

2. METHODS
We performed a review of the literature to address and explore the
mechanistic and statistical models of trachoma transmission and control
that have been developed to date. Our search was performed through
PubMed on the 24th of June, 2015, with no restriction on the year of pub-
lication or language applied. We employed a simple set of broad search terms
pairing the term ‘trachoma’ with the following key words: [model OR
Mathematical Modelling 9

modeling OR modelling], a second search using the keyword [mathemat-


ical] was also performed.
Studies were eligible for inclusion in the literature review if they devel-
oped a mechanistic transmission model or a statistical model which quanti-
fied or modelled the impact of an intervention, through regression analysis
or by fitting different statistical distributions to surveillance data. Statistical
models also eligible for inclusion were those estimating epidemiological pa-
rameters and inferring the underlying dynamics of infection through Mar-
kov models. All study types were eligible if they met these inclusion
criteria. Articles were selected for inclusion by first reviewing the full title
and abstracts of all studies that were identified in the initial review searches.
Studies that did not meet the inclusion criteria after this step were excluded
from the analysis. The full texts of potentially relevant articles were then
independently examined by AP, and data extracted were independently
verified by MG. MG was consulted by AP on queries relating to the inclu-
sion or exclusion of studies analyzed in the review.
Once a set of eligible articles was established, data from each article were
extracted. The data extracted were as follows: (1) the type and structure of
the mechanistic or statistical model applied and/or developed in the study;
(2) why the study was performed, and the data that were used to inform the
modelling analysis presented; (3) whether the study population was classified
as hyperendemic, mesoendemic or hypoendemic for trachoma and (4) key
findings from the study and how they related to the existing GET 2020 goals
(Table 1). We then assessed if epidemiological parameters had been re-
ported, and the values estimated or used in the model were noted. These
were: the transmission rate or transmission coefficient (commonly known
as the beta parameter), the age-specific duration of infection (or its recip-
rocal, the recovery rate), and the basic reproduction number R0. If R0
was not explicitly calculated, we used data on the transmission rate (param-
eter beta) and the recovery rate (denoted as parameter gamma in this review)
to estimate a crude R0 (where R0 is defined as the ratio between these latter
two parameters) (Ray et al., 2009) (Table 2).

3. RESULTS
3.1 Characteristics of identified studies
Our initial two searches yielded 172 publications, with 18 publications
being repeated across the two searches. Of the 154 remaining publications,
10
Table 1 Studies identified through the literature review that have developed mechanistic or statistical models of trachoma transmission
and/or control
Levels of
endemicity Relevance for GET 2020
References Model type Study purpose Type of data used evaluated goals

Assaad and Simple and two- Estimate FOI and Age prevalence of Hyperendemic Changes in transmission
Maxwell-Lyons stage catalytic rates of clinical trachoma have arisen due to
(1966) model acquisition and cases and active socioeconomic change.
loss of active disease in Taiwan These can be detected
disease before a (1960e61) with a catalytic model,
control which showed a
programme profound impact on
transmission.
Parthasarathy Catalytic model Estimate the FOI Data collected from Not stated Illustrates that catalytic
(1967) in the 15 Indian states models can be fitted to
population and between 1959 and age-specific disease
elucidate the 1963. Eye data. Estimates the FOI
epidemiological examination and identifies the
pattern of performed. maximum proportion
trachoma in a of the population likely
highly endemic to have been infected,
area giving insight on the
number of rounds of

A. Pinsent et al.
MDA that may be
required.
Mathematical Modelling
Sundaresan and Catalytic model Determine how Age prevalence of Hyperendemic Estimating the FOI may
Assaad (1973) FOI changes trachoma in be useful to measure
after Taiwan precontrol changes in transmission
implementation (1960e61) and in as a result of control
of school-based 1968e69
treatment
programme
Lietman et al. SIS deterministic, Explore how Endemic prevalence Hypo, meso, Achievable. Mass treat
(1999) partial immunity frequently from Egypt, hyperendemic every 12e24 months
in older age should mass Malawi, Tanzania, in hypo, every 6
groups antibiotic The Gambia and e12 months. In hyper/
treatment be Tunisia mesoendemic areas;
given for settings 100% coverage
with different
baseline
endemicity
levels
Lee et al. (2005) SIS deterministic Investigate Model prevalence Hyperendemic Easier to achieve
and stochastic, whether there over time. Rates of elimination if
only state I are optimal times transmission,b, and treatment is delivered
(infecteds) is of the year for recovery,g after high and before
modelled mass antibiotic calibrated with low season, when
distribution data for 24 transmission rates are at
Ethiopian villages their lowest
(Continued)

11
12
Table 1 Studies identified through the literature review that have developed mechanistic or statistical models of trachoma transmission
and/or controldcont'd
Levels of
endemicity Relevance for GET 2020
References Model type Study purpose Type of data used evaluated goals
Gambhir et al. SIS deterministic, Develop a model Kongwa district, Hyperendemic Need to consider age
(2007) ladder of to explain Tanzania infection profiles and
infection model observed different responders
distribution of (heterogeneity in
community exposure,
infection. predisposition)
Explain
heterogeneous
response to
treatment
Ray et al. (2007) SIS stochastic Investigate how 16 Ethiopian villages. Hyperendemic Biannual treatment
long should PCR data in implemented for
treatment be children aged 1 5 years will lead to
continued in e5 years. Baseline elimination in 95% of
high and follow-up data villages. Need to
transmission at 2, 6, 12, 18 and consider distribution of
settings to 24 months after infection that leads to

A. Pinsent et al.
eliminate mass drug hotspots and the role of
infection administration reintroduction from
untreated communities
Mathematical Modelling
Grassly et al. (2008) 2- and 4-state Infer parameters on 256 people from Mesoendemic Average duration of
Markov model. the natural Gambian villages, infection in young
Looking at history of Jali and Berending, children is long. This
infected and infection, and signs of infection contributes to
active disease how they relate checked every 2 persistence of infection
to demographic weeks over after treatment. Young
and baseline 6 months. children may need
immune Presence of more follow-up.
measurements chlamydial
lipopolysaccharide
antigen tested by
ELISA
Gambhir et al. SIS deterministic, Refine model to Data from The Hypo, meso, Captures age-specific
(2009) ladder of capture age- Gambia and hyperendemic prevalence patterns.
infection model specific Tanzania. Indicates roughly the
prevalence of Infection by number of prior
infection and qPCR. For infections which are
estimate the hyperendemic necessary for the
average number setting age-profiles development of TS and
of infections for of infection, TT
the development infection load
of disease (Tanzania) and rate
sequelae of recovery (The
Gambia)
(Continued)

13
14
Table 1 Studies identified through the literature review that have developed mechanistic or statistical models of trachoma transmission
and/or controldcont'd
Levels of
endemicity Relevance for GET 2020
References Model type Study purpose Type of data used evaluated goals
Blake et al. (2009) SIS stochastic Examine the Cross-sectional data Hypo, meso, Household transmission is
contribution of on prevalence of hyperendemic an important
transmission infection (by contributor to
between and PCR) from Upper incidence and repeat
within Saloum district and infections. Treatment
households Jali village in The only may not be
Gambia and Kahe sufficient. Not
Mpya and Maindi including household
villages in transmission can alter
Tanzania the expected result of
mass treatment
Ray et al. (2009) SIS stochastic Investigate two Data from three Hypo, meso, Graduating communities
different different regions. hyperendemic from a programme
treatment Collected at when infection is
strategies: three baseline, 3 and reduced below 5% is a
annual mass 6 months after reasonable strategy and
treatments in all treatment. Upper could reduce the

A. Pinsent et al.
communities (as conjunctiva swab amount of antibiotic
per WHO) and DNA tested distributed in some
versus three for with PCR. In areas by more than 2-
annual mass The Gambia 1- to fold
treatments but 5-year olds tested;
Mathematical Modelling
stopping In Tanzania and
treatment in Ethiopia data from
those 1-to 10-year olds
communities are used
whose
prevalence falls
<5%
Blake et al. (2010) SIS stochastic. Investigate West and East Africa Hypo, meso, Household-targeted
Differential whether (Upper Saloum hyperendemic treatment produces
transmission targeting District and Jali comparable results to
between antibiotics to village in The mass treatment. Active
children and households that Gambia; Kahe disease not a 100%
adults have at least one Mpya subvillage sensitive marker of
member with and Maindi village infection. Probability
active disease is in Tanzania). The of eliminating infection
effective in Presence of with 10 years of
preventing infection assessed biannual treatment not
infection with PCR. In very high
Calculate the cost- Maindi village, Goals not currently
effectiveness of quantitative PCR achievable
targeted to indicate
household presence of
treatment infection. Clinical
compared with observations also
mass antibiotic used
distribution
(Continued)

15
Table 1 Studies identified through the literature review that have developed mechanistic or statistical models of trachoma transmission
and/or controldcont'd

16
Levels of
endemicity Relevance for GET 2020
References Model type Study purpose Type of data used evaluated goals
Gambhir et al. SIS ladder of Examine the effect Same data as in Blake First model In low initial endemicity
(2010) infection model, of antibiotics on et al. (2009) hyperendemic areas, 1-2 annual mass
deterministic infection and Second model treatments may be
Second model to disease, mixture of sufficient for
look at a disease implications on endemicities elimination. In
only state programmes. hyperendemic areas,
Usefulness of three annual treatments
modelling for result in no lasting
control effect on infection or
programme disease sequelae. Hard
impact to achieve goals.
projections Implement full SAFE
to reduce transmission
Lietman et al. SIS stochastic Assess Prevalence of ocular Hypo (Nepal), Easier to achieve
(2011) reintroduction chlamydial meso elimination of infection
of infection infection by PCR (Tanzania), than previously
following in 24 Ethiopian hyperendemic thought when
treatment communities after (Ethiopia) considering positive
a single MDA feedback (the hazard of
round. a susceptible individual

A. Pinsent et al.
Communities becoming infected per
were retreated, by infectious case
design, after the increasing with
24-month survey prevalence).
Reintroduction is slow
enough to detect
through surveillance
Koukounari Latent Markov Infer the Same as Gambhir Hyperendemic The sensitivity and
et al. (2013) model population et al. (2009), but positive predictive
prevalence of only use data for values of clinical
infection and children aged examination for

Mathematical Modelling
active disease in <10 years. infection were low in
addition to the The Gambia but the
sensitivity and sensitivity of TI and
specificity of positive predictive
three diagnostic value of the clinical
tests examination for TF
and TI were low in
Tanzania. This could
result in communities
experiencing further
rounds of treatment
when it has been
eliminated from
communities.
Liu et al. (2013) SIS stochastic Estimate the 32 communities Mesoendemic The lack of change in R0,
change in R0 per cluster- small negative linear
year following randomized clinical trend and no evidence
treatment. Assess trial in Tanzania. of loss of immunity
the evidence Children 0e5 years suggest repeated
that treatment examined at treatment can help to
can cause loss of baseline and at eliminate infection
short-term 6e36 months after
immunity baseline. Swab for
the presence of
chlamydial DNA.
Mass treatment at
baseline, 12 and

17
24 months
(Continued)
Table 1 Studies identified through the literature review that have developed mechanistic or statistical models of trachoma transmission
and/or controldcont'd

18
Levels of
endemicity Relevance for GET 2020
References Model type Study purpose Type of data used evaluated goals
Liu et al. (2014) SIS stochastic Assess the Same data as in Liu Mesoendemic Efficacy ¼ 68% (95% CI:
prevalence of et al. (2013) 57e75%); no decrease
infection within in efficacy during trial;
the community 89% chance of
following elimination after
treatment. 10 years of annual
Calculate the treatment at 95%
effective field coverage
efficacy of
azithromycin in
clearing ocular
Chlamydia
Martin et al. (2015) Catalytic model Use seroconversion Tanzania Mesoendemic Absence of antibody
allowing for rate to estimate community, Kahe responses in children
change in FOI changes in the Mpya subvillage, born after
FOI after the 575 people. implementation of
implementation Detection of mass treatment reflects
of mass conjunctival swab lack of C. trachomatis
antibiotic DNA and eye transmission. FOI
distribution examination. modelling suggests

A. Pinsent et al.
Prevalence of TF, serology could play a
TI or both, and of role in posttreatment
TS/TT/CO. surveillance
Lietman et al. SIS stochastic Test the hypothesis Gurage, Ethiopia, Hyperendemic Surveying infection in
(2015) that a geometric and the TANA low transmission
distribution study in Amhara settings may be

Mathematical Modelling
describes the Ethiopia; from 24 difficult. Community-
prevalence of communities level cross-sectional
infection in monitored for prevalence may be
different infection. TEF approximated by a
communities study, 1- to 5-year geometric distribution.
where infection old children Its relatively heavy tail
is disappearing monitored suggests that presence
biannually. TANA of an occasional high-
study, 0- to 9-year prevalence community
old children is to be expected, not
monitored, treated necessarily reflecting
annually or transmission hotspots
biannually for or programme failure
42 months. PCR
data
Jimenez et al. Linear and logistic Determine how 283 cross-sectional Hypo, meso, Annual treatment alone is
(2015) regression many annual survey pairs with hyperendemic insufficient. More
statistical models mass treatments baseline and information needed on
are needed to follow-up data, the effects of baseline
achieve mass treatment prevalence, therapeutic
elimination. conducted in 170 coverage and
Assess factors districts. underlying
that affect the Prevalence of environmental and
success of active trachoma hygiene conditions.
reaching (TF or TF/TI) and When trachoma
elimination goals of trachomatous prevalence >30%,
trichiasis (TT) seven or more annual
treatments may be

19
required
(Continued)
Table 1 Studies identified through the literature review that have developed mechanistic or statistical models of trachoma transmission

20
and/or controldcont'd
Levels of
endemicity Relevance for GET 2020
References Model type Study purpose Type of data used evaluated goals
Shattock et al. Multistage Assess whether past 67 Australian Hypo, meso, Current intervention
(2015) infection SIS, trachoma communities hyperendemic strategy unlikely to
stochastic, intervention modelled using achieve 2020 goals.
individual-based efforts have been prevalence data. The likelihood of
model. effective. Data used for all 3 achieving this goal can
Household and Evaluate what different be significantly
community impact can be transmission increased by large-scale
transmission expected if settings. antibiotic distribution
current programmes
intervention accompanied by
strategies are screening, treatment,
maintained. facial cleanliness and
Investigate housing construction
whether shifts in
strategy or
increases in the
intensity of its
implementation
may lead to

A. Pinsent et al.
improved results
Rahman et al. Statistical model Test whether 75 communities in 8 Prevalence in Models correctly predict
(2015) C. trachomatis districts in 1999: 17e79%; that infection
prevalence Tanzania. Survey 2007/08: prevalence across
across 75 infection swab and 0e28% communities where

Mathematical Modelling
Tanzanian PCR. Preschool trachoma is
communities children aged disappearing can be
where trachoma 5 years and under described by an
has been were surveyed exponential
disappearing was distribution
exponentially
distributed
Gambhir et al. SIS deterministic, Explore the impact Maindi in Kongwa, Hyperendemic Increased intensity of all
(2015) ladder of of each of the Tanzania. interventions will
infection model components of Trachoma was reduce TT prevalence
SAFE on disease assessed according and CO incidence.
sequelae. to the WHO Surgery against TT
Investigate simplified grading alone is a stopgap until
whether the scheme. Specimens transmission is reduced
prevalence of were collected for and mass antibiotic
TT can be detection of ocular distribution is
reduced to 1 in C. trachomatis by enhanced. Goals are
1,000, and the real-time PCR achievable if all
incidence of CO interventions are
to 1 in 10,000 implemented
per annum simultaneously. High
transmission settings
may take 20 years to
achieve GET 2020
goals
CO, corneal opacity; ELISA, Enzyme-linked immunosorbent assay; FOI, force of infection; GET, Global Elimination of Blinding Trachoma; PCR, polymerase
chain reaction; q PCR, quantitative PCR; SIS, susceptibleeinfectedesusceptible; TANA, Trachoma Amelioration in Northern Amhara; TEF, Trachoma

21
Elimination Follow-up study; TS, trachomatous scarring; TF, trachomatous follicular; TI, trachomatous inflammatory; WHO, World Health Organization.
Table 2 Values of basic (R0) or effective (Re) reproduction numbers reported in the articles identified in this literature review or calculated

22
(as beta/gamma) where estimates of the rate of transmission (parameter beta) and of the rate of recovery (parameter gamma) were
provided
Beta (week1 unless
otherwise specified Gamma (week1 unless Parameters Reproduction R0,
in original article) otherwise specified) estimated in this number Country and Re,
References (95% CIs) (95% CIs) study (95% CIs) transmission setting R*a

Lee et al. 0.20 month1 0.10 month1 Beta: no, fixed 2.00 Ethiopia (hyper) R0b
(2005) Recovery: no,
fixed
Ray et al. 0.047 0.017 Beta: yes, MLE 3.16 Ethiopia (hyper) R0b
(2007) Recovery: yes,
MLE
Blake et al. 0.29 (0.16e0.51) 0.058 Beta: yes 1.25 The Gambia (meso) R *a
(2009),c 0.76 (0.39e1.40) Recovery: no 2.81 The Gambia (hyper)
(1/17.2 weeks)
1.73 (1.18e2.37) 1.18 Tanzania (meso)
1.70 (1.15e2.46) 2.65 Tanzania (hyper)
Ray et al. 0.052 0.052 (0.000e0.113) Beta: yes, MLE 1.01 (0.50e1.27) The Gambia (hypo) R0
(2009) 0.033 0.037 (0.001e0.073) Recovery: yes, 0.89 (0.66e1.36) Tanzania (meso)
MLE
0.039 0.0123 (0.005e0.02) 3.14 (2.51e3.77) Ethiopia (hyper)
Lietman 0.014 (0.007e0.021) 0.017 (0.01e0.024) Beta: yes, MLE 1.15 Ethiopia (prevalence of R0b
et al. 0.019 (0.016e0.052) 0.017 (0.013e0.041) Recovery: yes, 1.16 ocular Chlamydia

A. Pinsent et al.
(2011) MLE ranging from 5% to
0.014 (0.007e0.029) 0.014 (0.009e0.023) 1.11 75%)
Mathematical Modelling
0.019 (0.015, 0.037) 0.014 (0.011e0.024) 1.84
Liu et al. 0.233 (0.210e0.258) 0.167 month1 Beta: yes 1.40 (1.26e1.55) R0 for first year, prior to R0
(2013) month1 (1/6 months) Recovery: nod interventions.
Tanzania (hyper) at
baseline
Liu et al. 0.229 (0.202e0.262) 0.167 month1 Beta: yes 1.37 (1.21e1.57) Tanzania (hyper) at Re
(2014) month1 (1/6 months) Recovery: no baseline of trial
Gambhir 1.8 (1.6e2.1) year1 0.066 (0.043e0.154) Beta: Yes 1.10 The Gambia (hypo) Reb,e
et al. month1 (first
e
(2009), infection, for R0)
2.4 (2.0e2.9) year1 0.357 (0.313e0.417) Recovery: yes, 1.40 Tanzania (meso) Reb,e
month1 (after many NGMM
infections, Re)
27.7 (21.8e35.1) or 3.20e Tanzania (hyper) Reb,e
18 year1
R0, basic reproduction number; Re, effective reproduction number.
a
R*, Household basic reproduction number.
b
Derived from parameters given in the article but not explicitly calculated by the authors of the original study.
c
Values for beta are bG, Global (household/community) transmission coefficient; MLE, maximum likelihood estimation.
d
Liu et al. (2013) estimate beta for each year of the study assuming a fixed rate of recovery.
e
Values for Re calculated using the next-generation matrix method (NGMM) (unpublished calculation) from the best fit parameters estimated in Gambhir et al. (2009).

23
24 A. Pinsent et al.

24 articles were deemed appropriate after reviewing the title and abstracts of
all unique publications. The full text of these articles was obtained and
assessed. After this second screening stage, 23 articles were deemed to
meet the review’s inclusion criteria (Table 1). Twenty of these publications
presented mechanistic transmission models, and four presented statistical
models.
Of the 20 articles identified as describing transmission dynamics of
trachoma, nine used a deterministic framework (including four catalytic
models), nine used stochastic models, and one used both frameworks.
One of the statistical models used linear and logistic regression, two used a
latent/hidden Markov model and one fitted different statistical distributions
to surveillance data. We identified 11 studies that attempted to estimate the
force of infection (FOI) (the per susceptible incidence rate), the transmission
rate parameter and/or the recovery rate from infection. Although only five
studies calculated R0 explicitly, it was possible to calculate an approximate
value of R0 from the data presented in two other studies (Table 2). Only
five studies incorporated any form of acquired immunity as a result of suc-
cessive infections. We identified only one study that used serological data to
make inferences on historical patterns of transmission. All 23 studies had
conclusions that could provide insight into the achievability of the GET
2020 goals within the endemic locality analyzed. However, we only iden-
tified one study that fitted the model to epidemiological data collected
before, during and after an intervention had been implemented.

3.2 Deterministic and stochastic models


Two main research groups were identified which have developed mecha-
nistic models of trachoma, one group based at the University of California
San Francisco’s (UCSF) Francis Proctor Foundation and the second at Im-
perial College London (ICL). The mechanistic models developed by the
UCSF group have largely been stochastic, although their earlier work was
deterministic (Lee et al., 2005; Lietman et al., 1999). In these studies a single
state variable, P, representing the overall prevalence of infection in the com-
munity (or in a single constant sized age group, commonly 0- to 5-year olds)
is modelled (Lee et al., 2005; Ray et al., 2007). The hazard of infection (also
referred to as the FOI, commonly denoted by parameter lambda) for those
who are uninfected is proportional to the current prevalence of infection in
the population, which is multiplied by a composite parameter containing the
contact rate and the probability of transmission upon contact with an
infected person; this is often referred to as the transmission rate parameter
Mathematical Modelling 25

beta mentioned previously. Many of the group’s studies set up a stochastic


equation which permits determining analytically the equilibrium distribu-
tion of villages with a given prevalence level or determining numerically
the likely trajectory of the prevalence of infection for a given village over
time. These stochastic models capture and demonstrate the variability in
prevalence trajectories over time, given close baseline values and (antibiotic)
treatment coverage levels within small communities. In the context of the
studies, designed by the UCSF group, these small communities are often vil-
lages in Ethiopia. The prevalence levels of infection, measured by the
group’s field studies, can be represented by the stochastic model-generated
trajectories, with the mean prevalence calculated across several villages
appearing similar to a deterministic trajectory.
The mechanistic models developed at ICL have used primarily deter-
ministic frameworks (Gambhir et al., 2007, 2009, 2010) and have focussed
on developing age-structured models which capture the age-specific profiles
of infection observed within communities of different endemicities. A back-
bone of these models is a ladder of susceptibleeinfectedesusceptible (SIS)
infection (Fig. 1A) to account for key epidemiological features that change
with age with each successive infection (Fig. 1B). However, studies investi-
gating the relative contribution of household and community transmission
lend themselves to be more naturally simulated stochastically (Fig. 1C)
(Blake et al., 2009, 2010). We also identified a number of catalytic models
reported in the literature; the use of the catalytic model for epidemiological
purposes was initially presented by Muench (1934). These catalytic models
explore the change in age-specific prevalence of infection; here the FOI can
be estimated in a population from age-specific disease prevalence data, or
serological data. In these models individuals experience an FOI as suscepti-
ble, they then progress into the diseased class where they will stay until they
recover or die (Fig. 1D).

3.3 Data sets used and model fitting to data


The availability and collection of routine high-quality baseline (prior to
intervention), monitoring and evaluation (during the control
programme) and (postintervention) surveillance infection data to inform
mathematical modelling studies which aim to help and advice control pro-
grammes have been extremely limited. Consequently, the data analyzed by
modelling groups in the studies identified here have either been collected by
the modelling groups themselves (Lietman et al., 2011; Liu et al., 2013,
2014; Ray et al., 2007) or through the careful development and
26 A. Pinsent et al.

(A)
λ
S I
ɣ

(B)
λ ɣ1 λ ɣi λ
S1 I 1
Si Ii Sn In

(C) λ
λg S I
λ ɣ λg
S I λg
ɣ
λg λg λ
λg S I
ɣ
λ
S I
ɣ

(D)
λ
S δ
I

(E)
transmission develop active clear infection partial resolution
disease partially
infected no infected active resolved
susceptible disease
active with active active
disease disease only disease

re-infection

recovery

Figure 1 Diagrammatic overview of the four key model structures and state variables
that have been used for modelling trachoma transmission. Here S represents an indi-
vidual or individuals who are susceptible to infection and I represents those who are
infected with trachoma. (A) Classic SIS model structure; individuals move into the
infected compartment at a rate l (the FOI), and recover at a rate g (independent of
the number of previous infections), returning to the S compartment. (B) SIS ladder of
infection model (Gambhir et al., 2007); this framework tracks the number of infections
(i) that individuals experience throughout their lifetime as they age. They experience an
FOI lambda l and move into the I state; however, they recover at a rate gi dependent on
the number of previous infections they have had and move to state Si. Equally, the
infectivity of individuals decreases with each infection experienced. (C) Household
model (Blake et al., 2009, 2010); each oval represents either a household or a commu-
nity; transmission of infection is modelled within each household or community as in
(A) with local FOI lL; however, each community or household also provides and expe-
riences a (global) force of infection, lG from all other households or communities). (D)
Mathematical Modelling 27

management of collaborations with other research groups who have con-


ducted field studies (Blake et al., 2009; Gambhir et al., 2009, 2010). Given
the large range of settings around the world where trachoma is endemic
(with respect to climatic, ecological and social differences), it does not
seem appropriate that models parameterized primarily with data from Africa
be used to make accurate and informative projections about the transmission
dynamics of infection occurring in other endemic regions such as Southeast
Asia or Latin America (where infection is also endemic) (Fig. 2). We iden-
tified eight countries for which any modelling work had been conducted.
Data sets collected from Tanzania have been used in 11 studies analyzed
here. Datasets from The Gambia have been used in six studies, from Ethiopia
in five studies, Taiwan twice (no longer an endemic region), Egypt, Tunisia
(no longer an endemic region), Malawi and Australia and India just once
each. However, the same data set has repeatedly been used across a number
of studies. Data from Tanzania have come from four key regions: Morogoro,
Dodoma, Singida and Arusha. In The Gambia data have come from Jali,
Berenday and Upper Saloum. In Ethiopia, data sets have been analyzed
from the regions of Amhara and Gurage in Ethiopia. Therefore, given the
wide geographic distribution of trachoma (Fig. 2) and the considerable
within-country heterogeneity in transmission (http://www.trachomaatlas.
org/), a greater number of modelling studies should be looking to under-
stand the dynamics of infection at a more global scale, along with trying
to gain insight into the spatial heterogeneity in transmission that exists within
the same country. However, current progress remains limited given the
scarce availability of data from regions outside of a few key transmission
areas, primarily in sub-Saharan Africa, where research (rather than routine)
studies have been conducted. Moreover modelling is often the final activity
performed in such a partnership and constitutes a secondary analysis of the
data, following a primary analysis that includes the evaluation of risk factors
for disease, and a preintervention and postintervention description of infec-
tion and disease prevalence. This has meant that modelling and quantitative

=---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Simple catalytic model (Assaad and Maxwell-Lyons, 1966; Parthasarathy, 1967; Sundar-
easn and Assaad, 1973; Martin et al., 2015); individuals experience a force of infection
(or seroconversion rate) l, and move into the infected (seropositive), or active disease
state; however, once they are classified as such they do not serorevert or recover. In the
two-stage catalytic model (Assaad and Maxwell-Lyons, 1966), individual’s progress from
the S state to present with active trachoma disease (I state) at a rate l; individuals leave
the active disease compartment and recover (returning to the S state) at a rate d. (E) An
extension of the classic SIS model structure, which explicitly models individuals in the
disease state and after a certain period of recovery allows them to be re-infected while
they are still in the disease state.
28 A. Pinsent et al.

Figure 2 A map of the global distribution of trachoma endemic countries and those
currently under surveillance. Red indicates countries that are classified as endemic by
the WHO, and yellow indicates those currently under surveillance. Grey dots show a coun-
try where a transmission model of trachoma has been fitted to data from that country,
although Taiwan and Tunisia are no longer classified as endemic regions. The size of
the dots is not proportional to the number of studies. However, the total number of
data sets collected and used to inform modelling studies is not equal across all countries.
Data collected from Tanzania have been used in 11 modelling studies identified, data
from The Gambia have been used six times in the studies analyzed here, data from
Ethiopia have been used five times, data from Taiwan have been used twice and data
from Egypt, Tunisia, Malawi, India and Australia have all been used once. Data on
endemicity were inferred from (WHO, 2013) http://apps.who.int/neglected_diseases/ntddata/
trachoma/trachoma.html.

analysis have not been systematically used as an integrated part of trachoma


control programmes.
The data sets that have primarily been used for model calibration and
model-based analysis have come from long-term epidemiological studies
in Upper Saloum district in The Gambia (Burton et al., 2005), Rombo dis-
trict in Tanzania (Solomon et al., 2003, 2004), Kongwa district in Tanzania
(Burton et al., 2005) and Gurage zone of Ethiopia (Lietman et al., 2011,
2015; Liu et al., 2013, 2014; Ray et al., 2007, 2009). These data include lon-
gitudinal measurements of infection, clinical disease and associated risk fac-
tors. As most of these studies were begun at baselines where control
interventions had not yet been implemented, an ecological ‘steady-state’
(or endemic equilibrium) constitutes an appropriate initial condition for
modelling analyses. Nevertheless, the vast majority of articles identified in
this review estimate parameters explicitly from a very small number of
data sources. Studies for which infection, active disease and disease sequelae
Mathematical Modelling 29

status were all collected should be collated and shared in the future in order
to allow models to be simultaneously fitted to all of these (infection and
morbidity) data sources. In this study we only identified one article that
explicitly attempted to estimate the age-specific duration of active disease
(Grassly et al., 2008) through a hidden Markov model.
We identified one statistical study that used a substantial amount of pro-
grammatic clinical disease data, from a wide time and geographical range
collected by the International Trachoma Initiative (Jimenez et al., 2015).
This large data set included data on active infection (through PCR data),
prevalence data on active trachoma disease (TF or TF/TI) and TT (Table
1). These data were used to investigate the conditions under which control
programmes have been successful. The importance of this study lies not only
in its analysis of a large data set to draw its conclusions but also in its presen-
tation of these conclusions as a ‘decision tree,’ which is conceptually similar
to the way in which the WHO presents its recommendations for disease
control policy.
Mathematical models that are successfully fitted to epidemiological data
(be it PCR [infection] prevalence data or data on active disease prevalence),
endeavour to estimate just a few key epidemiological parameters. These are
most commonly the transmission rate (b), the basic reproduction number
(R0) and the duration of an individual’s infectious period (the reciprocal of
the recovery rate, gamma). Whether the methodology applied to estimate
these parameters is approached through Maximum Likelihood Estimation
(MLE) (Blake et al., 2009; Gambhir et al., 2009) or a combination of MLE
and Markov Chain Monte Carlo methods (Lietman et al., 2011), the aim
of these studies remains the same, which is, to understand the intensity of
transmission and the amount of effort required in order to control infection.

3.4 Transmission intensity and the basic reproduction


number
3.4.1 Estimating the force of infection
Intensity of transmission and hence the FOI experienced by a community can
be estimated through the careful collection of data on prevalence or through
serological data in the form of seroprevalence age profiles (Hens et al., 2012).
A study in the former category was reported by Sundaresan and Assaad
(1973) using data from Taiwan collected in the 1960s. These authors applied
deterministic FOI (catalytic) modelling to provide insight into the changes
in age-specific prevalence of trachoma over time and infer past patterns of
transmission in the country. Using survey data from 1960 to 1961dbefore
30 A. Pinsent et al.

the institution of a control, school-based programme with tetracycline treat-


ment (Assaad et al.,1966)dtrachoma prevalence increased with age in a
fashion compatible with the operation of a constant FOI (42/1000 year1).
However, using survey data from 1968 to 1969 revealed that the 0- to 9-
year olds had experienced a substantial decrease in the FOI (11/1000 year1)
(Sundaresan and Assaad, 1973).
The use of serological data to gain insight into the dynamics of transmission
has been a valuable tool across a number of infectious diseases (Badu et al.,
2015; Hens et al., 2010; Mladonicky et al., 2009; Wong et al., 2014; Wilkins
and Keystone, 2013; Yildiz Zeyrek et al., 2011). The analysis of serological
data has also been particularly useful in postelimination or low-transmission
settings (Bousema et al., 2010; Corran et al., 2007; Oguttu et al., 2014). How-
ever, our review only identified one article (a study conducted in Kongwa
district, Tanzania) that used a (reversible catalytic) model fitted to serological
data to understand how interventions may affect the serological status of a
community (Fig. 1D) (Martin et al., 2015) (Table 1) and how these data
can be used to shed light into the rate of acquisition of infection and anti-
bodies to infection. Reversible catalytic models (as illustrated by the first
two compartments of Fig. 1D) model seroprevalence in the population
over time. These models have two key parameters, the first is the seroconver-
sion rate l (Fig. 1D), the second is the seroreversion rate, denoted d in
Fig. 1D, which models the rate at which immunity in the population is
lost, hence the rate at which people serorevert. It is therefore akin to the
SIS model structure shown in Fig. 1D; however, in this instance the acquisi-
tion and loss of antibodies are modelled instead of the process of infection and
recovery. Martin et al. (2015) were able to identify a change in the intensity of
transmission from serological data between 10 and 15 years prior, which was
shown to coincide with a mass antibiotic administration programme that had
run between 2000 and 2002. Although Martin et al. (2015) were able to es-
timate a seroconversion rate, they were not able to estimate a seroreversion
rate as they did not identify any individuals in the data set who seroreverted.
As with other infectious diseases, a mounting evidence base for serological
testing, that is both sensitive and specific to trachoma infection has the poten-
tial to be used as a rich new data source for understanding the changing pat-
terns of transmission dynamics of trachoma (Martin et al., 2015).

3.4.2 Estimating the basic reproduction ratio


The most commonly used parameter to summarize the transmission poten-
tial of an infectious disease is the basic reproduction ratio or number, R0 (the
Mathematical Modelling 31

average number of secondary infectious cases that arise from an index case in
a wholly susceptible population). R0 was calculated (by the authors of the
papers) in only four of the trachoma modelling studies identified in this re-
view, despite the fact that it could have been estimated in several of the
remaining models as the ratio of the transmission coefficient and the recov-
ery rate from infection (beta/gamma). This has been done by us and the re-
sults are presented in Table 2. For the more complicated published models,
involving age-structure and functional forms for an individual’s acquired im-
munity and infectivity, the calculation of R0 is more involved and can most
easily be performed using the next-generation matrix method (Diekmann
et al., 2010). The acquisition and waning of immunity complicates the
calculation of the reproduction number, since R0’s definition refers to the
beginning of an epidemic, i.e., in a fully susceptible population (hence indi-
viduals have no prior immunity to infection). In these circumstances, while
the calculation process remains essentially the same, the parameter that is be-
ing calculated is correctly referred to as the effective reproduction number
(Re); in this review we identified only one study that estimated Re (Liu
et al., 2014).
As evidenced by the total number of studies that explicitly calculated R0
in this review, analysis and calculation of R0 in the trachoma modelling
community has remained very limited. However, recent work by Liu
et al. (2013) used a mechanistic model to perform a statistical analysis of
infection control and rebound data from several Ethiopian villages to deter-
mine whether there was a change in the value of the reproduction number
following each round of mass antibiotic administration. Changes in the
reproduction number over time could potentially be caused by the
decreasing efficacy of the antibiotic drug over time or the loss of short-
term immunity in the population, as the stimulus of exposure to C. tracho-
matis diminishes. Liu et al. (2013) found no statistically significant change
in the reproduction number in the Ethiopian data set analyzed and there-
fore, these authors concluded that changes in drug efficacy or in immunity
were not operating at an appreciable level in the epidemiological settings
evaluated in their study.
Comparing across the small number of studies that have attempted to
calculate a value of R0 or Re (or published values from which estimates could
be obtained), we observed limited heterogeneity in the range of estimates
from communities classified as having the same level of endemicity (Table 2).
For example, Ray et al. (2009) estimated R0 to be 3.14 (95% CI, 2.51e3.77)
for 16 hyperendemic communities in Ethiopia; in 2013 Liu et al. estimated
32 A. Pinsent et al.

an overall R0 of 1.39 (and yearly R0 values of 1.40, 1.38, 1.35) across 32


communities of Tanzania; Ray et al. (2007) provided values of beta
(0.047 week1) and gamma (0.017 week1) that yield a figure of 2.59 for
the 16 villages mentioned previously (which are located in the Gurage re-
gion of southern Ethiopia). We estimated Re using the next-generation ma-
trix method (unpublished estimate) using a previously published model and
best fit parameters from another hyperendemic area in Tanzania, and yielded
a value of 3.2 (Gambhir et al., 2009), an estimate similar to that of Ray et al.
(2009). Therefore, there is some consistency in the literature that values of
R0, even within high transmission settings are likely to be less than 5. These
estimates have been derived using different methodological approaches, sug-
gesting that the estimate is not specific to the methodology applied. How-
ever, it is worth noting that transmission of trachoma in some regions is
reported to be seasonal and that this source of heterogeneity has not been
incorporated into any of the R0 estimates identified in this review. There-
fore, current estimates of R0 may be pertinent to the particular times the
studies were conducted. Very few studies have collected prevalence data
of active trachoma and C. trachomatis infection at multiple time points
throughout consecutive years limiting the inference that can be made on
the role of seasonality in transmission.

3.5 Acquired immunity: recovery rate and infectivity


Immunity to trachoma with repeated infection is reported to decrease the
duration of an infected individual’s infectious period (Gambhir et al.,
2009, 2010) and also to reduce the bacterial load harboured by an individual
(West et al., 2005). However, in addition to whether any acquired immu-
nity affecting infectivity and/or recovery rates was incorporated into the
models reviewed here, we observed marked heterogeneity in the rates of re-
covery from infection reported and used in the modelling studies analyzed.
The models evaluated in this review that were published by the UCSF
group did not, in general, include the development of acquired immunity,
with the exception of one study in which the model assumed that older age
groups developed partial immunity to (re)infection either because of
decreased susceptibility to infection or because of clearing infection faster
than younger individuals (Lietman et al., 1999). In general, these models as-
sume a single value for the rate of recovery from infection, approximately
equivalent to a mean duration of infection of 6 months although there is
some heterogeneity between their studies; in the study by Lietman et al.
(1999) duration of infection ranges from 9 to 17 weeks. This is a somewhat
Mathematical Modelling 33

slower rate of recovery than that considered by the models of the ICL group,
which suggested that the mean duration of infection/infectivity after an in-
dividual has experienced multiple prior infections was approximately
2.8 monthsdbut duration of infection for the first episode was
15 monthsd(Gambhir et al., 2009). Additional work by the ICL group esti-
mated the duration of infection to be 16e17 weeks (c. 4 months) using
frequent follow-up data from The Gambia (Grassly et al., 2008), which
was subsequently used in a number of other studies (Blake et al., 2009;
Gambhir et al., 2010). However, the assumed duration of infectivity can
have substantial implications for the estimation of the feasibility of elimina-
tion of infection and the estimation of the transmission rate b. For models
that assume a longer duration of infection, the estimated value of b is likely
to be lower than for those assuming a shorter duration (although both types
of model would yield the same value of R0). From the perspective of
trachoma control, a lower value of b may suggest that the effort required
to eliminate infection through transmission reduction measures is less than
that needed when a shorter infectious period is assumed. However, the
assumption of a longer infectious period would increase the potential for
infection to persist within a community, especially if movement of infected
individuals between villages is considered, as this could result in the reintro-
duction of infection.
The models of the ICL group explicitly account for the development of
immunity to infection with an increasing number of infections experienced
(Gambhir et al., 2009, 2010). Acquired immunity to infection is modelled as
a function of the number of prior infections, i which is described by an
exponentially saturating rise in the recovery rate (or fall in the infectivity).
These functions, which have three parameters, are not the only choice of
functional form accounting for acquired immunity; for example, a logistic
function or a Hill function which allow for saturating effects may be
more appropriate (Regoes et al., 2004). However, the available epidemio-
logical data are at present unlikely to permit distinguishing between these
different functional forms. The importance of this process and its implication
for long-term transmission patterns remain poorly understood.

3.6 Infection and active disease, modelling of disease


sequelae
Most infectious disease models are based on the idea that contact between
individuals leads to the transmission of a pathogen from one individual to
another with a given probability; they are, therefore, infection based and
34 A. Pinsent et al.

not disease based. Surveillance for trachoma control is based upon disease,
and therefore, there is a need to link transmission models to active trachoma
and the later disease sequelae.
Due to the indirect relationship between active disease in trachoma and
C. trachomatis infection, PCR data (rather than data on active disease) have
most often been used to inform the vast majority of the models identified
in this review. Trachoma models have mostly been constructed as infection
models, as demonstrated by the large number of studies identified that have
used the classic SIS model structure (Fig. 1A) (also commonly used to model
genital Chlamydia infection). Notwithstanding the ubiquity of the SIS back-
bone, we only identified a small number of modelling studies which explic-
itly included age structure across the host population (Blake et al., 2010;
Gambhir et al., 2009, 2010; Grassly et al., 2008) and hence would have
the ability to assess the impact of infection (and disease) on an individual
throughout life, this represents a critical gap in the current literature.
Although the infectious load and duration of active infection decreases
with age, repeated infection throughout an individual’s life results in the
serious morbidities associated with trachoma infection. This is particularly
important as the first two GET 2020 goals seek to understand the impact
of the infection dynamics on the long-term patterns of active disease and
the elimination of blinding trachoma.
The successful economic evaluation of the impact of disease and disease-
associated morbidities through trachoma infection relies on accurately quan-
tifying the burden of disease (Lee et al., 2015). However, the development
of mathematical models for trachoma has remained primarily focused on the
use of the SIS model structure, which does not lend itself to the analysis and
understanding of the long-term implications of numerous active disease ep-
isodes with age. This represents a critical gap in the literature. However, we
identified three studies which used the ‘force of infection’ (FOI) modelling
approach (Assaad and Maxwell-Lyons, 1966; Parthasarathy, 1967;
Sundaresan and Assaad, 1973) to describe age profiles of total trachoma cases
(simple catalytic model, with constant FOI) and of active disease (two-stage
catalytic model) in Taiwan before the institution of control measures in the
early 1960s. In these models individuals become infected and clinically
diseased according to an FOI (which was estimated to be 0.042 year1);
this translated into a prevalence of trachoma that increased monotonically
with age. By contrast, the prevalence of active disease showed a humped
pattern, reaching its maximum at 15e20 years of age and decreasing after-
wards; in this case the two-stage catalytic model estimated a rate of
Mathematical Modelling 35

Figure 3 Flow diagram of the modelling study that developed the ladder of infection
model (Fig. 1B) in order to track and model progress to disease sequelae (Gambhir et al.,
2009). Progression into TS (trachomatous scarring) and TT (trachomatous trichiasis) is
determined by a threshold number of repeated infections. Once individuals reach a
threshold number of infections they are considered to have TS, they recover from infec-
tion and then get reinfected until they reach the threshold number of infections for TT
(dashed arrows), from here progression to CO (corneal opacity) occurs at a constant rate
s with no further infections required due to the damage already incurred by those with
TT. Disease sequelae classes are indicated with a dashed red box.

acquisition and of loss of active disease of 0.050 year1 and 0.072 year1,
respectively. Through these models it was possible to gain insight into the
dynamic nature of the burden of active disease with age prior to the imple-
mentation of an intervention.
Models which are intended to be used in close conjunction with control
programmes should, therefore, include active disease and disease sequelae
components if they are to be truly useful as tools to support decision-making
in the context of the GET 2020 goals. Although infection and active disease
are the proximate causes of scarring and the resulting sequelae, the timescales
associated with the development of scarring, trichiasis, corneal opacity and
blindness are different from one another and should be accounted for
when forecasting for control programmes aiming to achieve the GET
2020 goals (Gambhir et al., 2009, 2010, 2015; Shattock et al., 2015). To
date, very few models have done so. Gambhir et al. (2010) illustrated a first
attempt at adding active disease to a simple age-structured infection-based
model (Fig. 3). The model used a series of partial differential equations to
explore the impact of treatment on active disease, having previously fitted
the model to disease sequelae data. The authors identified that in their model
the prevalence of active disease was always higher than the prevalence of
infection and that following mass antibiotic treatment the decline in preva-
lence of active disease lagged behind the immediate drop in infection
prevalence (Gambhir et al., 2010). The mechanistic model of Shattock
et al. (2015) took this concept further by developing a stochastic agent-based
model to create a complex representation of the natural history of infection
(Fig. 1E) and disease for every individual within a defined community, and
by then fitting this model to active disease data from multiple Australian
36 A. Pinsent et al.

Aboriginal communities. Therefore, models which can be fitted to active


disease data are needed if we are to understand whether the GET 2020 goals
are being achieved. Modelling analysis of additional active disease data sets is
a pressing need in order to disentangle the complex relationship between the
dynamics of infection across different transmission settings and the preva-
lence of active disease/disease sequelae and to help the public health com-
munity to understand the feasibility of achieving the GET 2020 goals.
Here we suggest five crucial pieces of data that could be collected and that
would help to ensure more accurate parameterization of subsequent models
of trachoma infection and disease: (1) collection of data that help to inform
understanding of the age-specific duration of active disease (TI/TF) following
infection with C. trachomatis; (2) data on the duration of overlap between
active infection and active disease by age; (3) data on the age-specific duration
of infection alone; (4) data on how the bacterial load harboured with age
varies across different transmission settings and (5) data on how immunity
to trachoma wanes in the absence of frequent exposure.

3.7 Simulation of interventions, forecasting infection and


disease, and analysis of their cost-effectiveness
Mechanistic modelling in the context of control interventions was explored
in 13 of the 23 articles identified in this review, all of which examined the A
(antibiotics) component of the SAFE strategy. Lietman et al. (1999) showed
that annual mass drug administration (MDA) was not sufficient to eliminate
infectious trachoma from hyperendemic communities and that the fre-
quency of treatment required depends on the initial doubling time of a rein-
troduced epidemic; the latter does not decrease as prevalence of infection
decreases during a treatment programme. In order for the doubling time
of a reintroduced epidemic to be reduced, changes in the community
must occur, such as the implementation of hygiene measures (F and E com-
ponents), that can help to reduce the overall level of transmission. This study
also highlighted the need for biannual mass antibiotic treatment in hyperen-
demic communities. Gambhir et al. (2010) demonstrated a similar finding.
Ray et al. (2007) investigated the impact of annual and biannual rounds
of mass antibiotic treatment and showed that 5 years of biannual treatment
could be sufficient to eliminate infection in 95% of the hyperendemic Ethi-
opian villages simulated in the study. In a follow-up paper, Ray et al.
(2009) used the same modeldfitted to data from three different transmission
settings: Ethiopian (hyperendemic), Tanzanian (mesoendemic) and Gam-
bian (hypoendemic)dto demonstrate that an adjusted mass treatment
Mathematical Modelling 37

strategy, in which mass antibiotic administration was withheld from specific


communities whose infection prevalence levels fell below 5% (in children
aged 1e5 years), had a comparable impact to that of the WHO-endorsed
strategy and saved a considerable number of antibiotic doses. However,
the logistics and costs associated with testing infection at the community
level instead of the district level (outside of specific trial, research settings)
can be both expensive and challenging, making the practical implications
of these findings difficult to implement on a much wider, routine scale.
To overcome these difficulties, Ray et al. (2009) suggested that pooled
PCR sampling for entire communities could be performed, leading to sub-
stantial cost savings. However, a specific cost breakdown for the implemen-
tation of such an approach would be useful for understanding whether it
would be feasible from a programmatic perspective and for modelling its
cost-effectiveness.
An important factor which can impact the effectiveness of antibiotic
treatment is the level of coverage that can be achieved within a community.
If the level of coverage is poor few people will get treated, therefore, the
impact on transmission will be less than if the level of coverage achieved
were high. Within a trial setting, where much of the epidemiological data
about the success of MDA is collected, the level of coverage achieved can
be extremely high (>90%) (Lakew et al., 2009). However, outside of a trial
setting the level of coverage achieved is likely to be much lower, suggesting
that the findings identified in trials may not be fully reflective of the out-
comes observed during routine rounds of MDA. We did not identify any
studies that assessed the sensitivity of their findings to the level of MDA
coverage achieved. This represents an important gap in the modelling liter-
ature that needs to be explored more thoroughly in future studies if we are
to make realistic forecasts about the prospective impact of MDA.
One of the first questions public health workers ask disease modellers re-
lates to the forecasting of future disease levels. However, the majority of
trachoma modelling work has been conducted to improve the understand-
ing of trachoma epidemiology and not specifically to forecast disease levels.
Nevertheless, a well-parameterized mechanistic or statistical model is often
capable of being run forward in time, forecasting disease and uncertainty
bounds. Indeed, we identified three published studies that have been formu-
lated in order to make forecasts in the context of control interventions; these
were the studies by Gambhir et al. (2015), Ray et al. (2009) and Shattock
et al. (2015). Each of these three studies forecasts the prevalence of infection
and/or disease following various interventions (see Section 3.7). Shattock
38 A. Pinsent et al.

et al. (2015) fitted their individual-based model to clinical disease prevalence


surveillance data from 67 Australian Aboriginal communities, and then
investigated the possible future impact of different treatment combinations
on the long-term prevalence of trachoma infection. The authors identified
that under the current intervention strategies, control of infection within
these communities would not be achieved by 2020. Ray et al.
(2009) used a stochastic differential equation model to evaluate an antibi-
otic-sparing strategy that halted treatment when community infection
prevalence dropped below 5%, following the WHO recommended three
annual rounds of antibiotic administration. This was shown to be effective
in mesoendemic and hypoendemic communities [supported with evidence
from trial data from Tanzania (Solomon et al., 2008)]. However, the feasi-
bility of reducing the prevalence of infection in hyperendemic communities
was much lower. Gambhir et al. (2015) performed forecasts for a variety of
treatment interventions within the SAFE strategy, individually and in con-
cert with one another, with the specific aim of determining whether the
GET 2020 goals could be met within the next 5, 10 or 20 years (as of
2015). These studies highlight how modelling can be used to provide infor-
mation and insightful projections of public health importance in the context
of the GET 2020 goals.
In a modified approach to the classic SIS models previously discussed,
Blake et al. (2010) developed a model that incorporated both household
and community transmission which was used to explore cost-effective
means of distributing antibiotics, whereby treatment was tied to clinical dis-
ease diagnosis at the household level. The modelling analysis by Blake et al.
(2010) is the only study that features a health economics component, an area
that is clearly in need of attention as rationales for budgets to address blinding
disease elimination are made in the future. Research groups working on
several of the other NTDs have integrated health economics evaluations
of control interventions, largely pharmaceutical, into mechanistic modelling
frameworks resulting in studies that can immediately feed into policy [e.g.,
onchocerciasis (Turner et al., 2014a,b; 2015), lymphatic filariasis (Goldman
et al., 2007), soil-transmitted helminthiases (Turner et al., 2015a,b) and
schistosomiasis (Brooker et al., 2008; Guyatt et al., 1994)]. For a review out-
lining the use of mathematical modelling in economic evaluations of inter-
ventions against the NTDs included in the London Declaration see Turner
et al. (2014a,b), and for a review of economic evaluation approaches and
studies for the NTDs see Lee et al. (2015). Such approaches should also
be taken forward in trachoma.
Mathematical Modelling 39

In principle, the results of the statistical modelling paper by Jimenez et al.


(2015) could be used to understand the effectiveness of control interventions
for a variety of trachoma-endemic settings, whose previous intervention his-
tory varies from none to the implementation of the full WHO recommen-
ded SAFE strategy over time. The drawbacks of their statistical approach are
that it only allows for a small number of location-specific covariates to be
defined for each community and that while the results are often intuitive
(Table 1), the model is expressed as a set of correlations, without explicitly
accounting for the biological and epidemiological mechanisms underlying
their community classifications.
The majority of the modelling work identified in this review has
sought to understand and project the impact of antibiotic administration,
whether in the modality of mass treatment or through assessing the impact
of targeted treatment strategies to only those who present with active
infection or specific age groups (Blake et al., 2009; Gambhir et al.,
2010; Lakew et al., 2009; Lietman et al., 2011; Liu et al., 2014; Ray
et al., 2007, 2009).
However, antibiotics represent only one arm of the SAFE strategy. As
such, there is a pressing need for research to determine how best to incor-
porate nonpharmaceutical intervention (NPI) into mathematical models
of trachoma. In this review we identified only one study that explicitly
assessed the impact of both MDA and NPIs within communities (Shattock
et al., 2015). Indeed, the third GET 2020 goal states that the F and E com-
ponents of the SAFE strategy should be enhanced. However, accurate quan-
tification of the impact of these interventions in helping to reduce trachoma
transmission and incidence remains limited and poorly understood, making
it difficult to model accurately their potential impact. Ejere et al. (2012) con-
ducted a review of face washing promotion for the prevention of active
trachoma and concluded that although there is some evidence that face
washing combined with topical tetracycline can be effective in reducing se-
vere trachoma and increasing the prevalence of clean faces, face washing
alone or in combination with topical tetracycline did not significantly
reduce active trachoma. Rabiu et al. (2012) performed a review of environ-
mental interventions for preventing active trachoma and concluded that in-
secticides against the mechanical fly carriers may be effective in reducing
trachoma but provision of latrines as a fly control measure did not signifi-
cantly reduce trachoma transmission. Generally there is a dearth of data to
determine the effectiveness of all aspects of environmental sanitation in
the control of trachoma.
40 A. Pinsent et al.

Currently, the best available resource for understanding the possible


impact of NPIs on trachoma and incorporating these NPIs into trachoma
models is the meta-analysis presented by Stocks et al. (2014). This study
collated epidemiological evidence for each of the components of water,
sanitation, and hygiene (WASH) upon active trachoma and infection with
C. trachomatis. The odds ratios (ORs) for each WASH component provided
by this article could be translated into modified parameter values to be incor-
porated into appropriately defined mechanistic models; many interventions
would essentially alter the reproduction number by an amount directly
calculable from the relevant OR (generally by reducing the value of the
reproduction number in a linear manner).

4. DISCUSSION
4.1 Using modelling to determine the feasibility of the
GET 2020 goals
Studies focussing on the epidemiological modelling of trachoma have
been steadily published over the past 40 years. Collectively, they provide a
considerable amount of accumulated insight pertaining to the GET 2020
goals. The World Health Assembly passed resolution 51.11 in the year
1998 outlining the intention to end trachoma as a public health problem.
These goals included the Global Elimination of Blinding Trachoma by
the year 2020 (GET, 2020) (WHO, 2003). However, there has been limited
involvement of modellers until recently to help to understand if and
whether these goals are being/will be achieved within the proposed
timeframes.
The GET 2020 annual meeting is organized by the WHO and has taken
place since 1997. Prior to the main meeting, a Trachoma Scientific Informal
Workshop is held in which the latest scientific results pertaining to GET
2020 are shared. The results shared by this group comprise, for example,
data from intervention trials of azithromycin antibiotic distribution in a va-
riety of trachoma-endemic settings [primarily in The Gambia, Tanzania,
Ethopia, and more recently Niger (Cromwell., 2014) and Guinea Bissau
(Last et al., 2014)], and rates of success, reversion, and risk factors associated
with surgery for TT, among others (WHO, 2015). However, it is only
recently that modelling is being viewed as a tool to help to inform policy
and infection control strategies (see e.g., WHO, 2015).
The vast majority of modelling studies identified in this review have not
explicitly referred to the GET 2020 goals. However, they all contain
Mathematical Modelling 41

findings that pertain to the goals (Table 1). Therefore, despite the high rele-
vance of many of these articles to the questions posed and goals proposed by
GET 2020, modelling groups and their resulting studies have largely failed to
be integrated into the discussions leading to the development of control pro-
gramme policy. The body of model-based work that has been produced for
trachoma is not significantly different to that which has been published for
the other NTDs. However, the failure to integrate modelling into under-
standing the feasibility of achieving the GET 2020 goals across a range of
endemic settings may be due to the fact that trachoma has previously
been perceived as controllable and that the current intervention strategies
are working (House et al., 2007). As such, it may be felt that modelling
studies tend only to confirm intuition or are only taken into account as
part of the decision-support apparatus of public health workers when it is
deemed that current interventions are not working.
As described, a wide range of models has been developed to explore a
multitude of hypotheses. However, it could be suggested that many studies
seek not only to fit models to data to support or refute a hypothesis but
also aim to develop and apply innovative methodologies. This has potentially
led to the publication of a wide range of studies that from a programmatic
perspective may all appear quite different, leading to confusion in the
trachoma community. However, from the modelling perspective they are still
seeking to understand the same underlying processes. This in itself may be
leading to a barrier between modellers and those who work in the field.

4.2 Conclusions
Having assessed the modelling literature, we outline three key gaps that
future modelling work should address in order to help to understand if
and where the GET 2020 goals can be achieved. First, we would suggest
that modellers work to develop mathematical models that explicitly include
an active disease component which accounts for the fact that even if an in-
dividual does not have an active, PCR-detectable infection, they may still
have active disease. Mathematical models that are fitted to infection and
active disease data and validated with baseline and follow-up, longitudinal
active disease data will be valuable for forecasting purposes. Second, we
would encourage that a better quantitative understanding of the impact of
the F and E interventions be developed; this will allow NPIs to be included
more accurately into mathematical models, facilitating the long-term impact
of these interventions to be explored more thoroughly. Finally, we would
suggest that health economic analysis be integrated into scenario analysis,
42 A. Pinsent et al.

as not only is the understanding of the feasibility of the GET 2020 goals
vital but so is developing an understanding of the financial implications of
doing so.
Competing interests
M. Gambhir, M.G. Basan ~ez and I.M. Blake declare in accordance with
the ICMJE conflict of interest form that seven of the articles identified and
evaluated in this review have been authored or co-authored by them. AP
declares no competing interests.
Authors’ contributions
A. Pinsent and M. Gambhir designed the study and prepared the
first draft. I.M. Blake and M.G. Basan ~ez critically review the draft and
contributed with intellectual input. All authors contributed to the writing
of the final version of the manuscript and approved the final, submitted
version.

ACKNOWLEDGEMENTS
A. Pinsent, M.G. Basan ~ez and M. Gambhir gratefully acknowledge funding of the NTD
Modelling Consortium by the Bill and Melinda Gates Foundation in partnership with the
Task Force for Global Health. The views, opinions, assumptions or any other information
set out in this article are solely those of the authors. M. Gambhir also acknowledges funding
from the Australian NHMRC and Monash University.

REFERENCES
Abdou, A., Nassirou, B., Kadri, B., Moussa, F., Munoz, B.E., Opong, E., West, S.K., 2007.
Prevalence and risk factors for trachoma and ocular Chlamydia trachomatis infection in
Niger. Br. J. Ophthalmol. 91, 13e17.
Assaad, F.A., Maxwell-Lyons, F., 1966. The use of catalytic models as tools for elucidating
the clinical and epidemiological features of trachoma. Bull. World Health Organ. 34,
341e355.
Assaad, F.A., Sundaresan, T.K., Yang, C.Y., Yeh, L.J., 1971. Clinical evaluation of the
Taiwan trachoma control programme. Bull. World Health Organ. 45, 491e509.
Badu, K., Gyan, B., Appawu, M., Mensah, D., Dodoo, D., Yan, G., Drakeley, C., Zhou, G.,
Owusu-Dabo, E., Koram, K.A., 2015. Serological evidence of vector and parasite expo-
sure in Southern Ghana: the dynamics of malaria transmission intensity. Parasites Vectors
8, 251.
Bailey, R., Duong, T., Carpenter, R., Whittle, H., Mabey, D., 1999. The duration of hu-
man ocular Chlamydia trachomatis infection is age dependent. Epidemiol. Infect. 123,
479e486.
~ez, M.G., Anderson, R.M., 2015. Preface. Mathematical models for neglected tropical
Basan
diseases: essential tools for control and elimination. Part A. Adv. Parasitol. 87, xiiiexviii.
Blake, I.M., Burton, M.J., Bailey, R.L., Solomon, A.W., West, S., Mu~ noz, B., Holland, M.J.,
Mabey, D.C., Gambhir, M., Basan ~ez, M.G., Grassly, N.C., 2009. Estimating household
and community transmission of ocular Chlamydia trachomatis. PLoS Negl. Trop. Dis. 3,
e401.
Mathematical Modelling 43

Blake, I.M., Burton, M.J., Solomon, A.W., West, S.K., Basan ~ez, M.G., Gambhir, M.,
Bailey, R.L., Mabey, D.C., Grassly, N.C., 2010. Targeting antibiotics to households
for trachoma control. PLoS Negl. Trop. Dis. 4, e862.
Bousema, T., Youssef, R.M., Cook, J., Cox, J., Alegana, V.A., Amran, J., Noor, A.M.,
Snow, R.W., Drakeley, C., 2010. Serologic markers for detecting malaria in areas of
low endemicity, Somalia, 2008. Emerg. Infect. Dis. 16, 392e399.
Brooker, S., Kabatereine, N.B., Fleming, F., Devlin, N., 2008. Cost and cost-effectiveness of
nationwide school-based helminth control in Uganda: intra-country variation and effects
of scaling-up. Health Policy Plan. 23, 24e35.
Burton, M.J., Holland, M.J., Makalo, P., Aryee, E.A., Alexander, N.D., Sillah, A., Faal, H.,
West, S.K., Foster, A., Johnson, G.J., Mabey, D.C., Bailey, R.L., 2005. Re-emergence
of Chlamydia trachomatis infection after mass antibiotic treatment of a trachoma-endemic
Gambian community: a longitudinal study. Lancet 365, 1321e1328.
Burton, M.J., Mabey, D.C.W., 2009. The global burden of trachoma: a review. PLoS Negl.
Trop. Dis. 3, e460.
Corran, P., Coleman, P., Riley, E., Drakeley, C., 2007. Serology: a robust indicator of ma-
laria transmission intensity? Trends Parasitol. 23, 575e582.
Courtright, P., West, S.K., 2004. Contribution of sex-linked biology and gender roles to dis-
parities with trachoma. Emerg. Inf. Dis. 10, 2012e2016.
Cromwell, E.A., Amza, A., Kadri, B., Beidou, N., King, J.D., Sankara, D., Mosher, A.W.,
Hassan, S., Kane, S., Emerson, P.M., 2014. Trachoma prevalence in Niger: results of
31 district surveys. Trans. R. Soc. Trop. Med. Hyg. 108, 42e48.
Diekmann, O., Heesterbeek, J.A., Roberts, M.G., 2010. The construction of
next-generation matrices for compartmental epidemic models. J. R. Soc. Interface 7,
873e885.
Emerson, P.M., Lindsay, S.W., Alexander, N., Bah, M., Dibba, S.M., Faal, H.B.,
Lowe, K.O., McAdam, K.P., Ratcliffe, A.A., Walraven, G.E., Bailey, R.L., 2004.
Role of flies and provision of latrines in trachoma control: cluster-randomised controlled
trial. Lancet 363, 1093e1098.
Ejere, H.O., Alhassan, M.B., Rabiu, M., 2012. Face washing promotion for preventing
active trachoma. Cochrane Database Syst. Rev. 4, CD003659.
Evans, T.G., Ranson, M.K., 1995. The global burden of trachomatous visual impairment: II.
Assessing burden. Int. Ophthalmol. 19, 271e280.
Frick, K.D., Basilion, E.V., Hanson, C.L., Colchero, M.A., 2003. Estimating the burden and
economic impact of trachomatous visual loss. Ophthalmic Epidemiol. 10, 121e132.
Gambhir, M., Basan ~ez, M.G., Blake, I.M., Grassly, N.C., 2010. Modelling trachoma for con-
trol programmes. Adv. Exp. Med. Biol. 673, 141e156.
Gambhir, M., Basan ~ez, M.G., Burton, M.J., Solomon, A.W., Bailey, R.L., Holland, M.J.,
Blake, I.M., Donnelly, C.A., Jabr, I., Mabey, D.C., 2009. The development of an
age-structured model for trachoma transmission dynamics, pathogenesis and control.
PLoS Negl. Trop. Dis. 3, e462.
Gambhir, M., Basan ~ez, M.G., Turner, F., Kumaresan, J., Grassly, N.C., 2007. Trachoma:
transmission, infection, and control. Lancet Infect. Dis. 7, 420e427.
Gambhir, M., Grassly, N.C., Burton, M.J., Solomon, A.W., Taylor, H.R., Blake, I.M.,
Basan~ez, M.G., 2015. Estimating the future impact of a multi-pronged intervention strat-
egy on ocular disease sequelae caused by trachoma: a modeling study. Ophthalmic Epi-
demiol. 22, 394e402.
Goldman, A.S., Guisinger, V.H., Aikins, M., Amarillo, M.L.E., Belizario, V.Y., Garshong, B.,
Gyapong, J., Kabali, C., Kamal, H.A., Kanjilal, S., Kyelem, D., Lizardo, J., Malecela, M.,
Mubyazi, G., Nitiema, P.A., Ramzy, R.M.R., Streit, T.G., Wallace, A., Brady, A.M.,
Rheingans, R., Ottesen, E.A., Haddix, A.C., 2007. National mass drug administration
costs for lymphatic filariasis elimination. PLoS Negl. Trop. Dis. 1, e67.
44 A. Pinsent et al.

Grassly, N.C., Ward, M.E., Ferris, S., Mabey, D.C., Bailey, R.L., 2008. The natural history
of trachoma infection and disease in a Gambian cohort with frequent follow-up. PLoS
Negl. Trop. Dis. 2, e341.
Guyatt, H., Evans, D., Lengeler, C., Tanner, M., 1994. Controlling schistosomiasis: the cost-
effectiveness of alternative delivery strategies. Health Policy Plan. 9, 385e395.
Harding-Esch, E.M., Edwards, T., Sillah, A., Sarr, I., Roberts, C.H., Snell, P., Aryee, E.,
Molina, S., Holland, M.J., Mabey, D.C.W., Bailey, R.L., 2009. Active trachoma and
ocular Chlamydia trachomatis infection in two Gambian regions: on course for elimination
by 2020? PLoS Negl. Trop. Dis. 3, e573.
Hens, N., Aerts, M., Faes, C., Shkedy, Z., Lejeune, O., Van Damme, P., Beutels, P., 2010.
Seventy-five years of estimating the force of infection from current status data. Epide-
miol. Infect. 138, 802e812.
Hens, N., Shkedy, Z., Aerts, M., Faes, C., Van Damme, P., Beutels, P., 2012. Modeling In-
fectious Disease Parameters Based on Serological and Social Contact Data. A Modern
Statistical Perspective. Springer, London.
Hotez, P.J., Alvarado, M., Basan ~ez, M.G., Bolliger, I., Bourne, R., Boussinesq, M.,
Brooker, S.J., Brown, A.S., Buckle, G., Budke, C.M., Carabin, H., Coffeng, L.E.,
Fevre, E.M., F€ urst, T., Halasa, Y.A., Jasrasaria, R., Johns, N.E., Keiser, J., King, C.H.,
Lozano, R., Murdoch, M.E., O’Hanlon, S., Pion, S.D., Pullan, R.L., Ramaiah, K.D.,
Roberts, T., Shepard, D.S., Smith, J.L., Stolk, W.A., Undurraga, E.A., Utzinger, J.,
Wang, M., Murray, C.J., Naghavi, M., 2014. The global burden of disease study
2010: interpretation and implications for the neglected tropical diseases. PLoS Negl.
Trop. Dis. 8, e2865.
Hotez, P.J., Feck, A., Savioli, L., Molyneux, D.H., 2009. Rescuing the bottom billion
through control of neglected tropical diseases. Lancet 373, 1570e1575.
House, J., Gaynor, B., Taylor, H., Lietman, T.M., 2007. The real challenge: can we discover
why trachoma is disappearing before it’s gone? Int. Ophthalmol. Clin. 47, 63e76.
Jimenez, V., Gelderblom, H.C., Mann Flueckiger, R., Emerson, P.M., Haddad, D., 2015.
Mass drug administration for trachoma: how long is not long enough? PLoS Negl.
Trop. Dis. 9, e0003610.
Koukounari, A., Moustaki, I., Grassly, N.C., Blake, I.M., Basan ~ez, M.G., Gambhir, M.,
Mabey, D.C., Bailey, R.L., Burton, M.J., Solomon, A.W., Donnelly, C.A., 2013. Using
a nonparametric multilevel latent Markov model to evaluate diagnostics for trachoma.
Am. J. Epidemiol. 177, 913e922.
Lakew, T., House, J., Hong, K.C., Yi, E., Alemayehu, W., Melese, M., Zhou, Z., Ray, K.,
Chin, S., Romero, E., Keenan, J., Whitcher, J.P., Gaynor, B.D., Lietman, T.M., 2009.
Reduction and return of infectious trachoma in severely affected communities in
Ethiopia. PLoS Negl. Trop. Dis. 3, e376.
Last, A.R., Burr, S.E., Weiss, H.A., Harding-Esch, E.M., Cassama, E., Nabicassa, M.,
Mabey, D.C., Holland, M.J., Bailey, R.L., 2014. Risk factors for active trachoma and
ocular Chlamydia trachomatis infection in treatment-naïve trachoma-hyperendemic com-
munities of the Bijag os Archipelago, Guinea Bissau. PLoS Negl. Trop. Dis. 8, e2900.
Lee, B.Y., Bartsch, S.M., Gorham, K.M., 2015. Chapter eight e economic and financial
evaluation of neglected tropical diseases. Adv. Parasitol. 87, 329e417.
Lee, D.C., Chidambaram, J.D., Porco, T.C., Lietman, T.M., 2005. Seasonal effects in the
elimination of trachoma. Am. J. Trop. Med. Hyg. 72, 468e470.
Lietman, T.M., Porco, T.C., Dawson, C., Blower, S., 1999. Global elimination of
trachoma: how frequently should we administer mass chemotherapy? Nat. Med. 5,
572e576.
Lietman, T.M., Gebre, T., Abdou, A., Alemayehu, W., Emerson, P., Blumberg, S.,
Keenan, J.D., Porco, T.C., 2015. The distribution of the prevalence of ocular chlamydial
infection in communities where trachoma is disappearing. Epidemics 11, 85e91.
Mathematical Modelling 45

Lietman, T.M., Gebre, T., Ayele, B., Ray, K.J., Maher, M.C., See, C.W., Emerson, P.M.,
Porco, T.C., 2011. The epidemiological dynamics of infectious trachoma may facilitate
elimination. Epidemics 3, 119e124.
Liu, F.C., Porco, T.C., Ray, K.J., Bailey, R.L., Mkocha, H., Munoz, B., Quinn, T.C.,
Lietman, T.M., West, S.K., 2013. Assessment of transmission in trachoma programs
over time suggests no short-term loss of immunity. PLoS Negl. Trop. Dis. 7, e2303.
Liu, F.C., Porco, T.C., Mkocha, H.A., Munoz, B., Ray, K.J., Bailey, R.L., Lietman, T.M.,
West, S.K., 2014. The efficacy of oral azithromycin in clearing ocular chlamydia: math-
ematical modeling from a community-randomized trachoma trial. Epidemics 6, 10e17.
London Declaration on Neglected Tropical Diseases, 2012. Uniting to Combat Neglected
Tropical Diseases. Ending the Neglect and Reaching 2020 Goals. Available: http://
unitingtocombatntds.org/resource/london-declaration.
Martin, D.L., Bid, R., Sandi, F., Goodhew, E.B., Massae, P.A., Lasway, A., Philippin, H.,
Makupa, W., Molina, S., Holland, M.J., Mabey, D.C., Drakeley, C., Lammie, P.J.,
Solomon, A.W., 2015. Serology for trachoma surveillance after cessation of mass drug
administration. PLoS Negl. Trop. Dis. 9, e0003555.
Mladonicky, J.M., King, J.D., Liang, J.L., Chambers, E., Pa’au, M., Schmaedick, M.A.,
Burkot, T.R., Bradley, M., Lammie, P.J., 2009. Assessing transmission of lymphatic fila-
riasis using parasitologic, serologic, and entomologic tools after mass drug administration
in American Samoa. Am. J. Trop. Med. Hyg. 80, 769e773.
Murray, C.J., Vos, T., Lozano, R., Naghavi, M., Flaxman, A.D., Michaud, C., Ezzati, M.,
Shibuya, K., Salomon, J.A., Abdalla, S., Aboyans, V., Abraham, J., Ackerman, I.,
Aggarwal, R., Ahn, S.Y., Ali, M.K., Alvarado, M., Anderson, H.R., Anderson, L.M.,
Andrews, K.G., Atkinson, C., Baddour, L.M., Bahalim, A.N., Barker-Collo, S.,
Barrero, L.H., Bartels, D.H., Basan ~ez, M.G., Baxter, A., Bell, M.L., Benjamin, E.J.,
Bennett, D., Bernabé, E., Bhalla, K., Bhandari, B., Bikbov, B., Bin Abdulhak, A.,
Birbeck, G., Black, J.A., Blencowe, H., Blore, J.D., Blyth, F., Bolliger, I.,
Bonaventure, A., Boufous, S., Bourne, R., Boussinesq, M., Braithwaite, T.,
Brayne, C., Bridgett, L., Brooker, S., Brooks, P., Brugha, T.S., Bryan-Hancock, C.,
Bucello, C., Buchbinder, R., Buckle, G., Budke, C.M., Burch, M., Burney, P.,
Burstein, R., Calabria, B., Campbell, B., Canter, C.E., Carabin, H., Carapetis, J.,
Carmona, L., Cella, C., Charlson, F., Chen, H., Cheng, A.T., Chou, D.,
Chugh, S.S., Coffeng, L.E., Colan, S.D., Colquhoun, S., Colson, K.E., Condon, J.,
Connor, M.D., Cooper, L.T., Corriere, M., Cortinovis, M., de Vaccaro, K.C.,
Couser, W., Cowie, B.C., Criqui, M.H., Cross, M., Dabhadkar, K.C., Dahiya, M.,
Dahodwala, N., Damsere-Derry, J., Danaei, G., Davis, A., De Leo, D.,
Degenhardt, L., Dellavalle, R., Delossantos, A., Denenberg, J., Derrett, S., Des
Jarlais, D.C., Dharmaratne, S.D., Dherani, M., Diaz-Torne, C., Dolk, H.,
Dorsey, E.R., Driscoll, T., Duber, H., Ebel, B., Edmond, K., Elbaz, A., Ali, S.E.,
Erskine, H., Erwin, P.J., Espindola, P., Ewoigbokhan, S.E., Farzadfar, F., Feigin, V.,
Felson, D.T., Ferrari, A., Ferri, C.P., Fevre, E.M., Finucane, M.M., Flaxman, S.,
Flood, L., Foreman, K., Forouzanfar, M.H., Fowkes, F.G., Fransen, M.,
Freeman, M.K., Gabbe, B.J., Gabriel, S.E., Gakidou, E., Ganatra, H.A., Garcia, B.,
Gaspari, F., Gillum, R.F., Gmel, G., Gonzalez-Medina, D., Gosselin, R.,
Grainger, R., Grant, B., Groeger, J., Guillemin, F., Gunnell, D., Gupta, R.,
Haagsma, J., Hagan, H., Halasa, Y.A., Hall, W., Haring, D., Haro, J.M.,
Harrison, J.E., Havmoeller, R., Hay, R.J., Higashi, H., Hill, C., Hoen, B.,
Hoffman, H., Hotez, P.J., Hoy, D., Huang, J.J., Ibeanusi, S.E., Jacobsen, K.H.,
James, S.L., Jarvis, D., Jasrasaria, R., Jayaraman, S., Johns, N., Jonas, J.B.,
Karthikeyan, G., Kassebaum, N., Kawakami, N., Keren, A., Khoo, J.P., King, C.H.,
Knowlton, L.M., Kobusingye, O., Koranteng, A., Krishnamurthi, R., Laden, F.,
Lalloo, R., Laslett, L.L., Lathlean, T., Leasher, J.L., Lee, Y.Y., Leigh, J., Levinson, D.,
46 A. Pinsent et al.

Lim, S.S., Limb, E., Lin, J.K., Lipnick, M., Lipshultz, S.E., Liu, W., Loane, M.,
Ohno, S.L., Lyons, R., Mabweijano, J., MacIntyre, M.F., Malekzadeh, R.,
Mallinger, L., Manivannan, S., Marcenes, W., March, L., Margolis, D.J., Marks, G.B.,
Marks, R., Matsumori, A., Matzopoulos, R., Mayosi, B.M., McAnulty, J.H.,
McDermott, M.M., McGill, N., McGrath, J., Medina-Mora, M.E., Meltzer, M.,
Mensah, G.A., Merriman, T.R., Meyer, A.C., Miglioli, V., Miller, M., Miller, T.R.,
Mitchell, P.B., Mock, C., Mocumbi, A.O., Moffitt, T.E., Mokdad, A.A.,
Monasta, L., Montico, M., Moradi-Lakeh, M., Moran, A., Morawska, L., Mori, R.,
Murdoch, M.E., Mwaniki, M.K., Naidoo, K., Nair, M.N., Naldi, L., Narayan, K.M.,
Nelson, P.K., Nelson, R.G., Nevitt, M.C., Newton, C.R., Nolte, S., Norman, P.,
Norman, R., O’Donnell, M., O’Hanlon, S., Olives, C., Omer, S.B., Ortblad, K.,
Osborne, R., Ozgediz, D., Page, A., Pahari, B., Pandian, J.D., Rivero, A.P.,
Patten, S.B., Pearce, N., Padilla, R.P., Perez-Ruiz, F., Perico, N., Pesudovs, K.,
Phillips, D., Phillips, M.R., Pierce, K., Pion, S., Polanczyk, G.V., Polinder, S.,
Pope 3rd, C.A., Popova, S., Porrini, E., Pourmalek, F., Prince, M., Pullan, R.L.,
Ramaiah, K.D., Ranganathan, D., Razavi, H., Regan, M., Rehm, J.T., Rein, D.B.,
Remuzzi, G., Richardson, K., Rivara, F.P., Roberts, T., Robinson, C., De
Leon, F.R., Ronfani, L., Room, R., Rosenfeld, L.C., Rushton, L., Sacco, R.L.,
Saha, S., Sampson, U., Sanchez-Riera, L., Sanman, E., Schwebel, D.C., Scott, J.G.,
Segui-Gomez, M., Shahraz, S., Shepard, D.S., Shin, H., Shivakoti, R., Singh, D.,
Singh, G.M., Singh, J.A., Singleton, J., Sleet, D.A., Sliwa, K., Smith, E., Smith, J.L.,
Stapelberg, N.J., Steer, A., Steiner, T., Stolk, W.A., Stovner, L.J., Sudfeld, C.,
Syed, S., Tamburlini, G., Tavakkoli, M., Taylor, H.R., Taylor, J.A., Taylor, W.J.,
Thomas, B., Thomson, W.M., Thurston, G.D., Tleyjeh, I.M., Tonelli, M.,
Towbin, J.A., Truelsen, T., Tsilimbaris, M.K., Ubeda, C., Undurraga, E.A., van der
Werf, M.J., van Os, J., Vavilala, M.S., Venketasubramanian, N., Wang, M.,
Wang, W., Watt, K., Weatherall, D.J., Weinstock, M.A., Weintraub, R.,
Weisskopf, M.G., Weissman, M.M., White, R.A., Whiteford, H., Wiebe, N.,
Wiersma, S.T., Wilkinson, J.D., Williams, H.C., Williams, S.R., Witt, E., Wolfe, F.,
Woolf, A.D., Wulf, S., Yeh, P.H., Zaidi, A.K., Zheng, Z.J., Zonies, D., Lopez, A.D.,
Al Mazroa, M.A., Memish, Z.A., 2012. Disability-adjusted life years (DALYs) for 291
diseases and injuries in 21 regions, 1990e2010: a systematic analysis for the Global
Burden of Disease Study 2010. Lancet 380, 2197e2223.
Muench, H., 1934. Derivation of rates from summation data by the catalytic curve. JASA 29,
25e38.
~ez, M.G., Flaxman, S.R., Jonas, J.B., Keeffe, J., Leasher, J.L.,
Naidoo, K., Gichuhi, S., Basan
Pesudovs, K., Price, H., Smith, J.L., Turner, H.C., White, R.A., Wong, T.Y.,
Resnikoff, S., Taylor, H.R., Bourne, R.R., Vision Loss Expert Group of the Global
Burden of Disease Study, 2014. Prevalence and causes of vision loss in sub-Saharan Af-
rica: 1990e2010. Br. J. Ophthalmol. 98, 612e618.
Ngondi, J., Gebre, T., Shargie, E.B., Graves, P.M., Ejigsemahu, Y., Teferi, T., Genet, A.,
Mosher, A.W., Endeshaw, T., Zerihun, M., Messele, A., Richards Jr., F.O.,
Emerson, P.M., 2008. Risk factors for active trachoma in children and trichiasis in adults:
a household survey in Amhara Regional State, Ethiopia. Trans. R. Soc. Trop. Med. Hyg.
102, 432e438.
Oguttu, D., Byamukama, E., Katholi, C.R., Habomugisha, P., Nahabwe, C.,
Ngabirano, M., Hassan, H.K., Lakwo, T., Katabarwa, M., Richards, F.O.,
Unnasch, T.R., 2014. Serosurveillance to monitor onchocerciasis elimination: the
Ugandan experience. Am. J. Trop. Med. Hyg. 90, 339e345.
Parthasarathy, N.R., 1967. A simple catalytic model in trachoma epidemiology. J. All India
Ophthalmol. Soc. 15, 165e171.
Mathematical Modelling 47

Rabiu, M., Alhassan, M.B., Ejere, H.O., Evans, J.R., 2012. Environmental sanitary interven-
tions for preventing active trachoma. Cochrane Database Syst. Rev. 2, CD004003.
Rahman, S.A., West, S.K., Mkocha, H., Munoz, B., Porco, T.C., Keenan, J.D.,
Lietman, T.M., 2015. The distribution of ocular Chlamydia prevalence across Tanzanian
communities where trachoma is declining. PLoS Negl. Trop. Dis. 9, e0003682.
Ray, K.J., Lietman, T.M., Porco, T.C., Keenan, J.D., Bailey, R.L., Solomon, A.W.,
Burton, M.J., Harding-Esch, E., Holland, M.J., Mabey, D., 2009. When can antibiotic
treatments for trachoma be discontinued? Graduating communities in three African
countries. PLoS Negl. Trop. Dis. 3, e458.
Ray, K.J., Porco, T.C., Hong, K.C., Lee, D.C., Alemayehu, W., Melese, M., Lakew, T.,
Yi, E., House, J., Chidambaram, J.D., Whitcher, J.P., Gaynor, B.D., Lietman, T.M.,
2007. A rationale for continuing mass antibiotic distributions for trachoma. BMC Infect.
Dis. 7, 91.
Regoes, R.R., Wiuff, C., Zappala, R.M., Garner, K.N., Baquero, F., Levin, B.R., 2004.
Pharmacodynamic functions: a multiparameter approach to the design of antibiotic treat-
ment regimens. Antimicrob. Agents Chemother. 48, 3670e3676.
Shattock, A.J., Gambhir, M., Taylor, H.R., Cowling, C.S., Kaldor, J.M., Wilson, D.P.,
2015. Control of trachoma in Australia: a model based evaluation of current
interventions. PLoS Negl. Trop. Dis. 9, e0003474.
Solomon, A.W., Harding-Esch, E., Alexander, N.D.E., Aguirre, A., Holland, M.J.,
Bailey, R.L., Foster, A., Mabey, D.C.W., Massae, P.A., Courtright, P., Shao, J.F.,
2008. Two doses of azithromycin to eliminate trachoma in a Tanzanian community.
N. Engl. J. Med. 358, 1870e1871.
Solomon, A.W., Holland, M.J., Alexander, N.D., Massae, P.A., Aguirre, A., Natividad-
Sancho, A., Molina, S., Safari, S., Shao, J.F., Courtright, P., Peeling, R.W.,
West, S.K., Bailey, R.L., Foster, A., Mabey, D.C., 2004a. Mass treatment with single-
dose azithromycin for trachoma. N. Engl. J. Med. 351, 1962e1971.
Solomon, A.W., Holland, M.J., Burton, M.J., West, S.K., Alexander, N.D., Aguirre, A.,
Massae, P.A., Mkocha, H., Munoz, B., Johnson, G.J., Peeling, R.W., Bailey, R.L.,
Foster, A., Mabey, D.C., 2003. Strategies for control of trachoma: observational study
with quantitative PCR. Lancet 362, 198e204.
Solomon, A.W., Peeling, R.W., Foster, A., Mabey, D.C.W., 2004b. Diagnosis and assess-
ment of trachoma. Clin. Microbiol. Rev. 17, 982e1011.
Stocks, M.E., Ogden, S., Haddad, D., Addiss, D.G., McGuire, C., Freeman, M.C., 2014.
Effect of water, sanitation, and hygiene on the prevention of trachoma: a systematic re-
view and meta-analysis. PLoS Med. 11, e1001605.
Sundaresan, T.K., Assaad, F.A., 1973. The use of simple epidemiological models in the eval-
uation of disease control programmes: a case study of trachoma. Bull. World Health Or-
gan. 48, 709e714.
Taylor, H.R., Burton, M.J., Haddad, D., West, S., Wright, H., 2014. Trachoma. Lancet 384,
2142e2152.
Turner, H.C., Truscott, J.E., Hollingsworth, T.D., Bettis, A.A., Brooker, S.J.,
Anderson, R.M., 2015a. Cost and cost-effectiveness of soil-transmitted helminth treat-
ment programmes: systematic review and research needs. Parasites Vectors 8, 355.
Turner, H.C., Walker, M., Attah, S.K., Opoku, N.O., Awadzi, K., Kuesel, A.C.,
~ez, M.G., 2015b. The potential impact of moxidectin on onchocerciasis elimina-
Basan
tion in Africa: an economic evaluation based on the Phase II clinical trial data. Parasites
Vectors 8, 167.
Turner, H.C., Walker, M., Churcher, T.S., Basan ~ez, M.G., 2014a. Modelling the impact of
ivermectin on River blindness and its burden of morbidity and mortality in African
Savannah: EpiOncho projections. Parasites Vectors 7, 241.
48 A. Pinsent et al.

Turner, H.C., Walker, M., French, M.D., Blake, I.M., Churcher, T.S., Basan ~ez, M.G.,
2014b. Neglected tools for neglected diseases: mathematical models in economic
evaluations. Trends Parasitol. 30, 562e570.
West, S.K., 2003. Blinding trachoma: prevention with the safe strategy. Am. J. Trop. Med.
Hyg. 69, 18e23.
West, S.K., Munoz, B., Mkocha, H., Holland, M.J., Aguirre, A., Solomon, A.W., Foster, A.,
Bailey, R.L., Mabey, D.C., 2005. Infection with Chlamydia trachomatis after mass treat-
ment of a trachoma hyperendemic community in Tanzania: a longitudinal study. Lancet
366, 1296e1300.
West, S.K., Munoz, B., Turner, V.M., Mmbaga, B.B., Taylor, H.R., 1991. The epidemi-
ology of trachoma in central Tanzania. Int. J. Epidemiol. 20, 1088e1092.
World Health Organization (WHO), 2003. Report of the 2nd Global Scientific
Meeting on Trachoma. Available:http://www.who.int/blindness/2nd%20GLOBAL%
20SCIENTIFIC%20MEETING.pdf.
World Health Organization (WHO), 2012. Global WHO alliance for the elimination of
blinding trachoma by 2020. Wkly. Epidemiol. Rec. 87, 161e168.
World Health Organization (WHO), 2013. Trachoma. Status of Endemicity for Blinding
Trachoma, 2012. Available: http://apps.who.int/neglected_diseases/ntddata/trachoma/
trachoma.html.
World Health Organization (WHO), 2015c. Global Elimination of Trachoma Documents.
Available: http://www.who.int/blindness/publications/get2020/en/.
World Health Organization (WHO), 2015a. Neglected Tropical Diseases. Available: http://
www.who.int/neglected_diseases/diseases/en/.
World Health Organization (WHO), 2015b. Trachoma Fact Sheet No 382. Available:
http://www.who.int/mediacentre/factsheets/fs382/en/.
Wilkins, P.P., Keystone, J.S., 2013. Schistosomiasis serology is valuable and reliable. Clin.
Infect. Dis. 56, 312.
Wong, J., Hamel, M.J., Drakeley, C.J., Kariuki, S., Shi, Y.P., Lal, A.A., Nahlen, B.L.,
Bloland, P.B., Lindblade, K.A., Were, V., Otieno, K., Otieno, P., Odero, C.,
Slutsker, L., Vulule, J.M., Gimnig, J.E., 2014. Serological markers for monitoring histor-
ical changes in malaria transmission intensity in a highly endemic region of Western
Kenya, 1994e2009. Malar. J. 13, 451.
Wright, H.R., Taylor, H.R., 2005. Clinical examination and laboratory tests for estimation
of trachoma prevalence in a remote setting: what are they really telling us? Lancet Infect.
Dis. 5, 313e320.
Yildiz Zeyrek, F., Palacpac, N., Yuksel, F., Yagi, M., Honjo, K., Fujita, Y., Arisue, N.,
Takeo, S., Tanabe, K., Horii, T., Tsuboi, T., Ishii, K.J., Coban, C., 2011. Serologic
markers in relation to parasite exposure history help to estimate transmission dynamics
of Plasmodium vivax. PLoS One 6, e28126.
CHAPTER TWO

Progress in the Mathematical


Modelling of Visceral
Leishmaniasis
K.S. Rock*, R.J. Quinnellx, G.F. Medley{, O. Courtenay*, 1
*University of Warwick, Coventry, United Kingdom
x
University of Leeds, Leeds, United Kingdom
{
London School of Hygiene and Tropical Medicine, London, United Kingdom
1
Corresponding author: E-mail: orin.courtenay@warwick.ac.uk

Contents
1. Introduction 51
1.1 Geographical distribution and disease burden 51
1.2 Aetiological agents and transmission cycles 52
1.3 Disease in humans 52
1.3.1 The clinical spectrum 52
1.3.2 Natural history of VL disease 53
1.3.3 Transmission: xenodiagnosis vs. tissue and blood parasite loads 57
1.4 Disease in dogs 59
1.5 Sandfly biology 61
1.5.1 Life expectancy 62
1.5.2 Seasonality of sandflies 63
1.5.3 Feeding behaviour 63
1.6 Control of human VL 64
1.6.1 Treatment of VL 64
1.6.2 Parasite loads in response to treatment 66
1.6.3 Vector control 66
1.7 Control of ZVL 69
2. Mathematical Models of Visceral Leishmaniasis 71
2.1 Infection in humans 72
2.2 Infection in dogs 83
2.3 The role of sandflies 85
2.3.1 The basic reproduction number 89
2.3.2 Sandfly host choice 91
2.3.3 Seasonality of sandflies 92
3. DataeModel Interactions 96
3.1 Current data and initial models 96
3.2 Future data and alternative sources 99
4. Modelling Interventions 99
4.1 Humans 99
4.1.1 Diagnostics 99

Advances in Parasitology, Volume 94


© 2016 Elsevier Ltd.
j
ISSN 0065-308X
http://dx.doi.org/10.1016/bs.apar.2016.08.001 All rights reserved. 49
50 K.S. Rock et al.

4.1.2 Treatment 101


4.1.3 Human vaccine (hypothetical) 102
4.2 Vector 103
4.2.1 In the ISC 104
4.2.2 In Africa 104
4.2.3 In Brazil 105
4.3 Dogs 106
4.3.1 Diagnostics 106
4.3.2 Treatment 107
4.3.3 Dog culling 108
4.3.4 Dog vaccination 109
4.4 Other animal hosts 111
5. Conclusions 111
Acknowledgements 113
References 114

Abstract
The leishmaniases comprise a complex of diseases characterized by clinical outcomes
that range from self-limiting to chronic, and disfiguring and stigmatizing to life threat-
ening. Diagnostic methods, treatments, and vector and reservoir control options exist,
but deciding the most effective interventions requires a quantitative understanding of
the population level infection and disease dynamics. The effectiveness of any set of in-
terventions has to be determined within the context of operational conditions,
including economic and political commitment. Mathematical models are the best avail-
able tools for studying quantitative systems crossing disciplinary spheres (biology, med-
icine, economics) within environmental and societal constraints.
In 2005, the World Health Assembly and government health ministers of India, Nepal,
and Bangladesh signed a Memorandum of Understanding to eliminate the life threat-
ening form of leishmaniasis, visceral leishmaniasis (VL), on the Indian subcontinent by
2015 through a combination of early case detection, improved treatments, and vector
control. The elimination target is <1 case/10,000 population at the district or subdistrict
level compared to the current 20/10,000 in the regions of highest transmission.
Towards this goal, this chapter focuses on mathematical models of VL, and the
biology driving those models, to enable realistic predictions of the best combination
of interventions. Several key issues will be discussed which have affected previous
modelling of VL and the direction future modelling may take. Current understanding
of the natural history of disease, immunity (and loss of immunity), and stages of infec-
tion and their durations are considered particularly for humans, and also for dogs.
Asymptomatic and clinical infection are discussed in the context of their relative roles
in Leishmania transmission, as well as key components of the parasiteesandflyevector
interaction and intervention strategies including diagnosis, treatment and vector con-
trol. Gaps in current biological knowledge and potential avenues to improve model
structures and mathematical predictions are identified. Underpinning the marriage be-
tween biology and mathematical modelling, the content of this chapter represents the
first step towards developing the next generation of models for VL.
Progress in the Mathematical Modelling of Visceral Leishmaniasis 51

List of Abbreviations
CI Confidence interval
CL Cutaneous leishmaniasis
DALYs Disability adjusted life years
DAT Direct agglutination test
DMC Deltamethrin-impregnated collar
EIP Extrinsic incubation period
ELISA Enzyme-linked immunosorbent assay
EVM Environmental vector management
GDP Gross domestic product
ICER Incremental cost-effectiveness ratio
IRS Indoor residual spraying
ISC Indian subcontinent
ITNs Insecticide-treated nets
KA Kala-azar
LLINs Long-lasting insecticidal nets
LST Leishmania skin test
MCL Mucocutaneous leishmaniasis
MIL Miltefosine
NTD Neglected tropical disease
ODE Ordinary differential equations
PCR Polymerase chain reaction
PKDL Post-Kala-azar dermal leishmaniasis
PM Paromomycin
ROI Rate of infection
SEIR Susceptible, exposed, infected, recovered
SIR Susceptible, infected, recovered
SSG Sodium stibogluconate
VL Visceral leishmaniasis
WHO World Health Organization
ZVL Zoonotic visceral leishmaniasis

1. INTRODUCTION
1.1 Geographical distribution and disease burden
Endemic transmission of visceral leishmaniasis (VL) occurs in 79 coun-
tries on five continents where >58,000 new cases are officially reported each
year. Adjusting by the estimated levels of underreporting (1.2- to 8.0-fold),
the true annual case incidence is likely to be close to 0.2e0.4 million (Alvar
et al., 2012). Worldwide, 2.35 (95% CI 2.18e4.90) million disability
adjusted life years (DALYs) are lost due to leishmaniasis, placing leishmani-
asis second in mortality and fourth in morbidity burden among tropical
diseases, where VL is ranked as the second largest parasitic killer after malaria
(Mathers et al., 2007; Hotez et al., 2014).
52 K.S. Rock et al.

Most (90%) VL cases occur in six countries: India, Bangladesh, Sudan, South
Sudan, Brazil and Ethiopia (Alvar et al., 2012), and 60% of the global VL disease
burden is in the Indian subcontinent (ISC) (WHO, 2002), where intense trans-
mission is localized in Bihar state and Bengal (northeast [NE] India), in the
neighbouring Terai region of southeast (SE) Nepal, and in the Mymensingh
district in Central and West Bangladesh (Alvar et al., 2012). In Latin America,
VL is less geographically focal, traditionally rural but with recent expansion
into urbanized areas, notably in Brazil where approximately 90% of all VL cases
in the Americas occur (Maia-Elkhoury et al., 2008).

1.2 Aetiological agents and transmission cycles


Leishmaniasis in humans is caused by more than 20 species of vector-borne
protozoan flagellates of the genus Leishmania in two subgenera L. (Leishmania)
and L. (Viannia). Visceral leishmaniasis, also known as ‘Kala-azar’ (KA), is
usually caused by members of the Leishmania donovani complex (L. donovani
[syn. Leishmania archibaldi] in south Asia and east Africa, and Leishmania infan-
tum [syn. Leishmania chagasi] in central and south America, Mediterranean
countries, central Asia and parts of China) (Lukes et al., 2007). There are
also rare reports of VL caused by infection with Leishmania mexicana and
Leishmania tropica, which otherwise cause cutaneous leishmaniasis (Sacks
et al., 1995; Barral et al., 1986).
Leishmania is transmitted by blood-feeding females of at least 30 Phle-
botomine sandfly species in zoonotic transmission cycles involving one or
more mammal reservoir hosts, with the exception of two species (L. donovani
and L. tropica) that are transmitted in anthroponotic cycles. Domestic dogs
are the sole proven animal reservoir of VL due to L. infantum; however,
proven transmission between sandfly vectors and wild hares (Lepus granaten-
sis) in Madrid, indicates their potential as secondary local reservoirs (Molina
et al., 2012). Despite a growing number of reports of L. donovani infections
in dogs and other mammals in E. Africa and in the ISC (Bhattarai et al.,
2010; Dereure et al., 2003; Alam et al., 2011; Sharma et al., 2009; Kenubih
et al., 2014; Kassahun et al., 2015; Siriwardana et al., 2012) these hosts have
yet to be shown to be epidemiologically relevant in maintaining transmis-
sion; differentiating incidental or ‘sink’ hosts from reservoir or ‘source’ hosts
is required (reviewed by Quinnell and Courtenay, 2009).

1.3 Disease in humans


1.3.1 The clinical spectrum
Leishmaniasis is characterized by a diversity of aetiological agents causing
cutaneous (CL), mucocutaneous (MCL) and visceral (VL) presentations
dependent on the species or strain of Leishmania and on the integrity of
the host’s immune system. Infections may last for months to years, requiring
Progress in the Mathematical Modelling of Visceral Leishmaniasis 53

specific and varied disease management strategies. VL is a systemic infection


clinically characterized by prolonged fever, weight loss, splenohepatomegaly,
pancytopenia and hypergammaglobulinemia and anaemia. VL is almost al-
ways fatal if not treated, amounting to 20,000e40,000 deaths per year (Alvar
et al., 2012); however, the majority of VL-related deaths go unrecognized
(Ahluwalia et al., 2003; Barnett et al., 2005; Collin et al., 2006). The risk
of mortality from VL is exacerbated in immunocompromised (e.g. HIV
coinfected) patients with concomitant conditions (e.g. severe malnutrition)
(Alvar et al., 2008; Cerf et al., 1987; Anstead et al., 2001; Dye and Williams,
1993; Bern et al., 2007), and those with limited treatment access due to con-
flict, famine or population displacement (Reithinger et al., 2007). Even
among patients with access to VL treatment, the estimated case-fatality rate
is 10% (Alvar et al., 2012).
Post-Kala-azar dermal leishmaniasis (PKDL) is the late and chronic devel-
opment of skin lesions, usually following the apparent clinical cure of VL due
to L. donovani in South Asia and Africa (Zijlstra et al., 2003). PKDL can
develop months to years following apparent successful treatment for KA, or
can develop without a previous KA episode. PKDL is not life threatening
but is associated with disfigurement and social stigma (see Section 1.3.2).

1.3.2 Natural history of VL disease


Transitioning from being susceptible to developing disease is a continuous
process, but one of the challenges of mathematical modelling involves defi-
nition of discrete stages (compartments) and their interactions and overlaps.
These stages are outlined below.
Susceptible. All naïve individuals across all age groups are potentially
susceptible if exposed to Leishmania.
Individuals are not equally at risk of exposure and therefore infection,
with significant heterogeneities dependent on environmental and socioeco-
nomic factors, especially related to poverty (Boelaert et al., 2009; Hasker
et al., 2012; Alvar et al., 2012). Once infected the following clinical cate-
gories are recognized:
Asymptomatic infection. The majority of infections are clinically
asymptomatic (subclinical) in the long term. There is no standard definition
of asymptomatic infection (Singh et al., 2014; Carneiro et al., 2011), though
infection is indicated by detection of anti-Leishmania antibody by immuno-
logical tests (e.g. rK39 immunochromatographic test, direct agglutination
test [DAT], enzyme-linked immunosorbent assay [ELISA]), by parasite
detection by tissue culture, or by PCR. The variation in the sensitivity
and specificity of these tests with respect to asymptomatic infection, hence
their usefulness for surveillance, is difficult to determine (Cunningham
et al., 2012; de Ruiter et al., 2014; Adams et al., 2013). Antibody presence
54 K.S. Rock et al.

for example can indicate current infection, past exposure, or relatively recent
cure posttreatment for KA (see below). Thus, in a cross-sectional survey, in-
dividuals that mount an effective immune response will include individuals
who do not progress to clinical KA.
With the exception of sporadic outbreaks (Ritmeijer et al., 2007), clinical
development of leishmaniasis is rare relative to the number of infections. The
ratio of infection to disease varies for L. infantum in Brazil (6.5:1, 9:1, 18.5:1),
Spain (50:1), Iran (13:1); L. donovani in Bangladesh (4:1), India and Nepal
(8.9:1), and L. donovani in Sudan (7:1,1.5:1), Kenya (6:1, 4:1) and
Ethiopia (5.6:1, 11:1) (Evans et al., 1992; Badaro et al., 1986a,b; Ali and
Ashford, 1994; Schaefer et al., 1995; Zijlstra et al., 1994; Hailu et al., 2009;
Davies and Mazloumi Gavgani, 1999). Chapman et al. (2015) estimate that
20% of asymptomatic infections develop into KA in Bangladesh, which is
higher than other estimates, but utilizes individual, longitudinal data over
3 years. The collective figures suggest that a larger proportion of L. donovani
than L. infantum asymptomatic infections progress to disease but some of the
differences between species and studies come from different definitions of
asymptomatic infection and the use of different diagnostic tests; some studies
use the Leishmania skin test (LST) as a marker of infection; others use anti-
body detection tests of variable sensitivity and specificity. LST positivity is
considered indicative of delayed-type hypersensitivity associated with a pro-
tective T-cell mediated response, whereas anti-Leishmania antibody is not
considered protective. Immune markers characterizing likely progression to
disease are not fully defined but are influenced by host genetic factors, and
the degree of T-cell mediated response (reviewed by Kedzierski and Evans,
2014). Well-established risk factors influencing the probability of disease
development include malnutrition, HIV coinfection, and poverty-related
conditions (Bern et al., 2010; Boelaert et al., 2009; Alvar et al., 2008). Sero-
conversion or high Leishmania antibody titres indicate likely disease develop-
ment, but these are not specific markers (Hailu et al., 2009; Hasker et al.,
2014; Sudarshan et al., 2014; Carneiro et al., 2011; Badaro et al., 1986a;
Chapman et al., 2015). Additional challenges in defining asymptomatic status
are discussed elsewhere (Singh et al., 2014; Carneiro et al., 2011).
Estimates of the prepatent period, from detection of asymptomatic infec-
tion to development of KA vary between studies: in Sudan KA develops in
2e4 months (range: 2 weeks to 12 months) (Mueller et al., 2012); in
Ethiopia this period is typically 3e8 months (Hailu et al., 2009); in Brazil,
a mean of 4.9 months (standard deviation [SD] 3.6 months) (Badaro et al.,
1986a), and in the ISC 3e6 months (Topno et al., 2010) and 142 days
(Chapman et al., 2015).
Disease 1 e KA is defined on presentation of clinical signs, confirmed
by a positive serological or/and parasitological assay. In ISC operational
Progress in the Mathematical Modelling of Visceral Leishmaniasis 55

programmes, occurrence of at least 2 weeks fever and a positive rK39 test


result trigger treatment. The interactions between exposure and disease can
be inferred from the age distributions of KA cases. KA due to L. infantum
(also known as ‘infantile VL’) typically affects children <10 years (Badaro
et al., 1986a; Davies and Mazloumi Gavgani, 1999), whereas KA due to
L. donovani tends to occur in older children and young adults <30 years
in some locations, or older in others (Hailu et al., 2009; Mueller et al.,
2012; Burza et al., 2014; Hasker et al., 2013). The age distribution will
depend upon the exposure history of the population and will likely vary
over the course of an epidemic cycle which is thought to be about 10e
15 years in the ISC (Dye and Wolpert, 1988).
Disease 2 e PKDL occurs as a sequel to treatment and apparent cure of
KA caused by L. donovani; it is clinically characterized as dermal macular,
maculopapular or nodular lesions on the face, upper arms and trunk (Zijlstra
et al., 2003). People are usually otherwise well and the symptoms are less se-
vere and distinct from KA with no associated mortality. Laboratory diagnosis
of PKDL is challenging (Adams et al., 2013), and most cases are diagnosed on
clinical grounds. PKDL is more common, less chronic and faster to develop
the initial signs in parts of East Africa than in the ISC: up to 56% of treated
KA patients develop PKDL within 6 months (range: 0e13 months) in Sudan
(Zijlstra et al., 2003; Musa et al., 2013), compared to 2e18% within 2e
3 years (range: 6 months to 32 years) in the ISC (Rahman et al., 2010; Singh
et al., 2012; Uranw et al., 2011; Mondal et al., 2010a; Sultana et al., 2012). In
Sudan, skin lesions tend to heal spontaneously in around 85% of patients and
within 12 months, lasting a mean of 9.7 months (SD 4.7 months; range: 2e
28 months). In Bangladesh, around 50% (48/98) of untreated PKDL patients
showed spontaneous resolution within a median of 19 months (Islam et al.,
2013). Sudanese patients with lesions persisting >12 months show high
anti-Leishmania antibody titres (DAT) and negative LST (Zijlstra et al.,
2003; Musa et al., 2002). For 5e10% of PKDL cases, there is no prior history
of KA, and occasionally PKDL presents concomitantly with KA (Zijlstra
et al., 2003; Rahman et al., 2010; Islam et al., 2013). PKDL in Bangladeshi
KA patients is more likely in younger (<15 years) than in older patients,
with a median PKDL patient age of 12 years (Islam et al., 2013).
Infectious. For individuals to be confirmed as infectious/infective it is
necessary to demonstrate their ability to infect the blood-feeding sandfly
vector, which is known as xenodiagnosis. The infectious status of KA,
PKDL and asymptomatic patients is unclear, representing a large gap in
our current understanding of transmission pathways (see Section 1.3.3). A
proportion of symptomatic L. infantum infections are infectious to sandflies,
but their degree of infectiousness appears to be lower than for symptomatic
L. donovani infections and PKDL (Quinnell and Courtenay, 2009). The
56 K.S. Rock et al.

higher infectiousness of HIV-coinfected individuals (Molina et al., 2003)


suggests that this condition could become locally more relevant particularly
where HIV-coinfection case numbers are rising (Alvar et al., 2008, 2012).
KA patients are usually treated soon after presentation at health facilities,
so their period of potential infectiousness depends on care-seeking be-
haviour and diagnostic delays. However, relapses after treatment are not
uncommon, which could lengthen the infectious period. The proportion
of the population developing PKDL, though smaller, may delay seeking
health care especially when symptoms are mild; also the treatment may
not be completed, for example if considered by patients too expensive or
logistically difficult (Ozaki et al., 2011). In Bangladesh, current treatment
guidelines call for a 120-day course of intramuscular injections of sodium sti-
bogluconate (SSG) (Rahman et al., 2010). Thus a proportion of PKDL cases
may remain in a chronic state and are potentially infectious for long periods.
Recovered and immune. KA rarely occurs more than once in an in-
dividual, and treated cases appear to develop a strong immunity to disease (if
not infection). However, it is not clear whether asymptomatic infection
(that does not progress to disease) is protective against future infections
and future disease. The LST is often used as a measure of protective immu-
nity and shows a variable duration. A total of 42 out of 43 (98%) subjects
were still LST positive 10 years postcure from L. infantum in Maranh~ao,
Brazil (Viana et al., 2011). In other regions of Brazil, only 4% and 16%
were LST positive after 8e720 days and 4e9 years, respectively, postrecov-
ery from KA (Mayrink et al., 1971; Badaro et al., 1986a). Reasons for shorter
positive skin test durations are likely to include absence of re-exposure to
Leishmania to stimulate immunity, failing host immunity with age, and the
technical quality and specificity of heterologous vs. homologous LST anti-
gens (Bern et al., 2006). Few to no new KA episodes have been reported
among LST positives, and subsequent episodes are rare in others truly cured
of KA (Burza et al., 2014; Hailu et al., 2009). Nonetheless, after clinical re-
covery, Leishmania antibodies persist for variable periods of up to 12 years
(Gidwani et al., 2011; Bern et al., 2005), the variability partially determined
by the degree of continued exposure and any persistent infection.
On the other hand, relapses in response to less optimal treatment regimes
are not uncommon (see Section 1.6.1). To determine if immunocompetent
past cases return to be partially or fully susceptible to infection and disease, it
is essential to distinguish new infections from parasitological relapse, which is
rarely studied. Relapse has been observed in 119/8409 immunocompetent
KA cases in a median of 10.1 months (interquartile range: 7.1e13.8 months)
Progress in the Mathematical Modelling of Visceral Leishmaniasis 57

with 32% of these relapsing >12 months postsuccessful treatment with


20 mg/kg Ambisome, which is considered the current most effective treat-
ment against KA in the ISC (Burza et al., 2014). Longitudinal follow-up
studies are few and current parasite diagnostic tools may lack the required
sensitivity to detect low-level residual infections.

1.3.3 Transmission: xenodiagnosis vs. tissue and blood parasite


loads
Since most VL infections are asymptomatic, quantifying their transmission
potential relative to KA and PKDL cases is critical. Assuming that asymptom-
atically infected KA and PKDL cases all contribute towards infection in sand-
flies, it is possible that transmission is primarily driven by the asymptomatic
group, even if the relative infectivity of asymptomatics is much lower than
KA or PKDL (see Section 2.1). PKDL patients have also been suggested to
be an important reservoir based on observations that they consistently harbour
L. donovani parasites in skin lesions, often do not seek or comply with long
treatments, and can remain unsuccessfully treated for years (Nasreen et al.,
2012; Verma et al., 2010; Thakur and Kumar, 1992). Cryptic reservoirs
will become increasingly important to tackle as elimination is approached.
Quantifying infectiousness of host populations relies on xenodiagnosis or
artificial exposure of sandfly vectors to host blood, but xenodiagnosis studies
in particular are difficult on logistic grounds (e.g. maintaining sandfly
colonies) and ethical grounds (e.g. exposing asymptomatic infecteds).
Consequently xenodiagnosis studies are few, most are cross-sectional, and
involve small numbers of hosts or vectors (reviewed by Quinnell and
Courtenay, 2009). Historical xenodiagnosis studies demonstrate that both
KA and PKDL patients can be infectious to sandflies: 18/21 L. donovani
KA patients, and 4/4 PKDL (nodular lesions) infected 2e42% and 15e
53% of exposed Phlebotomus argentipes respectively, and 23/50 L. infantum
KA patients were infectious, on average, to 10e51% of exposed Lutzomiya
longipalpis/Phlebotomus perniciosus sandflies (reviewed by Quinnell and
Courtenay, 2009). In contrast, 0/27 asymptomatic L. infantum infected in-
dividuals exposed to Lu. longipalpis were infectious (Costa et al., 2000),
but being LST positive, it is likely that they had resolved their infections
and were more appropriately classified as recovered.
Molecular methods, particularly quantitative PCR (qPCR), to measure
parasite burdens in skin and peripheral blood, i.e. accessible to blood-feeding
sandflies, are more frequently being applied in population studies, and are
often assumed to be reliable markers of transmission potential.
58 K.S. Rock et al.

Unfortunately, the relationship between parasite load and xenodiagnosis


outcomes is not well defined, and is likely to vary with the Leishmaniae
host interaction, clinical stage and kinetics of parasite tropism. It can be diffi-
cult to compare studies as the number of target gene copies (e.g. kDNA for
preparing qPCR standards) varies between amastigotes and promastigotes,
and even between strains (Weirather et al., 2011). Studies suggest that there
exists a parasite density threshold above which canine and human infections
are disproportionately infectious (Courtenay et al., 2014; Miller et al., 2014).
In contrast, natural wildlife hosts (e.g. foxes and hares for L. infantum, rodents
for Leishmania braziliensis and L. tropica), in which infection is usually asymp-
tomatic, show no apparent relationship between parasite loads and
infectivity to sandflies (Courtenay et al., 2002a; Svobodova et al., 2003;
Kassahun et al., 2015; Andrade et al., 2015; Molina et al., 2012).
Leishmania burdens in blood may, to some extent, differentiate asymp-
tomatic and symptomatic infections and identify which infections are
more likely to progress to disease, or to relapse posttreatment (Sudarshan
et al., 2014; Mary et al., 2006). L. infantum (Pourabbas et al., 2013;
Mary et al., 2006; Moral et al., 2002; Biglino et al., 2010) and L. donovani
(Miller et al., 2014; Sudarshan and Sundar, 2014) have been detected in
blood in up to 58% of asymptomatic infections by a cross-sectional study.
Fourteen per cent of 4695 asymptomatic Ethiopians were qPCR-positive
in blood of which 3.2% had high genome equivalent counts (Miller et al.,
2014), suggesting that distribution of parasitaemia is skewed and that perhaps
the small proportion of asymptomatic infections with higher parasitaemia is
more likely to develop disease and has a disproportional role in onward
transmission. However of 44 Brazilian children with positive qPCR results
at baseline, only 10 remained qPCR-positive 12 months later, showing a
significant decrease in genome counts (dos Santos Marques et al., 2012).
Only 4 of 25 asymptomatic PCR-positive individuals remained PCR-
positive 12 months later in SE Brazil (de Gouvea Viana et al., 2008). In sharp
contrast, 83% (80/97) of asymptomatic individuals living in an active trans-
mission region of Brazil tested PCR-positive during 6 years follow-up
(tested three times by PCR) (Carneiro et al., 2011).
In PKDL patients, parasite loads are reported to be greater in nodular
compared to macular or papular lesions (Nasreen et al., 2012; Verma
et al., 2010, 2013), whereas samples from asymptomatic skin from PKDL
patients in Bangladesh (n ¼ 20) tested negative compared to the 95% posi-
tivity of active lesions (Nasreen et al., 2012). PCR positive results were ob-
tained in 35% (8/23) of slit skin samples taken from PKDL lesions in the
Progress in the Mathematical Modelling of Visceral Leishmaniasis 59

same region (Islam et al., 2013). Any contribution of PKDL to onward


transmission may differ substantially between Africa and the ISC due to dif-
ferences in PKDL onset, progression and clinical aspects as described above
(Section 1.3.2). Lesion type (i.e. parasite loads), treatment (see Section 1.6.1)
and health care-seeking behaviours are also likely to influence infectious
duration. Assuming nonbiased selection of study subjects, the majority
[77% (106/138) and 97% (107/110)] of PKDL patients’ cutaneous lesions
involved macules and papules, and only 3% (3/110 patients) and 4% (5/
138 patients) were nodular forms; the remainder were mixed or complex
lesion types (Nasreen et al., 2012; Islam et al., 2013). Also higher loads
were recorded in more chronic PKDL cases (>3 years history) compared
to acute (<3 years history) PKDL cases suggesting an increased parasite
load with disease duration (Verma et al., 2010).

1.4 Disease in dogs


Leishmania infantum infection in the domestic dog also causes a spectrum of
clinical disease, from asymptomatic infection to severe, fatal disease (reviewed
by Solano-Gallego et al., 2009). The burden of disease due to zoonotic
visceral leishmaniasis (ZVL) in dogs is difficult to quantify, since prevalence
varies at both large and small geographical scales, and diagnostic tests have
varying sensitivity and specificity. In Western Europe a review of canine sero-
logical surveys found the overall seroprevalence in 500,000 sampled dogs to
be 23%, with a median value of 10% (Franco et al., 2011). The prevalence
of disease will be lower, but the true prevalence of infection will be much
higher, since serology is known to have low sensitivity for detecting asymp-
tomatic infection (Baneth et al., 2008). The use of molecular methods has
shown that the real prevalence of infection can be up to 80%, in areas where
<30% were seropositive (Solano-Gallego et al., 2001; Lachaud et al., 2002).
Similarly, longitudinal studies using multiple diagnostic methods have shown
that the incidence of infection can be very high, with a mean time to infection
of 115 days in one Brazilian study (Quinnell et al., 1997), and incidence of
40e92% over the course of a transmission season in Europe (Dye et al.,
1993; Oliva et al., 2006). Indeed in some regions the incidence appears to
be steadily rising (Antoniou et al., 2009).
The course of infection is best examined using longitudinal studies of
natural infection. The few examples include a series of studies of ken-
nelled beagles exposed to natural infection in Italy, and a cohort study
of owned mixed-breed dogs in Amazonian Brazil (Quinnell et al.,
1997, 2001; Courtenay et al., 2002b; Oliva et al., 2006; Foglia Manzillo
60 K.S. Rock et al.

et al., 2013). Such studies have confirmed that infected dogs can be qual-
itatively divided into two groups: resistant and susceptible. In the Italian
studies, around 50% of infected dogs either showed no evidence of estab-
lished infection but had low levels of antileishmanial antibodies, or were
PCR-positive with low or no antileishmanial antibody response, and did
not progress to clinical disease. The other 50% of dogs, after a variable
period of asymptomatic infection, showed progressive infection, with
increasing parasite burdens and serological titres and development of clin-
ical signs (Gradoni, 2015). Clinical signs in canine VL are varied and
nonspecific: the most common signs are weight loss, lymphadenopathy
and a range of cutaneous signs (Baneth et al., 2008; Foglia Manzillo
et al., 2013). The proportion of dogs that develop clinical disease is harder
to estimate in Brazil, since dogs in many endemic areas are generally
malnourished and suffer from a range of other infections presenting similar
clinical signs. Resistance and susceptibility are associated with differences in
cellular immune responses, but little is known about the factors that under-
lie this variation in outcome between dogs (Baneth et al., 2008; Solano-
Gallego et al., 2009). Age, sex, nutrition and coinfection may play a
role, but variation in outcome is seen in even well-nourished experimental
dogs. Host genetics are known to be important, even within pedigree
breeds (Quilez et al., 2012).
Few studies have examined infectiousness, but it is likely that all suscep-
tible dogs become infectious, to a greater or lesser degree. In the Brazilian
study, about half of infected dogs became infectious, an average of
3 months after seroconversion, i.e. about 6 months after infection
(Courtenay et al., 2002b). This and other xenodiagnostic studies have
shown that the proportion of infectious dogs varies with clinical status
and also geographical area. Meta-analysis of seven published studies
showed that the proportion of polysymptomatic dogs that are infectious
is very high (0.80), but that a proportion of asymptomatic dogs are also in-
fectious (0.29) (Quinnell and Courtenay, 2009). The proportion of infec-
tious dogs was higher in European than in Latin American studies (0.86 vs.
0.45), which could indicate a greater susceptibility of an Old World sandfly
vector, Phlebotomus perniciosus, to infection compared to the major New
World vector, Lu. longipalpis. However, it remains unclear what the role
of resistant dogs is in disease transmission. Although some asymptomatic
dogs are infectious, it is not possible to tell from cross-sectional studies
whether these dogs are susceptible dogs in the early stages of infection,
or resistant dogs. The Brazil cohort study suggested that most
Progress in the Mathematical Modelling of Visceral Leishmaniasis 61

asymptomatic infectious dogs did go on to develop disease, and that truly


resistant dogs have only a small role in transmission (Courtenay et al.,
2002b).
Indeed there is a high degree of heterogeneity between symptomatic
dogs in their level of infectiousness, with very different contributions to
transmission. Of 42 infected dogs (18 infectious) in the Brazilian cohort
study, 7 (17%) highly infectious dogs were responsible for > 80% of the
transmission events (Courtenay et al., 2002b), and other studies show that
15e44% of infected dogs are responsible for >80% of transmission (Travi
et al., 2001; Molina et al., 1994). These results have been corroborated by
studies of tissue parasite burdens across dogs, which show similarly high
levels of heterogeneity (Courtenay et al., 2014).
Diagnosis of infection in dogs has been reviewed (Solano-Gallego
et al., 2009). PCR, particularly qPCR, is the most sensitive diagnostic
technique, and is usually the first test to be positive early after infection
(Oliva et al., 2006; Quinnell et al., 2001). Detecting infection in resistant
asymptomatic dogs is difficult, as a proportion of PCR-positive resistant
dogs do not go on to seroconvert, and antibody levels are low and vary
through time. Susceptible dogs appear to all seroconvert, but the time
taken for seroconversion can vary considerably. Clinical signs are not suf-
ficient for diagnosis due to their low specificity. Distinguishing infectious
from noninfectious dogs is also difficult; however, parasite burden assessed
by qPCR may provide a useful measure of the level of infectiousness
(Courtenay et al., 2014).
In addition to dogs and humans, a number of both domestic and wild
mammals can be naturally infected with L. infantum (reviewed by Quinnell
and Courtenay, 2009). In general, there is as yet no strong evidence that
hosts other than the domestic dog are responsible for significant human
infection, but further studies, particularly xenodiagnostic studies, are needed.
Other domestic or sylvatic reservoirs have thus been ignored in modelling to
date. However, there is recent evidence that hares, rather than dogs, have
been responsible for a localized outbreak of several hundred human cases
in Madrid (Molina et al., 2012).

1.5 Sandfly biology


Sandflies are small biting flies of the family Psychodidae. Only adult
females take blood meals, while both adult males and females feed on
plant sugars; larvae are terrestrial (see Killick-Kendrick, 1999 for a review
of sandfly biology). The vector of VL in the ISC is Phlebotomus argentipes,
62 K.S. Rock et al.

though Phlebotomus papatasi is suspected to be a secondary vector. In Sudan


the main vector is Phlebotomus orientalis, with other vector species occurring
in Kenya and China (Killick-Kendrick, 1999). The most widespread
vector of ZVL in the New World is Lu. longipalpis, which is likely to be
a complex of several species, while other species (Lutzomyia evansi and Lut-
zomyia cruzi) may be important locally (Lainson and Rangel, 2005;
Quinnell and Courtenay, 2009). In the Old World, there are at least a
dozen sandfly vectors for L. infantum, with Phlebotomus perniciosus and Phle-
botomus ariasi the major vectors in the western Mediterranean (Killick-
Kendrick, 1999).

1.5.1 Life expectancy


Sandflies have a relatively long generation time, taking at least a month to
develop to adulthood (Killick-Kendrick, 1999). A key epidemiological
parameter is the adult life expectancy, since to transmit infection female
sandflies need to survive long enough to take at least two blood meals
and to allow parasites to develop to infectivity (the extrinsic incubation
period). The length of the gonotrophic cycle varies between species, being
around 3 days in Lu. longipalpis, around a week for P. ariasi, 5 days for P.
argentipes and 5e11 days for P. papatasi (Srinivasan and Panicker, 1992;
Lainson et al., 1977; Palit et al., 1990).
There have been few studies of the life expectancy of adult sandflies of
any species under natural conditions. Adult life expectancy can be
approximated from field measures of physiological age using the parous
rate. The proportion of parous flies (flies that have laid eggs) approximates
the proportion of flies that are older than one gonotrophic cycle. This
method has been used to estimate the life expectancy of P. ariasi in France
as around 6 days, that of Lu. longipalpis in Brazil as 2.4 days, and that of P.
papatasi in India as 14 days (Dye et al., 1987, 1991; Srinivasan and Panicker,
1992). The parous rate of Lu. evansi in Colombia was comparable to that of
Lu. longipalpis in Brazil (Travi et al., 1996), while that of Lu. longipalpis in
Colombia was higher (Ferro et al., 1995), suggesting an average life expec-
tancy of up to a week. Markereleaseerecapture methods can also be used
to directly estimate life expectancy of sandflies, such as the L. braziliensis
vector Lu. (Nyssomyia) neivai (Casanova et al., 2009). However, for VL vec-
tors their use has been limited to studies of dispersal, which have shown
incidentally that adult Lu. longipalpis can survive for up to 7e8 days (Dye
et al., 1991; Morrison et al., 1993a), and P. ariasi for up to 28 days
(Killick-Kendrick and Rioux, 2002).
Progress in the Mathematical Modelling of Visceral Leishmaniasis 63

1.5.2 Seasonality of sandflies


There is a pronounced seasonal variation in numbers of VL vectors, both in
the Old and New World. In tropical regions, sandflies are usually present
every month, but with up to a 10-fold variation in sandfly numbers across
the year. In the ISC, numbers are lowest in December and January, when
temperatures are lowest, and typically highest in summer and the postmon-
soon period; numbers correlate positively with temperature, but correlations
with rainfall are more variable (Ghosh et al., 1999; Picado et al., 2010a). In
Amazonian Brazil, sandfly numbers peak at the end of the dry season, and
this variation in population size is reflected in a marked variation in the inci-
dence of canine infection across the year (Quinnell et al., 1997; Kelly et al.,
1997). The seasonal pattern in other areas of Brazil is different, with highest
numbers in the more arid NE Brazil typically occurring during or after the
rainy season (Sherlock, 1996; Deane and Deane, 1962), although some areas
show no clear seasonal pattern (Costa et al., 2013). In the Mediterranean area
seasonality is more marked, with no adults present during the winter months
and sandflies surviving this period as larvae in diapause. In Italy the period of
sandfly activity runs from May to October/November, with highest sandfly
numbers from June to August (Rossi et al., 2008). In mountainous north-
west Iran, the VL transmission season lasts for 4e5 months in summer, as
also observed in the Mediterranean (Gavgani et al., 2002). Seasonal patterns
are also often reflected in the parous rate, infection rate and other important
entomological parameters.

1.5.3 Feeding behaviour


The sandflies that transmit VL have fairly broad host choice. Lu. longipalpis
will readily feed on a range of vertebrate hosts, and host choice is determined
largely by host size and accessibility (Quinnell et al., 1992; Morrison et al.,
1993b). Lu. longipalpis is not considered to be strongly anthropophilic, but
this may vary across its range, as members of the Lu. longipalpis species com-
plex differ in their degree of attraction to human odours in the laboratory
(Rebollar-Tellez et al., 2006). In the Mediterranean, both P. perniciosus
and P. ariasi are similarly opportunistic feeders on mammalian and avian
hosts (Guy et al., 1984; de Colmenares et al., 1995). P. argentipes and P. ori-
entalis also feed on a range of mammals, both humans and domestic animals,
particularly cattle, buffalo and goats (Garlapati et al., 2012; Gebresilassie
et al., 2015; Palit et al., 2006), although P. papatasi appears to be predomi-
nantly anthropophagic (Palit et al., 2006). This lack of host specificity means
that domestic animals probably play an important role in the epidemiology
64 K.S. Rock et al.

either as reservoirs and/or as sources of blood. In the ISC, large numbers of


sandflies are found peridomestically in cattle sheds, while in Brazil, large
numbers are often found in animal shelters such as chicken sheds. Since
neither cattle nor chickens are hosts of Leishmania, this raises the possibility
that domestic animals may reduce the infection rate in humans by diverting
sandflies from humans or dogs and lowering both the feeding and infection
rate (zooprophylaxis) (Alexander et al., 2002; Bern et al., 2010). These ef-
fects could be magnified since both Lu. longipalpis and P. argentipes are
thought to have a lek-based mating system, where males aggregate on hosts
and attract females using pheromones (Dye et al., 1991; Lane et al., 1990).
Conversely, the presence of domestic animals could increase the biting
rate by acting as a maintenance host for the vector and thus increasing vector
numbers (Alexander et al., 2002; Bern et al., 2010) (zoopotentiation); both
positive and negative associations between cattle and risk of infection have
been reported in the ISC (Bern et al., 2010).

1.6 Control of human VL


1.6.1 Treatment of VL
The pentavalent antimonials [sodium stibogluconate (SSG) and meglumine
antimoniate], the mainstay treatment for VL since the 1940s, are generally
effective, with the notable exception of parts of south Asia where up to
60% treatment failure in Bihar, India (Mondal et al., 2010b; Sundar,
2001; Sundar et al., 2000) and neighbouring Nepal (Rijal et al., 2007,
2003) have been reported and associated with parasite resistance (Lira
et al., 1999; Downing et al., 2011). Antimony treatment requires 30-day
hospitalization for daily injections imposing a catastrophic economic burden
on affected families (Ozaki et al., 2011). Antimony is highly toxic, particu-
larly in malnourished and very young or old patients, hence diagnostic test
specificity is of particular importance. In the past two decades of drug
development (reviewed by den Boer et al., 2009), conventional forms of
amphotericin B, liposomal preparations of amphotericin B, miltefosine
(MIL), and paromomycin (PM) have been registered in various countries
to treat VL. Conventional amphotericin B has replaced antimonials as the
first-line treatment for VL in parts of India (Olliaro et al., 2005), following
trials showing that it is highly effective and safe (Sundar et al., 2003, 2004).
Paromomycin (formerly known as aminosidine) is an aminoglycoside
antibiotic with good antileishmanial activity, showing high levels of efficacy
and safety, and low relative cost, and was registered in India in 2006. MIL is
the only oral drug against VL registered for use in India since 2002 and has
Progress in the Mathematical Modelling of Visceral Leishmaniasis 65

been adopted by the Indian Ministry of Health in parts of Bihar (Matlashew-


ski et al., 2011) and more recently in Nepal. However, it requires 28 days of
oral treatment and as it is teratogenic it cannot be used to treat pregnant or
lactating women, which hinders its use in roll-out programmes. Reported
treatment failure rates in immunocompetent patients are up to 21% (and
33% among children) within 12 months in Nepal and 5e10% within
6 months in India, the former attributed to a low-dose response drug expo-
sure rather than in vitro parasite resistance (Dorlo et al., 2014; Sundar et al.,
2012; Rijal et al., 2013). The 2005 Memorandum of Understanding for a
‘VL Elimination Programme’ in the ISC was partially based on replacement
of antimony by MIL as the first-line treatment (Mondal et al., 2009). Single-
dose liposomal amphotericin B (AmBisome) showing significantly dimin-
ished renal toxicity is now considered the best available therapy against
L. donovani in the ISC (WHO, 2010a), at a negotiated affordable price
(Matlashewski et al., 2011; de Melo and Fortaleza, 2013), and has recently
been adopted in some regions of the ISC (Burza et al., 2014).
In east Africa, monotherapies have proven less efficacious and with
greater regional variability than seen in south Asia (den Boer et al., 2009;
Mueller et al., 2007; Khalil et al., 2014). Since AmBisome requires a
cold chain and is very expensive with limited availability, it is considered
a second-line treatment, but under continued evaluation (Khalil et al.,
2014). One concern with monotherapies is the potential development of
parasite resistance, particularly MIL and PM, because they require long treat-
ment courses, lowering compliance. MIL resistance is reported to result from
a single-point mutation in the MIL transporter at the Leishmania plasma
membrane, associated with reduced accumulation of cell reactive oxygen
which causes apoptotic death (Seifert et al., 2007; Das et al., 2013). Parasite
resistance against amphotericin B is thought to be less likely (Lachaud et al.,
2009). The safety, tolerance and effectiveness of combination therapies
(reviewed by van Griensven et al., 2010) now trialled in India and east Africa
(see also Musa et al., 2012; Sundar et al., 2008; Chunge et al., 1990; Thakur
et al., 2000; Melaku et al., 2007; Seaman et al., 1993) are encouraging, and
expected to help slow the rate of parasite resistance to treatment, to reduce
treatment duration, to enhance compliance, and to be cost-effective (Musa
et al., 2012; Hamad et al., 2010; Croft et al., 2006). Following further effi-
cacy trials (Khalil et al., 2014; Musa et al., 2012), the World Health Orga-
nization (WHO) recommended 17-day treatment with PM combined with
SSG as a first-line treatment against VL in east Africa (Khalil et al., 2014). In
Brazil, amphotericin B and liposomal amphotericin B are currently
66 K.S. Rock et al.

recommended by the Ministry of Health with multicentre clinical trials un-


derway (de Melo and Fortaleza, 2013). However, no current or proposed
treatment would be suitable for mass drug administration, which emphasizes
the importance of parallel development of accurate and operational
diagnostics.

1.6.2 Parasite loads in response to treatment


In relation to transmission, the choice of an appropriate VL treatment
regime to ensure parasite clearance and to eliminate potential infectivity
to sandflies is essential. Leishmania burdens in treated VL patients generally
decline during treatment (reviewed by Kip et al., 2014; see also Pourabbas
et al., 2013; Mour~ao et al., 2014; Mary et al., 2004, 2006; Aoun et al.,
2009; Sudarshan et al., 2011). However, inadequate drugs, doses and dura-
tions may result in parasite persistence (e.g., <50% and <60% tissue parasite
clearance, respectively, immediately after, and 6 months after, completion of
low dose PM treatment in Sudan (Hailu et al., 2010)). Relapses with a
resulting rise in parasite loads are not uncommon (e.g. Pourabbas et al.,
2013; Mour~ao et al., 2014), and even in KA patients considered clinically
cured without recorded relapse, residual parasites may persist, e.g. in 11/
21 Iranian patients who were treated with antimony (glucantime)
(Pourabbas et al., 2013); in 2/6 patients who were treated with sodium
antimony gluconate (SAG) (Verma et al., 2010); and in 4/48 Brazilian
children who were under various treatment regimes (Mour~ao et al., 2014).

1.6.3 Vector control


Indoor residual spraying (IRS). Vector control by IRS using dichlorodi-
phenyltrichloroethane (DDT) or pyrethroid insecticides is a key component
of national VL control campaigns, with mixed success in reducing infection
incidence via purposeful IRS against Leishmania vectors, or as a benefit of the
spraying campaigns against malaria vectors (Alexander and Maroli, 2003;
Ostyn et al., 2008). IRS is effective against adult vectors that blood-feed in-
doors (endophagic) or rest indoors between blood-feeding and searching for
an oviposition site (endophilic). IRS is not expected to have an impact on
exophilic and exophagic vectors, or immature stages, although sandfly
breeding sites are largely unknown (Alexander and Maroli, 2003). Measur-
able entomological indicators of control success include reductions in hu-
man landing rates, blood-feeding success, or human blood index, and
reduced vector survival (lower parous rates), as demonstrated for a wide
range of New World (Davies et al., 2000; Feliciangeli et al., 2003a, b;
Progress in the Mathematical Modelling of Visceral Leishmaniasis 67

Courtenay et al., 2007; Falcao et al., 1991) and Old World sandfly species
(Benzerroung et al., 1992; Reyburn et al., 2000; Joshi and Rai, 1994;
Kaul et al., 1994; Mukhopadhyay et al., 1996; Joshi et al., 2009). Insecticide
deterrency (sometimes called repellency cf. excito-repellency) induces vec-
tors to avoid contact with insecticide-treated surfaces, thus potentially
diverting infectious bites onto unprotected hosts; however, this is not a usual
feature of current insecticide products for vector control.
The only research-based cluster randomized trial to evaluate IRS was
conducted in India, Bangladesh and Nepal, though interventions varied
in insecticides used (DDT in India, deltamethrin in Bangladesh, and
alpha-cypermethrin in Nepal) and in the wall surfaces treated (Joshi
et al., 2009). House and cattle shed spraying reduced the indoor P. argen-
tipes overall density by 72%, an effect consistent across sites, albeit with
mixed results reported in other studies in India (Kumar et al., 2009; Picado
et al., 2010b; Dinesh et al., 2008a). Since 1991e1992, the Indian Kala-azar
Control Program has promoted two IRS interventions per year using
DDT; in Nepal IRS operational programmes use predominantly
lambda-cyhalothrin (Joshi et al., 2003), whereas in Bangladesh IRS was
very sporadic or nonexistent up to 2010 (Bern and Chowdhury, 2006;
Mondal et al., 2008). In Brazil, reactive IRS using lambda-cyhalothrin,
cypermethrin or deltamethrin is conducted on detection of a KA case
(da Sa ude Secretaria de Vigil^ancia em Saude Departamento de Vigil^ancia
Epidemiol ogica, 2006). IRS government-run campaigns against VL vec-
tors appear to be absent in east Africa.
Despite some encouraging entomological research outcomes, IRS as
deployed under the umbrella of operational VL control programmes has
had little to no impact on VL incidence, in India (Picado et al., 2012;
Ostyn et al., 2008), Nepal (Joshi et al., 2006), Bangladesh (Bern and
Chowdhury, 2006) or Brazil (de Souza et al., 2008). However, robust
IRS evaluation studies are limited (Quinnell et al., 2001; Picado et al.,
2012), though guidelines to monitor IRS in the context of the VL elimina-
tion programme are now in place (WHO, 2010b).
In India, resistance of P. argentipes to DDT and pyrethroids is patchy
(Ostyn et al., 2008; Dinesh et al., 2008b). The Indian and Nepalese routine
IRS activities are considered suboptimal due to financial and logistical con-
straints, poor quality spraying practices and low community compliance
(Chowdhury et al., 2011; Picado et al., 2012; Coleman et al., 2015). In
Brazil, spraying lambda-cyhalothrin in animal sheds where the vector Lu.
longipalpis congregates was shown to increase the numbers of Lu. longipalpis
68 K.S. Rock et al.

in household dining huts (attached to houses), indicating the importance of


complete coverage (Kelly et al., 1997).
Insecticide-impregnated nets and materials. Insecticide-treated
nets (ITNs) treated with synthetic pyrethroids have been shown to be
effective against mosquito vectors and malaria incidence (Lengeler, 2004).
ITNs against many species of sandflies similarly demonstrate knock-down
and toxicity (reviewed by Alexander and Maroli, 2003; Ostyn et al.,
2008), and have been shown to be efficacious in reducing the incidence
of both anthroponotic CL and zoonotic CL (Jalouk et al., 2007;
Alten et al., 2003; Reyburn et al., 2000; Moosa-Kazemi et al., 2007; Davies
et al., 2000; Kroeger et al., 2002; Yaghoobi-Ershadi et al., 2006; Tayeh
et al., 1997).
In eastern Sudan, lambda-cyhalothrin impregnated bednets against the
vector P. orientalis reduced the incidence of KA to 1.6% in a single interven-
tion village compared to 12.4% in the control village, but lacked replication
(Elnaiem, 1996). Provision of deltamethrin impregnated bednets to 114
follow-up Sudanese villages in response to the 1999e2001 VL epidemic,
reduced VL by 59%; the greatest impact was reported to be 17e20 months
after the distribution of the nets (Ritmeijer et al., 2007). An intervention trial
against Phlebotomus argentipes in Bangladesh indicated that widespread bednet
impregnation with slow-release insecticide may reduce the frequency of VL,
however, the study was inconclusive because it was not randomized and
lacked sufficient replicates (Mondal et al., 2013).
While few ITN trials have been conducted against VL vectors, a notable
exception is the cluster randomized controlled trial known as the ‘Kalanet
Trial’ (www.kalanet.info) which distributed deltamethrin impregnated
long-lasting bednets to 26 paired intervention and control villages in highly
endemic foci in India (16 pairs) and Nepal (10 pairs) (Picado et al., 2010b).
The control arm continued existing practices (irregular IRS, use of untreated
nets, and treatment of VL), which were accounted for in the final analyses.
After 24 months follow-up, there was no significant reduction observed in
seroconversion in intervention (347/6372 ¼ 5.4%) compared to control
(345/6319 ¼ 5.5%) residents, or in the risk of developing KA (37/
9829 ¼ 0.38%) versus controls (40/9981 ¼ 0.40%). However, the interven-
tion did significantly reduce the risk of malaria (relative risk ¼ 0.46 [95%
CI 0.28e0.77]). The authors suggest that possible reasons for the failure
to impact on VL transmission are that P. argentipes may be more exophilic
(lives outdoors) and exophagic (feeds outdoors), and zoophagic (prefers
feeding on animals) than previously considered as the effect of the
Progress in the Mathematical Modelling of Visceral Leishmaniasis 69

intervention on vector density was measured only indoors (Picado et al.,


2015). Thus, the entomological impact of bednets on sandflies depends
largely on both host and vector behaviours (Courtenay et al., 2007); further
issues about feasibility are discussed elsewhere (Courtenay et al., 2007; Ostyn
et al., 2008; Picado et al., 2015).
High coverage with ITNs may result in reduced vector infection rates
(i.e. a mass effect) in anthroponotic cycles but this has not been empirically
demonstrated. The role of ITNs in combating zoonotic transmission is
limited: for anthroponotic VL, ITNs are expected to provide personal pro-
tection against infective bites, and theoretically to impact on the transmission
cycle. For zoonotic organisms (where humans are not reservoirs) ITNs will
protect and impact on indoor vectors, but are not expected to directly
impact on the transmission cycle.

1.7 Control of ZVL


In areas where human cases are comparatively rare, such as the Mediterra-
nean region, ZVL is considered as a veterinary rather than a human public
health problem, and there are typically no national measures to control
infection. In other areas, such as Brazil, national ZVL programmes are in
place. In Brazil, the national control programme is focused on (1) detection
and culling of seropositive dogs, (2) indoor residual spraying (IRS) of houses
and animal shelters and (3) diagnosis and treatment of human cases.
These methods have been successfully used in China to control infection,
although the relative contributions of vector and dog control are not known
(Zhi-Biao, 1989; Costa, 2011). In Brazil such measures have shown little or
no efficacy, as the number of human and canine cases have not declined. As a
result, the rationale for these methods, especially dog culling, has been
increasingly questioned on theoretical, practical and ethical grounds
(Otranto and Dantas-Torres, 2013; Dye, 1996; Costa, 2011; Courtenay
et al., 2002b). A number of logistical problems have been identified with
the dog culling programme: coverage is often low, the serological diagnostic
tests used are relatively insensitive and there are long delays between sam-
pling and culling (reviewed by Quinnell and Courtenay, 2009). Culled
dogs are also rapidly replaced with naïve, uninfected dogs (Nunes et al.,
2008). These concerns have been partly addressed by the recent introduc-
tion of a new rapid diagnostic test, which should reduce delays and may
improve sensitivity. Unfortunately, the efficacy of either IRS or dog culling
remains unknown, due to a lack of well-designed, large field trials (Romero
and Boelaert, 2010; Quinnell and Courtenay, 2009).
70 K.S. Rock et al.

Other potential control methods include protection of dogs with topical


insecticides, protection of humans with bednets (see Section 1.6.3), phero-
mone traps for sandflies, dog treatment or canine vaccines. Synthetic pyre-
throids applied to dogs as topical formulations or in impregnated collars
have been shown to be highly effective in protecting dogs against sandflies,
with effects lasting for up to 8 months for collars but only 3e4 weeks for
topical ‘pour-on’ applications. Under field conditions, randomized controlled
trials have shown such methods also to be highly effective in protecting
individual dogs against infection, with efficacies of 50e100% (reviewed by
Quinnell and Courtenay, 2009; Otranto and Dantas-Torres, 2013). Howev-
er, only one randomized, community-level control trial against human infec-
tion has been carried out, showing deltamethrin-impregnated collars fitted to
dogs in NW Iran to reduce canine infection by 53% and by 38% in humans
(Gavgani et al., 2002). Lu. longipalpis males produce a pheromone that attracts
females to lekking sites on hosts, and field trials have shown that lures using
synthetic pheromone are highly attractive to both male and female sandflies
and could potentially be used in control programmes (Bray et al., 2014).
Treatment of dogs with antiparasitic drugs is widely used in a veteri-
nary context in Europe, but it is not used in control programmes. Indeed,
treatment of dogs with drugs used to combat human infection is banned
in Brazil to reduce the possibility of drug resistance. The clinical response
of dogs to treatment varies according to the initial condition of the dog,
but clinical cure is often seen in dogs with mild to moderate infection
(Solano-Gallego et al., 2009). However, treatment does not result in
parasitological cure, and dogs are likely to remain infectious, albeit at a
reduced level. In contrast to drugs, canine vaccines may provide a very use-
ful tool for disease control, and much effort has been invested in the
development of a canine ZVL vaccine. Three vaccines are now commer-
cially available: Leishmune and Leish-Tec in Brazil, and CaniLeish in
Europe. Relatively little has been published on the efficacy of these vac-
cines, particularly from phase III field trials, and a number of published
studies have methodological issues (Wylie et al., 2014). There is evidence
from several studies that Leishmune is effective against clinical disease, and
maybe also against infection, with reported efficacies of 76e100% (Oliveira
da Silva et al., 2001; Borja-Cabrera et al., 2002; Nogueira et al., 2005; Felix
de Lima et al., 2010). A single field study of CaniLeish shows no significant
effect on infection, but a 68% reduction in the proportion of dogs
with clinical disease (Oliva et al., 2014). There have been no published
phase III studies yet for Leish-Tec. An antidisease (as opposed to
Progress in the Mathematical Modelling of Visceral Leishmaniasis 71

anti-infection) vaccine would be suitable, provided that the vaccine


reduced parasite loads to a level below that at which parasite transmission
occurs. It has also been suggested that Leishmune may have specific trans-
mission-blocking effects (Saraiva et al., 2006). Vaccines may also have
immunotherapeutic effects when given to already infected animals
(Borja-Cabrera et al., 2004).

2. MATHEMATICAL MODELS OF VISCERAL


LEISHMANIASIS
The number of mathematical models of VL in either humans or animals
is low, in particular, for VL on the Indian subcontinent (see Table 5) where
the majority of elimination efforts are currently being targeted. The existing
models are quite varied both in the biological details that they capture, but
also in their underlying mathematical structure; however, it is noted that
they are mostly still ordinary differential equations (ODE) models which
ignore spatial effects. As might be expected with models of a neglected trop-
ical disease (NTD), many lack realistic parameterization that arises through
model fitting to high-quality data. Several of the models do address pertinent
questions related to the control of disease through intervention such as detec-
tion and treatment and vector control, but none directly address the WHO
goal of elimination of disease as a public health problem by 2020 or generate
the likely time to full elimination given the current prevalence of disease and
existing/proposed control strategies. These existing models of VL pave the
way for more sophisticated models of this disease that utilize and extend prior
frameworks.
The two forms of visceral leishmaniasis, caused by L. donovani and L.
infantum, are separated both geographically and by the way in which
they are modelled. L. donovani on the Indian subcontinent is considered
to be anthroponotic with only humans and sandflies playing a role in the
transmission cycle. Consequently models of KA in the ISC do not feature
nonhuman populations explicitly, although it is recognized that the pres-
ence and relative density of alternative hosts may ultimately affect the hu-
man biting rate. ZVL (caused by L. infantum) primarily affects Brazil in
addition to the Mediterranean region and parts of central Asia and China.
In these regions an animal population (dogs) not only forms part of the
transmission cycle, but it is also thought to drive it. Table 2 lists parameters
associated with VL progression in dogs. In regions of east Africa affected by
72 K.S. Rock et al.

Table 1 Parameters associated with human VL progression and their notation used
here and across modelling papers
ELmojtaba
Dye and et al. Mubayi Stauch et al.
Wolpert (2010a,b, et al. (2011, 2012,
Parameter Notation (1988) 2012) (2010) 2014)

Total birth/ (a þ e)N LH m hN h aHNH


recruitment rate
Death rate mH b mh mh mH
Rate of infection lH CI/Na lh lh lH
(ROI)
Rate of becoming sH - - g hb gHP þ fHSgHD
symptomatic
Probability of PKDL f - (1  s) - p2 þ p1p4c
after KA
Rate of developing xH - - - dHL
PKDL after
recovery
Disease-induced dH d d - mK
mortality
Waning immunity uH - - - rHC
Treatment rate (1st) gH - a1 a hb s1
Treatment rate (2nd) - - - - s2
Treatment rate hH - Bd - s3
(PKDL)
Probability of p - - p -
nonprivate
treatment
(1st) treatment- - - - - mT1
induced mortality
(2nd) treatment- - - - - mT2
induced mortality
All rates are per capita unless stated otherwise.
a
C is the vectorial capacity; see Section 2.3 for this derivation.
b
Erlang distributed.
c
Either with or without second treatment and conditional upon survival.
d
Spontaneous PKDL recovery is given at a rate a2.

L. donovani and in the ISC, other animals have been shown to be infected
although their contribution towards transmission remains untested.

2.1 Infection in humans


In order to understand the transmission dynamics of VL it is important to be
able to identify the natural progression of the disease in humans. From this
biological starting point, a mathematical model can be developed.
Table 2 Parameters associated with VL progression in dogs and their notation here and across modelling papers
Ben- Palatnik-
Hasibeder Salah Quinnell Courtenay de Sousa Ribas
Dye et al. et al. Dye et al. et al. Reithinger et al. et al. et al.
Parameter Notation (1988) (1992) (1994) (1996) (1997) (2002b) (2004) (2004) (2013) Costa et al. (2013)
h i
Total recruitment b DN D const const A1 bDD const const DN d þ a ðKD
K

bD D mdNd const
rate
Death rate mD b dD m1 d d d d d md mb
CI bFA I b1 Y2 a CD I CI CI a CD I a ½qa ðIa þ Da Þ
Rate of infection lD ðNM Þ ND N1 D NðNþab0=mÞ D abmFT D badmdS3 N
(ROI) þ qb ðIs þ Ds Þ
Inverse of latent sD l sA s1 s s s s s dd i
period
Disease-induced dD d dA a - a - dI - ad a
mortality
Waning immunity - - - - - - - r - sd -
Prob of ‘never- q - BB
BA þBB - 1  aD - 1a - 1  aD - -
infectious’
Prob of - - - - - - - - - - p
asymptomatic
Treatment rate (1st) gD - - - r - - - r ud -
Reduction in biting - - - - - - - annual - q -
(collar)
Cull rate - - - - k - pL,pI,pQb annual k x fu
Vaccination rate - - - - u - - - - Nc -
All rates are per capita unless stated otherwise.
a
Hostevector model. All others are based on vectorial capacity assumption for rate of infection (host-only model).
b
Annual cull of seropositive: latent, infectious and noninfectious dogs, respectively.
c
Transmission blocking vaccine.
74
Table 3 Parameters associated with VL in vectors and their notation here and across modelling papers where they are explicitly included
ELmojtaba
Hasibeder Ben-Salah Reithinger Mejhed et al. Mubayi Stauch et al. Hartemink Ribas
et al. et al. et al. et al. (2010a,b, et al. (2011, 2012, et al. et al.
Parameter Notation (1992) (1994) (2004) (2009) 2012) (2010) 2014) (2011) (2013)

Total birth rate mVNV dFNF A2(t) mFN varies LV - aFNF - ms(s2 þ s3)
Death rate mV dF m2 mF m2 mv mv mF msf ms
Total biting rate a - - a 2a 2a C b a ah þ ad
Human biting aa - - a(1  p) a a - ba - ah
rate
Dog/animal a(1  a) - - ap a a - b(1  a) h
a hþx ad
biting rate
Susceptibility to pV bAFa b2a c b1 c bvh pF1,.,pF4e c c
infection
mean EIP s1
V LF d sd sd - - g1
v
e
s1
F
c
EIPc sd
Ratio of vectors m ¼ NH NF/ND
NV
N2/N1 m H; R
V Vf n m NF/NH v/h mh,mdf
to hosts
EIP, extrinsic incubation period.
All rates are per capita unless stated otherwise.
a
Includes biting rate.
b
Dependent on status of infective human (i.e. symptomatic/asymptomatic etc.).
c
Exponentially distributed EIP.
d
Fixed duration EIP.

K.S. Rock et al.


e
Gamma distributed (n ¼ 25) EIP.
f
On humans and dogs respectively.
Progress in the Mathematical Modelling of Visceral Leishmaniasis 75

Table 4 Competing diagnostic and treatment strategies analysed by Boelaert et al.


(1999) in Sudan
Strategy Description

A Treat all clinical suspects


B Perform parasitological diagnosis on clinical suspects and treat if positive
C Perform DAT test on clinical suspects and treat if high titre. If
borderline DAT results then parasitological diagnosis and treatment if
positive.
D Perform DAT test on clinical suspects and treat if high titre
DAT, Direct Agglutination Test for Visceral Leishmaniasis.

The first model of VL in humans by Dye and Wolpert (1988) was a sim-
ple compartmental model in which individuals in the human population
were considered to be either susceptible to VL infection, infected (and infec-
tious) or have experienced the disease and were recovered and immune
from reinfection. This basic structure has been well studied for many diseases
and is known as an susceptible, - infected, - recovered (SIR) progression
model (see Fig. 1) (Keeling and Rohani, 2008). Despite its relative
simplicity, such a model can be analysed easily and provides clues towards
understanding observed phenomena. For example, Dye and Wolpert
were able to show that if VL is considered to have an immunizing effect,
this alone could generate the pronounced interepidemic periods seen histor-
ically (1880e1940) in India. Other factors such as treatment, disasters
(including earthquakes) and other diseases such as influenza governed the
size, but not the timing of these epidemics.
Although biological understanding of VL has advanced since 1988, it was
22 years after this original model that an improved dynamic transmission
model was developed to examine contemporary VL in the ISC. The model
by Mubayi et al. (2010) sought to quantify the underreporting bias of KA
cases in Bihar, India, by combining a transmission model with data (from
2003 to 2005) from the most affected 21 of 38 districts in Bihar. In addition
to the SIR compartments, the model includes stages for latent infection
where individuals have been infected with Leishmania but are not yet infec-
tious to sandflies; and treatment compartments where diseased patients
attend either public or private centres before recovery is possible (see
Fig. 2). This is a type of SEIR (susceptible, exposed, infected, recovered)
model. The proportion of those attending public facilities is essential to
answering the amount of underreporting as only public health centres
were required to report KA cases. Within the model, any privately treated
patients are assumed to be unreported. In 2006e2009 a retrospective study
Interventions Assumptions

Dog collar
Dog culling

Region
Dog vaccination
Dog treatment
Vector control
Human vaccine
Human treatment
Seasonality
Spatial aspects
Asympt. dogs
Humans
PKDL
Asympt. humans
Host-only
Assam Deterministic
Dyeand Wolpert (1988)
Sudan Boelaert et al. (1999)
Bihar Mubayi et al. (2010)
Meheus et al. (2010)
Stauch et al. (2011) A
Bihar Lee et al. (2012)
Bihar Stauch et al. (2012) A
Anthroponotic

Stauch et al. (2014) A


Ethiopia Miller et al. (2014)
Bangladesh Chapman et al. (2015)
Medley et al. (2016)
France Dye (1988)
- Hasibeder et al. (1992)

XX = ISC, XX = Brazil,
- Ben-Salah et al. (1994)
a /Malta Dye (1996)
- Williamsand Dye (1997)
Quinnell et al. (1997)
- Burattini et al. (1998)
Courtenay et al. (2002b)
= dead-end hosts,

Reithinger et al. (2004)


Natal Palatnik-de-Sousa et al. (2004)
Zoonotic

Morocco Mejhed et al. (2009)


Sudan ELmojtaba et al. (2010a) B
Table 5 Summary of mathematical modelling articles of VL extended from Rock et al. (2015a)

- ELmojtaba et al. (2010b) B


France Hartemink et al. (2011)
- ELmojtaba et al. (2012) B
Ribas et al. (2013)
= implicitly included in other terms

Costa et al. (2013)

K.S. Rock et al. 76


Progress in the Mathematical Modelling of Visceral Leishmaniasis 77

Figure 1 Simple model of SIR VL progression in humans as used by Dye and Wolpert
(1988) with progression between stages at rates lH and gH. Here the red box denotes
the only infectious stage.

Figure 2 Model of VL progression considered by Mubayi et al. (2010) including Erlang


distributed latent, infectious and treatment stages with purple denoting the infective
classes and yellow, the treatment in either public or private health centres.

indicated that the proportion of cases receiving treatment for VL in govern-


ment health facilities was low (less than 30% for adults, and as low as 13% for
5e10 year olds (Gidwani et al., 2011)) indicating that there could be a large
amount of underreporting.
To include more realistic latent, infectious and treatment periods,
Mubayi et al. (2010) make use of the ‘linear-chain trick’, a mathematical
technique where each of these compartments is partitioned into subcom-
partments. The use of this method generates Erlang-distributed (gamma-
distributed with shape parameter, k ˛ ℕ) periods rather than exponentially
distributed periods (MacDonald, 1978). The number of subcompartments
is dictated by the shape of the probability density function for that epidemi-
ological period; for example, Mubayi et al. used n ¼ 16 compartments for
the latent stage but only m ¼ 2 for the infectious stage because the variance
was much higher for the duration of the infectious period compared to the
variance of the latent period. Although this methodology provides a useful
mathematical tool for modelling the variability in the latent and infectious
periods, the subcompartments do not correspond directly to biological
processes.
Under this model the percentage of underreporting is 100(1  p) where
p is the proportion of patients who attend public health centres. The mean
number of reported cases per month is the influx into the first disease
78 K.S. Rock et al.

compartment of the model assuming that all new cases are immediately re-
ported from public health centres but never from private ones,
reported cases ¼ pmgH Im (1)
This equation may be rearranged to find p based on the endemic equi-
librium prevalence (denoted by stars),
reported cases

mgH Im
where Im can be expressed as a function of model parameters.
Using this methodology it was found that there was on average 88%
underreporting of infection in 2003 which decreased to 71% in 2005.
Underreporting reached as high as 96% (in Jahanabad district) and a positive
correlation was found between underreporting and population density
(Mubayi et al., 2010). This modelling demonstrates the huge extent of
underreporting that has likely occurred in the past due to differences in
reporting between health facilities and shows that caution must be taken
with standard reported incidence.
These first two dynamic models of VL progression in humans were
tailored to address specific questions; the first regarding interepidemic pe-
riods and the second addressing underreporting. However, there are other
aspects of L. donovani infection which are absent from these models, most
notably the inclusion of asymptomatic infections and PKDL.
The first transmission model to include this disease state was by
ELmojtaba et al. (2010a) who modelled infection in Sudan; this model
was later extended (ELmojtaba et al., 2010b, 2012). The compartmental
model was a typical SIR-type model with an extra infected (and infectious)
compartment for PKDL patients (see Fig. 3). In this framework, infected pa-
tients who are treated either fully recover (with probability (1  f )) or

Figure 3 Model of VL progression adopted by ELmojtaba et al. (2010a,b, 2012) with


both KA (IH) and PKDL (PH) classes and red denoting the infective classes. Following
KA, patients either develop PKDL with probability f or recover immediately.
Progress in the Mathematical Modelling of Visceral Leishmaniasis 79

Figure 4 Model of VL progression with both KA (IH) and PKDL (PH) classes with interme-
diate dormancy (DH). Red denotes the infective classes. The probability of progression
to PKDL from KA is the same as before (Fig. 3); however, the time from treatment to
PKDL onset is on average x1.

developed PKDL (with probability f ) following KA. The authors assumed


that PKDL was linked to effectiveness of treatment and so (1  f ) is propor-
tional to effectiveness. PKDL patients are assumed to be treated or to spon-
taneously recover (with rate hH) and once again become noninfectious.
By creating this extra compartment, the authors were able to analyse the
effect of treatment for both KA and PKDL, including both rate and effec-
tiveness (this is discussed further in Section 4). As this model was developed
for Sudan, it may be reasonable to model instantaneous transition from KA
infection to PKDL due to the relatively short time (0e6 months) between
the two, however, this can be amended to incorporate a dormancy period
between the two symptomatic stages (see Fig. 4) and would make the model
structure more appropriate for the ISC where the onset of PKDL can be
3 years after KA treatment.
The dynamic model (and variants thereof) by Stauch et al. (2011, 2012,
2014) includes a dormancy period between KA and PKDL, however, they
also add many more compartments to the human progression including clas-
ses for first- and second-line treatment and asymptomatic infection.
Conceptually, the Stauch et al. progression is not dissimilar to the approach
illustrated in Fig. 4, with an additional asymptomatic class occurring before
KA symptoms, possible spontaneous recovery from asymptomatic infection,
and waning immunity from the recovered class to susceptible. These addi-
tions are shown in Fig. 5.
Using this model of disease progression, Stauch et al. then create further
subcompartments which are tied to diagnostic test results, symptoms and
treatment. For example, asymptomatically infected people are now stratified
into two separate classes, denoted here by AH1 and AH2. The first of these
groups of people have positive PCR but negative DAT results,
whereas the second group tests positive for both. KA patients also have
PCR/DAT positive tests, however, the presence of symptoms is used to
80 K.S. Rock et al.

Figure 5 Model of VL progression with asymptomatic VL (AH), KA (IH) and PKDL (PH)
classes with intermediate dormancy (DH). Red denotes the infective classes.

distinguish them from asymptomatics, and the treatment they are receiving
(no treatment, first-line and second-line) assigns their subcompartment.
Fig. 6 shows how the new compartments are amended from the assumed
progression (Fig. 5) and the results of diagnostic tests.

Figure 6 Compartmental diagram showing VL progression paths in humans (adapted


from Stauch, A., Sarkar, R.R., Picado, A., Ostyn, B., Sundar, S., Rijal, S., Boelaert, M., Dujardin,
J.-C., Duerr, H.-P., 2011. Visceral leishmaniasis in the Indian subcontinent: modelling epide-
miology and control. PLoS Neglected Trop. Dis. 5, e1405.). Underneath their correspond-
ing results for three different tests for VL are shown (as given by Stauch, A., Sarkar, R.R.,
Picado, A., Ostyn, B., Sundar, S., Rijal, S., Boelaert, M., Dujardin, J.-C., Duerr, H.-P., 2011.
Visceral leishmaniasis in the Indian subcontinent: modelling epidemiology and control.
PLoS Neglected Trop. Dis. 5, e1405; Stauch, A., Duerr, H.-P., Dujardin, J.-C., Vanaerschot,
M., Sundar, S., Eichner, M., 2012. Treatment of visceral leishmaniasis: model-based analyses
on the spread of antimony-resistant L. donovani in Bihar, India. PLoS Neglected Trop. Dis. 6,
e1973; Stauch, A., Duerr, H.-P., Picado, A., Ostyn, B., Sundar, S., Rijal, S., Boelaert, M.,
Dujardin, J.-C., Eichner, M., 2014. Model-based investigations of different vector-related
intervention strategies to eliminate visceral leishmaniasis on the Indian subcontinent.
PLoS Neglected Trop. Dis. 8, e2810.). Blue boxes denote individuals who do not play a
role in transmission of disease to the vector. The red boxes denote those that do or
might (with asymptomatics contributing much less [40- to 80-fold less] than KA or
PKDL patients).
Progress in the Mathematical Modelling of Visceral Leishmaniasis 81

This model was developed from and fitted to cross-sectional data from
the Kalanet study in India and Nepal (see Section 1.6.3). The model does
not take into account the sensitivity or specificity of the three diagnostic tests
used in the trial. Unlike other models of VL, this model allows for waning
immunity, which is the loss of immunity to Leishmania infection over a
period of time. Through fitting to the data, the average time spent immune
(in the RH compartment) was just 307 days, which is shorter than might be
expected. The model also included the impact of HIV coinfection upon the
dynamics of VL, with a greater proportion of infected people developing
symptoms if they were HIV-positive compared to HIV-negative. As coin-
fection with these two pathogens was not part of the Kalanet study, the HIV
prevalence had to be taken from regional estimates.
Through tying human VL progression to results of diagnostic results,
Stauch et al. have raised an important question: how should asymptomati-
cally infected individuals (who are PCR positive) be modelled within the
transmission cycle (see Fig. 7)? Such people are infected with Leishmania par-
asites but their role in infection of sandflies is uncertain (see Section 1.3.3). In
the model of Stauch et al. ‘late-stage’ asymptomatics were assumed to have a
reduced infectivity by some factor, called here ε, compared to symptomat-
ically infected, KA individuals (who have infectivity pV towards sandflies).
‘Early stage’ asymptomatics were assumed to have even a lower infectivity
equal to 0.5ε, although the justification for this is not clear. Through model
fitting the reduced infectivity parameter, ε, was estimated to be 1/40 and so

Figure 7 Uncertainties in the human infectious reservoir. Here all individuals who may
contribute towards infection in the vector are shown in the large purple box. ‘Infectious’,
IH individuals are always assumed to be infective to sandflies and ‘dormant’, DH (before
relapsing to PKDL) individuals are thought to never be infective but do still have low
parasitaemia.
82 K.S. Rock et al.

this would indicate that asymptomatics are 40- to 80-fold less infectious than
KA or PKDL patients.
The model results suggest that even at this low infectivity, asymptomatic
individuals are a large enough group to sustain transmission and, therefore,
that treating KA and PKDL patients alone is unlikely to be sufficient to
reduce transmission enough to achieve elimination as a public health prob-
lem. This is the only dynamic model to examine asymptomatically infected
humans. It points toward a need to fit models that include asymptomatic in-
dividuals to other data sets, but also emphasizes the need for xenodiagnostics.
Ultimately, even if the individual infectivity of asymptomatics is low, this
should not be dismissed, as if there are many such individuals, it may hamper
intervention efforts.
Chapman et al. (2015) took an alternative approach to previous modellers
to infer both rates of disease progression and durations of disease states by
developing a multistage Markov model of the natural history of VL infection.
Utilizing data from a longitudinal study in Fulbaria Upazila of Bangladesh,
they concluded that disease durations were longer than those estimated by
Stauch et al. including a waning immunity of 1110 days (95% CI 988e
1247). Chapman et al. (2015) also found that progression from asymptomatic
infection to KA was high (14.7%) compared to Stauch et al. (0.33%).
Given these long duration estimates and high rates of progression to KA, it
would be expected that a dynamic transmission model similar to that of Stauch
et al. may produce different results when parameterized with the updated
values. In particular, asymptomatics may not be responsible for maintenance
of transmission as found by Stauch et al. (2011), however, as asymptomatic
infection is found to progress more often to KA, understanding the dynamics
of this subclinical, infected class of individuals will remain important.
Different models have made very different assumptions both about dis-
ease progression and parameterization; these differences and the notation
used in the original models are highlighted in Table 1. The huge variability
in parameters and their values may be attributed both to the underlying
model structures and also the uncertainty in many factors in transmission
and progression, such as rates or probabilities, which are either hard to mea-
sure or vary greatly by region. Consequently this will impact on the cross
applicability of results.
The mathematical modelling of L. infantum infection in humans is
limited to SIR-type models, where people are only considered to be
‘spill-over’ hosts from a main transmission cycle between sandflies and
dogs. As PKDL does not occur for this species of parasite, this human disease
Progress in the Mathematical Modelling of Visceral Leishmaniasis 83

progression is a reasonable assumption. In many endemic settings where L.


infantum is found, the prevalence in dogs is much greater than in humans,
and so little transmission is assumed to occur from humans.

2.2 Infection in dogs


Some models take a standard SI (Mejhed et al., 2009), SEI (Ben-Salah.,
1994; Reithinger et al., 2004) or SEIR (Ribas et al., 2013) compartmental
progression for dogs; however, the first model of VL in dogs (Dye, 1988)
takes the approach in which some dogs have an innate resistance which is
fixed at birth. These dogs, which will be referred to here as ‘never-
infectious’ (to use terminology coined later by Courtenay et al., 2002b),
play no role in the transmission cycle (see Fig. 8). This approach is also taken
by Dye (1996) and Palatnik-de Sousa et al. (2004). Table 2 summarizes
parameters associated with VL progression in dogs.
Hasibeder et al. (1992) assume a very similar progress to that of Dye (1988)
(Fig. 8); however, the model is subtly different by assuming that
never-infectious dogs can become seropositive, yet remain noninfective to
sandflies (see Fig. 9). These dogs may self-cure unlike their infective counter-
parts, returning to their original status of uninfected and never infectious. This
underlying model structure makes no difference to the disease dynamics
compared to Dye’s original model; however, including this extra infected
dog class may help to link seroprevalence data better to the model. This
model structure is also used later by Courtenay et al. (2002b).
As with human infection, infected dogs can exhibit a range of clinical signs
from completely asymptomatic to severe, fatal infection. Current diagnostics
through sera or blood eluates are able to detect latent and asymptomatic in-
fections (although with reduced sensitivity compared to infectious dogs
(Courtenay et al., 2002b)). Current tests cannot reliably identify infectious
dogs, but both the antiparasite IgG antibody level and the parasite load

Figure 8 Model of VL progression in dogs used by Dye (1988) with SD,ED,ID being the
susceptible, latent and infectious dogs, respectively, and the ‘never-infectious’ popula-
tion is denoted here by RD.
84 K.S. Rock et al.

Figure 9 Model of VL progression in dogs used by Hasibeder et al. (1992) with ‘never-
infectious’ population show as the bottom line. All dogs shown in the green circle are
infected and assumed seropositive; however, only dogs in the red box, ID, are infective
to sandflies.

have been shown to be highly correlated with probability of infection in sand-


flies during xenodiagnostic studies (Courtenay et al., 2002b, 2014).
The most complex structure for ZVL is given by Costa et al. (2013) in
their model which not only accounts for possible infective asymptomatic
dogs but also for imperfect screening due to diagnostic tools via the use of ex-
tra compartments (see Fig. 10). This modelling approach is more complex
than its predecessors, with model equations corresponding to this progression
allowing for delay between diagnosis and culling (discussed in Section 4).
The original notation and key aspects of each of these canine VL models
is summarised in Table 2.

Figure 10 Model of VL progression in dogs used by Costa et al. (2013). Here all dogs are
born susceptible but only some will develop symptoms after becoming infected and
passing through the incubation period. Dogs are screened at a rate r with sensitivity
d, and correctly categorized as diagnosed with symptomatic and asymptomatic infec-
tion, DS,DA, respectively. Due to imperfect specificity dZ of diagnostic tools, some nonin-
fected dogs are incorrectly diagnosed DZ. All dogs within the green ellipse test
seropositive. All red boxes correspond to infective dogs, although asymptomatic
dogs are around threefold less infective to sandflies than those which are symptomatic.
Progress in the Mathematical Modelling of Visceral Leishmaniasis 85

2.3 The role of sandflies


Sandflies are key in the transmission cycle but little is known about some
important factors. In particular with models of vector-borne disease, it is
necessary to be able to estimate certain parameters in order to mechanisti-
cally model the role of the vector as part of the transmission cycle. These
factors include vector density, life expectancy and relative biting rates on
both humans and animals (Table 3).
There are two different methods that are typically used to model
host-sandfly disease dynamics, both based on the RosseMacdonald model
(Ross, 1916; Macdonald, 1957), with the second method being a
simplification of the first. As is standard within vector-borne disease
modelling for the first type of RosseMacdonald model, both host and
vector equations are explicitly given. Several VL models (see
Table 3) use this explicit formulation with vector equations described by
the ODEs,
dSV
¼ BV  lV SV  mV SV
dt
dEV
¼ lV SV  sV EV  mV EV (2)
dt
dIV
¼ sV EV  mV IV
dt
which use an exponentially distributed extrinsic incubation period (EIP)
with mean duration 1/sV. Alternatively the delay ODEs,
dSV
¼ BV  lV SV  mV SV
dt
dEV
¼ lV SV  b
lV bS V expðmV sÞ  mV EV (3)
dt
dIV b
¼ lV b
S V expðmV sÞ  mV IV
dt
can be used where the caret denotes the variable evaluated s days ago, i.e.
b
S V ¼ SV ðt  sÞ. This formulation has a fixed duration EIP of s. In both
cases the rate of infection (ROI), often referred to as the force of infection,
towards vectors takes a similar form to,
ðIH þ PH Þ
lV ¼ apV (4)
NH
86 K.S. Rock et al.

for L. donovani in human populations where a is the biting rate per sandfly
on humans and pV is the infectivity towards the sandfly. With this form of
the ROI towards vectors, both symptomatic cases of VL and PKDL(IH and
PH, respectively) are assumed to contribute equally to infection in sandflies.
Parameter notation associated with vectors is found in Table 3.
In the case where PKDL does not occur (L. infantum) or for populations
of dogs (or other animals) the ROI is simply,
IH
lV ¼ apV (5)
NH
Other additions used by Stauch et al. (2011, 2012, 2014), including
infective asymptomatics and varying infectivity of HIV-positive individuals
are incorporated in a similar fashion but are not written here to retain a more
simple form. Amendments that can be made to consider the effect of addi-
tional animal populations are discussed later in Section 2.3.3.
The rates of infection for a ‘typical’ vector-borne disease are usually
given in the form,
IV
lH ¼ apH
NH
(6)
IH
lV ¼ apV
NH
towards hosts and vectors, respectively, for a system with just one host
involved in the transmission cycle. This formulation for lH arises through an
infected vector, IV, picking a single host out of NH available hosts and biting
at a rate a. The probability of the host being infected through a single in-
fectious bite is pH. The term lV is similar, with a single susceptible vector
picking an infected host from the total host population with probability
IH/NH at a rate a and the probability of infection for the vector is pV per bite
on an infected host.
However sometimes the ROI towards hosts and vectors is given by,
IV
lH ¼ bH
NV
(7)
IH
lV ¼ bV
NH
with the denominator in the host equation being NV instead of NH. This
form is used by Mubayi et al. (2010) in their hostevector model of VL in the
ISC and is one of the original variants of the RosseMacdonald model
Progress in the Mathematical Modelling of Visceral Leishmaniasis 87

(Smith et al., 2012). Despite the apparent difference in formulation, this


form is used if host and vector populations are assumed to be constant, and so
if bH ¼ apHNV/NH and bV ¼ apV, the systems (6) and (7) are equivalent.
A summary of the original notation used in VL models with explicit vec-
tor equations is given in Table 3.
The second type of model, used particularly in those articles pertaining to
canine VL, uses an approximation method to implicitly include vectors
without the need for separate vector equations. Using the fact that sandfly
life expectancy is vastly smaller than that of humans or even dogs
(mV [ mH,mD) it is possible to separate the vector and host time scales
with the vectors equilibrating much faster than hosts (Williams and Dye,
1997). This method takes one of the two forms above (either (2) or (3)) to
find the equilibrium number of infectious sandflies, and therefore a quasiequi-
librium (or pseudoequilibrium) approximation of the host infection dynamics.
The resulting approximation for the infectious vectors is,
lV
IV zNV  ℙðsurvive EIPÞ (8)
ðlV þ mV Þ
where the star denotes equilibrium, lV is as before (such as (4) or (5)) and the
probability of surviving the EIP is given by,
8
> 1 no incubation period
>
>
>
>
>
> expðmV sÞ
>
> fixed duration EIP
>
>
>
>
>
< sV
exponentially distributed EIP
ℙðsurvive EIPÞ ¼ sV þ mV
>
>
>
>  n
>
> nsV
>
>
>
> nsV þ mV
gamma distributed EIP
>
>
>
>
:
with shape parameter n
(9)
under various assumptions with the same mean duration used in the VL
modelling literature.
This may then be substituted into the ROI towards the host popula-
tion(s) by using,
IV
lH ¼ apH (10)
NH
88 K.S. Rock et al.

and so only host equations need to be explicitly written down.


In the case of VL, this approximation is almost always taken further by
considering the relative size of parameters in comparison to prevalence in
hosts, i.e. if mV [ lV. Under this new condition the equilibrium infection
in sandflies is given by,
lV
IV z NV  ℙðsurvive EIPÞ (11)
mV
In the case of endemic L. infantum infection in dogs in Brazil, dog infec-
tion prevalence is around 19%, with a sandfly biting rate of approximately
once every 3 days (Reithinger et al., 2004), sandfly life expectancy of
14 days (this is an upper estimate from the huge range of 2.4 (Dye, 1996)
to 16.7 days (Ribas et al., 2013)) and infectivity towards sandflies of
10.7% (Courtenay et al., 2002b). Under this parameterization
mV z 0.0714 per day and lV z 0.0068 per day and so mV is an order of
magnitude larger. Other less conservative estimates for these parameters in
this setting further increase the difference between these two values, making
the approximation quite reasonable. However, in some regions with higher
host prevalence or alternative parameterization this assumption would not
be appropriate. Care must also be taken to consider the expected impact
of control and seasonality on these approximations, which might make
them less appropriate.
This formulation is often favoured as it can be directly related to an epide-
miological measure, the vectorial capacity. The vectorial capacity is the num-
ber of new cases occurring from all the bites taken on a single infected person
on one day in an otherwise naïve population. It is usually denoted by,
NV a2 pH pV
C¼  ℙðsurvive EIPÞ (12)
m V NH
and so the ROI towards humans can be described in terms of the vectorial
capacity,
IV 
lH ¼ apH
NH
a2 NV pH pV IH
z  ℙðsurvive EIPÞ  (13)
m V NH NH
IH
¼C
NH
Progress in the Mathematical Modelling of Visceral Leishmaniasis 89

which leads to a host-only model with the same type of frequency-


dependent contact rate used in contagious disease modelling (Anderson and
May, 1992). Several models of VL (Dye and Wolpert, 1988; Dye, 1988,
1996; Courtenay et al., 2002b; Quinnell et al., 1997; Palatnik-de Sousa
et al., 2004; Costa et al., 2013) utilize this vectorial capacity quasiequilibrium
assumption.
Finally, as is often the case with cost-benefit analyses, some models
(Boelaert et al., 1999; Meheus et al., 2010; Lee et al., 2012) do not use a
dynamic, prevalence-dependent ROI and so the role of vectors is not
considered at all.

2.3.1 The basic reproduction number


Alongside the vectorial capacity, another epidemiological measure known as
the basic reproduction number, or basic reproduction ratio, R0, is often used
when comparing disease in different regions or examining elimination. Un-
like the vectorial capacity which is primarily related to the vector’s capacity
to transmit disease and therefore is necessarily restricted to vector-borne dis-
eases, R0 depends on parameters associated with all parts of the infection cy-
cle and is widely regarded as the most useful measure across a range of
diseases (Keeling and Rohani, 2008).
For contagious (directly-transmitted) disease in a homogeneously mix-
ing population, R0 summarizes the number of new cases occurring directly
from a single infected individual in a fully susceptible population without
intervention. However, for heterogeneous populations with structured
mixing or multiple host species the definition is slightly trickier (Rock
et al., 2014). In its formal mathematical derivation, R0 is defined using
the next generation matrix approach (Diekmann et al., 1990); this com-
putes the average number of secondary infections from a ‘typical’ infected
individual. Here a ‘typical’ individual may be a host or a vector and the sec-
ondary infections, or the next generation, for VL will always arise in the
opposite population as hosts can only infect vectors and vice versa. For
basic VL models this method is not too complex.
In the case where vector EIP is assumed to be exponentially distributed
(2) and there is an SEIR host progression, the next generation matrix, K, is
calculated by writing down the matrices of transmissions, T, and transitions,
S, associated with the variables EH, IH, EV and IV of infected and infectious
hosts and vectors.
90 K.S. Rock et al.

0 1
NH
B0 0 0 apH
B NH C
C
B C
B0 0 0 0 C
B C
T ¼B C (14)
B NV C
B0 apV 0 0 C
B NH C
@ A
0 0 0 0
0 1
ðsH þ mH Þ 0 0 0
B sH ðgH þ dH þ mH Þ 0 0 C
S¼B
@
C
0 0 ðsV þ mV Þ 0 A
0 0 sV mV
(15)
and the next generation matrix is given by,
K ¼ T S1 (16)
Finally R0NGM is the largest eigenvalue of the matrix, which is the spectral
radius of K,
R0NGM ¼ rðKÞ
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
NV a2 pH pV sV sH 1
¼
mV NH ðsV þ mV Þ ðsH þ mH Þ ðmH þ gH þ dH Þ
(17)
For VL that includes PKDL, alternative disease progressions or multiple
hosts this methodology can also be used. It can now be seen how R0 and the
vectorial capacity are linked,
 2 sH 1
R0NGM ¼ C (18)
ðsH þ mH Þ ðmH þ gH þ dH Þ
Confusingly, for vector-borne disease, the basic reproductive ratio is
often considered to be the expected number of new cases in the host popu-
lation generated by a single infected host (Smith et al., 2007). As this is two
infectious generations, rather than one, this expected number of cases is the
square of R0NGM given in equation (17). This version of the basic reproduc-
full
tive ratio, called R0 , links to a full host-to-host cycle and can also be
derived using the equations for the host-only model. Articles which use
Progress in the Mathematical Modelling of Visceral Leishmaniasis 91

full
this version include Dye (1996) and Quinnell et al. (1997) and arguably R0
could be considered to be a more intuitive metric as it has a relatively simple
biological interpretation.
Conversion between the two versions of the basic reproductive ratio is
trivial and they agree on the same critical threshold value of R0 ¼ 1, above
which a disease can invade a population and below which, it cannot.
Caution needs only be taken when using the value of R0 as metric to
compare either disease in different settings or between different models;
full
for example between that of Dye and Wolpert (1988) (R0 ¼ 2:13) and
full
Stauch et al. (2011) (R0NGM ¼ 3:94, which gives R0 ¼ 15:5) who both
model VL in the ISC.

2.3.2 Sandfly host choice


Vector choice of host can be important to the spread of vector-borne
disease. The presence of secondary hosts plays a key role in reducing
the biting pressure upon the primary species, however, if the secondary
hosts are also able to acquire and transmit the parasite they could become
a reservoir for infection. The concept of zooprophylaxis, where animals
which are not part of the transmission cycle are used to deflect biting
away from humans, has been suggested for other vector-borne diseases
including malaria and Chagas disease (WHO, 1982; Cruz-Pacheco
et al., 2012).
In the ISC where the disease is thought to be anthroponotic, it is reason-
able to expect that additional animal populations have protective effects for
the human population by moving the biting pressure away from humans.
However, the presence of animals close to the home may draw sandflies
to the area, resulting in higher vector densities and more risk. For the ISC
studies are conflicted about the protectiveness of cattle with some finding
significant protection and but others indicating an increased risk (Bern
et al., 2010). For L. infantum, dogs are a known reservoir of infection and
play a major role in transmission whereas chickens could have a zooprophy-
lactic role by shifting biting pressure away from humans and effectively
reducing the vector to host density.
Biting behaviour of sandflies (P. argentipes in the ISC and Lu. longipalpis
in the New World) seems to be primarily opportunistic; however, host
biomass is thought to play a role in host selection (Quinnell and Courtenay,
2009; Bern et al., 2010). In addition to relative humaneanimal and
humanevector densities, this sandfly host preference will impact upon
92 K.S. Rock et al.

transmission unless sandflies are motivated solely by convenience as has


been described for other vector-borne diseases (Rock et al., 2015b).
Despite this, not all modelling work integrates animal (or even multiple
animal) populations. Where animals are part of the transmission cycle, the
ROI towards the vector can be written,
aðIH þ PH Þ þ ð1  aÞIA
lV ¼ apV (19)
aNH þ ð1  aÞNA
where NA is the number of animals, IA is the number of infected animals,
and a is the relative propensity to feed on humans rather than animals. Stauch
et al. (2011, 2012, 2014) later on discount animal reservoirs by setting a ¼ 1,
but ELmojtaba et al. (2010a) use this type of ROI in their model analysis to
examine control via this reservoir (see Section 4.4). Even if animals play
no role, the same formulation can be used with a < 1 but IA ¼ 0 to model
the reduction in biting due to presence of animals (i.e. zooprophylatic
effects).

2.3.3 Seasonality of sandflies


Seasonality of sandfly populations is considered to be substantial, however, it
is not included in any models of VL in the ISC and only in a handful else-
where (Ben-Salah et al., 1994; Mejhed et al., 2009; Williams and Dye, 1997;
Quinnell et al., 1997). Some nonmodelling papers have sought to quantify
or qualitatively assess seasonal aspects of sandfly emergence or biting (Singh
et al., 2008; Rossi et al., 2008; Picado et al., 2010a) and a paper on cutaneous
leishmaniasis has incorporated detailed information about sandfly dynamics
into a transmission model (Bacaër and Guernaoui, 2006). Ben-Salah et al.
(1994) alter the hostevector transmission terms by means of a Boolean func-
tion so that transmission only occurs during the warmer 5 months of the year
and Mejhed et al. (2009) use seasonal data of sandfly population sizes from
Morocco to parameterize the model over the course of a year. The others,
Quinnell et al. (1997) and Williams and Dye (1997), describe how a host-
only model of VL may be adapted to include seasonality by using fluctuating
vectorial capacity.
In reality there is a complex relationship between vector mortality/
emergence, EIP, human biting rate and temperature (as discussed by
Hartemink et al., 2011), however, to facilitate the modelling process, sea-
sonality is usually associated with either change in vector emergence rate,
or instead by varying the transmission rate but keeping constant vector pop-
ulation size.
Progress in the Mathematical Modelling of Visceral Leishmaniasis 93

Fig. 11 shows how using a sinusoidal forcing function of the form


(1 þ a1cos(2pt)) on either the birth rate or the ROI affects the disease
about equilibrium for both host and vector populations. Although the
two types of forcing are fundamentally different the change in the resultant
dynamics and ROI between the two is slight (see Fig. 12B); as would be
expected, forcing vector and host ROI shifts the peak of infection a little
earlier than forcing vector births as there is a lag between the influx of
newly emerged vectors and them biting and becoming infected. The
most noticeable difference occurs within the vector dynamics; adjusting
the ROI in this way has no impact upon the vector population size, how-
ever, changing the birth rate does (see Fig. 11C). Consequently, while very
similar dynamics in the host population are seen in these two cases, the
prevalence measured in the vector population across the year shows a
marked difference, reaching very high prevalence just before the coldest
part of the year using forced births.
Under a forced birth rate R0NGM will be slightly underestimated (by
a1/8% where a1 is the amplitude of forcing in the function given above)
(c.f. Bacaër, 2007). Williams and Dye (1997) describe how this is related
to changes in incidence and how it will affect the net reproduction ratio
and its thresholds. In the typical case where per capita host death and recov-
ery rates are taken to be constant and forcing is on the vectorial capacity,
then the arithmetic mean dictates the overall dynamics and so the critical
threshold of R0 ¼ 1 still holds. Additionally, in the case of vector-borne dis-
eases including ZVL, knowledge of varying incidence (which can be
measured in the field) in the host population can be used to calculate R0
as average incidence corresponds to average vectorial capacity.
Seasonality in ROI can be critical in interpretation of surveillance data
and in model predictions of intervention outcomes. Repeated data on sand-
fly population size and infection prevalence are required to fit models that
include seasonality. The mechanistic functions suggested above could be
combined, and other mechanisms included. If seasonality is due to environ-
mental variables, such as rainfall and temperature, then there is no reason for
them to follow particular patterns, which further complicates data require-
ments and ability to forecast accurately.
Other options for season forcing include Boolean transmission where
transmission occurs only at certain times of the year and not in others, for
example the five warmest months out of the year (after Ben-Salah et al.,
1994). Fig. 12 demonstrates how altering the biting function in this way
compares to the previous sinusoidal biting function.
94 K.S. Rock et al.

(A) 1000
800

Hosts
600
400
200
0
0 1 2 3 4 5 6
Time

10000
8000
Vectors

6000
4000
2000
0
0 1 2 3 4 5 6
Time
Host and vector infection dynamics
(B) 0.014

0.012

0.01
FOI to hosts

0.008

0.006

0.004

0.002

0
0 1 2 3 4 5 6
Time

ROI towards humans


(C)
Total number of vectors

10000
8000
6000
4000
2000
0
0 1 2 3 4 5 6
Time
Percent prevalence in vectors

80

60

40

20

0
0 1 2 3 4 5 6
Time
ROI towards hosts
Figure 11 Simulations using no forcing and sinusoidal forcing on the birth rate and ROI
respectively. (A) Shows the expected change in level for susceptible (green) and
infected (red) host and vectors for a SEIR host/SEI vector model. (B) Shows how the
ROI will change under the three assumptions and (C) compares the population size
and per cent prevalence of infection in the vector population under forced births
and forced ROI.
Progress in the Mathematical Modelling of Visceral Leishmaniasis 95

(A)
1000
Sinusoidal bites
800 No forcing
Boolean bites

Hosts
600

400

200

0
0 1 2 3 4 5 6
Time
5000

4000
Vectors

3000

2000

1000

0
0 1 2 3 4 5 6
Time
Host and vector infection dynamics
(B)
0.018
Sinusoidal bites
No forcing
0.016 Boolean bites

0.014

0.012
FOI to hosts

0.01

0.008

0.006

0.004

0.002

0
0 1 2 3 4 5 6
Time
ROI towards hosts
Figure 12 Simulations using no forcing and sinusoidal/Boolean forcing on the
ROI respectively. (A) Shows the expected change in level for susceptible (green) and
infected (red) host and vectors for a SEIR host/SEI vector model. (B) Shows how the
ROI will change under the three assumptions.
96 K.S. Rock et al.

There is clearly a need for regular sampling to estimate vector parameters


across the seasons to better inform fluctuating transmission from sandflies.
Sampling frequency needs to be of the same order of magnitude as the gen-
eration time. Inadequate sampling frequency, e.g. annual, cannot discern
sandfly dynamics and could even give a distorted picture of average sandfly
prevalence or rate of infection towards host populations. Mejhed et al.
(2009) indicate that May/June and October are the two times of the year
when there is the highest values for R0 and so highest levels of transmission
in Marrakech, Morocco. Similarly, in the ISC, Picado et al. (2010a) argue
that IRS should be carried out in Feb/March and August/September based
on seasonal patterns observed in their sandfly density studies in India and
Nepal. However a study in Bihar, India which examined sandfly infection
as well as density found that while abundance was highest during August/
September, infection prevalence in sandflies was greatest in November/
December (Tiwary et al., 2013). Such results warrant further investigation
into the seasonality in all endemic regions, especially when planning the
timing of vector interventions such as IRS.

3. DATAeMODEL INTERACTIONS
3.1 Current data and initial models
In any model, individuals must have a defined infection state. For
compartmental models, these states are mutually exclusive, for example,
susceptible (uninfected) and infected. There are two general approaches
to fitting models to data. First, individuals are unambiguously assigned to
one compartment/state on the basis of diagnostic tests, and the numbers
of individuals in each diagnostic category are compared to model predic-
tion (e.g. Stauch et al., 2011, 2012, 2014; Chapman et al., 2015). Alterna-
tively, the model states can be related to population data through
diagnostic characteristics, i.e. the model can predict the prevalence of pos-
itives (true and false) for a given infection prevalence. The latter approach is
most often related to age groups (e.g. Rodriguez-Barraquer et al., 2014). In
either case, but especially the former, the information in the data, and
therefore the accuracy of parameter estimates, are greatly enhanced if the
data are longitudinal, i.e. multiple measurements from the same individuals
through time (e.g. Chapman et al., 2015). Time series data arise from
repeated measurements in the same population, but not necessarily the
Progress in the Mathematical Modelling of Visceral Leishmaniasis 97

same individuals; an example is data relating to cases in a defined popula-


tion over time. Age-related data are more informative than simple preva-
lence data, as age measures duration of exposure, and can serve as a proxy
for time. Having defined the states in the model that can be related to data,
the states have to be defined in terms of their duration and their infectious-
ness. These data can be inferred during a fitting process, or, better, esti-
mated independently from research data. As has already been discussed, a
major complication with Leishmania infection is that a precise, gold stan-
dard diagnostic does not exist. In a modelling context, a diagnostic is
good if it is accurate and matches well with compartments required in
the model. Of course, modellers will frequently choose compartments
that match with diagnostics, and Stauch et al. (2011, 2012, 2014) is one
such example. In contrast, diagnostics and vector measurement are more
advanced in malaria, which has inspired a plethora of modelling approaches
(Smith et al., 2014).
The dynamics of the vector population directly impact on the observed
VL transmission dynamics. However, several of the parameters associated
with the vector are hard to ascertain and consequently there are great uncer-
tainties in the parameterization of models, in particular those which explic-
itly model vector populations. The density of sandflies and their abundance
relative to both human and animal hosts are not only difficult to determine
but are known to vary greatly between different regions; for example, see
Hartemink et al. (2011) who used geographic sticky trap data to estimate
densities and map R0 in France. Likewise the seasonal fluctuations of sandfly
populations are hard to pin down and so are not featured in many transmis-
sion models of VL at present. Even something as fundamental as sandfly life
expectancy has large confidence intervals and while models typically use an
estimate of 2 weeks, they range from 2 days to an unlikely 2 months.
One way to make progress without this information is to create host-
only models which use the vectorial capacity, C, to infer the hostevector
component of transmission. Although this is useful for initial models, and
is consistent and efficient from a modelling view point, the value of C is still
reliant on estimating several vector parameters, and the usual assumption that
the quasiequilibrium assumption is valid. We are also missing accurate assess-
ment of the infectiousness of different host classes to sandflies, and again, to
make progress, simplifying assumptions have to be made, for example,
Stauch et al. (2011, 2012, 2014) use pV ¼ 1 for the probability of a sandfly
acquiring infection from an infected human. Of course, without
98 K.S. Rock et al.

information on sandfly biting rates, this assumption does not matter, as


doubling the sandfly population and halving pV will give the same results.
Parameter combinations such as R0 and C summarize both the model and
unknown data into a single number.
A feature of simple models of infectious disease is that apart from the
infection state, and perhaps age, all individuals are considered as equal rep-
resentatives of the average individual, and all are equally likely to transmit to
each other. However, there will be considerable heterogeneities in terms of,
for example, response to infection, and exposure to vectors. If this hetero-
geneity is great, then the concept of an average might be wrong (e.g.,
Dye, 1986; Smith et al., 2014). The consequence of mistakenly believing in-
dividuals are all essentially equal is usually to underestimate the control effort
required (i.e. to underestimate R0), as there are sections of the host-vector-
parasite population combinations that are transmitting at a much higher rate
than the rest. For Leishmania in particular, there is sufficient field evidence to
suggest that rates of transmission and disease vary significantly through both
time (Dye and Wolpert, 1988) and space (Bern et al., 2007; Woolhouse
et al., 1997), and between individuals (Courtenay et al., 2014) so that there
is unlikely to be a good ‘average’ situation. But without sufficient data to un-
derstand the important spatiotemporal scales, the correct scale of appropriate
spatiotemporal studies is unknown; for example, is annual assessment of a
village sufficient, or should studies be based on monthly assessment of
households in at least five villages? For malaria, spatiotemporal hotspots
have been identified at the household-month level (Bejon et al., 2014).
Broad estimates may still help in planning widespread interventions, but
more detailed information could help to target interventions in key foci at
optimal times.
As well as space and time, hosteparasiteevector populations can differ
by genetic, socioeconomic and environmental variables. The importance
of these dimensions is likely to be different at different times and places e
for example during an epidemic, the socioeconomic factors might change.
Unless the correct scales are known, the danger is that any one study gives
only a partial picture, and piecing the studies together is fraught with
difficulty due to, for example, difference between epidemiological stages
and variation in diagnostics. Additionally, highly variable diagnostics will
also potentially mask important heterogeneities, i.e. the noise will swamp
the signal.
Progress in the Mathematical Modelling of Visceral Leishmaniasis 99

3.2 Future data and alternative sources


Given the relative paucity of data that can be used to develop accurate
models, can we identify alternative sources of data, and define from current
models and understanding, what data ought to be collected?
Working with animal models of infection and disease has the consider-
able benefit that it is possible to conduct experiments (e.g. deliberate infec-
tion), but the disadvantage that the results might not transfer directly to
humans. Studying canine, zoonotic VL has the advantage that parameter es-
timates can directly inform models relevant to zoonotic human disease, as
well as, in the absence of direct data, informing anthroponotic transmission
models. While the quantitative parameter estimates are unlikely to be
directly transferable, the processes might be directly parallel to humans.
For example, infection rates of sandflies from xenodiagnosis experiments
in dogs have been shown to correspond to a threshold density of Leishmania
parasites in skin snips taken from ears (Courtenay et al., 2014); so while the
numerical values of transmission probabilities are for dogesandfly interac-
tions, it is likely that the same biological process occurs in humanesandfly
interactions. It is possible to perform xenodiagnosis directly in people, and
there are currently attempts to conduct such experiments (Miller et al.,
2014). Similarly, it is possible that the natural history of disease is related
to the size of the inoculating dose of infection (Bern et al., 2010).
Longitudinal and time series data will be critical to future model devel-
opment. The transmission dynamics at the population level (i.e. the pro-
cesses that determine the public health impact of VL) are derived from
the processes within individuals: infection, immunity, pathogenesis and
infectiousness. We need to know the natural history of infection, and partic-
ularly the performance of current and future diagnostics, over long periods
of time. This can come from cohort studies, but also from accurate case series
in defined locations, perhaps supplemented with cross-sectional data from
the same populations. To our knowledge, there is no data set available
from the ISC or elsewhere which has longitudinal measurements on
humans, and repeated measurements on vectors.

4. MODELLING INTERVENTIONS
4.1 Humans
4.1.1 Diagnostics
Diagnostic tools are a key part of the targeted control strategy for VL in or-
der to accurately diagnose and treat individuals and to monitor incidence
100 K.S. Rock et al.

and prevalence in the population. In addition to these primary uses, diag-


nostic tests for VL can also aid in developing and parameterizing mathemat-
ical models of disease; enabling individuals to be placed into one of the
mutually exclusive compartments.
For VL in humans some of the main diagnostic tools which can be used
to screen people are PCR, ELISA, DAT, LST and rK39 (a rapid diagnostic
test) as described in Section 1.3.2. While some study sites may use a combi-
nation of tools to determine presence of the parasite, antigens or antibodies,
often just a single test will be used to make a diagnosis. Given that mortality
is associated with clinical symptoms, only those displaying outwards signs of
infection are usually tested and subsequently treated.
Stauch et al. (2011, 2012, 2014) use three different diagnostics (PCR/
LST/DAT) to partition the human population under their model, however,
the model does not account for imperfect tests. Furthermore, while the ac-
curacy of tests such as PCR and DAT has been well analysed for detecting
KA, they have unknown sensitivity and specificity for asymptomatic
infection.
Using a cost-effectiveness analysis Boelaert et al. (1999) compared four
alternative diagnostic strategies to find the most effective (most deaths
averted compared to no intervention) and the most cost-effective (the
lowest cost per death averted). The strategies are outlined in Table 4 and
were based on treatment using antimonials. The authors found that for ex-
pected baseline estimates for prior probability in suspected clinical cases
(40%), that strategy A was most effective and B was worst. For lower prob-
abilities of infection the most effective strategy changed with A causing more
deaths (through possible treatment toxicity) than it averted. This result has
important implications: as prevalence reduces, moving forward to elimina-
tion, it suggests that strategies may have to change as the prevalence of dis-
ease in the population decreases. The most cost-effective strategy was B at a
cost of US$448 per death averted. This analysis was published in 1999 and
was based on the cost and accuracy and of drugs and diagnostics at the time.
With new diagnostic developments, lower incidences and newer drugs, the
available strategies and their (cost-)effectiveness need to be continually
reevaluated.
Recent work (Medley et al., 2016) highlights the importance of early
diagnosis and treatment of KA patients, if they are responsible for the major-
ity of transmission (i.e. if asymptomatics are not responsible for maintaining
transmission). It also potentially explains why current control programmes in
the ISC have been successful despite the apparent ineffectiveness of vector
Progress in the Mathematical Modelling of Visceral Leishmaniasis 101

control (Coleman et al., 2015). Results from this model suggest that a diag-
nostic test with only 50% sensitivity might result in a considerable reduction
(halving) of R0 and offer the potential for elimination. However, low
specificity of a diagnostic test (i.e. the false positive rate) would prevent it
being applied to individuals with nonspecific clinical symptoms. These
results suggest that a diagnostic test with high specificity, even at the expense
of relatively low sensitivity, is required for widespread intervention. As elim-
ination is approached, false positives will always become more significant,
and the specificity of tests more important. However, most consideration
of test characteristics during development concentrates on sensitivity.

4.1.2 Treatment
Treatment of KA plays an important part in control programmes, primarily
as the disease is usually fatal without it. However, as treatment reduces the
number of KA patients, it will also reduce the ROI towards the sandfly. The
shorter the time scale between onset of KA and successful treatment, the
smaller the infectious pool of people will be. However, this effect may be
small if asymptomatics are important in transmission (Stauch et al., 2011).
For Sudan, ELmojtaba et al. (2010a) examined the impact of first-line
treatment for KA; analysis varied both treatment rate and effectiveness of
preventing PKDL (i.e. f ). With high treatment alone there is a reduction
in symptomatic VL cases but since (1  f ) of these treatments produce a
PKDL case, infection rates were still high. Increasing both treatment and f
was unsurprisingly more effective. Unfortunately the PKDL treatment
rate was not varied to simulate extra treatment efforts with this class of
patients.
The model of Stauch et al. (2011, 2012, 2014) includes three treatment
types: first-line treatment of VL (at a rate s1), second-line treatment of VL
following first-line treatment failure (at a rate s2) and finally PKDL treat-
ment (at a rate s3). First-line treatment failure occurs with probability
p1 ¼ 5% and PKDL follows after treatment and remission with probability
p2 ¼ 3% from successful first-line treatments and p3 ¼ 3% from second-
line treatments (all conditional on surviving treatment with probability
(1  fT) z 0.95). Second-line treatment is always assumed successful upon
survival, in that individuals appear recovered at least for a time before
PKDL. The authors found that whilst treatment was clearly necessary
from an individual KA patient’s perspective and did decrease the prevalence
of KA, increasing effectiveness or time to treatment made a negligible differ-
ence to the underlying disease dynamics. If this is the case and the objective is
102 K.S. Rock et al.

to minimize occurrence of KA, then the benefit will only be gained while
treatment continues at a high level; once the treatment returns to the orig-
inal level, the prevalence will also bounce back. Stauch et al. concluded that
the limited effect of treatment alone on the incidence of KA was primarily
due to infective asymptomatics. In fact, under their parameterization, there
would still be over 15 KA cases per 10,000 in this endemic setting. The dif-
ference between the conclusion of Stauch et al. and that of Medley et al. is
entirely due to assumptions about the relative infectiousness of symptomatic
and asymptomatic cases. The apparent clustering of KA cases and recent suc-
cess of KA treatment in reducing VL in the ISC are both evidence for KA
cases being the predominant source of infection to sandflies. On the other
hand, asymptomatic cases are numerous and may be significantly infectious.
Only field data can resolve this difference.
A cost-effectiveness analysis by Meheus et al. (2010) compared the use of
different treatment options (either currently available or in Stage III trials) in
the ISC. This study differed from Boelaert et al. (1999) in geographic loca-
tion, but also Meheus et al. assumed patients were correctly identified prior
to treatment, whereas Boelaert et al. assumed all patients received the same
treatment. The 10 alternative treatment strategies included a range of mono-
and combination-therapies with each having associated pros and cons such as
the requirement for cold-chain storage, cost, length of treatment and
toxicity. The most cost-effective strategy was combination MIL and
PM treatment at a cost of US$92 per death averted. The authors found
that a possible future decrease in the cost of liposomal amphotericin B (L-
Amb) compared to MIL might change the most cost-effective strategy to
combination L-Amb and PM.

4.1.3 Human vaccine (hypothetical)


A prophylactic vaccine would reduce the human incidence by reducing the
numbers of susceptibles who are able to acquire Leishmania infection, or
reduce the proportion of those infected who develop disease. The former
vaccine provides protection to those who have been vaccinated, but also in-
direct protection to the nonvaccinated as there are fewer people infected.
The latter, disease preventing vaccine does not prevent the acquisition of
parasites, but alters the progression so that people do not develop KA.
This latter vaccine type is currently a promising candidate for use in the
near future; however asymptomatic individuals who do not develop KA
may still remain infectious, so there may possibly be less indirect protection
to the nonvaccinated.
Progress in the Mathematical Modelling of Visceral Leishmaniasis 103

Figure 13 The incremental cost-effectiveness ratio (ICER) compared to the per capita
gross domestic product (GDP) in Bihar. The diagram also shows the ranges given by
Lee et al. (2012) for ‘highly cost-effective’, ‘cost-effective’ and ‘not cost-effective’.

Lee et al. (2012) use a cost-effectiveness analysis to address the potential


impact of a human vaccine for VL. Unlike the other cost-effectiveness papers
which examine the cost-effectiveness per death averted (Boelaert et al., 1999;
Meheus et al., 2010), Lee et al. use DALYs avoided as their measure. The
incremental cost-effectiveness ratio (ICER) per DALY averted is compared
to the per capita gross domestic product (GDP) in Bihar of US$288 and is
designated ‘highly cost-effective’, ‘cost-effective’ or ‘not cost-effective’ (see
Fig. 13). In this analysis, the vaccine is assigned an efficacy, with 50% efficacy
meaning an individual has a probability of 0.5 of becoming immune to infec-
tion following vaccination. A successfully vaccinated individual, therefore,
will not only avoid KA but also possible PKDL.
Based on treatment with amphotericin B and an imperfect diagnostic test
(rK39), Lee et al. found that the (hypothetical) vaccine was cost-effective
when it cost US$100 or less per person and had at least 25% efficacy. If
the efficacy was increased to 50%, and the price was less than US$30,
then vaccination could even be cost-saving compared to treatment alone.
The analysis made use of a fixed ROI in simulations, as did Boelaert et al.
and Meheus et al. Such analyses will always underestimate the effectiveness
of a vaccine, as the model does not predict any indirect protection, i.e. they
assume that successfully vaccinating 50% of the population does not change
the risk in the nonvaccinated population (Edmunds et al., 1999). This
approach also ignores the longer-term dynamic impacts of vaccination and
the possibility of elimination. However, given the paucity of well-validated
dynamic models of VL, assuming a constant ROI is reasonable.

4.2 Vector
Interventions that target vectors generally change a combination of three
parameters: the biting rate per sandfly on the host population, a, vector
mortality, mV, and the number of vectors, NV. IRS is likely to increase
mortality and hence reduce the number of vectors but might not alter the
human biting rate of the remaining sandflies. The biting rate can be directly
104 K.S. Rock et al.

reduced by use of effective bed nets, which may be impregnated with long-
lasting insecticides to generate long-lasting insecticidal nets (LLINs).
Deltamethrin-impregnated collars (DMC) on dogs provide the dogs with
protection from sandfly bites via an excito-repellency effect while the insec-
ticidal effect kills sandflies. At present there are no proven effective methods
for reduction of sandfly populations by targeting nonadult stages. IRS,
LLINs and collars only likely impact on indoor populations but not on
the (outdoor) sandfly population as a whole; although this may be sufficient
to reduce biting pressure to humans.

4.2.1 In the ISC


While examining vector control was not the main goal of Stauch et al.
(2011), it was found that a reduction of 80% of vector density would be
needed in order to push R0 below one and achieve elimination. A second,
more detailed, study of vector interventions using the same model (Stauch
et al., 2014) considered the impact of different types of vector control on
sandfly dynamics. IRS and LLINs were both considered to kill vectors
directly, increasing their mortality rate, mV, which had a knock-on effect
by reducing vector density (NV/NH). Environmental vector management
(EVM), linked here to plastering of houses, was assumed to destroy available
breeding sites and therefore reduce the number, NV, of vectors.
Stauch et al. (2014) found that under the first strategy (increasing mor-
tality), elimination could be achieved if there was a reduction of 67% in
sandfly density, whereas if the direct approach (through EVM) was taken,
a higher reduction of 79% was needed. For the ISC in general, neither
LLIN nor EVM was found to be suitable interventions on their own, as field
estimates of expected density reduction were not high enough. LLIN in
Bangladesh could be viable with reductions in intradomestic sandfly density
reaching 70e80% (as measured by CDC light traps) (Joshi et al., 2009), but
in India and Nepal bed nets have not had such success. IRS was found to be
a feasible intervention strategy because it has been shown to reduce densities
by 72% (higher than the required 67%). The model also compared sole in-
terventions to a combined approach and found that, for example, a 42%
reduction via EVM (as seen in the field) would require a further 50% reduc-
tion via IRS to reach the critical threshold.

4.2.2 In Africa
ELmojtaba et al. (2010a) examine control of the vector population in their
model through reducing the biting rate corresponding to interventions
Progress in the Mathematical Modelling of Visceral Leishmaniasis 105

including bed nets or ensuring that clothes cover the body. ELmojtaba et al.
found that vector control was by far the most effective intervention
compared to various human treatment strategies (although the biting rate
was reduced for both humans and reservoirs) as this control reduces not
only the infectious pressure on susceptible humans but also the ROI on
the sandflies from all infective individuals (including PKDL). While this
model does not include asymptomatics, it is interesting to note that this logic
extends to them as well and it should be expected that bed nets work well at
reducing the contribution of infection from all humans, regardless of
whether their disease status is known/treated.

4.2.3 In Brazil
Dye (1996) and Ribas et al. (2013) used another similar approach to model
control for humans. Both incorporated (direct) insecticidal vector control
such as IRS by increasing vector mortality in their models. The results are
congruent; it was found that vector control has a large impact upon preva-
lence in humans and was by far the best control method out of the several
suggested. Even though Dye found that prevalence in the dog population
was lower under treatment than under quite modest insecticide interven-
tions, the reduction in the total number of bites taken caused by increasing
vector mortality was sufficient that insecticide was still the best option for
reducing transmission to humans where they are a spill-over host.
Ribas et al. (2013) examine the use of DMC in their model by altering
the ROIs to and from dogs,
aD IV aD ID
lD ¼ pD and lV ¼ pV (20)
q ND q ND
or under reparameterization,
IV ID
lD ¼ aD ð1  BÞpD and lV ¼ aD ð1  BÞpV (21)
ND ND
where 1q ¼ ð1  BÞ and 0  B  1.
Here B can be thought of as the proportional reduction in biting. Using
this model structure, Ribas et al. found that collaring was reasonably effec-
tive at preventing infection and could dramatically reduce prevalence in
humans if the protection offered by the collars were high.
Reithinger et al. (2004) took an alternative approach by adding an
additional compartment to their model (see Fig. 14). Here pulse collaring
of 80% of dogs was simulated either one or two times per year with DMC
106 K.S. Rock et al.

Figure 14 Model of collaring by Reithinger et al. (2004). The subscript CD represents


those dogs which have been collared, which occurs at rC ¼ 80% of the dog population
per year (in a single pulse). Collars loose effectiveness or are removed/lost at a contin-
uous rate of xC ¼ 0.001 per day. Collared dogs have 90% protection from sandfly bites.

losing effectiveness continuously over 8 months. The results were found to be


dependent upon the demography of the dog population with a lower average
life expectancy making collars a more effective strategy. Simulated DMC
effects did not vary greatly with change in endemicity. In contrast, baseline
prevalence levels impacted quite substantially upon culling success.
These DMC models examined a reduction in biting for dogs with collars
but did not consider the possible effects of either an associated increase in
mortality of sandflies from feeding attempts on collared dogs, or displaced
bites (and subsequently higher biting) on uncollared dogs. These two addi-
tional effects might be expected to decrease or increase the total prevalence
respectively. Insecticidal effects of collars would not stop transmission from
infected flies to susceptible dogs but would prevent dog to fly transmission as
sandflies would either not feed or die before completing the parasite’s EIP.

4.3 Dogs
4.3.1 Diagnostics
Courtenay et al. (2002b) examined the infectious states of a naturally
exposed cohort of dogs by xenodiagnosis to appropriately model the dog
population by partitioning into never/ever-infectious categories (used by
Hasibeder et al., 1992). They used the sensitivity/specificity of the different
diagnostic tools to simulate which dogs would be identified as seropositive
during screening. The authors found that insensitivity of the diagnostic test,
as well as delays between diagnosis and culling, could explain why a culling
intervention might fail. In a follow-on study, Courtenay et al. (2014) used
qPCR to quantify heterogeneities in L. infantum burdens in canine tissues to
relate to their infectiousness to sandflies, and demonstrated that qPCR could
identify the majority of infectious dogs and potentially distinguish highly
Progress in the Mathematical Modelling of Visceral Leishmaniasis 107

infectious from noninfectious animals. In their model they demonstrated


that a small proportion of dogs were responsible for the majority of transmis-
sion events.
Costa et al. (2013) use the sensitivity and specificity of diagnostics in their
model which includes infectious symptomatic and asymptomatic dogs (with
asymptomatic dogs assumed to be three fold less infective to sandflies than
those with symptoms), as described previously and visualized in Fig. 10. Un-
der the model, the sensitivity and specificity of the diagnostic were both
assumed to be within a range of 80e100% and were assumed to be the
same for symptomatic and asymptomatic dogs (which is unlikely to be the
case). The model results indicate that the presence of asymptomatic dogs
impacted on the control strategies and time scales negatively; the authors
concluded that models without asymptomatics might overestimate the
effectiveness of interventions reliant on diagnostic tools such as culling. In
higher prevalence areas the control of the asymptomatic dog population
was necessary to reduce transmission below the threshold level. In this
case high sensitivity tests would be needed to find the asymptomatic reser-
voir, while high specificity tests would reduce the numbers of dogs needed
to be culled and critically, reduce the unnecessary culling of false positives.
This indicates that the optimal strategy is likely to depend upon regional
prevalence. As screening is costly, it is likely that this would only be cost-
effective in higher prevalence areas.

4.3.2 Treatment
Unlike VL in humans, treatment of disease in dogs is not part of national control
programmes; nevertheless it is possible to treat dogs for infection with drugs.
Only two articles pertaining to dogs included the effect of treatment of infected
dogs in their models (Dye, 1996; Ribas et al., 2013). In both articles, the same
approach was taken, whereby a constant treatment rate was assumed, presum-
ably corresponding to ad hoc detection of infected dogs via their symptoms.
However Ribas et al. assume that dogs will become susceptible to disease
once more posttreatment rather than retaining immunity from re-infection
as assumed by Dye. This difference has marked implications for modelling re-
sults and explains the stark discrepancy in conclusions drawn by these authors;
Ribas et al. found that treatment of infected dogs had virtually no impact on
overall prevalence of infection (in humans) whereas Dye found it to be
moderately beneficial at reducing prevalence. In reality, dogs may relapse
to a diseased state, not through reinfection, but rather through recrudescence
of parasites posttreatment as treatment is seldom curative.
108 K.S. Rock et al.

4.3.3 Dog culling


Unlike the strategies to control VL in humans, there are alternative options for
intervention in animal populations. The first of these interventions is to cull
infected dogs in order to reduce the ROI towards the sandfly; this interven-
tion has been posed as a viable strategy in Brazil and has had seemingly
beneficial effects in China (when employed concurrently with insecticide)
(Zhi-Biao, 1989). Some authors (Dye, 1996; Palatnik-de Sousa et al., 2004;
Costa et al., 2013) implement culling in models in a similar way to which
treatment is modelled by Ribas et al. (2013), that is, a diagnosis of infection
is made either via symptoms or ongoing screening and culling will subse-
quently occur at some constant rate throughout the year with the removed
dogs being replaced with an influx of susceptible dogs (or puppies). In
contrast, other authors (Courtenay et al., 2002b; Reithinger et al., 2004)
opt to model a pulse culling campaign where dogs are annually removed after
mass screening (50% screening in Reithinger and 100% in Courtenay). Cour-
tenay et al. assume that removed dogs are ‘rapidly’ replaced by susceptible
ones, while Reithinger et al. use a dog recruitment method in which the pop-
ulation returns to equilibrium over time at a rate described by,
 
NV
m V NV þ b 1  NV (22)
K
where mV is the weighted average mortality rate from both natural death and
infection, b is the density-dependent growth rate of the population and K is
the carrying capacity of the population at equilibrium.
The second form of replacement is more realistic as an area could not
replace all their dogs instantly. As culled dogs will be replaced with suscep-
tible ones, unsurprisingly most authors conclude that culling is not likely to
be the best option available. Dye (1996) found it to be the worse option
among all proposed strategies by using a maximum cull rate of infectious
dogs of around 4% per month in his analysis. Costa et al. used a 4% screening
of symptomatic dogs per month (which would result in an even smaller pro-
portion being culled) with a 4-month delay before culling. They found that
while this was sufficient to reduce transmission to less than 1% in low trans-
mission settings, high-endemicity regions would need to cull a minimum of
30% of asymptomatically (but infectious) dogs to achieve the same results.
Costa et al. also emphasized how, if screening the whole dog population,
79% of those culled would have been removed unnecessarily using a diag-
nostic test with 80% specificity.
Progress in the Mathematical Modelling of Visceral Leishmaniasis 109

Palatnik-de-Souza et al. and Ribas et al. examine much higher rates of


culling assuming homogeneity in infectiousness with the former deter-
mining that 7% of infectious dogs must be removed per day to reduce R0
below the critical threshold of one, and the latter concluding that below
18% culling per day there would be virtually no impact on disease transmis-
sion to humans.
One group, Reithinger et al. (2004), found culling to be moderately
effective at reducing prevalence of infection in dogs. In particular, increasing
the percentage of infected dogs culled and/or the frequency of screening
campaigns (from one to two pulses per year), made a significant reduction
in prevalence. Culling had a higher relative effectiveness in low transmission
settings compared to high transmission areas.
In summary, culling can be made more effective by reducing time be-
tween diagnosis and culling, and by improving the sensitivity so that a higher
percentage of infectious dogs are identified. There needs to be a (perhaps
unrealistically) high screening coverage and rate to achieve substantial reduc-
tion in disease burden. Specificity of diagnostics must be increased if all dogs
are to be screened in order to reduce incorrect diagnosis and culling of
noninfectious dogs which in turn should increase dog owner compliance
(Costa et al., 2013; Courtenay et al., 2014). In particular it would useful
to have a tool to differentiate between infected and infectious dogs to
minimize the level of culling needed and target this control effectively
(Courtenay et al., 2002b). Culling of healthy uninfectious dogs should be
avoided from an ethical perspective.

4.3.4 Dog vaccination


Often when vaccination is considered in models it is assumed to be prophy-
lactic with individuals gaining immunity (or partial immunity) with success-
ful inoculation. Such a (fully) prophylactic vaccine model is used by Dye
(1996) to examine the potential effect of immunization of a dog population;
he found that at low levels of intervention, vaccination would be one of the
best strategies, although insecticides may more easily eliminate disease at
higher levels of intervention. Under Dye’s parameterization, around 90%
of the susceptible dog population would have to be successfully vaccinated
in order to achieve elimination.
Aside from prophylactic vaccines, other types of vaccine which have
been developed for other diseases could potentially provide an alternative
intervention for VL. One which has been modelled in a dog population is
110 K.S. Rock et al.

a transmission-blocking vaccine, in which a vaccinated dog may still acquire


infection, but upon becoming infected they will have a reduced infectivity
towards sandflies through reduced development of the parasite within the
vector (Ribas et al., 2013).
In the model by Ribas et al. (2013), the ROI towards the vector is
amended from,
ID
lV ¼ aD pV (23)
ND
to include this reduced infectivity,
pV ID
l V ¼ aD (24)
n ND
or under re-parameterization,
ID
lV ¼ aD ð1  T ÞpV (25)
ND
where 1n ¼ ð1  T Þ and 0  T  1.
Here, T might represent the vaccine efficacy, i.e. T ¼ 1 is 100% effective
at blocking transmission and T ¼ 0 is 0% effective given all the dog popu-
lation is vaccinated. Alternatively T could be the proportion of successfully
vaccinated dogs in the population. In the first case the infectivity parameter,
pV is essentially reduced by a factor of (1  T ), whereas in the latter, just a
proportion (1  T ) of dogs remain unvaccinated and since vaccination does
not prevent dogs becoming infected, the same proportion, (1  T ) are fully
infective to vectors while the rest are noninfective. Either of these results in
the same mathematical formulation of the ROI.
Under this formulation, the transmission-blocking vaccine was simulated
and found to be marginally better at reducing prevalence in humans (consid-
ered to be a spill-over host) than treatment of dogs; however, as the inter-
vention impacts only on infectivity of some blood meals and does not
directly affect sandfly feeding rates or mortality, the intervention did not
have a pronounced effect upon prevalence of infection in humans.
In the case of VL, transmission-blocking vaccines may also prevent dis-
ease in dogs (but not infection) as both symptoms and infectiousness are
correlated with parasite burden. Consequently there may be less of a distinc-
tion between prophylactic and transmission-blocking vaccines for ZVL.
Future canine vaccines may also have a curative effect if given to a diseased
dog; this effect has not been explicitly modelled.
Progress in the Mathematical Modelling of Visceral Leishmaniasis 111

4.4 Other animal hosts


ELmojtaba et al. (2010a) found that the presence of other animal popula-
tions which are able to acquire and transmit VL would increase the preva-
lence in humans, however, at present no other animals have been shown
to be reservoirs (Quinnell and Courtenay, 2009). The presence of additional
host populations and their impact on the primary population is nontrivial; in
large enough numbers the secondary host population removes biting pres-
sure away from the primary hosts (humans) and consequently this may in
fact reduce the ROI upon primary hosts. Conversely the additional popu-
lation(s) may drive infection in the primary hosts, or, in this case, stifle the
effects of control upon just the primary population through treatment of
infective VL humans. Under the parameterization of ELmojtaba et al.
(2010a) the removal of this secondary population was beneficial to humans
although the reduction in R0NGM was relatively paltry (3.4 down to 2.3) and
indicates that while infection is boosted by the animal reservoir, it is not sus-
tained by it.

5. CONCLUSIONS
Despite model predictions based on well-defined (average) parameter
estimates and scenarios, the operational realities on the ground are somewhat
different, adding a variable complexity that is difficult to capture. VL disease
in the ISC has been targeted for elimination as a public health problem, but
there are still many facets of this infection which are not well understood. In
particular, the progression of disease through different stages, including
asymptomatic infection and PKDL, is unclear, especially in relation to diag-
nosis and infectivity. Recent efforts to control VL in the ISC seem to have
had success, prompting the original goal of elimination as a public health
problem to be brought forward from 2020 to 2017 (WHO, 2013; Boni
et al., 2014). However, there are also indications that caution as well as opti-
mism should be exercised; modelling and transmission studies have indicated
a credible possibility that asymptomatically infected individuals might hinder
elimination efforts or obscure the true number of those with Leishmania
infection.
Treatment of people presenting with KA is central to the current inter-
ventions. Models disagree on whether this intervention alone is adequate to
achieve control and sufficiently reduce incidence. This disagreement hinges
on the relative proportion of transmission to sandflies due to symptomatic
112 K.S. Rock et al.

people who are then treated. If KA patients are responsible for the majority
of transmission, then treatment will have a big impact on onward transmis-
sion. The evidence for this situation is indirect and largely reliant on the
spatiotemporal clustering of KA cases and the long period cycles seen at
regional levels; both of these are more easily explained if KA cases are
responsible for transmission. However, if the majority of transmission is
due to asymptomatic infections, then treating KA patients promptly will
have little impact on transmission. There is less evidence for this situation,
but it is not unlikely if there are large numbers of asymptomatic infections
that transmit to sandflies at a low rate. This situation highlights the need
for more accurate data on the natural history and relative infectiousness of
different infection states.
Looking forward, model predictions of vector control consistently result
in effective control of disease, not just in the ISC but in regions affected by
ZVL too. IRS seems to have the maximal benefit among strategies including
bed net use and other EVM. However, it is emphasized that these modelling
results do depend critically on the infectivity of different disease states, the
effectiveness of IRS and the vector response to EVM. Again, there is a
paucity of data to support parameterization of these processes in any model.
With control of sandfly populations appearing pivotal in the elimination
campaign, it will be especially important to improve both our understanding
of sandfly biology and how transmission may change through the course of a
year. Future modelling of VL in humans could utilize the basic frameworks
set out by those who have previously modelled seasonality of sandflies in
ZVL as well as the wider literature on seasonal dynamics of other disease
vectors.
Modelling of VL in dogs has shown that the sensitivity and specificity of
diagnostics are likely to have a direct impact on the success of culling strate-
gies for dogs, and hints at the wider importance of accurate diagnostic tools
in reporting and controlling human disease. Caution must be taken with
modelled intervention comparisons because, in reality, some interventions
may be far easier to deploy (to the extent needed) than others. This is high-
lighted in the canine VL papers which have mixed suggestions for the best
intervention.
The current intervention programmes focus on eliminating VL disease as
a public health problem rather than considering achieving a break in trans-
mission. Elimination as a public health problem is technically described as
control in that continued efforts are required to maintain the lower inci-
dence. As the numbers of cases in the ISC falls towards the target of one
Progress in the Mathematical Modelling of Visceral Leishmaniasis 113

per 10,000, clinical awareness is likely to fall. Additionally, if there is popu-


lation (herd) immunity that currently reduces transmission (because a pro-
portion of the population are immune), then this too may wane over
time. As a consequence it is possible that there will be an increasingly suscep-
tible population with a less alert medical and public health system and
ongoing transmission, all within a matter of years. This is the situation in
which transmission can resurge and potentially large epidemics occur.
Therefore, it is important to understand transmission rather than focus solely
on diseased individuals to ensure that recrudescence does not occur.
There are several key knowledge and data gaps which have been iden-
tified within this chapter. There is a need to better understand the biology of
disease including disease progression (by determining markers to differen-
tiate self-clearing asymptomatic infection from those that progress to dis-
ease), infectivity of all infected individuals, and prospects for human
vaccine development. However, there is also a need to better inform models
with data; models need to consider time-dependent transients, which em-
phasizes the importance of longitudinal data. The important dimensions
of heterogeneity, including individual-level differences and larger-scale
spatial aspects, need to be incorporated in models if they are to capture
critical aspects of transmission dynamics.
Unsurprisingly, the number of models of VL is extremely low, especially
when compared to the vast modelling literature for other vector-borne dis-
ease such as malaria and dengue. However, the existing models do address a
range of important questions such as the level of underreporting of VL dis-
ease and relative impact of different interventions. Future models must be
fitted to high quality contemporary data sets in order to ensure the highest
level predictive value of the results and hopefully, through good communi-
cation between modellers and the larger VL community, the crucial
outstanding questions may be addressed.

ACKNOWLEDGEMENTS
The authors would like to thank Dr Lloyd A. C. Chapman for his comments on the manu-
script and Dr T. Deirdre Hollingsworth for her involvement with VL research under the Di-
agnostics Modelling Consortium and NTD Modelling Consortium. KSR was funded for this
work under the Diagnostics Modelling Consortium funded by the Bill & Melinda Gates
Foundation, and OC and RJQ acknowledge the long-standing support for field studies by
the Wellcome Trust. KSR, GFM and OC gratefully acknowledge funding by the Bill and
Melinda Gate Foundation in partnership with the Task Force for Global Health. The views,
opinions, assumptions or any other information set out in this article are solely those of the
authors.
114 K.S. Rock et al.

REFERENCES
Adams, E.R., Versteeg, I., Leeflang, M.M.G., 2013. Systematic review into diagnostics
for post-Kala-azar dermal leishmaniasis (PKDL). J. Trop. Med. 2013, 150746.
Ahluwalia, I.B., Bern, C., Costa, C.H.N., Akter, T., Chowdhury, R., Ali, M., Alam, D.,
Kenah, E., Amann, J., Islam, M., Wagatsuma, Y., Haque, R., Breiman, R.F.,
Maguire, J.H., 2003. Visceral leishmaniasis: consequences of a neglected disease in a Ban-
gladeshi community. Am. J. Trop. Med. Hyg. 69, 624e628.
Alam, M.S., Ghosh, D., Khan, G.M., Islam, M.F., Mondal, D., Itoh, M., Islam, N.,
Haque, R., June 2011. Survey of domestic cattle for anti-Leishmania antibodies and
Leishmania DNA in a visceral leishmaniasis endemic area of Bangladesh. BMC Vet.
Res. 7, 27.
Alexander, B., de Carvalho, R.L., McCallum, H., Pereira, M.H., 2002. Role of the domestic
chicken (Gallus gallus) in the epidemiology of urban visceral leishmaniasis in Brazil.
Emerg. Infect. Dis. 8, 1e6.
Alexander, B., Maroli, M., 2003. Control of phlebotomine sandflies. Med. Vet. Entomol. 17,
1e18.
Ali, M., Ashford, R.W., 1994. Visceral leishmaniasis in Ethiopia. IV. Prevalence, incidence
and relation of infection to disease in an endemic area. Ann. Trop. Med. Parasitol. 88,
289e293.
Alten, B., Caglar, S.S., Kaynas, S., Simsek, F.M., 2003. Evaluation of protective efficacy of
K-OTAB impregnated bednets for cutaneous leishmaniasis control in Southeast Anato-
lia-Turkey. J. Vector Ecol. 28, 53e64.
Alvar, J., Aparicio, P., Aseffa, A., den Boer, M., Canavate, C., Dedet, J.-P., Gradoni, L.,
Ter Horst, R., Lopez-Velez, R., Moreno, J., April 2008. The relationship
between leishmaniasis and AIDS: the second 10 years. Clin. Microbiol. Rev. 21,
334e359.
Alvar, J., Vélez, I.D., Bern, C., Herrero, M., Desjeux, P., Cano, J., Jannin, J.G., Boer, M.d.,
the WHO Leishmaniasis Control Team, May 2012. Leishmaniasis worldwide and global
estimates of its incidence. PLoS One 7, e35671.
Anderson, R.M., May, R.M., 1992. Infectious Diseases of Humans: Dynamics and Control.
Oxford University Press.
Andrade, M.S., Courtenay, O., Brito, M.E.F., Carvalho, F.G., Carvalho, A.W.S.,
Soares, F.C.S., Carvalho, S.M.S., Costa, P.L., Zampieri, R.A., Floeter-Winter, L.M.,
Shaw, J.J., Brandao-Filho, S.P., 2015. Infectiousness of sylvatic and synanthropic small
rodents implicates a multi-host reservoir of Leishmania (Viannia) braziliensis). PLoS
Neglected Trop. Dis. 9, e4137.
Anstead, G.M., Chandrasekar, B., Zhao, W., Yang, J., Perez, L.E., Melby, P.C., August
2001. Malnutrition alters the innate immune response and increases early visceralization
following Leishmania donovani infection. Infect. Immun. 69, 4709e4718.
Antoniou, M., Messaritakis, I., Christodoulou, V., Ascoksilaki, I., Kanavakis, N.,
Sutton, A.J., Carson, C., Courtenay, O., June 2009. Increasing incidence of zoonotic
visceral leishmaniasis on Crete, Greece. Emerg. Infect. Dis. 15, 932e934.
Aoun, K., Chouihi, E., Amri, F., Alaya, N.B., Raies, A., Mary, C., Bouratbine, A.,
December 2009. Contribution of quantitative real-time polymerase chain reaction to
follow-up of visceral leishmaniasis patients treated with meglumine antimoniate. Am.
J. Trop. Med. Hyg. 81, 1004e1006.
Bacaër, N., January 2007. Approximation of the basic reproduction number R0 for vector-
borne diseases with a periodic vector population. Bull. Math. Biol. 69, 1067e1091.
Bacaër, N., Guernaoui, S., July 2006. The epidemic threshold of vector-borne diseases with
seasonality. J. Math. Biol. 53, 421e436.
Progress in the Mathematical Modelling of Visceral Leishmaniasis 115

Badaro, R., Jones, T.C., Carvalho, E.M., Sampaio, D., Reed, S.G., Barral, A., Teixeira, R.,
Johnson, W.D., 1986a. New perspectives on a subclinical form of visceral leishmaniasis. J.
Infect. Dis. 154, 1003e1011.
Badaro, R., Jones, T.C., Teixeira, H.R.R., Johnson, W.D., 1986b. A prospective study of
visceral leishmaniasis in an endemic area of Brazil. J. Infect. Dis. 154, 639e649.
Baneth, G., Koutinas, A.F., Solano-Gallego, L., Bourdeau, P., Ferrer, L., July 2008. Canine
leishmaniosis e new concepts and insights on an expanding zoonosis: part one. Trends
Parasitol. 24, 324e330.
Barnett, P.G., Singh, S.P., Bern, C., Hightower, A.W., Sundar, S., 2005. Virgin soil: the
spread of visceral leishmaniasis into Uttar Pradesh, India. Am. J. Trop. Med. Hyg. 73,
720e725.
Barral, A., Badaro, R., Barral-Netto, M., Grimaldi, G., Momem, H., Carvalho, E.M., 1986.
Isolation of Leishmania mexicana amazonensis from the bone marrow in a case of American
visceral leishmaniasis. Am. J. Trop. Med. Hyg. 35, 732e734.
Bejon, P., Williams, T.N., Nyundo, C., Hay, S.I., Benz, D., Gething, P.W., Otiende, M.,
Peshu, J., Bashraheil, M., Greenhouse, B., Bousema, T., Bauni, E., Marsh, K.,
Smith, D.L., Borrmann, S., 2014. A micro-epidemiological analysis of febrile malaria
in coastal Kenya showing hotspots within hotspots. eLife 3, e02130.
Ben-Salah, A., Smaoui, H., Mbarki, L., Anderson, R.M., Ismail, R.B., 1994. Development
of a mathematical model on the dynamics of canine leishmaniasis transmission. Arch.
l’Institut Pasteur Tunis 71, 431e438.
Benzerroung, E.H., Benhabylles, N., Izri, M.A., Belahcene, E.K., 1992. Les pulverisations
intra- et peri-domiciliaires de DDT dans la lutte contre la leishmaniose cutanee zoono-
tique en Algerie. Ann. Soc. Belge Med. Trop. 72, 5e12.
Bern, C., Amann, J., Haque, R., Chowdhury, R., Ali, M.Y., Kurkijian, K.M., Vaz, L.,
Wagatsumam, Y., Breiman, R.F., Secor, W.E., Maguire, J., 2006. Loss of Leishmanin
skin test antigen sensitivity and potency in a longitudinal study of visceral leishmaniasis
in Bangladesh. Am. J. Trop. Med. Hyg. 75, 744e748.
Bern, C., Chowdhury, R., 2006. The epidemiology of visceral leishmaniasis in Bangladesh:
prospects for improved control. Indian J. Med. Res. A 123, 275e288.
Bern, C., Courtenay, O., Alvar, J., February 2010. Of cattle, sand flies and men: a systematic
review of risk factor analyses for South Asian visceral leishmaniasis and implications for
elimination. PLoS Neglected Trop. Dis. 4, e599.
Bern, C., Haque, R., Chowdhury, R., Ali, M., Kurkijian, K.M., Vaz, L., Amann, J.,
Wahed, M.A., Wagatsumam, Y., Breiman, R.F., Williamson, J., Secor, W.E.,
Maguire, J.H., 2007. The epidemiology of visceral leishmaniasis and asymptomatic leish-
manial infection in a highly endemic Bangladeshi village. Am. J. Trop. Med. Hyg. 76,
909e914.
Bern, C., Hightower, A.W., Chowdhury, R., Ali, M., Amann, J., 2005. Risk factors for
Kala-azar in Bangladesh. Emerg. Infect. Dis. 11, 655e662.
Bhattarai, N.R., Van der Auwera, G., Rijal, S., Picado, A., Speybroeck, N., Khanal, B., De
Doncker, S., Das, M.L., Ostyn, B., Davies, C.R., Coosemans, M., Berkvens, D.,
Boelaert, M., Dujardin, J.-C., February 2010. Domestic animals and epidemiology of
visceral leishmaniasis, Nepal. Emerg. Infect. Dis. 16, 231e237.
Biglino, A., Bolla, C., Concialdi, E., Trisciuoglio, A., Romano, A., Ferroglio, E., January
2010. Asymptomatic Leishmania infantum infection in an area of northwestern Italy (Pied-
mont Region) where such infections are traditionally nonendemic. J. Clin. Microbiol.
48, 131e136.
Boelaert, M., Lynen, L., Desjeux, P., Van der Stuyft, P., August 1999. Cost-effectiveness of
competing diagnostic-therapeutic strategies for visceral leishmaniasis. Bull. World Health
Organ. 77, 667e674.
116 K.S. Rock et al.

Boelaert, M., Meheus, F., Sanchez, A., Singh, S.P., Vanlerberghe, V., Picado, A.,
Meessen, B., Sundar, S., June 2009. The poorest of the poor: a poverty appraisal of
households affected by visceral leishmaniasis in Bihar, India. Trop. Med. Int. Health
14, 639e644.
Boni, M., Pratlong, F., El Hadi Osman, M., Bucheton, B., El-Safi, S., Feugier, E.,
Musa, M.K., Davoust, B., Dessein, A., Dedet, J.-P., 2014. Report of the 67th session.
WHO Regional Comm. South-East Asia 1e157.
Borja-Cabrera, G.P., Correia Pontes, N.N., da Silva, V.O., Paraguai de Souza, E.,
Santos, W.R., Gomes, E.M., Luz, K.G., Palatnik, M., Palatnik-de Sousa, C.B.,
2002. Long lasting protection against canine Kala-azar using the FML-QuilA saponin
vaccine in an endemic area of Brazil (Sao Goncalo do Amarante, RN). Vaccine 20,
3277e3284.
Borja-Cabrera, G.P., Cruz Mendes, A., Paraguai de Souza, E., Hashimoto Okada, L.Y., de
A Trivellato, F.A., Kawasaki, J.K.A., Costa, A.C., Reis, A.B., Genaro, O.,
Batista, L.M.M., Palatnik, M., Palatnik-de Sousa, C.B., June 2004. Effective immuno-
therapy against canine visceral leishmaniasis with the FML-vaccine. Vaccine 22, 2234e
2243.
Bray, D.P., Carter, V., Alves, G.B., Brazil, R.P., Bandi, K.K., Hamilton, J.G.C., March
2014. Synthetic sex pheromone in a long-lasting lure attracts the visceral leishmaniasis
vector, Lutzomyia longipalpis, for up to 12 Weeks in Brazil. PLoS Neglected Trop. Dis.
8, e2723.
Burattini, M.N., Coutinho, F., Lopez, L.F., Massad, E., 1998. Modelling the dynamics of
leishmaniasis considering human, animal host and vector populations. J. Biol. Syst. 6,
337e356.
Burza, S., Sinha, P.K., Mahajan, R., Lima, M.A., Mitra, G., Verma, N., Balsegaram, M.,
Das, P., January 2014. Risk factors for visceral leishmaniasis relapse in immunocompetent
patients following treatment with 20 mg/kg liposomal amphotericin B (Ambisome) in
Bihar, India. PLoS Neglected Trop. Dis. 8, e2536.
Carneiro, M., Moreno, E.C., Goncalves, A.V., Lambertucci, K.R., Antunes, C.M.F., 2011.
Visceral leishmaniasis: challenges in identifying subclinical Leishmania infection. Drug
Dev. Res. 72, 442e450.
Casanova, C., Natal, D., Santos, F.A.M., 2009. Survival, population size, and gonotrophic
cycle duration of Nyssomyia neivai (Diptera: Psychodidae) at an endemic area of American
cutaneous leishmaniasis in Southeastern Brazil. J. Med. Entomol. 46, 42e50.
Cerf, B.J., Jones, T.C., Badaro, R., Sampaio, D., Teixeira, R., Johnson, W.D., 1987. Malnu-
trition as a risk factor for severe visceral leishmaniasis. J. Infect. Dis. 156, 1030e1033.
Chapman, L.A., Dyson, L., Hollingsworth, T.D., Courtenay, O., Bern, C., Medley, G.F.,
2015. Quantification of the natural history of visceral leishmaniasis and consequences
for control. Parasites Vectors 22.
Chowdhury, R., Huda, M.M., Kumar, V., Das, P., Joshi, A.B., Banjara, M.R., Akhter, S.,
Kroeger, A., Krishnakumari, B., Petzold, M., Mondal, D., Das, M.L., January 2011.
The Indian and Nepalese programmes of indoor residual spraying for the elimination
of visceral leishmaniasis: performance and effectiveness. Ann. Trop. Med. Parasitol.
105, 31e35.
Chunge, C.N., Owate, J., Pamba, H.O., Donno, L., March 1990. Treatment of visceral
leishmaniasis in Kenya by aminosidine alone or combined with sodium
stibogluconate. Trans. R. Soc. Trop. Med. Hyg. 84, 221e225.
Coleman, M., Foster, G.M., Deb, R., Singh, R.P., Ismail, H.M., Shivam, P., Ghosh, A.K.,
Dunkley, S., Kumar, V., Coleman, M., Hemingway, J., Paine, M.J.I., Das, P., July 2015.
DDT-based indoor residual spraying suboptimal for visceral leishmaniasis elimination in
India. PNAS 112, 8573e8578.
Progress in the Mathematical Modelling of Visceral Leishmaniasis 117

Collin, S.M., Coleman, P.G., Ritmeijer, K., Davidson, R.N., March 2006. Unseen Kala-
azar deaths in south Sudan (1999e2002). Trop. Med. Int. Health 11, 509e512.
Costa, C.H.N., 2011. How effective is dog culling in controlling zoonotic visceral leishman-
iasis? A critical evaluation of the science, politics and ethics behind this public health
policy. Rev. Soc. Bras. Med. Trop. 44, 232e242.
Costa, C.H.N., Gomes, R.B.B., Silva, M.R.B., Garcez, L.M., Ramos, P.K.S., Santos, R.S.,
Shaw, J.J., David, J.R., Maguire, J.H., 2000. Competence of the human host as a reser-
voir for Leishmania chagasi. J. Infect. Dis. 182, 997e1000.
Costa, D.N.C.C., Codeço, C.T., Silva, M.A., Werneck, G.L., 2013. Culling dogs in
scenarios of imperfect control: realistic impact on the prevalence of canine visceral
leishmaniasis. PLoS Neglected Trop. Dis. 7, e2355.
Courtenay, O., Carson, C., Calvo-Bado, L., Garcez, L.M., Quinnell, R.J., January 2014.
Heterogeneities in Leishmania infantum infection: using skin parasite burdens to identify
highly infectious dogs. PLoS Neglected Trop. Dis. 8, e2583.
Courtenay, O., Gillingwater, K., Gomes, P.A., Garcez, L.M., Davies, C.R., 2007. Delta-
methrin-impregnated bednets reduce human landing rates of sandfly vector Lutzomyia
longipalpis in Amazon households. Med. Vet. Entomol. 21, 168e176.
Courtenay, O., Quinnell, R.J., Garcez, L.M., Dye, C., 2002a. Low infectiousness of a wild-
life host of Leishmania infantum: the crab-eating fox is not important for transmission.
Parasitology 125, 407e414.
Courtenay, O., Quinnell, R.J., Garcez, L.M., Shaw, J.J., Dye, C., 2002b. Infectiousness in a
cohort of Brazilian dogs: why culling fails to control visceral leishmaniasis in areas of high
transmission. J. Infect. Dis. 186, 1314e1320.
Croft, S.L., Sundar, S., Fairlamb, A.H., January 2006. Drug resistance in leishmaniasis. Clin.
Microbiol. Rev. 19, 111e126.
Cruz-Pacheco, G., Esteva, L., Vargas, C., June 2012. Control measures for Chagas disease.
Math. Biosci. 237, 49e60.
Cunningham, J., Hasker, E., Das, P., El-Safi, S., Goto, H., Mondal, D., Mbuchi, M.,
Mukhtar, M., Rabello, A., Rijal, S., Sundar, S., Wasunna, M., Adams, E., Menten, J.,
Peeling, R., Boelaert, M., for the WHO/TDR Visceral Leishmaniasis Laboratory
Network, October 2012. A global comparative evaluation of commercial immunochro-
matographic rapid diagnostic tests for visceral leishmaniasis. Clin. Infect. Dis. 55, 1312e
1319.
da Saude Secretaria de Vigil^ancia em Sa ude Departamento de Vigil^ancia Epidemiol ogica,
2006. Manual de Vigilancia e Controle da Leishmaniose Visceral (Tech. rep).
Das, M., Saudagar, P., Sundar, S., Dubey, V.K., August 2013. Miltefosine-unresponsive
Leishmania donovani has a greater ability than miltefosine-responsive L. donovanito resist
reactive oxygen species. FEBS J. 280, 4807e4815.
Davies, C.R., Llanos-Cuentas, E.A., Campos, P., Monge, J., Leon, E., Canales, J., 2000.
Spraying houses in the Peruvian Andes with lambda-cyhalothrin protects
residents against cutaneous leishmaniasis. Trans. R. Soc. Trop. Med. Hyg. 94,
631e636.
Davies, C.R., Mazloumi Gavgani, A.S., 1999. Age, acquired immunity and the risk of
visceral leishmaniasis: a prospective study in Iran. Parasitology 119, 247e257.
de Colmenares, M., Portus, M., Botet, J., Dobano, C., Gallego, M., Wolff, M., Segui, G.,
1995. Identification of blood meals of Phlebotomus perniciosus (Diptera: Psychodidae) in
Spain by a competitive enzyme-linked immunosorbent assay biotin/avidin method. J.
Med. Entomol. 32, 229e233.
de Gouvea Viana, L., de Assiss, T.S., da Silva, A.R., de Souza, G.F., Caligiorne, R., da
Silva, A.C., Peruhype-Magalhaes, V., Marciano, A.P., Martins-Filho, O.A.,
Rabello, A., 2008. Combined diagnostic methods identify a remarkable proportion of
118 K.S. Rock et al.

asymptomatic Leishmania (Leishmania) chagasi carriers who present modulated cytokine


profiles. Trans. R. Soc. Trop. Med. Hyg. 102, 548e555.
de Melo, E.C., Fortaleza, C.M.C.B., 2013. Challenges in the therapy of visceral leishmaniasis
in Brazil: a public health perspective. J. Trop. Med. 2013. Article ID 319234.
de Ruiter, C.M., van der Veer, C., Leeflang, M.M.G., Deborggraeve, S., Lucas, C.,
Adams, E.R., August 2014. Molecular tools for diagnosis of visceral leishmaniasis: sys-
tematic review and meta-analysis of diagnostic test accuracy. J. Clin. Microbiol. 52,
3147e3155.
de Souza, V.M.M., da Silva Juliao, F., Neves, R.C.S., Magalhaes, P.B., Bisinotto, T.V., de
Souza Lima, A., de Oliveira, S.S., Junior, E.D.M., 2008. Communitary assay for assess-
ment of effectiveness of strategies for prevention and control of human visceral leishman-
iasis in the municipality of Feira de Santana, State of Bahia, Brazil. Epidemiol. Serviços
Saude 17, 97e106.
Deane, L.M., Deane, M.P., 1962. Visceral leishmaniasis in Brazil: geographical distribution
and transmission. Rev. Inst. Med. Trop. S~ao Paulo 4, 198e212.
den Boer, M.L., Alvar, J., Davidson, R.N., Ritmeijer, K., Balasegaram, M., September 2009.
Developments in the treatment of visceral leishmaniasis. Expert Opin. Emerg. Drugs 14,
395e410.
Dereure, J., El-Safi, S.H., Bucheton, B., Boni, M., Kheir, M.M., Davoust, B., Pratlong, F.,
Feugier, E., Lambert, M., Dessein, A., Dedet, J.-P., October 2003. Visceral leishmaniasis
in eastern Sudan: parasite identification in humans and dogs; host-parasite relationships.
Microbes Infect. 5, 1103e1108.
Diekmann, O., Heesterbeek, J.A.P., Metz, J.A.J., 1990. On the definition and the compu-
tation of the basic reproduction ratio R0 in models for infectious diseases in heteroge-
neous populations. J. Math. Biol. 28, 365e382.
Dinesh, D.S., Das, P., Picado, A., Davies, C., Speybroeck, N., Boelaert, M., Coosemans, M.,
2008a. The efficacy of indoor CDC light traps for collecting the sandfly Phlebotomus
argentipes, vector of Leishmania donovani. Med. Vet. Entomol. 22, 120e123.
Dinesh, D.S., Das, P., Picado, A., Davies, C.R., Speybroeck, N., Ostyn, B., Boelaert, M.,
Coosemans, M., May 2008b. Long-lasting insecticidal nets fail at household level to
reduce abundance of sandfly vector Phlebotomus argentipes in treated houses in Bihar
(India). Trop. Med. Int. Health 13, 953e958.
Dorlo, T.P.C., Rijal, S., Ostyn, B., de Vries, P.J., Singh, R., Bhattarai, N., Uranw, S.,
Dujardin, J.-C., Boelaert, M., Beijnen, J.H., Huitema, A.D.R., June 2014. Failure of
miltefosine in visceral leishmaniasis is associated with low drug exposure. Int. J. Infect.
Dis. 210, 146e153.
dos Santos Marques, L.H., Gomes, L.I., da Rocha, I.C.M., da Silva, T.A.M., Oliveira, E.,
Morais, M.H.F., Rabello, A., Carneiro, M., December 2012. Low parasite load esti-
mated by qPCR in a cohort of children living in urban area endemic for visceral leish-
maniasis in Brazil. PLoS Neglected Trop. Dis. 6, e1955.
Downing, T., Imamura, H., Decuypere, S., Clark, T.G., Coombs, G.H., Cotton, J.A.,
Hilley, J.D., De Doncker, S., Maes, I., Mottram, J.C., Quail, M.A., Rijal, S.,
Sanders, M., Schonian, G., Stark, O., Sundar, S., Vanaerschot, M., Hertz-Fowler, C.,
Dujardin, J.-C., Berriman, M., December 2011. Whole genome sequencing of multiple
Leishmania donovani clinical isolates provides insights into population structure and mech-
anisms of drug resistance. Genome Res. 21, 2143e2156.
Dye, C., 1986. Population dynamics of mosquito-borne disease: effects of flies which bite
some people more frequently than others. Trans. R. Soc. Trop. Med. Hyg. 80, 69e77.
Dye, C., February 1988. The epidemiology of canine visceral leishmaniasis in southern
France: classical theory offers another explanation of the data. Parasitology 96, 19e24.
Dye, C., August 1996. The logic of visceral leishmaniasis control. Am. J. Trop. Med. Hyg.
55, 125e130.
Progress in the Mathematical Modelling of Visceral Leishmaniasis 119

Dye, C., Davies, C.R., Lainson, R., 1991. Communication among phlebotomine sandflies: a
field study of domesticated Lutzomyia longipalpis populations in Amazonian Brazil. Anim.
Behav. 42, 183e192.
Dye, C., Guy, M.W., Elkins, D.B., Wilkes, T.J., Killick-Kendrick, R., 1987. The life expec-
tancy of phlebotomine sandflies: first field estimates from southern France. Med. Vet.
Entomol. 1, 417e425.
Dye, C., Vidor, E., Dereure, J., 1993. Serological diagnosis of leishmaniasis: on detecting
infection as well as disease. Epidemiol. Infect. 103, 647e656.
Dye, C., Williams, B.G., 1993. Malnutrition, age and the risk of parasitic disease: visceral
leishmaniasis revisited. Proc. R. Soc. B Biol. Sci. 254, 33e39.
Dye, C., Wolpert, D.M., 1988. Earthquakes, influenza and cycles of Indian Kala-azar. Trans.
R. Soc. Trop. Med. Hyg. 82, 843e850.
Edmunds, W.J., Medley, G.F., Nokes, D.J., 1999. Evaluating the cost-effectiveness of vacci-
nation programmes: a dynamic perspective. Statistic Med. 18, 3263e3282.
ELmojtaba, I.M., Mugisha, J.Y.T., Hashim, M.H.A., 2010a. Mathematical analysis of the dy-
namics of visceral leishmaniasis in the Sudan. Appl. Math. Comput. 217, 2567e2578.
ELmojtaba, I.M., Mugisha, J.Y.T., Hashim, M.H.A., June 2010b. Modelling the role of
cross-immunity between two different strains of leishmania. Nonlinear Anal. Real World
Appl. 11, 2175e2189.
ELmojtaba, I.M., Mugisha, J.Y.T., Hashim, M.H.A., 2012. Vaccination model for visceral
leishmaniasis with infective immigrants. Math. Methods Appl. Sci. 36, 216e226.
Elnaiem, D.A., 1996. Use of Pyrethroid Impregnated Bednets for the Control of Visceral
Leishmaniasis in Eastern Sudan. WHO. Tech. Rep. WHO ID L3/1881/47.
Evans, T.G., Teixeira, M.J., McAuliffe, I.T., de Alencar Barros Vasconcelos, I.,
Vasconcelos, A.W., de Queiroz Sousa, A., de Oliveira Lima, J.W., Pearson, R.D.,
1992. Epidemiology of visceral leishmaniasis in northeast Brazil. J. Infect. Dis. 166,
1124e1132.
Falcao, A.L., Falcao, A.R., Pinto, C.T., Gontijo, C.M.F., Falqueto, A., 1991. Effect of del-
tamethrin spraying on the sandfly populations in a focus of American cutaneous
leishmaniasis. Mem orias do Inst. Oswaldo Cruz 86, 399e404.
Feliciangeli, M.D., Mazzarri, M.B., Blas, S.S., Zerpa, O., 2003a. Control trial of Lutzomyia long-
ipalpis s.l. in the Island of Margarita, Venezuela. Trop. Med. Int. Health 8, 1131e1136.
Feliciangeli, M.D., Mazzarri, M.B., Campbell-Lendrum, D.H., Maroli, M., Maingon, R.,
2003b. Cutaneous leishmaniasis vector control perspectives using lambdacyhalothrin re-
sidual house spraying in El Ingenio, Miranda State, Venezuela. Trans. R. Soc. Trop.
Med. Hyg. 97, 641e646.
Felix de Lima, V.M., Ikeda, F.A., Rossi, C.u. N., Feitosa, M.M., de Oliveira Vasconcelos, R.,
Nunes, C.M., Goto, H., June 2010. Diminished CD4þ/CD25þ T cell and increased
IFN-g levels occur in dogs vaccinated with Leishmune in an endemic area for visceral
leishmaniasis. Vet. Immunol. Immunopathol. 135, 296e302.
Ferro, C., Morrison, A.C., Torres, M., Pardo, R., Wilson, M.L., Tesh, R.B., September
1995. Age structure, blood-feeding behavior, and Leishmania chagasi infection in Lutzo-
myia longipalpis (Diptera: Psychodidae) at an endemic focus of visceral leishmaniasis in
Colombia. J. Med. Entomol. 32, 618e629.
Foglia Manzillo, V., Di Muccio, T., Cappiello, S., Scalone, A., Paparcone, R., Fiorentino, E.,
Gizzarelli, M., Gramiccia, M., Gradoni, L., Oliva, G., May 2013. Prospective study on
the incidence and progression of clinical signs in naïve dogs naturally infected by Leish-
mania infantum. PLoS Neglected Trop. Dis. 7, e2225.
Franco, A.O., Davies, C.R., Mylne, A., Dedet, J.-P., Gallego, M., Ballart, C., Gramiccia, M.,
Gradoni, L., Molina, R., Galvez, R., Morillas-Marquez, F., Baron-Lopez, S.,
Pires, C.A., Afonso, M.O., Ready, P.D., Cox, J., September 2011. Predicting the
120 K.S. Rock et al.

distribution of canine leishmaniasis in western Europe based on environmental variables.


Parasitology 138, 1878e1891.
Garlapati, R.B., Abbasi, I., Warburg, A., Poché, D., Poché, R., May 2012. Identification of
bloodmeals in wild caught blood fed Phlebotomus argentipes (Diptera: Psychodidae)
using cytochrome b PCR and reverse line blotting in Bihar, India. J. Med. Entomol.
49, 515e521.
Gavgani, A.S., Hodjati, M.H., Mohite, H., Davies, C.R., 2002. Effect of insecticide-impreg-
nated dog collars on incidence of zoonotic visceral leishmaniasis in Iranian children: a
matched-cluster randomised trial. Lancet 360, 374e379.
Gebresilassie, A., Yared, S., Aklilu, E., Kirstein, O.D., Moncaz, A., Tekie, H., Balkew, M.,
Warburg, A., Hailu, A., Gebre-Michael, T., 2015. Host choice of Phlebotomus orientalis
(Diptera: Psychodidae) in animal baited experiments: a field study in Tahtay Adiyabo dis-
trict, northern Ethiopia. Parasites Vectors 8, 190.
Ghosh, K., Mukhopadhyay, J., Desai, M.M., Senroy, S., Bhattacharya, A., August 1999.
Population ecology of Phlebotomus argentipes (Diptera: Psychodidae) in West Bengal,
India. J. Med. Entomol. 36, 588e594.
Gidwani, K., Picado, A., Ostyn, B., Singh, S.P., Kumar, R., Khanal, B., Lejon, V.,
Chappuis, F., Boelaert, M., Sundar, S., January 2011. Persistence of Leishmania donovani
antibodies in past visceral leishmaniasis cases in India. Clin. Vaccine Immunol. 18, 346e
348.
Gradoni, L., February 2015. Canine Leishmania vaccines: still a long way to go. Vet. Parasitol.
208, 94e100.
Guy, M.W., Killick-Kendrick, R., Gill, G.S., Rioux, J.-A., Bray, R.S., 1984. Ecology of
leishmaniasis in the south of France. 19. Determination of the hosts of Phlebotomus ariasi
Tonnoir, 1921 in the Cevennes by bloodmeal analyses. Ann. de Parasitol. Humaine
Comp. (Paris) 59, 449e458.
Hailu, A., Gramiccia, M., Kager, P.A., December 2009. Visceral leishmaniasis in Aba-Roba,
south-western Ethiopia: prevalence and incidence of active and subclinical infections.
Ann. Trop. Med. Parasitol. 103, 659e670.
Hailu, A., Musa, A., Wasunna, M., Balasegaram, M., Yifru, S., Mengistu, G., Hurissa, Z.,
Hailu, W., Weldegebreal, T., Tesfaye, S., Makonnen, E., Khalil, E., Ahmed, O.,
Fadlalla, A., El-Hassan, A., Raheem, M., Mueller, M., Koummuki, Y., Rashid, J.,
Mbui, J., Mucee, G., Njoroge, S., Manduku, V., Musibi, A., Mutuma, G., Kirui, F.,
Lodenyo, H., Mutea, D., Kirigi, G., Edwards, T., Smith, P., Muthami, L., Royce, C.,
Ellis, S., Alobo, M., Omollo, R., Kesusu, J., Owiti, R., Kinuthia, J., for the Leishmaniasis
East Africa Platform (LEAP) group, October 2010. Geographical variation in the
response of visceral leishmaniasis to paromomycin in East Africa: a multicentre, open-la-
bel, randomized trial. PLoS Neglected Trop. Dis. 4, e709.
Hamad, S.H., Khalil, E.A.G., Musa, A.M., Ibrahim, M.E., Younis, B.M., Elfaki, M.E.E., El-
Hassan, A.M., The Leishmaniasis Research Group, August 2010. Leishmania donovani:
genetic diversity of isolates from Sudan characterized by PCR-based RAPD. Exp. Para-
sitol. 125, 389e393.
Hartemink, N., Vanwambeke, S.O., Heesterbeek, H., Rogers, D., Morley, D., Pesson, B.,
Davies, C., Mahamdallie, S., Ready, P., August 2011. Integrated mapping of establish-
ment risk for emerging vector-borne infections: a case study of canine leishmaniasis in
southwest France. PLoS One 6, e20817.
Hasibeder, G., Dye, C., Carpenter, J., 1992. Mathematical modelling and theory for estimating
the basic reproduction number of canine leishmaniasis. Parasitology 105, 43e53.
Hasker, E., Kansal, S., Malaviya, P., Gidwani, K., Picado, A., Singh, R.P., Chourasia, A.,
Singh, A.K., Shankar, R., Menten, J., Wilson, M.E., Boelaert, M., Sundar, S., February
2013. Latent infection with leishmania donovani in highly endemic villages in Bihar,
India. PLoS Neglected Trop. Dis. 7, e2053.
Progress in the Mathematical Modelling of Visceral Leishmaniasis 121

Hasker, E., Malaviya, P., Gidwani, K., Picado, A., Ostyn, B., Kansal, S., Singh, R.P.,
Singh, O.P., Chourasia, A., Singh, A.K., Shankar, R., Wilson, M.E., Khanal, B.,
Rijal, S., Boelaert, M., Sundar, S., January 2014. Strong association between serological
status and probability of progression to clinical visceral leishmaniasis in prospective cohort
studies in India and Nepal. PLoS Neglected Trop. Dis. 8, e2657.
Hasker, E., Singh, S.P., Malaviya, P., Picado, A., Gidwani, K., Singh, R.P., Menten, J.,
Boelaert, M., Sundar, S., 2012. Visceral leishmaniasis, rural Bihar, India. Emerg. Infect.
Dis. 18, 1662e1664.
Hotez, P.J., Alvarado, M., Basan ~ez, M.-G., Bolliger, I., Bourne, R., Boussinesq, M.,
Brooker, S.J., Brown, A.S., Buckle, G., Budke, C.M., Carabin, H., Coffeng, L.E.,
Fevre, E.M., F€urst, T., Halasa, Y.A., Jasrasaria, R., Johns, N.E., Keiser, J., King, C.H.,
Lozano, R., Murdoch, M.E., O’Hanlon, S., Pion, S.D.S., Pullan, R.L.,
Ramaiah, K.D., Roberts, T., Shepard, D.S., Smith, J.L., Stolk, W.A.,
Undurraga, E.A., Utzinger, J., Wang, M., Murray, C.J.L., Naghavi, M., July 2014.
The global burden of disease study 2010: interpretation and implications for the
neglected tropical diseases. PLoS Neglected Trop. Dis. 8, e2865.
Islam, S., Kenah, E., Bhuiyan, M.A.A., Rahman, K.M., Goodhew, B., Ghalib, C.M.,
Zahid, M.M., Ozaki, M., Rahman, M.W., Haque, R., Luby, S.P., Maguire, J.H.,
Martin, D., Bern, C., August 2013. Clinical and immunological aspects of
post-Kala-azar dermal leishmaniasis in Bangladesh. Am. J. Trop. Med. Hyg. 89,
345e353.
Jalouk, L., Al Ahmed, M., Gradoni, L., Maroli, M., 2007. Insecticide-treated bednets to pre-
vent anthroponotic cutaneous leishmaniasis in Aleppo Governorate, Syria: results from
two trials. Trans. R. Soc. Trop. Med. Hyg. 101, 360e367.
Joshi, A.B., Bhatt, L.R., Regmi, S., Ashford, R.W., 2003. An assessment of the effectiveness
of insecticide spray in the control of visceral leishmaniasis in Nepal. J. Nepal Health Res.
Counc. 1, 1e6.
Joshi, A.B., Das, M.L., Akhter, S., Chowdhury, R., Mondal, D., Kumar, V., Das, P.,
Kroeger, A., Boelaert, M., Petzold, M., October 2009. Chemical and environmental
vector control as a contribution to the elimination of visceral leishmaniasis on the Indian
subcontinent: cluster randomized controlled trials in Bangladesh, India and Nepal. BMC
Med. 7, 54.
Joshi, D.D., Sharma, M.C., Bhandari, S., 2006. Visceral leishmaniasis in Nepal during 1980e
2006. J. Commun. Dis. 38, 139e148.
Joshi, R.D., Rai, R.N., 1994. Impact of DDT spraying on populations of P. argentipes and P.
papatasi in Varanasi district, Uttar Pradesh. J. Commun. Dis. 26, 56e58.
Kassahun, A., Sadlova, J., Dvorak, V., Kostalova, T., Rohousova, I., Frynta, D., Aghova, T.,
Yasur-Landau, D., Lemma, W., Hailu, A., Baneth, G., Warburg, A., Volf, P.,
Votypka, J., 2015. Detection of Leishmania donovani and L. tropica in Ethiopian wild
rodents. Acta Trop. 145, 39e44.
Kaul, S.M., Sharma, R.S., Dey, K.P., Rai, R.N., Verghese, T., 1994. Impact of DDT indoor
residual spraying on Phlebotomus argentipes in a Kala-azar endemic village in eastern Uttar
Pradesh. Bull. World Health Organ. 72, 79e81.
Kedzierski, L., Evans, K.J., July 2014. Immune responses during cutaneous and visceral
leishmaniasis. Parasitology 141, 1544e1562.
Keeling, M.J., Rohani, P., 2008. Modeling Infectious Diseases in Humans and Animals.
Princeton University Press.
Kelly, D.W., Mustafa, Z., Dye, C., 1997. Differential application of lambda-cyhalothrin to
control the sandfly Lutzomyia longipalpis. Med. Vet. Entomol. 11, 13e24.
Kenubih, A., Dagnachew, S., Almaw, G., Abebe, T., Takele, Y., Hailu, A., Lemma, W.,
2014. A preliminary survey of domestic animal visceral leishmaniasis and risk factors in
North West Ethiopia. Afr. J. Parasitol. Res. 1, 1e5.
122 K.S. Rock et al.

Khalil, E.A.G., Weldegebreal, T., Younis, B.M., Omollo, R., Musa, A.M., Hailu, W.,
Abuzaid, A.A., Dorlo, T.P.C., Hurissa, Z., Yifru, S., Haleke, W., Smith, P.G.,
Ellis, S., Balasegaram, M., El-Hassan, A.M., Schoone, G.J., Wasunna, M.,
Kimutai, R., Edwards, T., Hailu, A., January 2014. Safety and efficacy of single dose
versus multiple doses of AmBisome for treatment of visceral leishmaniasis in eastern Af-
rica: a randomised trial. PLoS Neglected Trop. Dis. 8, e2613.
Killick-Kendrick, R., 1999. The biology and control of Phlebotomine sand flies. Clin. Der-
matol. 17, 279e289.
Killick-Kendrick, R., Rioux, J.-A., 2002. Mark-release-recapture of sand flies fed on leish-
manial dogs: the natural life-cycle of Leishmania infantum in Phlebotomus ariasi. Parassitol.
44, 67e71.
Kip, A.E., Balasegaram, M., Beijnen, J.H., Schellens, J.H.M., de Vries, P.J., Dorlo, T.P.C.,
December 2014. Systematic review of biomarkers to monitor therapeutic response in
leishmaniasis. Antimicrob. Agents Chemother. 59, 1e14.
Kroeger, A., Avila, E.V., Morison, L., 2002. Insecticide impregnated curtains to control do-
mestic transmission of cutaneous leishmaniasis in Venezuela: cluster randomised trial. Br.
Med. J. 325, 810e813.
Kumar, V., Kesari, S., Dinesh, D.S., Tiwari, A.K., Kumar, A.J., Singh, V.P., Das, P., 2009. A
report on the indoor residual spraying (IRS) in the control of Phlebotomus argentipes, the
vector of visceral leishmaniasis in Bihar (India): an initiative towards total elimination tar-
geting 2015 (Series-1). J. Vector-Borne Dis. 46, 225e229.
Lachaud, L., Bourgeois, N., Plourde, M., Leprohon, P., Bastien, P., Ouellette, M., January
2009. Parasite susceptibility to amphotericin B in failures of treatment for visceral leish-
maniasis in patients Coinfected with HIV type 1 and leishmania infantum. Clin. Infect.
Dis. 48, e16ee22.
Lachaud, L., Chabbert, E., Dubessay, P., Dereure, J., Lamothe, J., Dedet, J.-P., Bastien, P.,
2002. Value of two PCR methods for the diagnosis of canine visceral leishmaniasis and
the detection of asymptomatic carriers. Parasitology 125, 197e207.
Lainson, R., Rangel, E.F., 2005. Lutzomyia longipalpis and the eco-epidemiology of Amer-
ican visceral leishmaniasis, with particular reference to Brazil - a review. Mem orias do
Inst. Oswaldo Cruz 100, 811e827.
Lainson, R., Ward, R.D., Shaw, J.J., April 1977. Experimental transmission of Leishmania cha-
gasi, causative agent of neotropical visceral leishmaniasis, by the sandfly Lutzomyia
longipalpis. Nature 266, 628e630.
Lane, R.P., Pile, M.M., Amerasinghe, F.P., 1990. Anthropophagy and aggregation behav-
iour of the sandfly Phlebotomus argentipes in Sri Lanka. Med. Vet. Entomol. 4, 79e88.
Lee, B.Y., Bacon, K.M., Shah, M., Kitchen, S.B., Connor, D.L., Slayton, R.B., March 2012.
The economic value of a visceral leishmaniasis vaccine in Bihar state, India. Am. J. Trop.
Med. Hyg. 86, 417e425.
Lengeler, C., 2004. Insecticide-treated nets for malaria control: real gains. Bull. World Health
Organ. 82, 84.
Lira, R., Sundar, S., Makharia, A., Kenney, R.T., Gam, A., Saraiva, E.M., 1999. Evidence
that the high incidence of treatment failures in Indian Kala-azar is due to the emergence
of antimony-resistant strains of Leishmania donovani. J. Infect. Dis. 180, 564e567.
Lukes, J., Mauricio, I.L., Schonian, G., Dujardin, J.-C., Soteriadou, K., Dedet, J.-P.,
Kuhls, K., Tintaya, K.W.Q., Jirku, M., Chocholova, E., Haralambous, C.,
Pratlong, F., Obornik, M., Horak, A., Ayala, F.J., Miles, M.A., 2007. Evolutionary
and geographical history of the Leishmania donovani complex with a revisionof current
taxonomy. PNAS 104, 9375e9380.
Macdonald, G., 1957. The Epidemiology and Control of Malaria. Oxford.
MacDonald, N., 1978. Time Lags in Biological Models, 27. Springer Verlag.
Progress in the Mathematical Modelling of Visceral Leishmaniasis 123

Maia-Elkhoury, A.N.S., Alves, W.A., de Sousa-Gomes, M.L., de Sena, J.M., Luna, E.A.,
2008. Visceral leishmaniasis in Brazil: trends and challenges. Cad. Sa ude P ublica 24,
2941e2947.
Mary, C., Faraut, F., Drougoul, M.-P., Xeridat, B., Schleinitz, N., Cuisenier, B.,
Dumon, H., November 2006. Reference Values for Leishmania infantum parasitemia in
different clinical presentations: quantitative polymerase chain reaction for therapeutic
monitoring and patient follow-up. Am. J. Trop. Med. Hyg. 75, 858e863.
Mary, C., Faraut, F., Lascombe, L., Dumon, H., 2004. Quantification of Leishmania infantum
DNA by a real-time PCR assay with high sensitivity. J. Clin. Microbiol. 42, 5249e5255.
Mathers, C.D., Ezzati, M., Lopez, A.D., November 2007. Measuring the burden of
neglected tropical diseases: the global burden of disease framework. PLoS Neglected
Trop. Dis. 1, e114.
Matlashewski, G., Arana, B., Kroeger, A., Battacharya, S., Sundar, S., Das, P., Sinha, P.K.,
Rijal, S., Mondal, D., Zilberstein, D., Alvar, J., March 2011. Visceral leishmaniasis: elim-
ination with existing interventions. Lancet Infect. Dis. 11, 322e325.
Mayrink, W., Magalh~aes, P.A., Batista, S.M., da Costa, C.A., 1971. Diagnosis of Kala-azar. II.
Study of Montenegro’s test and detection of Leishmania in skin material from patients
with Kala-azar, before and after antimony treatment. Rev. Inst. Med. Trop. S~ao Paulo
13, 268e271.
Medley, G.F., Hollingsworth, T.D., Olliaro, P., Adams, E.R., 2015. Health-seeking behav-
iour, diagnostics and transmission dynamics in the control of visceral leishmaniasis in the
Indian subcontinent. Nature 528 (7580). S102-S108.
Meheus, F., Balasegaram, M., Olliaro, P., Sundar, S., Rijal, S., Faiz, M.A., Boelaert, M.,
September 2010. Cost-effectiveness analysis of combination therapies for visceral leish-
maniasis in the Indian Subcontinent. PLoS Neglected Trop. Dis. 4, e818.
Mejhed, H., Boussa, S., El Houda Mejhed, N., 2009. Development of mathematical models
predicting the density of vectors: case of sandflies vectors of leishmaniasis. In: Proceedings
of the 10th WSEAS International Conference on Mathematics and Computers in
Biology and Chemistry, pp. 62e67.
Melaku, Y., Collin, S.M., Keus, K., Gatluak, F., Ritmeijer, K., Davidson, R.N., 2007. Treat-
ment of Kala-azar in southern Sudan using a 17-day regimen of sodium stibogluconate
combined with paramomycin: a retrospective comparison with 30-day sodium stibo-
gluconate monotherapy. Am. J. Trop. Med. Hyg. 77, 89e94.
Miller, E., Warburg, A., Novikov, I., Hailu, A., Volf, P., Seblova, V., Huppert, A.,
October 2014. Quantifying the contribution of hosts with different parasite concentra-
tions to the transmission of visceral leishmaniasis in Ethiopia. PLoS Neglected Trop.
Dis. 8, e3288.
Molina, R., Amela, C., Nieto, J., San-Andres, M., Gonzales, F., Castillo, J.A., Lucientes, J.,
Alvar, J., 1994. Infectivity of dogs naturally infected with Leishmania infantum to colo-
nized Phlebotomus perniciosus. Trans. R. Soc. Trop. Med. Hyg. 88, 491e493.
Molina, R., Gradoni, L., Alvar, J., October 2003. HIV and the transmission of Leishmania.
Ann. Trop. Med. Parasitol. 97, 29e45.
Molina, R., Jiménez, M.I., Cruz, I., Iriso, A., Martín-Martín, I., Sevillano, O., Melero, S.,
Bernal, J., November 2012. The hare (Lepus granatensis) as potential sylvatic reservoir
of Leishmania infantum in Spain. Vet. Parasitol. 190, 268e271.
Mondal, D., Alam, M.S., Karim, Z., Haque, R., Boelaert, M., Kroeger, A., 2008. Present
situation of vector-control management in Bangladesh: a wake up call. Health Policy
87, 369e376.
Mondal, D., Huda, M.M., Karmoker, M.K., Ghosh, D., Matlashewski, G., Nabi, S.G.,
Kroeger, A., July 2013. Reducing visceral leishmaniasis by insecticide impregnation of
bed-nets, Bangladesh. Emerg. Infect. Dis. 19, 1131e1134.
124 K.S. Rock et al.

Mondal, D., Nasrin, K.N., Huda, M.M., Kabir, M., Hossain, M.S., Kroeger, A.,
Thomas, T., Haque, R., October 2010a. Enhanced case detection and improved diag-
nosis of PKDL in a Kala-azar-endemic area of Bangladesh. PLoS Neglected Trop. Dis.
4, e832.
Mondal, D., Singh, S.P., Kumar, N., Joshi, A., Sundar, S., Das, P., Siddhivinayak, H.,
Kroeger, A., Boelaert, M., January 2009. Visceral leishmaniasis elimination programme
in India, Bangladesh, and Nepal: reshaping the case finding/case management strategy.
PLoS Neglected Trop. Dis. 3, e355.
Mondal, S., Bhattacharya, P., Ali, N., August 2010b. Current diagnosis and treatment of
visceral leishmaniasis. Expert Rev. Anti-Infective Ther. 8, 919e944.
Moosa-Kazemi, S.H., Yaghoobi-Ershadir, M.R., Akhaven, A.A., Abdoli, H., Zahraei-
Ramazani, A.R., Jafari, R., Houshmand, B., Nadim, A., Hosseini, M., 2007. Deltameth-
rin-impregnated bed nets and curtains in an anthroponotic cutaneous leishmaniasis
control program in northeastern Iran. Ann. Saudi Med. 27, 6e12.
Moral, L., Rubio, E.M., Moya, M., 2002. A leishmanin skin test survey in the human pop-
ulation of l’Alacantí region (Spain): implications for the epidemiology of Leishmania
infantum infection in southern Europe. Trans. R. Soc. Trop. Med. Hyg. 96, 129e132.
Morrison, A.C., Ferro, C., Morales, A., Tesh, R.B., Wilson, M.L., 1993a. Dispersal of the
sand fly Lutzomyia longipalpis (Diptera: Psychodidae) at an endemic focus of visceral leish-
maniasis in Colombia. J. Med. Entomol. 30, 427e435.
Morrison, A.C., Ferro, C., Tesh, R.B., 1993b. Host preferences of the sand fly Lutzomyia
longipalpis at an endemic focus of American visceral leishmaniasis in Colombia. Am. J.
Trop. Med. Hyg. 49, 68e75.
Mour~ao, M.V.A., Toledo Jr., A., Gomes, L.I., Freire, V.V., Rabello, A., 2014. Parasite load
and risk factors for poor outcome among children with visceral leishmaniasis. A cohort
study in Belo Horizonte, Brazil, 2010-2011. Mem orias do Inst. Oswaldo Cruz 109,
147e153.
Mubayi, A., Castillo-Chavez, C., Chowell, G., Kribs-Zaleta, C., Ali Siddiqui, N.,
Kumar, N., Das, P., January 2010. Transmission dynamics and underreporting of
Kala-azar in the Indian state of Bihar. J. Theor. Biol. 262, 177e185.
Mueller, M., Ritmeijer, K., Balasegaram, M., Koummuki, Y., Santana, M.R., 2007. Unre-
sponsiveness to AmBisome in some Sudanese patients with Kala-azar. Trans. R. Soc.
Trop. Med. Hyg. 101, 19e24.
Mueller, Y.K., Nackers, F., Ahmed, K.A., Boelaert, M., Djoumessi, J.-C., Eltigani, R.,
Gorashi, H.A., Hammam, O., Ritmeijer, K., Salih, N., Worku, D., Etard, J.-F.,
Chappuis, F., 2012. Burden of visceral leishmaniasis in villages of eastern Gedaref State,
Sudan: an exhaustive cross-sectional survey. PLoS Neglected Trop. Dis. 6, e1872.
Mukhopadhyay, A.K., Hati, A.K., Chakraborty, S., Saxena, N.B., 1996. Effect of DDT on
Phlebotomus sandflies in Kala-azar endemic foci in West Bengal. J. Commun. Dis. 28,
171e175.
Musa, A., Khalil, E., Hailu, A., Olobo, J., Balasegaram, M., Omollo, R., Edwards, T.,
Rashid, J., Mbui, J., Musa, B., Abuzaid, A.A., Ahmed, O., Fadlalla, A., El-Hassan, A.,
Mueller, M., Mucee, G., Njoroge, S., Manduku, V., Mutuma, G., Apadet, L.,
Lodenyo, H., Mutea, D., Kirigi, G., Yifru, S., Mengistu, G., Hurissa, Z., Hailu, W.,
Weldegebreal, T., Tafes, H., Mekonnen, Y., Makonnen, E., Ndegwa, S., Sagaki, P.,
Kimutai, R., Kesusu, J., Owiti, R., Ellis, S., Wasunna, M., June 2012. Sodium stibogluc-
onate (SSG) & paromomycin combination compared to SSG for visceral leishmaniasis in
East Africa: a randomised controlled trial. PLoS Neglected Trop. Dis. 6, e1674.
Musa, A.M., Khalil, E.A.G., Raheem, M.A., Zijlstra, E.E., Ibrahim, M.E., Elhassan, I.M.,
Mukhtar, M.M., El-Hassan, A.M., December 2002. The natural history of Sudanese
post-Kala-azar dermal leishmaniasis: clinical, immunological and prognostic features.
Ann. Trop. Med. Parasitol. 96, 765e772.
Progress in the Mathematical Modelling of Visceral Leishmaniasis 125

Musa, A.M., Khalil, E.A.G., Younis, B.M., Elfaki, M.E.E., Elamin, M.Y., Adam, A.O.A.,
Mohamed, H.A.A., Dafalla, M.M.M., Abuzaid, A.A., El-Hassan, A.M., 2013. Treat-
ment-based strategy for the management of post-Kala-azar dermal leishmaniasis patients
in the Sudan. J. Trop. Med. 2013. Article ID 708391.
Nasreen, S.A., Hossain, M.A., Paul, S.K., Mahmud, M.C., Ahmed, S., Ghosh, S.,
Kobayashi, N., 2012. PCR-based detection of leishmania DNA in skin samples of
post Kala-azar dermal leishmaniasis patients from an endemic area of Bangladesh. Jpn.
J. Infect. Dis. 65, 315e317.
Nogueira, F.S., Moreira, M.A.B., Borja-Cabrera, G.P., Santos, F.N., Menz, I., Parra, L.E.,
Xu, Z., Chu, H.J., Palatnik-de Sousa, C.B., Luvizotto, M.C.R., September 2005.
Leishmune vaccine blocks the transmission of canine visceral leishmaniasis. Vaccine
23, 4805e4810.
Nunes, C.M., Lima, V.M.F.d., Paula, H.B.d., Perri, S.H.V., Andrade, A.M.d., Dias, F.E.F.,
Burattini, M.N., May 2008. Dog culling and replacement in an area endemic for visceral
leishmaniasis in Brazil. Vet. Parasitol. 153, 19e23.
Oliva, G., Nieto, J., Foglia Manzillo, V., Cappiello, S., Fiorentino, E., Di Muccio, T.,
Scalone, A., Moreno, J., Chicharro, C., Carrillo, E., Butaud, T., Guegand, L.,
Martin, V., Cuisinier, A.-M., McGahie, D., Gueguen, S., Canavate, C., Gradoni, L.,
October 2014. A randomised, double-blind, controlled efficacy trial of the LiESP/
QA-21 vaccine in naíve dogs exposed to two Leishmania infantum transmission seasons.
PLoS Neglected Trop. Dis. 8, e3213.
Oliva, G., Scalone, A., Foglia Manzillo, V., Gramiccia, M., Pagano, A., Di Muccio, T.,
Gradoni, L., April 2006. Incidence and time course of Leishmania infantum infections
examined by parasitological, serologic, and nested-PCR techniques in a cohort of
naïve dogs exposed to three consecutive transmission seasons. J. Clin. Microbiol. 44,
1318e1322.
Oliveira da Silva, V., Borja-Cabrera, G.P., Correia Pontes, N.N., Paraguai de Souza, E.,
Luz, K.G., Palatnik, M., Palatnik-de Sousa, C.B., 2001. A phase III trial of efficacy of
the FML-vaccine against canine Kala-azar in an endemic area of Brazil (Sao Goncalo
do Amaranto, RN). Vaccine 19, 1082e1092.
Olliaro, P., Guerin, P.M., Gerstl, S., Haaskjold, A.A., Rottingen, J.-A., Sundar, S.,
November 2005. Treatment options for visceral leishmaniasis: a systematic review of
clinical studies done in India, 1980e2004. Lancet 5, 763e774.
Ostyn, B., Vanlerberghe, V., Picado, A., Dinesh, D.S., Sundar, S., Chappuis, F., Rijal, S.,
Dujardin, J.-C., Coosemans, M., Boelaert, M., Davies, C.R., August 2008. Vector con-
trol by insecticide-treated nets in the fight against visceral leishmaniasis in the Indian sub-
continent, what is the evidence? Trop. Med. Int. Health 13, 1073e1085.
Otranto, D., Dantas-Torres, F., July 2013. The prevention of canine leishmaniasis and its
impact on public health. Trends Parasitol. 29, 339e345.
Ozaki, M., Islam, S., Rahman, K.M., Rahman, A., Luby, S.P., Bern, C., September 2011.
Economic consequences of post-Kala-azar dermal leishmaniasis in a rural Bangladeshi
community. Am. J. Trop. Med. Hyg. 85, 528e534.
Palatnik-de Sousa, C.B., Batista-de Melo, L.M., Borja-Cabrera, G.P., Palatnik, M.,
Lavor, C.C., September 2004. Improving methods for epidemiological control of canine
visceral leishmaniasis based on a mathematical model. Impact on the incidence of the
canine and human disease. An. Acad. Bras. Ciencias 76, 583e593.
Palit, A., Bhattacharya, S.K., Kundu, S.N., 2006. Host preference of Phlebotomus argentipes
and Phlebotomus papatasi in different biotopes of West Bengal, India. Int. J. Environ.
Health Res. 15, 449e454.
Palit, A., Kishore, K., Sen, A.B., 1990. Gonotrophic cycles of Phlebotomus argentipes in nature
in Bihar - preliminary experiences. Indian J. Parasitol. 14, 121e123.
126 K.S. Rock et al.

Picado, A., Das, M.L., Kumar, V., Dinesh, D.S., Rijal, S., Singh, S.P., Das, P.,
Coosemans, M., Boelaert, M., Davies, C.R., March 2010a. Phlebotomus argentipes seasonal
patterns in India and Nepal. J. Med. Entomol. 47, 283e286.
Picado, A., Dash, A.P., Bhattacharya, S., Boelaert, M., July 2012. Vector control interven-
tions for visceral leishmaniasis elimination initiative in South Asia, 2005e2010. Indian
J. Med. Res. 136, 22e31.
Picado, A., Ostyn, B., Rijal, S., Sundar, S., Singh, S.P., Chappuis, F., Das, M.L., Khanal, B.,
Gidwani, K., Hasker, E., Dujardin, J.-C., Vanlerberghe, V., Menten, J., Coosemans, M.,
Boelaert, M., April 2015. Long-lasting insecticidal nets to prevent visceral leishmaniasis
in the indian subcontinent; methodological lessons learned from a cluster randomised
controlled trial. PLoS Neglected Trop. Dis. 9, e0003597.
Picado, A., Singh, S.P., Rijal, S., Sundar, S., Ostyn, B., Chappuis, F., Uranw, S.,
Gidwani, K., Khanal, B., Rai, M., Paudel, I.S., Das, M.L., Kumar, R., Srivastava, P.,
Dujardin, J.-C., Vanlerberghe, V., Andersen, E.W., Davies, C.R., Boelaert, M.,
2010b. Longlasting insecticidal nets for prevention of Leishmania donovani infection in In-
dia and Nepal: paired cluster randomised trial. Br. Med. J. 341, c6760.
Pourabbas, B., Ghadimi Moghadam, A., Pouladfar, G., Rezaee, Z., Alborzi, A., May 2013.
Quantification of Leishmania infantum kinetoplast DNA for monitoring the response to
meglumine antimoniate therapy in visceral leishmaniasis. Am. J. Trop. Med. Hyg. 88,
868e871.
Quilez, J., Martínez, V., Woolliams, J.A., Sanchez, A., Pong-Wong, R., Kennedy, L.J.,
Quinnell, R.J., Ollier, W.E.R., Roura, X., Ferrer, L., Altet, L., Francino, O., April
2012. Genetic control of canine leishmaniasis: genome-wide association study and
genomic selection analysis. PLoS One 7, e35349.
Quinnell, R.J., Courtenay, O., 2009. Transmission, reservoir hosts and control of zoonotic
visceral leishmaniasis. Parasitology 136, 1915e1934.
Quinnell, R.J., Courtenay, O., Davidson, S., Garcez, L., Lamothe, J., Ramos, P.K.S.,
Shaw, J.J., Shaw, M.-A., Dye, C., 2001. Detection of Leishmania infantum by PCR,
serology and cellular immune response in a cohort study of Brazilian dogs. Parasitology
122, 253e261.
Quinnell, R.J., Courtenay, O., Garcez, L., Dye, C., August 1997. The epidemiology of
canine leishmaniasis: transmission rates estimated from a cohort study in Amazonian
Brazil. Parasitology 115, 143e156.
Quinnell, R.J., Dye, C., Shaw, J.J., 1992. Host preferences of the phlebotomine sandfly Lut-
zomyia longipalpis in Amazonian Brazil. Med. Vet. Entomol. 6, 195e200.
Rahman, K.M., Islam, S., Rahman, M.W., Kenah, E., Galive, C.M., Zahid, M.M.,
Maguire, J., Rahman, M., Haque, R., Luby, S.P., Bern, C., January 2010. Increasing
incidence of post-Kala-azar dermal leishmaniasis in a population-based study in
Bangladesh. Clin. Infect. Dis. 50, 73e76.
Rebollar-Tellez, E.A., Hamilton, J.G.C., Ward, R.D., March 2006. Genetic inherence of
the response to human kairomones by two allopatric members of the Lutzomyia longipal-
pis complex. Physiol. Entomol. 31, 94e97.
Reithinger, R., Brooker, S.J., Kolaczinski, J.H., 2007. Visceral leishmaniasis in eastern Africa
e current status. Trans. R. Soc. Trop. Med. Hyg. 101, 1169e1170.
Reithinger, R., Coleman, P.G., Alexander, B., Vieira, E.P., Assis, G., Davies, C.R.,
January 2004. Are insecticide-impregnated dog collars a feasible alternative to dog cull-
ing as a strategy for controlling canine visceral leishmaniasis in Brazil? Int. J. Parasitol.
34, 55e62.
Reyburn, H., Ashford, R., Mohsen, M., Hewitt, S., Rowland, M., July 2000. A randomized
controlled trial of insecticide-treated bednets and chaddars or top sheets, and residual
spraying of interior rooms for the prevention of cutaneous leishmaniasis in Kabul,
Afghanistan. Trans. R. Soc. Trop. Med. Hyg. 94, 361e366.
Progress in the Mathematical Modelling of Visceral Leishmaniasis 127

Ribas, L.M., Zaher, V.L., Shimozako, H.J., Massad, E., 2013. Estimating the optimal control
of zoonotic visceral leishmaniasis by the use of a mathematical model. TheScientific-
WorldJournal 2013, 810380e810386.
Rijal, S., Chappuis, F., Singh, R., Bovier, P.A., Acharya, P., Karki, B.M.S., Das, M.L.,
Desjeux, P., Loutan, L., Koirala, S., May 2003. Treatment of visceral leishmaniasis in
south-eastern Nepal: decreasing efficacy of sodium stibogluconate and need for a policy
to limit further decline. Trans. R. Soc. Trop. Med. Hyg. 97, 350e354.
Rijal, S., Ostyn, B., Uranw, S., Rai, K., Bhattarai, N.R., Dorlo, T.P.C., Beijnen, J.H.,
Vanaerschot, M., Decuypere, S., Dhakal, S.S., Das, M.L., Karki, P., Singh, R.,
Boelaert, M., Dujardin, J.-C., May 2013. Increasing failure of miltefosine in the treat-
ment of Kala-azar in Nepal and the potential role of parasite drug resistance, reinfection,
or noncompliance. Clin. Infect. Dis. 56, 1530e1538.
Rijal, S., Yardley, V., Chappuis, F., Decuypere, S., Khanal, B., Singh, R., Boelaert, M., De
Doncker, S., Croft, S.L., Dujardin, J.-C., April 2007. Antimonial treatment of visceral
leishmaniasis: are current in vitro susceptibility assays adequate for prognosis of in vivo
therapy outcome? Microbes Infect. 9, 529e535.
Ritmeijer, K., Davies, C.R., Van Zorge, R., Wang, S.-J., Schorscher, J., Dongu’du, S.I.,
Davidson, R.N., February 2007. Evaluation of a mass distribution programme for
fine-mesh impregnated bednets against visceral leishmaniasis in eastern Sudan. Trop.
Med. Int. Health 12, 404e414.
Rock, K.S., Brand, S., Moir, J., Keeling, M.J., January 2014. Dynamics of infectious diseases.
Rep. Prog. Phys. 77, 026602.
Rock, K.S., le Rutte, E.A., de Vlas, S.J., Adams, E.R., Medley, G.F., Hollingsworth, T.D.,
April 2015a. Uniting mathematics and biology for control of visceral leishmaniasis.
Trends Parasitol. 31, 1e9.
Rock, K.S., Stone, C.M., Hastings, I.M., Keeling, M.J., Torr, S.J., Chitnis, N., March
2015b. Mathematical models of human African trypanosomiasis epidemiology. Adv. Par-
asitol. 87, 53e133.
Rodriguez-Barraquer, I., Buathong, R., Iamsirithaworn, S., Nisalak, A., Lessler, J.,
Jarman, R.G., Gibbons, R.V., Cummings, D.A.T., January 2014. Revisiting Rayong:
shifting seroprofiles of dengue in Thailand and their implications for transmission and
control. Am. J. Epidemiol. 179, 353e360.
Romero, G.A.S., Boelaert, M., January 2010. Control of visceral leishmaniasis in Latin
America e a systematic review. PLoS Neglected Trop. Dis. 4, e584.
Ross, R., February 1916. An application of the theory of probabilities to the study of a priori
pathometry. Part I. Proceedings of the Royal Society A: Mathematical. Phys. Eng. Sci.
92, 204e230.
Rossi, E., Bongiorno, G., Ciolli, E., Di Muccio, T., Scalone, A., Gramiccia, M., Gradoni, L.,
Maroli, M., February 2008. Seasonal phenology, host-blood feeding preferences and nat-
ural Leishmania infection of Phlebotomus perniciosus (Diptera, Psychodidae) in a high-
endemic focus of canine leishmaniasis in Rome province. Italy. Acta Trop. 105, 158e165.
Sacks, D.L., Kenney, R.T., Kreutzer, R.D., Jaffe, C.L., Gupta, A.K., Sharma, M.C.,
Sinha, S.P., Neva, F.A., Saran, R., 1995. Indian Kala-azar caused by Leishmania tropica.
Lancet 345, 959e961.
Saraiva, E.M., Barbosa, A.d. F., Santos, F.N., Borja-Cabrera, G.P., Nico, D.,
Souza, L.O.P., Mendes-Aguiar, C.d. O., de Souza, E.P., Fampa, P., Parra, L.E.,
Menz, I., Dias, J.G., de Oliveira, S.M., Palatnik-de Sousa, C.B., March 2006. The
FML-vaccine (Leishmune) against canine visceral leishmaniasis: a transmission block-
ing vaccine. Vaccine 24, 2423e2431.
Schaefer, K.U., Kurtzhals, J.A., Gachihi, G.S., Muller, A.S., Kager, P.A., 1995. A prospective
sero-epidemiological study of visceral leishmaniasis in Baringo District, Rift Valley Prov-
ince, Kenya. Trans. R. Soc. Trop. Med. Hyg. 89, 471e475.
128 K.S. Rock et al.

Seaman, J., Pryce, D., Sondorp, H.E., Moody, A., Bryceson, A.D., Davidson, R.N., 1993.
Epidemic visceral leishmaniasis in Sudan: a randomized trial of aminosidine plus sodium
stibogluconate versus sodium stibogluconate alone. J. Infect. Dis. 168, 715e720.
Seifert, K., Perez-Vitoria, F.J., Stettler, M., Sanchez-Canete, M.P., Castanys, S., Gamarro, F.,
Croft, S.L., 2007. Inactivation of the miltefosine transporter, LdMT, causes miltefosine
resistance that is conferred to the amastigote stage of Leishmania donovani and persists in
vivo. Int. J. Antimicrob. Agents 30, 229e235.
Sharma, N.L., Mahajan, V.K., Negi, A.K., Verma, G.K., 2009. The rK39 immunochromatic
dipstick testing: a study for K39 seroprevalence in dogs and human leishmaniasis patients
for possible animal reservoir of cutaneus and visceral leishmaniasis in endemic focus of
Satluj river valley of Himachal Pradesh (India). Indian J. Dermatol. Venereol. Leprol.
75, 52e55.
Sherlock, I.A., 1996. Ecological interactions of visceral leishmaniasis in the state of Bahia,
Brazil. Mem orias do Inst. Oswaldo Cruz 91, 671e683.
Singh, O.P., Hasker, E., Sacks, D., Boelaert, M., Sundar, S., April 2014. Asymptomatic Leish-
mania infection: a new challenge for Leishmania control. Clin. Infect. Dis. 58, 1424e
1429.
Singh, R., Lal, S., Saxena, V.K., August 2008. Breeding ecology of visceral leishmaniasis vec-
tor sandfly in Bihar state of India. Acta Trop. 107, 117e120.
Singh, R.P., Picado, A., Alam, S., Hasker, E., Singh, S.P., Ostyn, B., Chappuis, F.,
Sundar, S., Boelaert, M., August 2012. Post-Kala-azar dermal leishmaniasis in visceral
leishmaniasis-endemic communities in Bihar, India. Trop. Med. Int. Health 17,
1345e1348.
Siriwardana, H.V.Y.D., Chandrawansa, P.H., Sirimanna, G., Karunaweera, N.D., August
2012. Leishmaniasis in Sri Lanka: a decade old story. Sri Lankan J. Infect. Dis. 2, 2e12.
Smith, D.L., Battle, K.E., Hay, S.I., Barker, C.M., Scott, T.W., McKenzie, F.E., April 2012.
Ross, Macdonald, and a theory for the dynamics and control of mosquito-transmitted
pathogens. PLoS Pathog. 8, e1002588.
Smith, D.L., McKenzie, F.E., Snow, R.W., Hay, S.I., 2007. Revisiting the basic reproduc-
tive number for malaria and its implications for malaria control. PLoS Biol. 5, e42.
Smith, D.L., Perkins, T.A., Reiner, R.C., Barker, C.M., Niu, T., Chaves, L.F., Ellis, A.M.,
George, D.B., Le Menach, A., Pulliam, J.R.C., Bisanzio, D., Buckee, C., Chiyaka, C.,
Cummings, D.A.T., Garcia, A.J., Gatton, M.L., Gething, P.W., Hartley, D.M.,
Johnston, G., Klein, E.Y., Michael, E., Lloyd, A.L., Pigott, D.M., Reisen, W.K.,
Ruktanonchai, N., Singh, B.K., Stoller, J., Tatem, A.J., Kitron, U., Godfray, H.C.J.,
Cohen, J.M., Hay, S.I., Scott, T.W., March 2014. Recasting the theory of mosquito-
borne pathogen transmission dynamics and control. Trans. R. Soc. Trop. Med. Hyg.
108, 185e197.
Solano-Gallego, L., Koutinas, A., Mir o, G., Cardoso, L., Pennisi, M.G., Ferrer, L.,
Bourdeau, P., Oliva, G., Baneth, G., October 2009. Directions for the diagnosis, clinical
staging, treatment and prevention of canine leishmaniosis. Vet. Parasitol. 165, 1e18.
Solano-Gallego, L., Morell, P., Arboix, M., Alberola, J., Ferrer, L., February 2001.
Prevalence of Leishmania infantum infection in dogs living in an area of canine leishman-
iasis endemicity using PCR on several tissues and serology. J. Clin. Microbiol. 39,
560e563.
Srinivasan, R., Panicker, K.N., 1992. Seasonal abundance, natural survival and resting behav-
iour of Phlebotomus papatasi (Diptera: Phlebotomidae) in Pondicherry. Indian J. Med.
Res. A 95, 207e211.
Stauch, A., Duerr, H.-P., Dujardin, J.-C., Vanaerschot, M., Sundar, S., Eichner, M.,
December 2012. Treatment of visceral leishmaniasis: model-based analyses on the
spread of antimony-resistant L. donovani in Bihar, India. PLoS Neglected Trop. Dis.
6, e1973.
Progress in the Mathematical Modelling of Visceral Leishmaniasis 129

Stauch, A., Duerr, H.-P., Picado, A., Ostyn, B., Sundar, S., Rijal, S., Boelaert, M.,
Dujardin, J.-C., Eichner, M., April 2014. Model-based investigations of different vec-
tor-related intervention strategies to eliminate visceral leishmaniasis on the Indian
subcontinent. PLoS Neglected Trop. Dis. 8, e2810.
Stauch, A., Sarkar, R.R., Picado, A., Ostyn, B., Sundar, S., Rijal, S., Boelaert, M.,
Dujardin, J.-C., Duerr, H.-P., November 2011. Visceral leishmaniasis in the Indian sub-
continent: modelling epidemiology and control. PLoS Neglected Trop. Dis. 5, e1405.
Sudarshan, M., Singh, T., Singh, A.K., Chourasia, A., Singh, B., Wilson, M.E.,
Chakravarty, J., Sundar, S., December 2014. Quantitative PCR in epidemiology for
early detection of visceral leishmaniasis cases in India. PLoS Neglected Trop. Dis. 8,
e3366.
Sudarshan, M., Sundar, S., September 2014. Parasite load estimation by qPCR differentiates
between asymptomatic and symptomatic infection in Indian cisceral leishmaniasis. Diag-
nostic Microbiol. Infect. Dis. 80, 40e42.
Sudarshan, M., Weirather, J.L., Wilson, M.E., Sundar, S., July 2011. Study of parasite kinetics
with antileishmanial drugs using real-time quantitative PCR in Indian visceral
leishmaniasis. J. Antimicrob. Chemother. 66, 1751e1755.
Sultana, A., Zakaria, S.M., Bhuiyan, S.I., Habib, A., Dey, S.K., Rahman, M., Basher, A., July
2012. Spectrum of skin lesions of post-Kala-azar dermal leishmaniasis in Kala-azar
endemic areas of Bangladesh. Mymensingh Med. J. 21, 529e532.
Sundar, S., 2001. Drug resistance in Indian visceral leishmaniasis. Trop. Med. Int. Health 6,
849e854.
Sundar, S., Jha, T.K., Mishra, M., Singh, V.P., Buffels, R., 2003. Single-dose liposomal
amphotericin B in the treatment of visceral leishmaniasis in India: a multicenter study.
Clin. Infect. Dis. 37, 800e804.
Sundar, S., Mehta, H., Suresh, A.V., Singh, S.P., Rai, M., Murray, H.W., 2004. Amphoter-
icin B treatment for indian visceral leishmaniasis: conventional versus lipid formulations.
Clin. Infect. Dis. 38, 377e383.
Sundar, S., More, D.K., Singh, M.K., Singh, V.P., Sharma, S., Makharia, A., Kumar, P.C.,
Murray, H.W., 2000. Failure of pentavalent antimony in visceral leishmaniasis in India:
report from the center of the Indian epidemic. Clin. Infect. Dis. 31, 1104e1107.
Sundar, S., Rai, M., Chakravarty, J., Agarwal, D., Agrawal, N., Vaillant, M., Olliaro, P.,
Murray, H.W., October 2008. New treatment approach in indian visceral leishmaniasis:
single-dose liposomal amphotericin B followed by short-course oral miltefosine. Clin.
Infect. Dis. 47, 1000e1006.
Sundar, S., Singh, A., Rai, M., Prajapati, V.K., Singh, A.K., Ostyn, B., Boelaert, M.,
Dujardin, J.-C., Chakravarty, J., July 2012. Efficacy of miltefosine in the
treatment of visceral leishmaniasis in India after a decade of use. Clin. Infect. Dis. 55,
543e550.
Svobodova, M., Votypka, J., Nicolas, L., Volf, P., 2003. Leishmania tropica in the black rat
(Rattus rattus): persistence and transmission from asymptomatic host to sand fly vector
Phlebotomus sergenti. Microbes Infect. 5, 361e364.
Tayeh, A., Jalouk, L., Al-Khiami, A.M., 1997. A Cutaneous Leishmaniasis Control Trial Us-
ing Pyrethroid-impregnated Bednets in Villages Near Aleppo, Syria. WHO. Tech. Rep.
WHO/LEISH/97.41.
Thakur, C.P., Kanyok, T.P., Pandey, A.K., Sinha, G.P., Zaniewski, A.E., Houlihan, H.H.,
Olliaro, P., July 2000. A prospective randomized, comparative, open-label trial of the
safety and efficacy of paromomycin (aminosidine) plus sodium stibogluconate versus so-
dium stibogluconate alone for the treatment of visceral leishmaniasis. Trans. R. Soc.
Trop. Med. Hyg. 94, 429e431.
Thakur, C.P., Kumar, K., August 1992. Post Kala-azar dermal leishmaniasis: a neglected
aspect of Kala-azar control programmes. Ann. Trop. Med. Parasitol. 86, 355e359.
130 K.S. Rock et al.

Tiwary, P., Kumar, D., Mishra, M., Singh, R.P., Rai, M., Sundar, S., April 2013. Seasonal
variation in the prevalence of sand flies infected with Leishmania donovani. PLoS One 8,
e61370.
Topno, R.K., Das, V.N.R., Ranjan, A., Pandey, K., Singh, D., Kumar, N., Siddiqui, N.A.,
Singh, V.P., Kesari, S., Bimal, S., Kumar, A.J., Meena, C., Kumar, R., Das, P.,
September 2010. Asymptomatic infection with visceral leishmaniasis in a disease-
endemic area in Bihar, India. Am. J. Trop. Med. Hyg. 83, 502e506.
Travi, B.L., Montoya, J., Gallego, J., Jaramillo, C., Llano, R., Velez, I.D., 1996. Bionomics of
Lutzomyia evansi (Diptera: Psychodidae) vector of visceral leishmaniasis in Northern
Colombia. J. Med. Entomol. 33, 278e285.
Travi, B.L., Tabares, C.J., Cadena, H., Ferro, C., Osorio, Y., 2001. Canine visceral leishman-
iasis in Colombia: relationship between clinical and parasitologic status and infectivity for
sand flies. Am. J. Trop. Med. Hyg. 64, 119e124.
Uranw, S., Ostyn, B., Rijal, A., Devkota, S., Khanal, B., Menten, J., Boelaert, M., Rijal, S.,
December 2011. Post-Kala-azar dermal leishmaniasis in Nepal: a retrospective cohort
study (2000e2010). PLoS Neglected Trop. Dis. 5, e1433.
van Griensven, J., Balasegaram, M., Meheus, F., Alvar, J., Lynen, L., Boelaert, M., March
2010. Combination therapy for visceral leishmaniasis. Lancet Infect. Dis. 10, 184e194.
Verma, N., Singh, D., Pandey, K., Das, V.N.R., Lal, C.S., Verma, R.B., Sinha, P.K.,
Das, P., November 2013. Comparative evaluation of PCR and imprint smear micro-
scopy analyses of skin biopsy specimens in diagnosis of macular, papular, and mixed
papulo-nodular lesions of post-Kala-azar dermal leishmaniasis. J. Clin. Microbiol. 51,
4217e4219.
Verma, S., Kumar, R., Katara, G.K., Singh, L.C., Negi, N.S., Ramesh, V., Salotra, P.,
April 2010. Quantification of parasite load in clinical samples of leishmaniasis patients:
IL-10 level correlates with parasite load in visceral leishmaniasis. PLoS One 5, e10107.
Viana, G.M.C., Nascimento, M.D.S.B., Neto, D.J.A., Rabelo, E.M.F., Binda J unior, J.R.,
Santos J unior, O.M., Santos, A.C., Galv~ao, C.S., Guimar~aes, R.S., 2011. Anti-Leish-
mania titers and positive skin tests in patients cured of Kala-azar. Braz. J. Med. Biol.
Res. 44, 62e65.
Weirather, J.L., Jeronimo, S.M.B., Gautam, S., Sundar, S., Kang, M., Kurtz, M.A.,
Haque, R., Schriefer, A., Talhari, S., Carvalho, E.M., Donelson, J.E., Wilson, M.E.,
October 2011. Serial quantitative PCR assay for detection, species discrimination,
and quantification of Leishmania spp. in human samples. J. Clin. Microbiol. 49, 3892e
3904.
WHO, 1982. Manual on Environmental Management for Mosquito Control. Tech.
Rep. 66.
WHO, 2002. Global Burden of Disease in 2002: Data Sources, Methods and Results. WHO.
Tech. Rep. 54.
WHO, 2010a. Control of the Leishmaniasis. WHO. Tech. Rep. 949.
WHO, 2010b. Monitoring and Evaluation Tool Kit for Indoor Residual Spraying. WHO.
Tech. rep.
WHO, February 2013. Sustaining the Drive to Overcome the Global Impact of Neglected
Tropical Diseases. WHO. Tech. Rep. WHO/HTM/NTD/2013.1.
Williams, B.G., Dye, C., 1997. Infectious disease persistence when transmission varies
seasonally. Math. Biosci. 145, 7e88.
Woolhouse, M.E., Dye, C., Etard, J.-F., Smith, T., Charlwood, J.D., Garnett, G.P.,
Hagan, P., Hii, J.L., Ndhlovu, P.D., Quinnell, R.J., Watts, C.H., Chandiwana, S.K.,
Anderson, R.M., January 1997. Heterogeneities in the transmission of infectious agents:
implications for the design of control programs. PNAS 94, 338e342.
Wylie, C.E., Carbonell-Anto~ nanzas, M., Aiassa, E., Dhollander, S., Zagmutt, F.J.,
Brodbelt, D.C., Solano-Gallego, L., November 2014. A systematic review of the efficacy
Progress in the Mathematical Modelling of Visceral Leishmaniasis 131

of prophylactic control measure for naturally-occurring canine leishmaniosis, part I:


Vaccinations. Prev. Vet. Med. 117, 7e18.
Yaghoobi-Ershadi, M.R., Moosa-Kazemi, S.H., Zahraei-Ramazani, A.R., Jalai-Zand, A.R.,
Akhaven, A.A., Arandian, M.H., Abdoli, H., Houshmand, B., Nadim, A., Hosseini, M.,
2006. Evaluation of deltamethrin-impregnated bed nets and curtains for control of zoo-
notic cutaneous leishmaniasis in a hyperendemic area of Iran. Bull. Soc. Pathol. Exot. 99,
43e48.
Zhi-Biao, X., 1989. Present situation of visceral leishmaniasis in China. Parasitol. Today 5,
224e228.
Zijlstra, E.E., El-Hassan, A.M., Ismael, A., Ghalib, H.W., December 1994. Endemic Kala-
azar in Eastern Sudan: a longitudinal study on the incidence of clinical and subclinical
infection and post-Kala-azar dermal leishmaniasis. Am. J. Trop. Med. Hyg. 51, 826e
836.
Zijlstra, E.E., Musa, A.M., Khalil, E.A.G., El-Hassan, I.M., El-Hassan, A.M., February 2003.
Post-Kala-azar dermal leishmaniasis. Lancet Infect. Dis. 3, 87e98.
CHAPTER THREE

Soil-Transmitted Helminths:
Mathematical Models of
Transmission, the Impact of Mass
Drug Administration and
Transmission Elimination Criteria
J.E. Truscott*, x, 1, H.C. Turner*, x, S.H. Farrell*, x, R.M. Anderson*, x
*London Centre for Neglected Tropical Disease Research, London, United Kingdom
x
School of Public Health, Imperial College London, London, United Kingdom
1
Corresponding author: E-mail: j.truscott@imperial.ac.uk

Contents
1. Introduction 134
1.1 Soil-transmitted helminth life cycles 136
1.2 Diagnosis of infection 140
2. Key Epidemiological Features and Processes 141
2.1 Age intensity profiles and mixing 141
2.2 Parasite aggregation within hosts 143
2.3 Density-dependent processes e fecundity 145
2.4 Density-dependent processes e sexual reproduction 147
2.5 Parameter assignments for the key biological processes in transmission 148
and treatment
2.6 Parameter estimation 151
2.7 Control policy for soil-transmitted helminth treatment by mass drug 157
administration
3. Mathematical Models 158
3.1 Environmental contamination 160
3.2 Age structure 161
3.3 Models of drug treatment at a population level 162
3.4 Acquired immunity 165
3.5 Stochastic models 165
3.6 Model predictions on the control of disease and the likelihood that 167
mass drug administration alone can break transmission for different
soil-transmitted helminth species
3.6.1 Formulation of a stochastic model 177
3.7 Health economics 180
3.8 Cost data 185
3.9 Effectiveness metrics 185
Advances in Parasitology, Volume 94
© 2016 Elsevier Ltd.
j
ISSN 0065-308X
http://dx.doi.org/10.1016/bs.apar.2016.08.002 All rights reserved. 133
134 J.E. Truscott et al.

3.10 Interventions to add to community-based drug treatment 185


4. Conclusions 186
Acknowledgements 189
References 189

Abstract
Infections caused by soil-transmitted helminthias (STHs) affect over a billion people
worldwide, causing anaemia and having a large social and economic impact through
poor educational outcomes. They are identified in the World Health Organization
(WHO) 2020 goals for neglected tropical diseases as a target for renewed effort to
ameliorate their global public health burden through mass drug administration
(MDA) and water and hygiene improvement. In this chapter, we review the underlying
biology and epidemiology of the three causative intestinal nematode species that are
mostly considered under the STH umbrella term. We review efforts to model the
transmission cycle of these helminths in populations and the effects of preventative
chemotherapy on their control and elimination. Recent modelling shows that the
different epidemiological characteristics of the parasitic nematode species that make
up the STH group can lead to quite distinct responses to any given form of MDA.
When connected with models of treatment cost-effectiveness, these models are poten-
tially a powerful tool for informing public policy. A number of shortcomings are iden-
tified; lack of critical types of data and poor understanding of diagnostic sensitivities
hamper efforts to test and hence improve models.

List of Abbreviations
DALY disability-adjusted life year
epg eggs per gram of faeces
MDA mass drug administration
NTD neglected tropical disease
PCR polymerase chain reaction
pre-SAC pre-school age children
qPCR quantitative PCR
R0 basic reproduction ratio or number
SAC school age children
STH soil-transmitted helminth
WASH water, sanitation and hygiene
WHO World Health Organization

1. INTRODUCTION
The most common neglected tropical diseases (NTDs) are the soil-
transmitted helminthiases, which are caused by the intestinal parasitic
nematodes Ascaris lumbricoides, Trichuris trichiura and the hookworm species,
Ancylostoma duodenale and Necator americanus. (Within this chapter, Ascaris and
Soil-Transmitted Helminths 135

Trichuris will be taken to refer to the human species, A. lumbricoides and


T. trichiura, respectively and hookworm will denote Ancylostoma duodenale
and/or N. americanus, which are rarely distinguished in population-level
studies.) It is estimated that 5.3 billion people worldwide, including one
billion school-aged children (SAC), live in areas of endemic infection for
at least one of these soil-transmitted helminth (STH) species (Pullan and
Brooker, 2012). The impact of infection on the host is rarely acute and usu-
ally long term and cumulative in nature. STH infections rarely cause death,
but chronic and intense infections can contribute to malnutrition, anaemia,
and can also adversely affect physical and cognitive development in child-
hood (Brooker et al., 2010; World Bank, 2003; Albonico et al., 2008). As
a consequence, measuring the burden of morbidity is difficult, and further
complicated by the fact that the highest prevalences are in low-income
countries (Brooker et al., 2006). Within these countries, spatial distribution
of STH burden remains highly heterogeneous. This aspect is well illustrated
by the maps to be found at the Global Atlas of Helminth Infections (http://
www.thiswormyworld.org/). The Global Burden of Disease 2010 Study
estimated that 5.19 million disability-adjusted life years (DALYs) are attrib-
utable to STH infections (Pullan et al., 2014b; Murray et al., 2012).
The World Health Organization’s (WHO) policy for STH control
focusses on mass drug administration (MDA) to control and ultimately
eliminate STHs, although efforts are also being made to improve access to
clean water and hygiene (Strunz et al., 2014). MDA strategies identify three
groups, preschool-aged children (pre-SAC), SAC, and women of
childbearing age, on the basis that heavy infection in these groups will
have a detrimental impact on anaemia, child growth, and development.
The current WHO guidelines focus on SAC, both for monitoring infection
and as a target for treatment, although treatment of pre-SAC and women of
childbearing age is also recommended where sustainable delivery
mechanisms exist, especially in areas of intense transmission (WHO,
2012). The guidelines recommend treating SAC annually where any STH
prevalence falls between 20% and 50% and twice a year where it exceeds
50% (WHO, 2006).
The long-term nature of STH-associated morbidity and the clustered
nature of STH worm burden in hosts mean that measuring the impact of
treatment is difficult. A recent systematic review found very few studies of
the highest quality (randomized or controlled clinical trials) to demonstrate
any cognitive or educational benefits of regular chemotherapy (Taylor-
Robinson et al., 2015). A number of studies that fall outside this strict
136 J.E. Truscott et al.

selection criterion offer evidence of long-term educational and economic


benefits due to deworming (Hicks et al., 2015; Ahuja et al., 2015). Howev-
er, to date, there are insufficient high-quality studies to confirm a
correlation.
In 2012, the WHO announced its intention of scaling up MDA for soli-
transmitted helminthiases to treat 75% of the pre-SAC and SAC population
in need by 2020 (WHO, 2012). This decision was further endorsed by the
London Declaration on Neglected Tropical Diseases (Uniting to Combat
NTDs, 2012). Progress has been good in some areas, but less so in others.
In 2013, global therapeutic coverage of those in need was 39% for SAC
and 49% for pre-SAC (World Health Organization, 2015). Data for more
recent years are yet to be published by the WHO, but a huge gain in
coverage is not expected. Recent research has suggested a change in policy
for the control of STHs by MDA, to broaden coverage to include adults
who often harbour significant reservoirs of infection especially in the case
of hookworm (Anderson et al., 2015).
This chapter examines recent progress in the development and applica-
tion of mathematical models in the study of the transmission dynamics and
control by MDA of the STHs. Work in this area up to the 1990s is reviewed
by Anderson and May (1992). More recent developments are described in
Anderson et al. (2013) and Truscott et al. (2014b). Our focus in this paper
is the development of probability models to relate STH worm burdens in
hosts to egg counts in faeces, deterministic elimination criteria to break
transmission and the associated coverage of MDA, and comparisons between
the predictions of individual-based stochastic models of transmission with
their deterministic counterparts.

1.1 Soil-transmitted helminth life cycles


Soil-transmitted helminths live in the intestine of their (human) hosts and
their eggs are passed in the faeces of infected persons (Fig. 1). If an infected
person defecates not using a toilet or latrine (near bushes, in a garden or field)
or if the faeces of an infected person are used as fertilizer, eggs are deposited
on soil. The STH parasites have a direct life cycle which requires no inter-
mediate hosts or vectors, so it is also possible for eggs to be transmitted
through direct contact or food preparation as a consequence of poor hygiene
as well as within the household.
The STH species differ markedly in their behaviour outside the host.
Eggs initially undergo a period of maturation for 2e3 weeks, after which
they become infectious. Ascaris and Trichuris eggs can remain viable in the
Soil-Transmitted Helminths
137
Figure 1 Diagrammatic representation of the life cycles of Ascaris, Trichuris and hookworm. From CDC, Creative Commons (Center for Disease
Control and Prevention, 2015. Parasites [Online]. Available at: http://www.cdc.gov/parasites/.).
138 J.E. Truscott et al.

soil for several months. Hookworm eggs hatch into larvae which can survive
for several weeks without finding a new host, depending on environmental
conditions. To infect a new host (or reinfect the original host), Ascaris and
Trichuris eggs need to be ingested. This can happen when hands or fingers
that have contaminated dirt on them are put in the mouth or by consuming
vegetables and fruits that have not been carefully cooked, washed or peeled.
The larvae hatch in the intestine and penetrate the intestinal wall into the
bloodstream (Jia et al., 2012). In the case of Ascaris, the larvae are carried
via the portal and systemic circulation to the lungs, from which they ascend
the bronchial tree, reach the throat, and are swallowed. Upon reaching the
small intestine, they develop into adult worms (https://www.cdc.gov/
parasites/ascariasis/biology.html). In the case of Trichuris, there is no pulmo-
nary passage, but the larvae mature and establish themselves as adults in the
colon (http://www.cdc.gov/parasites/whipworm/biology.html). Hook-
worm larvae must enter through the skin and infection is transmitted
primarily by walking barefoot on contaminated soil (although An. duodenale
can also be transmitted through the ingestion of larvae) (http://www.cdc.
gov/parasites/hookworm/biology.html). Figure 1 illustrates the life cycle
of the species considered under the STH term.
The species also show differing responses to environmental conditions.
Humidity is an important factor for all species, with Ascaris and Trichuris
eggs being unable to embryonate at less than 50% humidity. Maximum
survival rates for hookworm larvae occur between 20 C and 30 C and
development rates for eggs peak close to 30 C (Brooker et al., 2006). Since
survival to infect a new host is an essential part of the transmission cycle,
environmental conditions can be expected to have a strong impact on
the components of the species’ transmission cycles that take place outside
the human host. Work examining the correlations between environmental
conditions and infection prevalence indicates that hookworm is
considerably more temperature tolerant than Ascaris or Trichuris.
Prevalences for the latter fall off rapidly for mean land surface temperatures
above 30 C, while hookworm maintains high prevalences up to about
45 C (Brooker et al., 2006). This hardiness can probably be ascribed to
the ability of the larvae, being motile, to find protection from local
extremes of temperature. This temperature sensitivity can be used to
predict large-scale spatial distributions of infection from satellite-derived
environmental data. Temperature and humidity in a given location can
also vary strongly on an annual cycle. Helminth infection rates of domestic
animals have been noted to increase markedly during rainy seasons
Soil-Transmitted Helminths 139

(Lima, 1998; Sissay et al., 2007; Nwosu et al., 2007). Although little
corresponding work has so far been directed towards human hosts, there
is some evidence that similar effects apply (Niangaly et al., 2012). This
has clear implications for the timing of treatment and monitoring efforts
in regions with strongly seasonal climates.
From ingestion to establishment and maturity takes 2e3 months in all
species, although the details of the process vary between species (Bethony
et al., 2006). Both hookworm and Ascaris reach the lungs via the blood-
stream, before being swallowed and establishing themselves in the small in-
testine. Trichuris eggs are conveyed directly to the small intestine, where they
hatch and then establish themselves in the colon (Knopp et al., 2012). All
three species are dioecious with separate sexes and the production of fertil-
ized eggs requires the presence of male and female worms within the same
host. Eggs are expelled in the faeces to complete the infection cycle. Female
Ascaris worms release eggs into the intestine whether they are fertilized or
not, whereas Trichuris and hookworm are thought to only release fertilized
eggs. This biological property is relevant to a number of issues including the
assessment of the impact of control programmes on sustained transmission
and the inference of worm load from egg counts in faeces.
The human host is also exploited by not only the species already
mentioned, but also by others adapted to other hosts species, giving rise to
the potential for animal reservoirs of infection. Ascaris also infects pigs, in
the form of the species Ascaris suum. However, there is considerable uncer-
tainty as to whether A. suum and A. lumbricoides qualify as truly different spe-
cies. Cross-infection is known to be possible and the cross-infected parasites
are capable of completing their life cycle (Nejsum et al., 2012). Hybridization
has also been observed (Criscione et al., 2007). Given the close domestic
proximity of pigs and humans, it has been argued that there is little justifica-
tion in recognizing two distinct species (Leles et al., 2012). For Trichuris, the
similarity between T. trichiura and Trichuris suis (infecting pigs) makes it diffi-
cult to identify cross-infection, but cases of patent infection of humans with
T. suis have been recorded. Eggs of Trichuris vulpis have also been detected in
human faeces, indicating a zoonosis in dogs. Several hookworm species can
also infect humans. Ancylostoma ceylanicum (natural host: dogs and cats) can
successfully complete its life cycle in humans. Ancylostoma caninum can also
infect humans, but there is no evidence of egg production.
It is not known what contribution zoonosis makes to the overall trans-
mission of STHs, but as the human transmission cycle is controlled by
chemotherapy and improvements in hygiene, a zoonotic infection cycle
140 J.E. Truscott et al.

may allow the parasite to persist. This process is well illustrated by the impact
of canine zoonosis on efforts to eradicate Guinea worm (Dracunculus
medinensis) (Callaway, 2016).

1.2 Diagnosis of infection


Identification of infected persons is traditionally achieved by the examina-
tion of stool samples for the eggs of the three STH species. The most com-
mon method is the Kato-Katz technique, based on a faecal smear followed
by microscopy examination. This was first developed in Japan in the 1950s
(Holland and Kennedy, 2002) and is still by far the commonest diagnostic
technique in use for population-scale studies (see Medley et al., 2016; this
volume). Not coincidently, it is also the diagnostic technique recommended
by the WHO for detection of STH infections (WHO, 2006). A range of
other faecal egg count techniques exist, such as ether concentration (Garcia,
2007), FLOTAC (Cringoli, 2006), mini-FLOTAC (Barda et al., 2013), and
while some have superior sensitivity, they generally require an additional
overhead in terms of equipment and training over Kato-Katz variants.
The sensitivity of the various tests is hard to assess in the absence of a
‘gold standard’ for the presence of infection or the concentration of eggs
in the stool, and studies have used aggregate measures or latent class
methods. Sensitivities for Kato-Katz are generally assessed to be in the range
50e90% for Ascaris, with lower values (20e40%) for Trichuris and hook-
worm. Sensitivity drops markedly at low intensity of infection and there is
much variability between studies (Nikolay et al., 2014; Glinz et al., 2010;
Tarafder et al., 2010). PCR and qPCR techniques are now becoming avail-
able and show much higher levels of sensitivity and specificity than micro-
scopy techniques, pointing to their possible adoption as a gold standard
(Becker et al., 2015; Easton et al., 2016).
Quantitative information on intensity of infection can also be recovered
from microscopy techniques in the form of eggs per gram of faeces (epg).
These show considerable variability across successive measurements
(Anderson and Schad, 1985; Krauth et al., 2012; Sinniah, 1982; Croll
et al., 1982). Intensity of infection can also be directly measured by counting
worms expelled in the days after treatment (Bundy et al., 1987; Elkins et al.,
1986; Bradley et al., 1992). Simple models are available to link an individ-
ual’s worm burden with their egg output. A comparison of model fits indi-
cates that there is considerable variability in egg output per worm as
measured across different studies, possibly arising from lack of diagnostic
rigour or poor standardization (Hall and Holland, 2000). These two sources
Soil-Transmitted Helminths 141

of uncertainty make inferring worm burden from egg output difficult.


qPCR techniques have been shown to correlate well with egg output in
some circumstances and so may become a useful indicator of individual
worm burden (Easton et al., 2016).

2. KEY EPIDEMIOLOGICAL FEATURES AND


PROCESSES
2.1 Age intensity profiles and mixing
Cross-sectional epidemiological profiles of infection prevalence and
intensity (measured by eggs per gram of faeces or worm expulsion methods)
with host age reveal different and characteristic patterns for the three major
STH infections (Fig. 2). Age profiles in both infections with Ascaris and Tri-
churis exhibit convex curves, in which infection intensity peaks in children
and declines in adults, usually more so in the case of Trichuris. In hookworm
infection, the age-specific intensity trend is much more monotonic in struc-
ture, being generally low in children before plateauing or rising more slowly
in adults. Assuming that worm life span is not significantly affected by the
age of the host, these patterns must result from a combination of age-
dependent force of infection or failure of worms to establish, due to acquired
immunity. There is considerable uncertainty about the role played by im-
munity in the life cycle of STHs. All three species trigger a strong immune
response in the host. However, there is no clear evidence that the responses
generated offer any protection, as illustrated by the ability of hosts to become
repeatedly infected throughout their lives (Lamberton and Jourdan, 2015;
Loukas and Prociv, 2001; Barda et al., 2015). In the case of Ascaris, there
is evidence that antibody levels track infection status rather than control it
and that age-specific patterns reflect changes in risk behaviour with time
(Bundy and Medley, 1992). This is in contrast to the case for Schistosoma hae-
matobium, in which the protective nature of the immune response is reflected
in a negative correlation between infection intensity and antibody status
with age, indicating a protective aspect to immune response (Mutapi
et al., 1998). In animal models, protective immunity has been observed
for hookworm reinfection (Davey et al., 2013). Ultimately, the role played
by immunity in the establishment and persistence of soil-transmitted
helminthiases in humans is unclear. Detailed data linking infection intensity
to immune status and the response of both to chemotherapy would greatly
improve our current understanding.
142 J.E. Truscott et al.

Figure 2 Typical ageeintensity profiles for (A) Ascaris, (B) Trichuris and (C) hookworm in
terms of worm burden. Data sources: (A) Ascaris data from Pulicat, South India study
(Elkins, D.B., Haswell-Elkins, M., Anderson, R.M., 1986. The epidemiology and control of
intestinal helminths in the Pulicat Lake region of Southern India. I. Study design and pre-
and post-treatment observations on Ascaris lumbricoides infection. Trans. R. Soc. Trop.
Med. Hyg. 80, 774e792); (B) Trichuris data from St. Lucia Island study (Bundy, D.A.,
Thompson, D.E., Cooper, E.S., Golden, M.H., Anderson, R.M., 1985a. Population dynamics
and chemotherapeutic control of Trichuris trichiura infection of children in Jamaica and St.
Lucia. Trans. R. Soc. Trop. Med. Hyg. 79, 759e764); (C) Hookworm data sets collated by
Brooker and co-workers (Brooker, S., Bethony, J., Hotez, P.J., 2004. Human hookworm
infection in the 21st century. Adv. Parasitol. 58, 197e288), supplemented with data from
West Bengal study (Nawalinski, T., Schad, G.A., Chowdhury, A.B., 1978. Population biology
of hookworms in children in rural West Bengal. I. General parasitological observations. Am.
J. Trop. Med. Hyg. 27, 1152e1161).
Soil-Transmitted Helminths 143

The age profiles observed are therefore most likely the result of different
levels of contact with the infectious material in the environment for hosts
with respect to age class. Hence, the profiles reflect the social structure of
the host community, where eggs are deposited and who comes in contact
with them, as well as the details of the nature of the infectious material. It
may be significant that hookworm, which has a fairly short-lived larval in-
fectious stage, has a different profile to the parasites that are transmitted
through longer-lived eggs.
The significance of the different infection age profiles of the STH species
for the effectiveness of age-group targeted MDA was examined by Anderson
et al. (2013). Intervention strategies are frequently targeted at SAC (5e
15 years of age). Typically, only 30% or less of populations in STH-affected
areas fall into this age category, rising to 50% if pre-SAC are included.
Combining this with mean worm burdens by age group shows that the frac-
tion of the worm population reachable by such strategies may be strongly
dependent on species. Only 15% or less of hookworm burden will fall in
the treated age group, while for Ascaris, this may reach 50%. Treatment
coverage of SAC is usually achieved through the agency of local schools,
but these routinely enrol only 40e90% of the target population (Anderson
et al., 2013). As a result, it is clear that the impact of treatment targeted at
SAC will vary widely across different causative species and, in the case of
hookworm, potentially have a very limited impact (Anderson et al.,
2015). Within models, age-dependent contact and parasite establishment
rates are described by a composite parameter b(a). Given an environmental
intensity of infectious material L(t), individuals at a given time (t) and age (a)
experience a force of infection, L(t)b(a). The worm burden of an individual
of a given age is a result of the worms they have acquired up to that age, less
those that have already died. Hence, the profile reflects both the age-specific
contact rate and the worm life span.

2.2 Parasite aggregation within hosts


A common feature of a wide range of parasitic helminths is a pronounced
aggregation of worms amongst hosts where the variance in the numbers
per host greatly exceeds the mean value (that is, some hosts have far more
worms than others). Some typical patterns based on worm expulsion studies
are recorded in Fig. 3. Whereas for diseases like schistosomiasis and oncho-
cerciasis, worm distributions are difficult to record, except via autopsy
studies (Cheever et al., 1977) or nodulectomy studies (Duerr et al., 2001),
respectively, worm expulsion techniques can be used for the intestinal
144 J.E. Truscott et al.

Figure 3 Frequency distributions of worm numbers per person. (A) Hookworm bur-
dens from West Bengal study (Anderson and Schad, 1985); (B) Ascaris worm burdens
from Pulicat Lake study (Elkins et al., 1986), both studies conducted in India.

helminths and the degree of aggregation measured directly. The process of


expulsion is necessarily more involved and time-consuming than egg-
counting techniques. Studies show that to achieve a high sensitivity for
the test, it is necessary to collect all stool output from the subject over a num-
ber of days; estimates include from the second to the sixth day for 80% of the
burden and up to the seventh day for 97% (Easton et al., 2016; Forrester and
Scott, 1990). These factors are likely to depend also on the type and quantity
of drug given and the species being collected (Williams-Blangero et al.,
1999; Bundy et al., 1985a). Specificity of expulsion is likely to be very
high. A further important point, particularly in the context of detailed
studies, is that, unlike egg counting techniques, expulsion is both a measure-
ment and a treatment and consequently destabilizes the parasite population
being studied.
The distribution of worms amongst hosts is typically negative binomial
in character. The shape parameter of the distribution, k, varies inversely
with the degree of worm clumping (a value above 5 denotes a Poisson
distribution). Fig. 3 shows worm count data for hookworm and Ascaris.
Aggregation is also observed within individual age groups of the popula-
tion, although with slightly varying degrees of aggregation across age
groups (Anderson and May, 1985; Holland et al., 1989; Hall et al.,
1999). Combining a number of individual studies, further details of
worm aggregation emerge. The degree of aggregation appears to be a
Soil-Transmitted Helminths 145

function of the overall worm prevalence in the population. The higher the
prevalence, the larger the value of k and the less aggregated the worm
population is among hosts. This effect has been observed for both
hookworm and Ascaris (Lwambo et al., 1992; Guyatt et al., 1990). As a
result, as prevalence drops, worms appear to be increasingly concentrated
in just a few individuals.
The main mechanisms proposed to account for aggregation are host
predisposition to heavy or light infection and/or environmental heteroge-
neity (Anderson and Medley, 1985). The predisposition model assumes
that individual hosts have a wide range of rates of parasite exposure/estab-
lishment, either as a result of genetic or immunological factors in the pop-
ulation, or through experiencing a different force of infection through
different patterns of behaviour. If the rates of parasite establishment are
gamma-distributed within the population and parasite death rates are con-
stant with time, then a negative binomial distribution of parasites will arise
across the host population. Environmental heterogeneity can assume that
the variability lies in the clustering of infectious material, such that the
number of worms established from an infection event is aggregated
(Walker et al., 2010a,b). Other forms of environmental heterogeneity in
exposure may arise, where, for example, people are exposed nonrandomly
depending on spatial location and other behavioural factors, but in this
case, it will typically manifest itself as predisposition to infection dependent
on behaviour and spatial location. Theses generating processes have been
investigated by Anderson and Medley using an individual-based stochastic
model, where they recover the negative binomial distribution through a
log-distribution of egg clumping (Anderson and Medley, 1985). A poten-
tially important difference between the two approaches to aggregation is
that an egg-clumping paradigm leads to significantly slower recovery of
worm burden in the host community after treatment. In reality, however,
both these mechanisms may be playing a role in any given community.

2.3 Density-dependent processes e fecundity


For STH species, it is believed that the main density-dependent process acts
on the parasite’s fecundity based on patterns observed in epidemiological
studies involving faecal egg counts and worm expulsion. As the number
of worms increases within a host, the rate at which each female worm
produces eggs decreases. The overall egg output may continue to rise
with female worm burden or may begin to fall, for some forms of density
dependence. This negative density dependence is probably the main
146 J.E. Truscott et al.

mechanism that limits the overall worm burden in a community (Walker


et al., 2009). A secondary effect is that the efficacy of chemotherapeutic in-
terventions on transmission can be highly nonlinear, where, for example,
decreasing the parasite load by 50%, will not decrease transmission by
50% due to the presence of density-dependent egg production. Conse-
quently, the indirect benefits of treatment on the reduction in transmission
will depend on the number and distribution of parasites between hosts
(Anderson and May, 1992).
Data illustrating this process are recovered from individuals in a two-
stage process, whereby an initial round of egg intensity measurements (typi-
cally using the Kato-Katz technique) is followed up with worm expulsion
and counting. Fig. 4 shows such a data sets for hookworm and Ascaris. Aver-
aged over discrete age groups, the downward trend in fecundity is clear for
the Ascaris data. Looking at the raw data, as exemplified by the hookworm
data in panel A, the large variability in egg output for a given worm burden is
evident.

Figure 4 (A) Egg output (epg) per female hookworm plotted against total worm
burden. (B) Egg output per female Ascaris worm plotted against total female worm
population in a given host. (A) Data from Zimbabwe study (Bradley, M., Chandiwana,
S.K., Bundy, D.A., Medley, G.F., 1992. The epidemiology and population biology of Necator
americanus infection in a rural community in Zimbabwe. Trans. R. Soc. Trop. Med. Hyg.
86, 73e76). (B) Data from Nigeria study (Holland, C.V., Asaolu, S.O., Crompton, D.W.,
Stoddart, R.C., Macdonald, R., Torimiro, S.E., 1989. The epidemiology of Ascaris lum-
bricoides and other soil-transmitted helminths in primary school children from Ile-Ife,
Nigeria. Parasitology 99, 275e285).
Soil-Transmitted Helminths 147

Within most of the models described here, an exponential description is


used, such that the mean egg production from a single female worm among
a population of n females is given by legn. Hence, the mean egg production
from a host with n females would be lnegn. Other models of fecundity
have been used, such as a power law dependence on worm burden (egg pro-
duction/female ¼ lng) and may in some instances fit the data slightly bet-
ter (Croll et al., 1982). However, they are considerably more difficult to
integrate into a simple closed-form model framework.

2.4 Density-dependent processes e sexual reproduction


The soil-transmitted helminth parasites exist as distinct sexes within the host,
so it is necessary for a female worm to share a host with male worms in order
for viable, fertilized eggs to be generated. As a result, the probability of being
co-established with a male of the species can be very low when worm bur-
dens are low in the population, giving a positive density dependence on
mean worm burden for fertilized egg output. Fig. 5 illustrates the resulting
relationship between worm burden and egg fertilization for Ascaris (from
Seo et al. (1979)). The fraction of unfertilized egg passers drops rapidly to
less than 10% for more than five detected worms. A consequence of this

Figure 5 Fraction of people passing fertilized and unfertilized egg as a function of the
measured worm burden of Ascaris lumbricoides. Data taken from Seo, B.S., Cho, S.Y., Chai,
J.Y., 1979. Egg discharging patterns of Ascaris lumbricoides in low worm burden cases.
Korean J. Parasitol. 17, 98e104.
148 J.E. Truscott et al.

dependency is the theoretical existence of a breakpoint for the parasite in a


community such that for average worm burden below a critical level, there
are insufficient fertilized female worms to support the parasite population
and it will be eliminated from the host population in time (Anderson and
May, 1992; Macdonald, 1965; May, 1977). The form of mating for STH
parasites is generally assumed to be polygamous, since, unlike schistosomes,
the parasites are not found as maleefemale pairs as established adults. How-
ever, the existence of a breakpoint is quite general across different forms of
mating.
Mathematically, the dependence of fertile egg production on parasite ag-
gregation and mode of reproduction has been analysed in detail, leading to a
mating probability factor multiplying the total egg output from a given pop-
ulation, 4 (May, 1977; Leyton, 1968); effectively, the fraction of egg output
that is fertilized. For STHs, the reproductive strategy is usually assumed to be
polygamous (see disease description section), where the presence of a single
male allows the fertilization of all females present,
 
1 þ M ð1  zÞ=k kþ1
4ðM ; z; kÞ ¼ 1  (1)
1 þ M ð2  zÞ=k
With the inclusion of sexual reproduction, it becomes important to be
clearer about what the variable M represents. In the above and all equations
within this chapter, M refers to the female worm burden, with the assump-
tion that any given worm in a host has an equal probability of being male
or female. Parameter z in Eqn (1) is defined in terms of density-dependent
fecundity, as explained in the text following the expressions given in
Eqn (7).
In general, given the aggregated nature of STH parasites, the mating
probability term is close to 1 for all but the lowest worm burdens and is
therefore often ignored. However, as we have recently shown, sexual repro-
duction can have a significant impact in the context of regular mass drug
treatment and elimination (Anderson et al., 2013; Truscott et al., 2014a).

2.5 Parameter assignments for the key biological processes


in transmission and treatment
In terms of structure, the mathematical models describing the transmission
dynamics of these different STH species are identical due to the common
direct life cycle structure. Differences in the mechanisms of worm estab-
lishment are not significant in the context of population-level descriptions
Soil-Transmitted Helminths 149

of changing parasite burden, although this might not be the case if immu-
nity or some form of prophylactic treatment (immunization) were to have
a significant effect on the parasite establishment process in the human host.
In any event, it is notable that the time delay from infection to establish-
ment in the intestine as an adult worm is similar in all three species and
generally much shorter than the life span of the worm in the host and so
would contribute little to differentiating the species (Anderson and May,
1985).
A key problem in estimating parameter values for STH parasites is the
‘indirect’ nature of the majority of the data. The central feature of any
model is the worms; their acquisition, death and their production of
new infectious material. The majority of data, however, concerns the
detection of eggs in stool samples. Most frequently, epidemiological
data are available only in the form of summary statistics, such as the
prevalence (fraction infected) or the mean intensity in terms of egg counts
within a population or age category. Such data are typically cross-
sectional by age or age group. It is seldom longitudinal over time.
More usefully, this kind of data is sometimes available in its raw form
as readings or sets of readings from individuals of egg counts (usually
only directly from the original authors). In the case of Ascaris, some of
the available data can also distinguish between fertilized and unfertilized
eggs. The paucity of quality data in this area of epidemiological study,
especially raw data from each person sampled, highlights the need for
international electronic registers of collected information as is often
organized in other fields of epidemiological study.
Much more rarely, worm expulsions are carried out on a subset of the
human population, subsequent to egg intensity measurements, providing
worm counts for individuals. In some cases, these data can be further
stratified by worm sex and weight. However, the time, expense and la-
bour intensiveness of this process means that expulsions are rarely done
as a matter of course in MDA impact monitoring programmes, but
only within the context of specific research epidemiological studies. As
a result, little of the available data are of great value in estimating key pa-
rameters that influence the impact of MDA programmes and define
breakpoints in transmission or therapeutic coverage levels required to
halt transmission.
Tables 1 and 2 document current estimates of the key population param-
eters that influence the transmission dynamics of STH species. Drug efficacy
150 J.E. Truscott et al.

Table 1 Population parameters, development rates and life expectancies of


parasites and free-living infective stages
Ascaris Trichuris
Parameter lumbricoides trichiura Hookworm Sources

Infective stage Ova Ova Larvae


Egg production 10,000e 2,000e 3,000e Anderson and May
(eggs/female 200,000 20,000 20,000 (1982), Bundy
worm/day) and Cooper
(1989)
and Crompton
(2001)
Life expectancy 28e84 days 10e30 days 3e10 days Anderson and May
of free-living (1982), Bundy
infective stages and Cooper
(1989)
and Crompton
(2001)
Adult life span 1e2 years 1e2 years 3e4 years Anderson and May
(1992)
Pre-patency (adult 50e80 days 50e84 days 28e50 days Anderson and May
development to (1992)
sexual maturity)
Larval development 8e37 days 20e100 2e14 days Smith and Schad
time to infective days (1989), Nwosu
stage (1978) and Beer
(1976)
Maximum 35e39 C 37e39 C 40 C Smith and Schad
temperature of (1989), Nwosu
viable (1978) and Beer
development (1976)
Basic reproduction 1e5 4e6 2e3 Anderson and May
number (R0) (1992)
Maximum mean 650e9,900 370 200 Bradley et al.
no. eggs per (1992), Bundy
gram of faeces et al.
per female (1985a), Croll
worm, l et al.
(1982), Holland
et al. (1989)
and Martin et al.
(1983)
Density-dependent 0.25e0.6 e e Hall and Holland
fecundity, g (2000)
(power law)
Soil-Transmitted Helminths 151

Table 1 Population parameters, development rates and life expectancies of


parasites and free-living infective stagesdcont'd
Ascaris Trichuris
Parameter lumbricoides trichiura Hookworm Sources
Density-dependent 0.05e0.0035 0.01 0.03e0.08 Bradley et al.
fecundity, g (1992), Bundy
(exponential) et al.
[/female worm] (1985a), Elkins
et al.
(1986), Sinniah
(1982) and Ye
et al. (1994)
Negative binomial 0.57e0.75 0.3e0.4 0.23e0.64 Anderson and
aggregation Schad
parameter, k (1985), Bradley
et al.
(1992), Bundy
et al. (1985b)
and Ye et al.
(1994)

Table 2 Reported cure rates (a measure of drug efficacy) of albendazole and


mebendazole (Keiser and Utzinger, 2008)
Cure rates (95% Confidence Interval)
Albendazole Mebendazole

Ascaris lumbricoides 88% (79e93%) 95% (91e97%)


Hookworms 72% (59e81%) 15% (1e27%)
Trichuris trichiura 28% (13e39%) 36% (16e51%)
Cure rate: the percentage of individuals who became helminth egg negative following
treatment with an anthelmintic drug. Values are taken from a meta-analysis performed by
Keiser and Utzinger (2008).

figures are given in Table 2 for the three main species (Keiser and Utzinger,
2008).

2.6 Parameter estimation


Cross-sectional study data, in which individual data are supplemented with
host age, is essential to parameterize the age-dependent aspects of a model.
Within the model, age-related infection patterns arise from age-dependent
contact of the host with infectious material in the environment and the
152 J.E. Truscott et al.

probability of parasite establishment upon contact, represented by the


composite parameter, b(a). In practice, a limited number of beta values
are used, corresponding to critical age categories of the host population (in-
fant, pre-SAC, SAC and adult), as dictated by the data. The role of b(a) in
the model is fully elaborated in Section 3.2. If the data can be considered to
come from an undisturbed baseline, then parameters can be estimated using
the equilibrium solution of the model. The generic baseline individual data
set will comprise three possible types of age-specific data; worm counts
{ai,wi}, egg intensities {ai,Ei} and paired egg and worm data {ai,wi,Ei}.
If we define Iw,IwE,IE as the indices of individuals with worm
counts, worm and egg counts and egg counts only, we can write an overall
likelihood for the complete data set {D} given the theta parameters, as the
product of three terms,
LðfDgjqÞ ¼ Lw ðfwgjqw ÞLwE ðfw; EgjqE ÞLE ðfEgjqw ; qE Þ (2)
where
Y
Lw ðfwgjqÞ ¼ Lw ðwi jM ðai ; qw Þ; kÞ
Iw
Y
LwE ðfw; EgjqÞ ¼ LEjw ðEi jwi ; qE Þ
IwE
YX
Xmax
LE ðfEgjqÞ ¼ LEjw ðEi jX; qE ÞLw ðXjM ðai ; qw Þ; kÞ
IE X¼0

In the above equation, qw ¼ {R0,b(a),r(a),g,k,s}, where r (a) is the


age-specific relative contribution of infectious stages to the environmental
reservoir and sigma is the per capita mortality rate of adult worms, is the
set of parameters that control the worm burden age profile for the endemic
(¼ the equilibrium where parasites persist) state of the model, and M(ai,qw) is
the model equilibrium worm burden at age ai (see Eq. (11)). The grouping
qE ¼ {l,g,kE} describes the relationship between egg output of an individ-
ual host and their worm burden. Of the three terms in the likelihood expres-
sion, only the worm burden term, Lw, directly relates worm burden data at a
given age to the model worm burdens. The term LE describes the likelihood
of age-related egg output data in terms of the model worm burdens and the
relationship of worm burden to egg output. Since the transmission model is
formulated in terms of the mean worm burden by age, it is necessary to sum
the weighted contributions from all possible worm burdens in an individual.
The uncertainty in the number of female worms in the host reduces the
Soil-Transmitted Helminths 153

inferential power of egg count data. This effect is further compounded by


the variance in the submodel that connects egg output to worm burden.
Given paired data, {Ei,wi} for individual i, the mean egg output is given
by E ¼ lw40 ðw; gÞ, where 40 (w,g) is the reduction in fecundity due to
worm burden. If pðE; E ; kE Þ is the probability for a measurement E, where
kE parameterizes the variance of the distribution, we can write the likelihood
as
LEjw ðEi jwi ; qE Þ ¼ pðEi ; E ðwi ; l; gÞ; kE Þ (3)
Previous studies of variability in egg output for individuals across multi-
ple measurements show high variance to mean ratios, characteristic of nega-
tive binomial distributions (Anderson and Schad, 1985; Croll et al., 1982).
Preliminary analysis of paired data from expulsion studies for the three
STH species shows that this distribution also accounts well for the variability
in egg output across the population. These high variance to mean ratios also
contribute to the ‘broadness’ of the LE likelihood term, further reducing its
capacity to inform parameter values.
The likelihood term, Lw, which appears also in LE, is not equally infor-
mative with regard to all parameters in qE. In particular, it is particularly
poor at independently specifying values for g and the basic reproduction
number, R0. The reason for this can be seen in the simplest models for
mean worm burden. Anderson and May (1992) quote the following for
mean worm burden in a model without age structure (where z is a function
of density-dependent worm fecundity as explained below),
   
1=ðkþ1Þ 1=ðkþ1Þ
k R0 1 k R0 1
M¼ x (4)
1z g
In practice, maximizing Lw with respect to qE is dominated by matching
the mean worm burden of the observed data to that of the model. As a result,
worm burden data can specify the value of the function of R0, k and g as
shown in Eq. (4), but is very poor at inferring their individual values. For
example, increasing both k and g by 20% would leave M unchanged and
have little effect on the quality of model fit. This effect can be seen clearly
in three different fits to worm burden data from the baseline of the Pulicat
Lake study (Elkins et al., 1986), shown in Table 3. The three columns give
maximum likelihood estimators and 95% credible intervals for three
different scenarios. In the first two columns, values for k and g are taken
from independent fits to egg output/worm data (from Pulicat Lake, India
154 J.E. Truscott et al.

Table 3 Maximum likelihood estimates (MLEs) of model parameters for fits to


baseline data for Ascaris lumbricoides from the Pulicat data set from India (a), and
the Ile-Ife dataset from Nigeria (b). The values in square brackets indicate the 95%
Credible Intervals.
Fit to worm burden Fit to worm burden Fit to complete data
Parameter (units) g ¼ 0.04a g ¼ 0.08b setc

R0 1.7 [1.57e2.28] 2.45 [2.17e4.09] 2.12 [1.68e3.22]


b0e1 0.47 [0.28e1.31] 0.47 [0.29e1.31] 0.22 [0.13e0.55]
b2e4 1.86 [1.21e4.4] 1.86 [1.18e5.4] 1.88 [1.07e3.19]
b15þ 0.56 [0.38e0.92] 0.56 [0.38e1.0] 0.53 [0.36e0.79]
g 0.04 0.08 0.07 [0.048e0.098]
k 0.7 0.7 0.9 [0.76e1.12]
s (/yr) 1 1 1
l (epg/female NA NA 3893 [3227e5146]
worm)
kE NA NA 0.88 [0.73e1.0]
Age categories for the contact and contamination parameter b are 0e1, 2e4, 5e14, 15þ years with
b5e14 ¼ 1. Parameter g estimated from (a) Pulicat study independently, (b) Nigeria study (Holland
et al., 1989) and (c) Pulicat study, jointly with other parameters. Square brackets contain 95% credible
intervals assuming uniform priors, where relevant.

and Ile-Ife, Nigeria, respectively), while in the third, all available data are
included in Eq. (2). While the values for the age-specific contact parameters
are largely preserved across different fits, a range of values for R0 are recov-
ered, dependent on which values of gamma and k are used. In all three cases,
the value of M predicted by Eq. (4) is almost identical ðM x6:4Þ. In general,
an increase in the value of the fecundity parameter is matched by an increase
in R0 and/or k. Note also the (Bayesian-derived) credible intervals associated
with these parameter estimates. Intervals tend to be wide, reflecting the high
variance associated with the negative binomial distribution which underlies
both the distribution of worms in hosts and also egg output from a single
host. It also reflects the quantity of data points, with short age range b values
for infants and pre-SAC being poorly specified. Estimates from fits to the full
data set (column three) show tighter intervals in general.
The corresponding fits to data, based on the maximum likelihood esti-
mates (MLEs), can be judged in Fig. 6, which shows the model equilibrium
worm burden from the model against the worm expulsion data, along with
5- and 95-percentiles for the underlying negative binomial distribution. The
different fits are effectively identical and the predicted 5 and 95% intervals
contain approximately 90% of the data points.
Soil-Transmitted Helminths 155

Figure 6 Equilibrium worm burden generated by the model for different fits to Ascaris
data from the Pulicat study data. Parameters are taken from Table 3. (A) g ¼ 0.04,
(B) g ¼ 0.08, (C and D) Worm burden and egg output fit respectively to full data set
(Table 3 column 4).

Table 4 records the maximum likelihood parameter estimates for paired


data sets for hookworm and Trichuris. The original data come from the
epidemiological studies described in the publications of Bundy et al.
(1985a) on Trichuris and Bradley et al. (1992) on hookworm. In the case
of Trichuris, an individual-based expulsion data set has been used from a
St. Lucia-based study, comprising 119 individual records (Bundy et al.,
1985a). For hookworm, raw data from the original study (based in
Zimbabwe) cannot be recovered and only age group averaged burdens
are available. The lack of individual data means that the aggregation param-
eter, k, cannot be simultaneously calculated and the value is taken from a
separate expulsion arm of the same study. Expulsion data are the source of
the fecundity parameters used.
156 J.E. Truscott et al.

Table 4 MLE parameter values for fit of model equilibrium to worm burden data for
hookworm (Bradley et al., 1992) and Trichuris (Bundy et al., 1985a)
Fit to hookworm worm Fit to Trichuris worm
Parameter (units) data data

R0 2.34 1.77 [1.44e2.5]


binf 0.03 0.3 [0.22e1.48]
bpre-SAC 0.09 1.28 [1.09e5.9]
badults 2.5 0.17 [0.12e1.03]
k 0.35 (not fitted) 0.38 [0.3e0.48]
g 0.08 (not fitted) 0.0035 (not fitted)
s (/yr) 0.5 (not fitted) 1 (not fitted)
Different contact age groups are used for each species. For Trichuris, age breaks are 0, 2, 7, 12, 75; for
hookworm, 0, 2, 5, 15, 75. Square brackets contain 95% credible intervals assuming uniform priors,
where relevant. Intervals not given for hookworm due to poor data.

The quality of the fits for the two species is shown in Fig. 7. Note that
the mean worm burdens across the three species vary by two orders of
magnitude. Of the three parameters in Eq. (4) that govern the value
of the mean worm burden, the bulk of the variation is found in estimates
of the fecundity parameter.

Figure 7 Equilibrium worm burden generated by model and fitted to data for (A) hook-
worm and (B) Trichuris. Parameters in Table 4. Hookworm data are taken from Zimbabwe
study and Trichuris data from St. Lucia study (Bradley, M., Chandiwana, S.K., Bundy, D.A.,
Medley, G.F., 1992. The epidemiology and population biology of Necator americanus
infection in a rural community in Zimbabwe. Trans. R. Soc. Trop. Med. Hyg. 86, 73e76;
Bundy, D.A., Thompson, D.E., Cooper, E.S., Golden, M.H., Anderson, R.M., 1985a. Population
dynamics and chemotherapeutic control of Trichuris trichiura infection of children in Ja-
maica and St. Lucia. Trans. R. Soc. Trop. Med. Hyg. 79, 759e764).
Soil-Transmitted Helminths 157

It is clear that age-stratified egg output or even worm output baseline


data are insufficient to parameterize a transmission model. Different values
of g and k will lead to different fitted values of R0 and these will, in turn,
lead to different conclusions with regard to the resilience of the parasite in
the presence of chemotherapy. Values of g and l, calculated from expul-
sion data, typically have quite a wide range of values, probably arising from
differences in skill and practice in egg and worm counting (Hall and
Holland, 2000). This is particularly true of l, a further strike against the
usefulness of egg output data. Hence, the best data sets to fit to are those
that contain some worm expulsion data within them, ensuring more uni-
form diagnostic practices. Alternatively, high-quality reinfection studies
should provide useful independent data to estimate R0, although these
are scarce.

2.7 Control policy for soil-transmitted helminth treatment


by mass drug administration
The main control strategies for STH infections are regular periodic
MDA targeting pre-SAC and SAC using anthelmintics (predominantly
albendazole and mebendazole). Lymphatic filariasis control, where the
whole community is treated with two drugs including albendazole, also
impacts STH infections and contributes to their control (WHO, 2006;
Keiser and Utzinger, 2008). STH control programmes, which originally
used mobile teams to distribute the drugs, are now predominantly centred
around school-based delivery systems (WHO, 2002; Hotez et al., 2006).
This enables the programmes to be linked with the school educational sys-
tem (WHO, 2002), which has been shown to be highly cost-effective
(Hotez et al., 2006), and a practical method of reaching children in poor
rural areas. In addition, SAC are believed to be most at risk for a large share
of the overall morbidity and associated developmental consequences of
STH infections (Brooker et al., 2010; World Bank, 2003). There has
been a growing recognition of the disease burden in and potential benefit
of treating pre-SAC (Albonico et al., 2008) and more broadly the whole
community especially for the control of hookworm with its predominance
in adult age groups (Anderson et al., 2015).
Although both albendazole and mebendazole have a good efficacy
against A. lumbricoides, mebendazole fails to effectively clear hookworm in-
fections, and neither drug has an adequate efficacy against T. trichiura (Keiser
and Utzinger, 2008; Vercruysse et al., 2011) e with cure rates of 28 and
36%, respectively (see Table 2).
158 J.E. Truscott et al.

The recommendations for treatment of soil-transmitted helminthiases


through MDA by the WHO are summarized in Table 5. Aside from
SAC, WHO also recommends the treatment of pre-SAC, women of child-
bearing age, and adults in certain high-risk occupations (such as tea-pickers
and miners) (WHO, 2006).
Many questions remain regarding how best to deliver STH treatment
programmes to achieve the greatest impact; these include which age groups
should be targeted, how often and how this should change in areas with
different predominant STH species (Anderson et al., 2012). Mathematical
models, along with epidemiological studies, provide a template to investi-
gate optimal control programme design in defined settings.

3. MATHEMATICAL MODELS
The earliest work on the theory of helminth infection was published
in the late 1960s by Tallis and Leyton through the development of stochastic
models of transmission targeted to nematode parasites of sheep and cattle.
These had little impact on practice due to an absence of connections to
data or field epidemiological observations (Leyton, 1968; Tallis and Leyton,
1966, 1969). Using probability generating functions, they derived key pa-
rameters in the distribution of parasite numbers per host (e.g., mean, vari-
ance, skewness measures) and their output of infectious material. They
included in their model stochastic descriptions of worm establishment

Table 5 WHO recommendations for the treatment of STHs with mass drug
administration to SAC and pre-SAC age groups
Control strategy
Prevalence of
any STH infection Preventive Additional
Category at baseline chemotherapy interventions

Schools in 50% Treat all school-age Improve sanitation and


high-risk children (enrolled and water supply; provide
areas nonenrolled) twice a health education
yeara
Schools in 20% Treat all school-age Improve sanitation and
low-risk and <50% children (enrolled and water supply; provide
areas nonenrolled) once a health education
year
pre-SAC, preschool-aged children; SAC, school-aged children; STH, soil-transmitted helminth.
a
If the resources are available and the prevalence is towards the higher end of the interval, a third drug
distribution round might be added (in this case, the frequency will be every 4 months).
Soil-Transmitted Helminths 159

processes, mating dynamics and the acquisition of immunity by the host.


While this approach is general, analytical results could not be obtained
due to the highly nonlinear nature of the stochastic model excepting for
some closed-form expressions for extinction. At about the same time, Mac-
donald identified that a consequence of sexually reproducing parasites
distributed among individual hosts was an inability to generate fertile infec-
tious material when prevalence is low (Macdonald, 1965). This phenome-
non introduces the idea of two stable states, endemic infection and
extinction separated by a breakpoint; namely, a level of infection intensity
or prevalence below which insufficient fertile infectious material is gener-
ated by the parasites to maintain a viable transmission cycle.
Anderson and May introduced much more general descriptions of hel-
minth population dynamics and melded into the model descriptions of
host age, the distribution of worm numbers per host, density dependence
in egg production and sexual mating functions dependent on worm distri-
butions and mating habits (Anderson and May, 1982). The first adaptation of
this helminth population dynamics model to soil-transmitted helminthiases
was by Anderson in 1980 (Anderson, 1980). The widely observed negative
binomial distribution of parasites per hosts can be dynamically generated by
assuming a gamma-distributed distribution for host infectious contact rate. A
simple exponential survival function for the parasites allows a simpler differ-
ential equation model for the evolution of the mean female worm burden,
averaged across the population, M ðtÞ. The balance of parasite acquisition
and loss within a host is described by
dM
¼ bL  sM (5)
dt
where b is the mean infectious contact and parasite establishment rate across
the population, s is the reciprocal of the mean adult worm life span (A), and
L represents the concentration of female eggs or larvae in the environment.
The dynamics of the environmental stage of the parasite are represented by
(where parameter psi represents the rate at which infectious stages enter the
environmental reservoir),
dL jlzM
¼ ðkþ1Þ
 m2 L (6)
dt ½1 þ ð1  zÞM =k
A number of these parameters can be compressed into the basic repro-
duction number for macroparasites, R0, describing transmission intensity
160 J.E. Truscott et al.

and being defined as the average number of female parasites produced by a


female worm that themselves infect hosts and survive to reproductive matu-
rity in a susceptible host population in the absence of density-dependent
processes (e.g. sexual reproduction). The two equations are as follows,
dM
¼ sM ½R0 f ðM ; k; zÞ  1;
dt
ðkþ1Þ
f ðM ; k; zÞ ¼ ½1 þ ð1  zÞM =k (7)
jzlb
R0 ¼
sm2
where parameter psi is as defined above and z ¼ exp(g) represents the
impact of density-dependent egg production, k is the aggregation
parameter for the underlying negative binomial distribution, s is the
reciprocal (1/A) of the mean life span of the adult worm in the human host,
and m2 is the per capita mortality rate of the infective stages (eggs or larvae)
in the environment. It is important that g be in terms of the density of
female worms in the host for the current formulation. If g is in terms of
total worm burden, it will need to be doubled to take account of the
assumed one-to-one sex ratio in the host. The function f is derived from
considering the mean egg output generated by a negatively binomially
distributed worm burden with an exponential dependence of fecundity on
host worm burden. This approach can be considered a pseudo-probabilistic
model (a hybrid model), in that while the mean worm burden evolves
deterministically, the model describes the distribution of worms in an in-
dividual at a given age.
The basic model shown above stands as a point of origin for most models
of STH infection used to examine issues of control and elimination today
and as such we examine in more detail the assumptions implicit within it.

3.1 Environmental contamination


The dynamics of infectious material in the environment are generally ignored.
The timescale of external infectious stages is short compared to the life span of
the parasite in the human host and is thus assumed to be in equilibrium.
Additionally, current interventions for soil-transmitted helminthiases are pre-
dominantly chemotherapeutic and hence only target the adult worms in their
parasitic stage. For strategies involving water, sanitation and hygiene
(WASH), explicit modelling of an environmental reservoir may be necessary
(Brooker et al., 2006; Cundill et al., 2011; Pullan and Brooker, 2012).
Soil-Transmitted Helminths 161

3.2 Age structure


The age profile of infection intensity in hosts is a fairly consistent distinguish-
ing feature between the three main STH species and tells us about the vary-
ing rate of parasite acquisition with age (Fig. 2). Moreover, since
chemotherapy is often targeted at particular age groups, modelling age struc-
ture is essential in analysing control and elimination. The model described
above can be simply extended to a partial differential equation for changes
over time and by age,
vM vM
þ ¼ LðaÞ  sM ðaÞ (8)
vt va
where M(a,t) is the mean worm burden of individuals of age a and L(a) is the
age-dependent force of infection (Anderson and May, 1992). The presence
of age structure requires us to consider how hosts of different ages contribute
to and have contact with the infectious material in the environment. These
interactions are described by a contribution function, r(a), and a contact and
establishment rate, b(a). The parameter r(a) can be thought of as a function
describing the relative contribution with age since its magnitude is indis-
tinguishable from the parameter j (the rate at which infectious stages enter
the environmental reservoir) and the absolute quantity is embedded within
the definition of R0. The concentration of infectious material in the reser-
voir, L, is given by
ZN
dL jl
¼ M ðaÞf ðM ðaÞ; z; kÞ4ðM ðaÞ; z; kÞrðaÞSðaÞda  m2 L (9)
dt a
a¼0

and L(a) ¼ Lb(a); S (a) is the probability of a host being alive at age a. The
integral in this expression sums the contribution from hosts of each age to
the total amount of infectious material entering the infectious reservoir. The
function f, describing the output of eggs from a host aged a, is multiplied by
the fraction of those eggs that are fertilized, 4. Contributions are further
weighted according to the host survival curve, S(a). The parameter a is the
mean host age in the population.
For the age-structured model, the basic reproduction number is given by
ZN Za
zlj
R0 ¼ rðaÞSðaÞ bðxÞesðaxÞ dxda (10)
m2 a
a¼0 x¼0
162 J.E. Truscott et al.

The stable endemic solution for mean worm burden with age is given by
Za
0
 
M ðaÞ ¼ L bða0 Þesðaa Þ da0 ¼ L  QðaÞ (11)
a0 ¼0

Note that the infection age profile shape is determined only by the infec-
tion contact parameter and the worm death rate. The equilibrium reservoir
state, L*, is a solution of
ZN
jl
L  QðaÞf ðL  QðaÞ; z; kÞ4ðL  QðaÞ; z; kÞrðaÞSðaÞda ¼ L  (12)
am2
a¼0

There are either three solutions for the above (disease-free, breakpoint
and endemic equilibrium) or one (disease-free) for low R0 values.
The relationship between the age-structured model (9) and the basic
model (6) can be seen by multiplying the age-structured model by SðaÞ=a
and integrating out age. If S(a) ¼ exp(ma), then
ZN
dM L
¼ bðaÞSðaÞda  ðs þ mÞM ¼ L b  ðs þ mÞM (13)
dt a
a¼0

where b is the mean contact and establishment rate over all ages, and where
Z N
SðaÞ
M ðtÞ ¼ M ða; tÞ da. It is clear that the term arising from the host
a¼0 a
death rate has been omitted from the original formulation, but since this is a
small fraction of the worm death rate for STHs, the effect is negligible.

3.3 Models of drug treatment at a population level


Anderson and May (1992) use the model described by Eq. (7) to examine
the impact of treatment on prevalence and intensity of infection and average
egg output of the population. The authors develop a continuous model for
the impact of a long sequence of regular treatment cycles. The impact is
analogous to an additional death rate for the parasite population, c, where
c ¼ lnð1 ghÞ=s (14)
Here g is the proportion treated per round, h is the efficacy of the drug
and s is the interval between treatments. This formulation assumes that those
treated are randomly chosen from the population and that transient
Soil-Transmitted Helminths 163

disturbances in the parasite population following rounds of treatment have


no long-term effects. Note that the term gh is the effective coverage.
Anderson (1980) and Anderson and May (1982) were the first studies to
use models to address the impact of control programmes of drug treatment
based on an individual’s parasite burden and infection prevalence. The
model used is deterministic and not specific to a particular helminth species.
Results show that selective treatment [based on epidemiological evidence of
predisposition to heavy or light infection (see Schad and Anderson, 1985)]
can be effective in terms of reducing worm burdens per unit of drug sup-
plied, particularly if the aggregation within hosts is high and the effect is
not particularly sensitive to the selection threshold used. However, this anal-
ysis needs to be extended to take account of the additional cost of screening
implicit in implementing a selective treatment programme. The detection of
people with high worm burdens may also be problematic. High worm
burden is assessed through correspondingly high egg output detected by
diagnostic techniques such as Kato-Katz. Since egg output is highly variable
within a stool sample, and from day-to-day in the same patient (see Ander-
son and Schad, 1985), and since Kato-Katz has poor sensitivity, the ability to
select suitable candidates for treatment may be subject to error. In general,
however, individual-based models are more appropriate for simulating se-
lective treatment schemes and these are discussed later.
A range of models and concomitant numerical analyses were published
in the 1990s by Chan, Guyatt and colleagues which are extensions of the
models of the Anderson and May framework. Medley et al. (1993) con-
structed a model that allows host worm burden to adapt dynamically to
treatment and reinfection processes. This also has the advantage of being
able to distinguish between treated and untreated individuals and hence be-
tween treatment coverage and drug efficacy. The model does not include
sexual reproduction dynamics or age structure. In Chan et al. (1994a), the
authors added a simple age structure to the basic model with compartments
representing worm burden in children and adults, respectively. This struc-
ture is later further refined to include a preschool-age group as well
(Chan, 1997). Such a structure allows for the different worm burdens in
children and adults, characteristic of STHs, by letting the two groups acquire
worms from and discharge infectious material to the environment at
different rates. This approach also facilitates the modelling of both MDA
and treatment aimed at children only.
Medley et al. (1993) implemented the concept of a worm burden
threshold as a measure of disease prevalence in a population. This metric
164 J.E. Truscott et al.

can then be used as a measure of effective control and hence to assess the
success of an intervention within a model framework. The concept is orig-
inally introduced in Guyatt and colleagues and then further elaborated into a
set of age-specific worm thresholds for each of the three STH species
(Guyatt and Bundy, 1991; Chan et al., 1994b). Data exist relating morbidity
measures such as growth stunting (Ascaris, Trichuris) and anaemia (hook-
worm) to epg output, and these can be back-calculated through estimated
egg output per female worm to give a worm burden threshold. The consid-
erable uncertainties in each stage of this process make the resultant thresholds
approximate. This model is further employed to examine the cost-
effectiveness of repeated rounds of treatment for Ascaris over a 5-year period
and an equal period after treatment ends (Guyatt et al., 1993). The effects of
a range of treatment frequencies on metrics of worm burden, infection and
disease prevalence are examined using cost data from an actual chemother-
apeutic control programme. The authors find that it is more cost-effective to
intervene in high transmission areas than low, and that for low transmission
areas, lower treatment frequencies are more cost-effective. This is a result of
the slower bounce back time from treatment under lower transmission
resulting in rapid follow-up finding few worms to treat.
Chan et al. (1994a) extended the use of disease burden as a metric by us-
ing a two-age group model (adults and children) to study the impact of tar-
geted deworming. The authors validate their model against longitudinal
treatment data from both Ascaris and Trichuris studies. The model demon-
strates the interdependence of different age groups through the shared reser-
voir of infectious material, with treatment in one age group reducing worm
burden and disease prevalence in another. In particular, they demonstrate
that, over 5 years of treatment, annual treatment of children with a single
initial round for adults is more beneficial than biannual treatment of children
for the same period. This is the first demonstration that direct treatment of
the risk population may not be the most effective mode of intervention for
the whole community or the risk population in question. Guyatt et al.
(1995) extend the investigation to these different forms of treatment consid-
ering their cost-effectiveness, using the same costing data. They found that,
while treatment of the whole population is more effective by all metrics, in-
terventions targeted at children are more cost-effective, due to the concen-
tration of parasites in this subpopulation in the case of Ascaris. To some
extent, the cost-effectiveness of an intervention was dependent on the
metric applied. When an infection prevalence metric was used the result
Soil-Transmitted Helminths 165

indicated that annual treatment was optimal, whereas a high worm burden
‘disease’ metric suggested an interval of 2 years. Given that opinions may
differ as to the goals of treatment (e.g., elimination of disease vs reduction
in morbidity in children), the metric chosen can change the predicted
optimal strategy. Chan (1997) extended the analysis of morbidity by using
a three-age group model to assess the impact of SAC-targeted annual treat-
ment on morbidity measured in DALYs. A recent study has used similar
models to look at integrated cost-effectiveness across all three STH species
and schistosomiasis (Lo et al., 2015).
The models discussed up to this point have all represented infectious ma-
terial in the environment, be it eggs or larvae, as a single reservoir of material
which decays exponentially. Chan et al. (1997) further extended the two-
age group model by introducing more heterogeneous infectious contact
structure. The environmental reservoir is divided into two independent parts
and different contact patterns between age groups and reservoirs are
explored. The different models are fitted to longitudinal hookworm data
under treatment and show that for the given data, the standard assumption
of a common reservoir is optimal over more complex structures.

3.4 Acquired immunity


Anderson and May (1992) also developed a partial differential equation for
the acquisition of immunity, represented as an integral over past experience
of infection or exposure to infection. At equilibrium, this model facilitated
some analytical exploration for the age distribution of infection and how the
convexity of the profile changes under different rates of acquiring immunity
(Anderson and Medley, 1985).
As mentioned earlier, however, the evidence for immunity playing an
important role in STH infection is limited despite evidence of immunolog-
ical markers. Reinfection studies show that all age groups reacquire parasites
despite long past exposure. As such, in this paper we focus on age-related
exposure as opposed to acquired immunity, being the dominant factor in
shaping ageeintensity profiles.

3.5 Stochastic models


Anderson and Medley (1985) extended the basic principles described above
to a stochastic individual-based model. The births and deaths of individual
hosts are governed by a realistic age distribution for a typical low-income
country. For each host, the processes of acquisition, maturation and death
166 J.E. Truscott et al.

of the individual’s worms are modelled. The negative binomial aggregation


of parasites in hosts is generated through the statistics of acquisition. Two
models of acquisition were investigated: predisposition, in which an individ-
uals’ rate of parasite acquisition is particular to them and drawn from a
gamma distribution, and environmental in which eggs/larvae are clumped
in the environment (described by a log-normal distribution), but individuals
have no predisposition. Hosts have an underlying age-dependent contact
rate, allowing the model to match the characteristic ageeintensity profiles
observed in cross-sectional studies (as described in the Section 2.6). Egg
production in hosts was modelled by an exponential model of density-
dependent fecundity, as described above. However, the dynamics of sexual
reproduction were omitted.
The individual-based nature of the model allows a much more precise
description of treatment than the continuous approximation described by
the earlier Anderson and May (1982) model. Each round of chemotherapy
is treated as a discrete event and the choice of individuals to be treated can be
a function of age, treatment compliance and past or current infection status.
The drug efficacy is used as a treatment survival probability for individual
worms in a treated host.
Anderson and Medley (1985) use their detailed individual-based stochas-
tic model to examine the impact of different generators of parasite aggrega-
tion under treatment and the impact of different forms of selective
treatment. Results show that recovery posttreatment is significantly faster
with predisposition than with environmentally generated heterogeneity,
driven by the rapid recovery of hosts with a strong predisposition and that
selective treatment is improved if the most heavily infected people are
repeatedly identified rather than identified only once. Such an approach is
clearly much more laborious and expensive, even before considering how
the wormiest people can be reliably identified.
While considerably more computationally intensive, this approach
includes much more of the natural heterogeneity inherent in demog-
raphy, human behaviour, environmental factors and parasite population
processes. Output from the model is probabilistic in nature, so variability
in results can be directly examined and compared with observed patterns
of the distribution of parasite numbers per host. It also facilitates a much
more detailed view of the population in which individuals can also be
followed to look at details of reinfection and also compliance to
treatment.
Soil-Transmitted Helminths 167

3.6 Model predictions on the control of disease and the


likelihood that mass drug administration alone can
break transmission for different soil-transmitted
helminth species
We employ two examples (see Figs. 8 and 9) illustrating the differences be-
tween Ascaris and hookworm in the effect of control on heavy worm burden
under different treatment regimes. The baseline age distributions for these
both diseases are fitted to data (see Section 2.6 for the details) and have similar
transmission intensities, as indicated by the value of the basic reproduction
number, R0. They also have similar patterns of density dependence in fecun-
dity. As is clear from Figs. 2 and 3, their baseline age profiles and degrees of
underlying parasite aggregation are quite different from each other. The
threshold for high worm burden in each species is the lower of the two
defined in the literature. We compare three different patterns of treatment:
annual and biannual treatment of both pre-SAC and SAC, and annual mass
treatment of all individuals. Effective coverage is 75%, reflecting a 95%
drug efficacy and approximately 80% coverage. Results are generated by
the age-structured model described in Truscott et al. (2014b) and averaged
across age categories. At present, the primary goal of treatment is controlling
morbidity in children, and all regimes examined are effective at this, although
less so for Ascaris due to the greater prevalence of high worm burdens in that
age group at baseline. For hookworm, model simulations suggest that annual
mass treatment is preferable to biannual treatment of children. Treatment tar-
geted at children has a direct impact on their morbidity whereas mass treat-
ment has both a direct effect of child morbidity and an indirect effect on

Figure 8 The effect of three different regimes of coverage on heavy worm burdens of
hookworm. Mean prevalence of heavy worm burdens across (A) whole host population,
(B) pre-SAC and SAC and (C) adult age groups. Parameters as in Table 4.
168 J.E. Truscott et al.

Figure 9 The effect of three different regimes of coverage on heavy worm burdens of
Ascaris. Mean prevalence of heavy worm burdens across (A) whole host population, (B)
pre-SAC and SAC and (C) adult age groups. Parameters taken from Table 3 (column 4).

the underlying force of infection. The age-intensity profile indicates that the
majority of the worms are in adults, and treatment of the whole population
also serves to reduce the force of infection on children, reducing their
morbidity as well as that experienced by adults. For Ascaris, the highest infec-
tion intensity is in the SAC and pre-SAC age groups, and hence focussing on
treatment of children targets the majority of the worms. As a result, morbidity
in children is rapidly reduced along with the overall force of infection,
benefitting adults as well without directly targeting them. The importance
of the direct and indirect impact of interventions on the morbidity risk group
(usually children) is a common consideration in microparasitic disease control
(e.g., vaccination against childhood diseases) but is rarely discussed in the
context of helminthiases (Medley et al., 1993).
Discussion of the possibility of the elimination of transmission occurs in
some of the earliest work on helminth transmission models. It is first iden-
tified by Macdonald in the context of such models having an endemic and a
worm-free equilibrium simultaneously, separated by an unstable breakpoint
(Macdonald, 1965). Anderson and May (1982) point out that the aggregated
nature of worms amongst hosts means that the breakpoint will occur at very
low mean worm burdens and will therefore not play an important role in
parasite dynamics. Anderson and May discuss a second, dynamic, form of
elimination arising as the product of a treatment programme and which
can be achieved without the effect of sexual reproduction based on reducing
the basic reproduction number below unity in value. Using their continuous
definition of regular chemotherapeutic treatment, they identify a critical
proportion (chosen at random at each round of treatment) who must be
_
treated per unit time, g , that will reduce the mean worm burden in the
Soil-Transmitted Helminths 169

host population to zero, where tau is the interval between treatments and A
is the life expectancy of adult worms, 1 over sigma,
_
g ¼ ½1  expðð1  R0 Þs=AÞ=h (15)
Here h denotes drug efficacy. This formulation is derived on the assump-
tion that the influence of sexual reproduction is negligible. Based on similar
reasoning, they also develop an expression for a critical level of vaccine
coverage p to achieve elimination p ¼ ½1  1=R0 =n, where n is the average
duration of vaccine protection for a vaccinated person. It should be noted
that this type of elimination is only stable if treatment continues indefinitely
unless sexual reproduction is present in the model.
The possibility of elimination is not considered in the work by Medley,
Bundy, Guyatt and colleagues from the 1990s, as it was seen as not feasible,
given the resources available to provide drugs at the time (Medley et al.,
1993). Additionally, the models employed lacked treatment of sexual repro-
duction and hence would be unable to generate unstable elimination equi-
libria over finite periods of treatment.
Truscott et al. (2014a) extend the analytical approach to parasite elimi-
nation introduced by Anderson and May (1982). In the context of simple
age-structured models, they find five key dimensionless parameter groupings
that control whether elimination by repeated SAC-targeted chemotherapy
can be achieved. These include the following:
• R0, the basic reproduction number for the parasite in the population as a
whole;
• rc, the fraction of the transmission cycle attributable to SAC;
• tl, the effective treatment interval, defined as the interval between regular
rounds of treatment as a fraction of the mean worm life span;
• gh, the effective treatment coverage for the population, the product of
the therapeutic coverage and drug efficacy; and
• ε, the life span of infectious material in the environment as a fraction of
the mean worm life span.
The parameter grouping rc encapsulates the social structure within the
model in that it contains all the information on the age-specific contact
and deposition rates. As such, it is one of the main discriminants between
different species. For both Ascaris and Trichuris rc  0.65, whereas for hook-
worm, its value will be significantly less than 0.2.
Fig. 10(A) shows the impact on parasite population growth rate at low
parasite populations of effective coverage and the contribution of school
children to transmission, rc. The growth rate can be seen as a proxy for
170 J.E. Truscott et al.

the resilience of the parasite population to treatment. It is clear that for high
values of rc, at a given level of treatment, the parasite is much less resilient
than at low levels, illustrating that elimination through school-based
deworming will be much easier for Ascaris and Trichuris than for hookworm
for a given level of effective coverage. Fig. 10(B) shows critical values of R0
and effective coverage at which elimination occurs and illustrates how the
dynamics of the infectious reservoir can impact elimination. The parameter
ε represents the timescale for the turnover of infectious material in the reser-
voir. The ε ¼ 0 curve corresponds to the standard assumption of a reservoir
at equilibrium (Anderson and May, 1982). The more realistic value of 0.2
(for Ascaris) gives significantly more resilience to the parasite, arising from
the fact that infectious material is able to ‘shelter’ in the environment, where
treatment cannot reach it, and reinfect hosts later.
As the impact of mating among parasites is most marked at low worm
prevalences, it is not surprising that sexual reproduction should have an
impact on elimination. The inclusion of sexual reproduction makes the
elimination at a given coverage level possible at a much higher R0 value
(see Anderson and May, 1992). The mechanism by which this happens
can be seen in Fig. 11, which contrasts the time series of child mean
worm burden under annual treatment with and without parasite sexual
reproduction in place (Truscott et al., 2014a). Sexual reproduction acts to
reduce the output of fertile infectious material even considerably above
the point at which the ‘breakpoint’ burden occurs. The effect is that with

Figure 10 (A) The effects of the contribution of children to transmission, rc, on resil-
ience to the impact of treatment and (B) the effect of reservoir timescale, ε, on the pos-
sibility of elimination. Treatment of children is annual.
Soil-Transmitted Helminths 171

Figure 11 (A) Critical treatment effectiveness (coverage x efficacy) to reach elimination


for STH sexual reproduction and nonsexual reproduction dynamics and different treat-
ment intervals (annual vs. biennial) (annual vs. biennial). (B) Evolution of worm burden
in children under annual treatment with and without sexual reproduction dynamics
(Truscott et al., 2014a). SR, sexual reproduction.

sexual reproduction, the parasite population’s recovery is always reduced


with respect to that without sexual reproduction, and over many rounds
of treatment this accumulated reduction finally leads to elimination.
The impact of low worm burden on parasite fertility also has conse-
quences for how a given quantity of drugs should be administered in a com-
munity. Fig. 12(A) shows the mean worm burden in children over time
under three different scenarios, all of which deliver the same quantity of
drugs per unit time. Delivery with four or six monthly intervals, leads to
elimination at approximately the same time, whereas annual treatment is
significantly faster. The larger instantaneous delivery of drugs forces the
parasite population temporarily into the regime of limited fecundity,
hampering its bounce back and allowing the effect to accumulate over
time. The continuous description originally devised by Anderson and
May is formally the limit of shorter and shorter treatment intervals as
depicted in Fig. 12(A), and as such its continuous nature represents an inef-
ficient delivery method of a defined quantity of drug (Anderson and May,
1982). It is always better to give all at once rather than spreading it over
time. In the absence of sexual reproduction, all treatment intervals will
average the same over time. Fig. 12(B) shows the minimum number of
rounds to achieve elimination at different treatment frequencies (effective
coverage 80%). The optimal number of rounds varies between treatment
172 J.E. Truscott et al.

Figure 12 (A) Time series showing the effect of different intervention frequencies with
same annual treatment rate. (B) Minimum number of treatment rounds necessary to
achieve elimination (with sexual reproduction) as a function of R0 and the interval be-
tween treatments (Truscott et al., 2014a).

frequencies as R0 representing transmission intensity is varied (Truscott et al.,


2014a). The implication is that the optimal number of rounds is found at
different treatment intervals for different transmission intensities. Since the
number of rounds is a strong indicator of the cost of a programme, the
cost-effectiveness of elimination programs may be improved by careful
choice of the treatment interval and measurement of the intrinsic transmis-
sion intensity (R0) in a defined setting. Ideally, the latter should be measured
prior to the initiation of MDA.
We can illustrate the dependence of elimination on effective coverage
levels in different age groups for the population by mapping out the surface
of critical levels of coverage that lead to elimination, within a given number
of rounds (Anderson et al., 2014; Truscott et al., 2014b). Fig. 13 shows sur-
faces of the minimum coverage levels for annual treatment of pre-SAC,
SAC and adults to achieve elimination of Ascaris within 25 years for esti-
mated R0 values between 1.7 and 2.5 as in Table 3. The number of rounds
to elimination is defined as the least number of rounds of treatment from
which the parasite population will not recover. The four panels represent
different parameterizations of the model to the same data set (the Pulicat
Lake study) as described in the earlier Section 2.6. Fig. 13A and B, are
derived based on fits to the worm burden age profile (see Table 3) with
g ¼ 0.04 and 0.08, respectively. Panel C shows the parameterization for
panel B, but with g ¼ 0.04, while panel D is parameterized using both
egg and worm burden profile data simultaneously.
Soil-Transmitted Helminths 173

Figure 13 Elimination surfaces for Ascaris. The surface shows the minimum levels of
effective coverage needed to effect elimination within 25 years. Values above the sur-
face are predicted to always achieve elimination on the basis of the predictions of the
deterministic models. Parameter values are derived from fits to baseline data from
Table 3. (A) Fecundity parameter g ¼ 0.04 (Table 3, column 2), (B) g ¼ 0.08 (Table 3, col-
umn 3), (C) all parameters fitted together (Table 3, column 4), (D) parameters as for B,
but with g ¼ 0.04. Model as described in Truscott et al. (2015).

All four panels show surfaces that are at approximately the same angle to
the axes, indicating that the incremental effects of changing coverage in any
of the age groups are approximately the same across them all. The impact of
treatment of pre-SAC and adults is comparable, while treatment of SAC is
approximately twice as effective. This reflects the lack of variability in the
infectious contact parameters across the different fits to the baseline data.
The position of the surface changes markedly between different parameter-
izations. The natural correlation between g and R0, discussed earlier, means
that Fig. 13B has a higher R0 value (2.46) than Fig. 13A (1.7). As a result, the
parasite is significantly more resilient to elimination with the higher R0
value. However, elimination is also very sensitive to the value of the fecun-
dity parameter, g. Fig. 13C has the same parameter values as Fig. 13B, but g
is reduced to 0.04. Although transmission intensity is the same, the parasite
population is considerably more resilient. A possible explanation for this
174 J.E. Truscott et al.

effect is that the lower severity of density dependence allows larger mean
worm burdens in the population. The effect of treatment on the worm pop-
ulation is proportional, so the impact of a given round should not be affected
significantly. However, the breakpoint below which the parasite population
can no longer support itself is largely unchanged and hence requires greater
effective coverage and/or more rounds of treatment to achieve. For the fit to
both egg and worm data, the value of R0 (2.1) is higher than in panel A, but
the higher value of g reduces the resilience to give a similar elimination
response.
The differences in response to treatment among the three STH species
can be clearly seen by comparing the elimination surfaces for Ascaris in
Fig. 13 with those for hookworm and Trichuris in Fig. 14. Parameters for
the latter two diseases are defined in Table 4, generated from fitting to base-
line worm burden data sets described in the earlier Section 2.6. The surfaces
illustrate the strong differences between the three species in terms of their
response to regular treatment. Hookworm distribution among hosts is
dominated by the worm burden in adults. As a result, elimination is most
sensitive to changes in coverage of the adult population (Fig. 14A). The
low worm burdens in pre-SAC and SAC mean that treatment of these
age groups has little effect on transmission in the community as a whole,
although it may have a large effect on the morbidity of the groups treated.
For Trichuris (Fig. 14B), the low worm burden in adults as compared to
the other two species means that treatment of adults is not very effective.
Elimination of Trichuris is best achieved by a combination of coverage of
pre-SAC and SAC age groups. Note also that although Trichuris has a lower
R0 in this example than hookworm, it is much more resilient to elimination.
This is a consequence of the comparatively low density-dependent

Figure 14 Elimination surfaces for hookworm (A) and Trichuris (B). Parameters are taken
from Table 4. Model as described in Truscott et al. (2015).
Soil-Transmitted Helminths 175

fecundity for Trichuris (0.0035/female worm) with respect to hookworm


(0.08/per female worm). This accounts for the relatively high mean
worm burden and the difficulty in achieving elimination, which is the
same effect observed for Ascaris above. Given the low drug efficacies of
albendazole and mebendazole monotherapy for Trichuris (Table 2), this sug-
gests that elimination will prove particularly difficult to achieve.
The parameters that determine the predictions of the model’s response to
regular treatment are g, R0, the life span of the parasite and the age-specific
contact rates that control the force of infection and generate the character-
istic age intensity profile observed in epidemiological studies (see Fig. 2). Of
these, contact rates and g are possibly the most variable between species, and
between studies within a given species.
Predictions about elimination by MDA can be extended to derive the
predicted number of rounds of treatment at a defined coverage level, for
defined age groups and for a given frequency of treatment (e.g., annual or
twice a year). Some sample calculations are shown in the tables displayed
in Fig. 15 for Ascaris and Trichuris and for hookworm in Fig. 16. Each table
in these two figures shows the minimum number of rounds of treatment
necessary to eliminate the parasite as a function of the effective coverage
(therapeutic coverage times drug efficacy) in adults, and SAC and pre-
SAC jointly. The scan is effectively a cross-sectional slice through the equiv-
alent surfaces in Figs. 13 and 14 in which SAC and pre-SAC coverage are
equal to each other. The basic structure of these tables reflects that of the
equivalent elimination surfaces. Note that close to the elimination threshold

Figure 15 Annual rounds of treatment predicted to lead to the elimination of transmis-


sion for (A) Ascaris and (B) Trichuris. Parameters as defined for Figs. 13 and 14. NA indi-
cates more than 25 rounds of annual treatment required. Model as described in
Truscott et al. (2015). pre-SAC, preschool-aged children; SAC, school-aged children.
176 J.E. Truscott et al.

Figure 16 (A) Annual treatment rounds to elimination of hookworm and (B) 6-monthly
rounds of treatment predicted to be required for the elimination of Trichuris. Model as
described in Truscott et al. (2015). pre-SAC, preschool-aged children; SAC, school-aged
children.

(breakpoint) it can take a long time to achieve a break in transmission. The


models predict that it needs high levels of coverage to achieve elimination in
the short term defined as between 5 and 10 years.
The predicted elimination surfaces and the predicted number of rounds
required to achieve elimination are subject to a number of sources of vari-
ability. The parameter values used here are the maximum likelihood estima-
tors derived from fitting to specific epidemiological data sets. It is clear that
inferred values for a given species of STH vary considerably between studies.
For example, hookworm worm burdens across a range of studies show large
variations in magnitude, indicating matching variations in g and/or R0, if
not in the age-dependent contact rates (Fig. 2). This suggests variation not
only in transmission intensity but also in the hosts’ response to the parasite.
For a given data set, there will be uncertainty in the parameter values as a
function of the quality and the quantity of data collected. A fully Bayesian
approach will yield a posterior distribution of parameter values which will
necessarily blur the surfaces represented by the MLE values. This approach
will be addressed in future work.
A further source of variation is the underlying randomness of host
demography and worm acquisition and death, as discussed in the following
section on the formulation of a stochastic model. This is partially addressed
by the negative binomial distribution of worm burdens, but the impact will
be particularly important when addressing elimination effects. Under regular
treatment, the infection state of individuals in the population will determine
whether the disease is able to recover in the population or not. To
Soil-Transmitted Helminths 177

investigate the influence of demographic variability, we compare the behav-


iour of the deterministic model with a fully stochastic version.

3.6.1 Formulation of a stochastic model


Our stochastic model is an individual-based simulation operating at the level
of individual worms within hosts and is closely based on the one developed
by Anderson and Medley (1985). (Note that all the worms harboured by an
individual host are identical yet their number is tracked within their hosts.)
Table 6 lists the stochastic events involved and their associated rates. The
events refer to an individual with index i, out of a population of H hosts,
and age ai with Ni worms in total, ni of which are female. Each individual
i has a predisposition to infection drawn from a gamma distribution with
shape parameter k and mean 1.
li wGðk; 1Þ (16)
Treatment events are predetermined, occurring at times, tj, where
j ¼ 1.NT, where NT is the number of treatments, but are still stochastic
in their effect on a host and its worm burden. The dynamics of infectious
material is governed by a deterministic formulation. Total rate of production
of new infectious material, ET, from the population is
X
H
ET ¼ jl ni egni 1ðNi sni Þ (17)
i¼1

Table 6 Table of events for the stochastic model, giving the event type, its rate per
individual and its effect on the state of the model
Event Rate Effect

Worm acquisition by b(ai)liL Ni /Ni þ 1;


host i, aged a ni /ni þ Berð0:5Þ
Worm death in host i s/Worm Ni /Ni  1; ni /ni 
Berð0:5Þ; ni  0

Host birth/death for m(ai) Host age ai ¼ 0


host aged a Ni ¼ ni ¼ 0
Treatment of host i, d(t  tj)g(ai) Individual i is treated
aged a
Effect of treatment on e ni /n0i wBðni ; 1  hÞ
worms in host, i Ni /n0i þ BðNi  n0i ; 1  hÞ
Ber( ) represents a Bernoulli distributed random variable. The function d(.) represents a delta function.
178 J.E. Truscott et al.

Here, 1(.) is the index function evaluating to 0 or 1 depending on the


argument, ensuring that infectious material is only generated by individuals
with both male and female worms. The dynamics of the material in the
reservoir is then described by
dL
¼ ET  m2 L (18)
dt
A degree of variability within the model is lost at this stage, since ET is a
sum of means and does not include the additional variability in egg output
around the mean. However, for a host population of several hundred, it is
assumed that the variability will be small in comparison to the mean.
The definition of elimination used for the deterministic model can be
carried over in probabilistic terms to the stochastic model environment.
In the deterministic case, if the numbers of rounds of treatment are less
than the critical number, the probability of elimination is one. For rounds
equal to or greater than the critical value, the parasite is eliminated and the
probability is zero. In the latter case, there is a finite probability of elimi-
nation at any given number of treatment rounds. Fig. 17A shows the prob-
ability of elimination as a function of the number of rounds of annual
treatment. The parameters used are the MLE values from fitting to the
Pulicat data set (Table 3, column 4). Treatment coverage is 75% of SAC
and pre-SAC with a drug efficacy of 99%, giving an effective coverage
of just under 75%. In this scenario, the deterministic model achieves elim-
ination after nine rounds. In the stochastic model, this corresponds to a
probability of about 75% of elimination. To achieve a 95% probability
of elimination, about 12 rounds are necessary. The relative ease of elimina-
tion between the deterministic and stochastic models can be understood by
considering the variability in the infectious reservoir. In the deterministic
model, if the amount of infectious material in the reservoir falls below
the critical breakpoint value, elimination will occur and not otherwise.
In the stochastic case, levels of infectious material vary around the deter-
ministic mean. Hence, there is an additional mechanism by which the para-
site population can cross the breakpoint and be eliminated. As a result,
elimination within a stochastic paradigm is easier to achieve than within
a deterministic description.
Fig. 17B shows a representative sample of 10 stochastic realizations as a
time series. For populations that recover, the process takes approximately
10 years. The location of the breakpoint is clear at approximately one
worm per individual on average. The large variation in recovery time
Soil-Transmitted Helminths 179

Figure 17 (A) Probability of elimination as a function of the number of annual rounds


of treatment. Vertical dashed shaded bar indicates the minimum number of rounds to
achieve elimination according to the equivalent deterministic model. (B) Time series
showing total worm burden, averaged across the population, from a representative
sample of stochastic runs used to calculate the probability of elimination for eight
rounds of treatment delivered annually. Note that in some simulations elimination is
not achieved.

reflects not only the randomness of worm establishment but also the vari-
ability in the proportion of the population in high infection exposure (a
high force of infection) age categories.
A much fuller set of analyses based on the stochastic model will be given
in a future publication, but a few key points are worth noting aside from the
distribution of treatment rounds required to cross the transmission break-
point. Figure 18 shows clearly the variation in worm load person by person
and within an individual over time. We can also see the footprint of a chang-
ing rate of exposure to infection as an individual ages and moves from one
age grouping to the next. It is also possible to observe from Fig. 18 the
patchiness of treatment coverage due to the assumption of random selection
of individuals at each round of treatment to meet a defined coverage level.
Given enough repeat simulations, the mean worm burdens in each age
group over time approach the deterministic predictions as is expected in
models of this character where the existence of two stable equilibria and
one unstable boundary is set deterministically by the biology of parasite
reproduction. Future studies using the stochastic version of the age-
structured deterministic model will examine other sources of variation
including predisposition to infection, nonrandom compliance to treatment,
immigration and emigration of people from communities and spatial struc-
ture in host population distribution and movement.
180 J.E. Truscott et al.

Figure 18 Five individual host worm burdens generated by a stochastic model during
a 4-year annual treatment programme in which coverage is 75% for SAC and pre-SAC.
(A) For five hosts eligible for treatment during the program and (B) five adults not
eligible for treatment. Large dots represent birth (A) or death (B) of individuals in the
community. The grey region indicates the time period during which treatment occurs,
for comparison. pre-SAC, preschool-aged children; SAC, school-aged children.

3.7 Health economics


Mathematical models can be particularly useful tools for investigating the
cost-effectiveness of interventions (a summary of the economic evaluations
for soil-transmitted helminthiases using models is presented in Table 7
(Turner et al., 2015)). This is because models can be used to make projec-
tions over long time horizons and can, therefore, capture the long-term ben-
efits of interventions e empirical approaches using primary data from the
field (due to practical and time constraints) often have a limited time horizon
Soil-Transmitted Helminths 181

of a few years. Furthermore, models can be used to quantify the impact of


different epidemiological and programmatic settings on the generalizability
of the results. This is particularly important when investigating the cost-
effectiveness of alternative interventions.
In a recent systematic review on this area (Turner et al., 2015), only two
studies (Guyatt et al., 1993, 1995) were identified that investigated the cost-
effectiveness of alternative STH treatment strategies using a dynamic model.
This is important as dynamic transmission models couple the rate of infec-
tion and the population abundance of infection by explicitly modelling
the transmission cycle of the disease (Turner et al., 2014b; Brennan et al.,
2006; Kim and Goldie, 2008; Edmunds et al., 1999). Consequently, these
models can capture the so-called ‘herd effect’ or indirect effects of interven-
tions, whereby individuals can benefit from an intervention even if they are
not directly targeted. Fig. 19 illustrates this concept by showing that a
school-based MDA programme treating children for Ascaris can also have
a notable indirect benefit for the adults (whose worm burden is also reduced
over time due to the reductions in transmission e even though they are not
treated) (Turner et al., 2014b; Medley et al., 1993). Accounting for these
herd/indirect effects of interventions can be crucial to the validity of the
conclusions drawn from cost-effectiveness evaluations of interventions
against infectious diseases (Brennan et al., 2006; Kim and Goldie, 2008;
Edmunds et al., 1999). A further advantage of dynamic models is that
they can be used to evaluate the possibility of breaking transmission. This
is important given the recent interest in breaking transmission for STHs
(Bill & Melinda Gates Foundation, 2014; Brooker et al., 2015b) and that
potentially, new but more expensive interventions may be cost-saving/
cost-effective in the long term, only because they enable achieving elimina-
tion faster (Geoffard and Philipson, 1997; Turner et al., 2014a,b). Despite
this, static models (such as decision trees) are more widely used for economic
evaluations (Lugner et al., 2010; Sonnenberg and Beck, 1993; Briggs and
Sculpher, 1998; Bala and Mauskopf, 2006) e though this does not just apply
for NTDs (Turner et al., 2014b). In these models, individual hosts acquire
infection at a rate (i.e., the force of infection) which is not linked to the
abundance of infection in the population as a whole). Consequently, they
often assume that an individual’s probability of being exposed to an infection
is unaffected by an intervention (Lugner et al., 2010). This is often unrealistic
for infectious diseases, where an individual’s probability of being exposed to
an infection can change over the course of an intervention (as it is dependent
on the amount of infection in the population).
Table 7 Summary of the identified cost-effectiveness studies

182
Target of
Study Question intervention Effects Primary conclusions Source of the costs

Chan (1997) Cost-effectiveness of Ascaris • DALY The analysis indicates that Unpublished
school-based treating SAC is highly cost- data
Ascaris control effective; US$8 per DALY
(dynamic model). averted (for a high prevalence
community).
Guyatt et al. Cost-effectiveness Ascaris • Unit reductions If the aim of an intervention is to Unpublished
(1993) analysis of mass in mean worm reduce Ascaris-related data
anthelmintic burden morbidity using mass
treatment: Effects • Infection cases treatment, then it is more cost-
of treatment averted effective to intervene in higher
frequency on • Disease cases transmission areas.
Ascaris infection averted Furthermore, relatively long
(dynamic model). intervals between treatments
offer the most cost-effective
strategy.
Guyatt et al. Options for the Ascaris • Infection cases Child-targeted treatment can be Guyatt et al. (1993)
(1995) chemotherapeutic averted more cost-effective than mass e Which was
control of Ascaris • Disease cases treatment in reducing the based on
(dynamic model). averted number of disease cases. The unpublished data
results also imply that (with
the assumed circumstances)

J.E. Truscott et al.


enhancing coverage is more
cost-effective than increasing
frequency of treatment.
Soil-Transmitted Helminths
Hall et al. The cost- STH • Cost per infected This analysis suggests that a new PCD (1998), PCD
(2009) effectiveness of person treated three-tier treatment for (1999), Brooker
using different • Cost per deciding initial treatment et al. (2008)
thresholds for moderately/ frequency (if the combined and Fiedler and
determining the heavily infected prevalence is above 40%, treat Chuko (2008)
treatment person treated all children once a year; above
frequency (static • Cost per diseased 60% treat twice a year; and
distribution person treated above 80% treat three times a
model). year), would be more cost-
effective than the current
WHO recommended
thresholds.
Lee et al. The potential cost- Hookworm • DALY A hookworm vaccine may Hotez et al.
(2011) effectiveness of a provide not only cost savings (2006), Guyatt
hookworm but potential health benefits to (2003)
vaccine (static both SAC and non-pregnant and Brooker et al.
model). women of childbearing age. (2006)
The most cost-effective
strategy may be to combine
vaccination with the current
drug treatment.
DALY, disability-adjusted life years; SAC, school-aged children; STH, soil-transmitted helminths.
Adapted from Turner, H.C., Truscott, J.E., Hollingsworth, T.D., Bettis, A.A., Brooker, S.J., Anderson, R.M., 2015. Cost and cost-effectiveness of soil-transmitted
helminth treatment programmes: systematic review and research needs. Parasit. Vectors 8, 355.

183
184 J.E. Truscott et al.

Figure 19 The model-projected indirect benefit of treating children (2e15 year olds)
for Ascaris lumbricoides on the untreated adults (15 year olds). The results were ob-
tained using a fully aged structured deterministic dynamic transmission model,
described in more detail in Truscott et al. (2014b) and Anderson et al. (2014). Results
assume a low transmission setting (R0 ¼ 2) and a high coverage (80%).
Figure adapted from Turner, H.C., Walker, M., French, M.D., Blake, I.M., Churcher, T.S.,
Basan
~ez, M.G., 2014b. Neglected tools for neglected diseases: mathematical models in
economic evaluations. Trends Parasitol. 30, 562e570.

Published studies examining the cost-effectiveness of alternative


STH treatment strategies use a dynamic model focussing on Ascaris con-
trol and assume a mobile team distributed the drugs (the main delivery
method at the time of the studies) (Guyatt et al., 1993, 1995). Conse-
quently, the potential influence of the other STHs and the current
school-/community-based delivery systems on the cost-effectiveness of
different strategies has not been explored. This is particularly important
for hookworm, which has a different age-profile of infection than Ascaris
and Trichuris, with the adults usually having a larger proportion of the over-
all worm burden (Fig. 2). Consequently, ignoring this feature will signifi-
cantly underestimate the cost-effectiveness of expanding MDA
programmes to the whole community. Furthermore, the current treat-
ments for STH have a much lower efficacy against Trichuris (Table 2).
Models can be used to investigating the potential cost-effectiveness of
alternative treatment regimens.
Soil-Transmitted Helminths 185

3.8 Cost data


Currently, the overwhelming majority of STH treatment cost data pertains to
programmes targeting SAC once a year (Turner et al., 2015) (despite the cur-
rent goals focussing on scaling up treatment for both pre-SAC and SAC). The
absence of cost data for other age-groups (pre-SAC and adults) is a major bar-
rier for further research regarding how best to optimize STH control, and
there is an urgent need for further studies to investigate the costs for targeting
different combinations of age groups at different treatment frequencies
(Turner et al., 2015; Lee et al., 2015) It will be important, that these studies
consider possible economies of scale (and scope), particularly when comparing
the costs of different strategies and between different studies/settings (Mansley
et al., 2002; Turner et al., 2015). Such economies of scale have been found to
have significant implications when investigating the cost-effectiveness of STH
control (Turner et al., 2016a; Brooker et al., 2008).

3.9 Effectiveness metrics


Due to the difficulties in developing statistical models that link the different
clinical disease measures of the impact of STH infection to their population
dynamics, most of the published modelling studies use infection-based effec-
tiveness metrics (Table 7) (Turner et al., 2015). These studies defined disease
as having a modelled worm burden above age and STH species-specific
thresholds (Chan et al., 1994b; Medley et al., 1993; Turner et al., 2016a;
Guyatt et al., 1993, 1995). It should be acknowledged that these intensity
thresholds are uncertain and would be influenced by several host-specific
factors such as nutritional status (Mascarini-Serra, 2011), history of infection,
and for anaemia the initial iron balance (Gilles et al., 1964). Furthermore, the
thresholds for hookworm were informed by a study in which N. americanus
was the predominant species (Lwambo et al., 1992). Consequently, they
may not be accurate for An. duodenale (which is associated with higher rates
of blood loss) (Lwambo et al., 1992). Only two modelling studies (Lee et al.,
2011; Chan, 1997) were found that used DALYs as the effectiveness mea-
sure. Though it should be acknowledged that the methodology and key as-
sumptions surrounding DALY calculations are often unclear, and
surrounded by notable uncertainty (GiveWell, 2010, 2011).
3.10 Interventions to add to community-based drug
treatment
Models show that elimination by MDA alone is not possible in some
settings. It is, therefore, necessary to consider other complementary control
186 J.E. Truscott et al.

interventions such as WASH (Pullan et al., 2014a) and improvements and


health education. In addition in the future, there may be the possibility of
a vaccine (perhaps partially efficacious). Models can be useful tools in
informing how such interventions could impact transmission with and
without MDA.
Mathematical models can also be useful in investigating potential alterna-
tive treatment regimens against T. trichiura (such as ivermectin coadministra-
tion (Turner et al., 2016b) triple dosing with standalone with the current
monotherapy’s, papaya cysteine proteinases (Levecke et al., 2014), oxantel
pamoate-albendazole (Keiser et al., 2014; Speich et al., 2015) and pyran-
tel-oxantel (Albonico et al., 2002)). Previous analysis indicates it would be
highly advantageous for pre-SAC to be eligible for any treatment combina-
tion against T. trichiura e particularly if policy goals for MDA shift to trying
to break transmission (Turner et al., 2016b; Anderson et al., 2015).

4. CONCLUSIONS
Current demands from control implementers for the development of
mathematical models of STH control by MDA centre on providing predic-
tions to guide policy formulation on who to treat, how often to treat and for
how long, in various transmission settings. The focus has recently shifted
from the need for broad qualitative predictions on what age groups to treat
(treating SAC will rarely interrupt transmission, except in low exposure set-
tings, and never for hookworm with significant burdens in adults), to a num-
ber of more specific questions.
The first is that of model validation via field epidemiological studies to test
prediction against observed outcomes. The conceptual challenge is obvious
and important but the practice is more challenging. If models predict how
many to treat, in which age classes and for how long, longitudinal epidemi-
ological studies must be conducted with a variety of arms representing, for
example, treating SAC only, treating pre-SAC and SAC, and treating the
whole community, in three different transmission settings (low, medium
and high). On top of this, randomization between and within arms must
take place to manage heterogeneities such as social, cultural or environmental
differences between different villages in the trial. Furthermore, given current
predictions, such studies will have to be conducted over a 5- to 10-year ho-
rizons under high coverage levels to test if transmission is interrupted. One
such trial has started in early 2015 in villages in a coastal region south of
Soil-Transmitted Helminths 187

Mombasa in Kenya. The trial is called TUMIKIA (Interrupting transmission


of soil-transmitted helminths: a study protocol for cluster-randomized trials
evaluating alternative treatment strategies and delivery systems in Kenya)
and is in its early stages to test SAC treatment versus whole community treat-
ment. An initial baseline survey of 20,104 individuals (of all ages) has been
completed and involves 110 field officers and 40 laboratory technicians.
The study will run over a 5-year period on a similar scale with components
of the epidemiological study focussing on epidemiological parameter estima-
tion, gathering cost data and recording compliance (Brooker et al., 2015a).
Published modelling work has thus far largely ignored the impact of uncer-
tainty in model behaviour. There are two main sources; stochastic uncertainty
which arises from the fundamentally random processes of worm acquisition,
death and host demographic processes, and uncertainty in the parameter values
that underpin the models. As is shown in the stochastic model section, the
probabilistic nature of the results of stochastic simulations leads to a different
interpretation of results, with firm boundaries for types of behaviour replaced
with probability distributions and risks for given outcomes. These distributions
could be expanded to include parameter uncertainty by drawing parameters
values for individual stochastic runs from the appropriate posterior distributions
arising from Bayesian parameter estimation.
Certain features are not dealt with in detail by the models described in
this chapter. While seasonality of the environmental phase of STH species
is sensitive to seasonal changes in temperature and humidity, we have not
discussed seasonal movements of the hosts. Many communities experience
seasonal patterns of migration to find work. These movements can be the
source of parasite transmission over long distances and also lead to biases
in disease monitoring, due to parts of the population being excluded from
monitoring.
The details of the person-to-person transmission network are not
included, as they are not currently known. An understanding of these pro-
cesses would probably also lead to an understanding of the characteristic ag-
gregation of parasites within hosts. Genetic data from studies of individuals
within a community (Criscione et al., 2010) could possibly be used to
develop an understanding of transmission dynamics in a similar fashion to
the use of phylogenetic trees to understand outbreak dynamics for micropar-
asites (e.g., Colijn and Gardy, 2014).
Insights from modelling also need to play a part in the design of moni-
toring and evaluation programmes. It is common in school-based deworm-
ing studies for follow-up monitoring to be confined to the age groups that
188 J.E. Truscott et al.

have been treated, largely to assess the direct impact of treatment. Howev-
er, as is clear from modelling or just from looking at the infection age
profiles for different species, these data do not give a clear impression of
the infection status of the community as a whole. To gauge the optimal
distribution of treatment in the population, particularly as elimination be-
comes a possible target, better quality and more comprehensive data are
needed.
A further key issue is the currently poor understanding of the relationship
between individual worm burden and the outcome of diagnostic tests. Egg
counting and PCR methods are indirect and hence require additional models
that will connect the transmission model (in terms of worms) to the observed
data. Measures of prevalence using Kato-Katz are known to have often poor
sensitivities that depend on the species being detected and its prevalence.
PCR techniques appear to have much higher sensitivities, although these
have not been characterized yet. Robust probabilistic models are needed to
describe the relationship between ‘true’ prevalence and that measured by
PCR and Kato-Katz that will function at the low prevalences that will be
encountered as elimination is approached (see Medley et al., 2016; this
volume). Infection intensity data from egg count and qPCR has the possibil-
ity of providing a better picture of the underlying force of infection which
governs the elimination process. Data connecting qPCR and epg is now
becoming available, and models of epg output as a function of worm burden
exist but are insufficiently developed at present. The high degree of biological
variability in egg output among worms and the range of parameter values
across studies mean that it is unclear what the inferential value of intensity
data will be. There is a lot of work to be done developing and testing prob-
abilistic models for diagnostic tests and then evaluating which should be used
and how to optimize the power of the study. A further dimension to consider
is how these decisions would be affected by economic constraints.
The understanding that comes from epidemiological models needs to be
extended by grafting on a layer of heath economics. Cost functions can be
added to encompass the costs of treating under different strategies (such as
age targeted vs mass preventive chemotherapy) while accounting of econo-
mies of scale (Turner et al., 2016a). Models are being developed as described
in this paper, but the limiting factor is good cost data for the different poten-
tial strategies.
The final area is that of the impact on community-wide MDA on the
likelihood of the emergence of drug-resistant strains of the parasites and their
spread in the human population. Little is known about the genetic
Soil-Transmitted Helminths 189

determinants of resistance although in the veterinary field it is known that


wide scale use of anthelmintics can induce resistance in nematode parasites
(McManus et al., 2014). Some work has been done to use models to assess
the implications of spread and how best to slow the spread of resistant genes
(Churcher and Basan ~ez, 2008), but much remains to be done in the STH
field. One impediment to progress at present is the absence of data on the
parasite genes that are important and the identification of markers (but see
Diawara et al., 2013). There is no great difficult in grafting on population
genetic terms for a diploid organism on to the population dynamic frame-
work (Anderson and May, 1992), but in the absence of both biological
and epidemiological data, model predictions cannot be tested. One argu-
ment against community-wide treatment is that unlike SAC-targeted treat-
ment it leaves no reservoir of parasites not exposed to the drug. So far,
however, the experience in LF community-based treatment programmes
is encouraging. As far as can be assessed these programmes have not induced
the rapid rise of resistant parasite variants. The need is for careful monitoring
of drug efficacy as the scale of treatment coverage rises.
All these questions are now a focus of modelling efforts and the coming
year should see new insights and new models created by current research
funded under the Gates Foundation NTD Modelling Consortium. The
most important immediate task is the one of validation, namely testing
model predictions in various field settings with different treatment strategies.
Longitudinal community-based studies, if well designed, will be able to
determine what else must be addressed in model formulation to take account
of the many heterogeneities present in communities within given country
settings. They will also help to improve parameter estimates of the key vari-
ables, determine what needs to be better measured to help policy formula-
tion (e.g., cost and compliance data) evaluate what are the costs of different
control options and how best to monitor and evaluate the impact of current
control efforts based on MDA.

ACKNOWLEDGEMENTS
We thank the Bill and Melinda Gates Foundation for research grant support and GlaxoS-
mithKline for core support to the London Centre for Neglected Tropical Disease Research.

REFERENCES
Albonico, M., Allen, H., Chitsulo, L., Engels, D., Gabrielli, A.F., Savioli, L., 2008. Control-
ling soil-transmitted helminthiasis in pre-school-age children through preventive
chemotherapy. PLoS Negl. Trop. Dis. 2, e126.
190 J.E. Truscott et al.

Albonico, M., Bickle, Q., Haji, H.J., Ramsan, M., Khatib, K.J., Montresor, A., Savioli, L.,
Taylor, M., 2002. Evaluation of the efficacy of pyrantel-oxantel for the treatment of
soil-transmitted nematode infections. Trans. R. Soc. Trop. Med. Hyg. 96, 685e690.
Anderson, R.M., 1980. The dynamics and control of direct life cycle helminth parasites. In:
Barigozzi, C. (Ed.), Lecture Notes in Biomathematics. Springer, Berlin, Heidelberg.
Anderson, R.M., May, R.M., 1982. Population dynamics of human helminth infections:
control by chemotherapy. Nature 297, 557e563.
Anderson, R.M., May, R.M., 1985. Herd immunity to helminth infection and implications
for parasite control. Nature 315, 493e496.
Anderson, R.M., Medley, G.F., 1985. Community control of helminth infections of man by
mass and selective chemotherapy. Parasitology 90, 629e660.
Anderson, R.M., Schad, G.A., 1985. Hookworm burdens and faecal egg counts: an analysis
of the biological basis of variation. Trans. R. Soc. Trop. Med. Hyg. 79, 812e825.
Anderson, R.M., May, R.M., 1992. Infectious Diseases of Humans: Dynamics and Control.
Oxford Science Publications, Oxford.
Anderson, R., Hollingsworth, T.D., Truscott, J., Brooker, S., 2012. Optimisation of mass
chemotherapy to control soil-transmitted helminth infection. Lancet 379, 289e290.
Anderson, R.M., Truscott, J.E., Pullan, R.L., Brooker, S.J., Hollingsworth, T.D., 2013.
How effective is school-based deworming for the community-wide control of soil-
transmitted helminths? PLoS Negl. Trop. Dis. 7, e2027.
Anderson, R.M., Truscott, J.E., Hollingsworth, T.D., 2014. The coverage and frequency of
mass drug administration required to eliminate persistent transmission of soil-transmitted
helminths. Philos. Trans. R. Soc. Lond. B Biol. Sci. 369, 20130435.
Anderson, R.M., Turner, H.C., Truscott, J.E., Hollingsworth, T.D., Brooker, S., 2015.
Should the goal for the treatment of Soil Transmitted Helminth (STH) infections be
changed from morbidity control in children to community wide transmission
elimination? PLoS Negl. Trop. Dis. 9, e3897.
Ahuja, A., Baird, S., Hicks, J.H., Kremer, M., Miguel, E., Powers, S., 2015. When should
governments subsidize health? The case of mass deworming. World Bank Econ. Rev.
29 (Suppl. 1), S9eS24.
Bala, M.V., Mauskopf, J.A., 2006. Optimal assignment of treatments to health states using a Mar-
kov decision model: an introduction to basic concepts. Pharmacoeconomics 24, 345e354.
Barda, B.D., Keiser, J., Albonico, M., 2015. Human trichuriasis: diagnostics update. Curr.
Trop. Med. Rep. 2, 201e208.
Barda, B.D., Rinaldi, L., Ianniello, D., Zepherine, H., Salvo, F., Sadutshang, T.,
Cringoli, G., Clementi, M., Albonico, M., 2013. Mini-FLOTAC, an innovative direct
diagnostic technique for intestinal parasitic infections: experience from the field. PLoS
Negl. Trop. Dis. 7, e2344.
Becker, S.L., Piraisoody, N., Kramme, S., Marti, H., Silué, K.D., Panning, M., Nickel, B.,
Kern, W.V., Herrmann, M., Hatz, C.F., N’Goran, E.K., Utzinger, J., von M€ uller, L.,
2015. Real-time PCR for detection of Strongyloides stercoralis in human stool samples
from C^ ote d’ivoire: diagnostic accuracy, inter-laboratory comparison and patterns of
hookworm co-infection. Acta Trop. 150, 210e217.
Beer, R.J., 1976. The relationship between Trichuris trichiura (Linnaeus 1758) of man and Tri-
churis suis (Schrank 1788) of the pig. Res. Vet. Sci. 20, 47e54.
Bethony, J., Brooker, S., Albonico, M., Geiger, S.M., Loukas, A., Diemert, D., Hotez, P.J.,
2006. Soil-transmitted helminth infections: ascariasis, trichuriasis, and hookworm. Lancet
367, 1521e1532.
Bill & Melinda Gates Foundation, 2014. Global Partners Are Taking the “Neglect” Out of
“Neglected Tropical Diseases” [Online]. Available: http://www.gatesfoundation.org/
Media-Center/Press-Releases/2014/04/Global-Partners-Are-Taking-the-Neglect-out-
of-Neglected-Tropical-Diseases.
Soil-Transmitted Helminths 191

Bradley, M., Chandiwana, S.K., Bundy, D.A., Medley, G.F., 1992. The epidemiology and
population biology of Necator americanus infection in a rural community in Zimbabwe.
Trans. R. Soc. Trop. Med. Hyg. 86, 73e76.
Brennan, A., Chick, S.E., Davies, R., 2006. A taxonomy of model structures for economic
evaluation of health technologies. Health Econ. 15, 1295e1310.
Briggs, A., Sculpher, M., 1998. An introduction to Markov modelling for economic
evaluation. Pharmacoeconomics 13, 397e409.
Brooker, S., Bethony, J., Hotez, P.J., 2004. Human hookworm infection in the 21st century.
Adv. Parasitol. 58, 197e288.
Brooker, S., Clements, A.C., Bundy, D.A., 2006. Global epidemiology, ecology and control
of soil-transmitted helminth infections. Adv. Parasitol. 62, 221e261.
Brooker, S., Kabatereine, N.B., Fleming, F., Devlin, N., 2008. Cost and cost-effectiveness of
nationwide school-based helminth control in Uganda: intra-country variation and effects
of scaling-up. Health Policy Plan. 23, 24e35.
Hotez, P.J., Bundy, D.A., 2010. The global atlas of helminth infection: mapping the
way forward in neglected tropical disease control. PLoS Negl. Trop. Dis. 4, e779.
Brooker, S.J., Mwandawiro, C.S., Halliday, K.E., Njenga, S.M., Mcharo, C., Gichuki, P.M.,
Wasunna, B., Kihara, J.H., Njomo, D., Alusala, D., Chiguzo, A., Turner, H.C., Teti, C.,
Gwayi-Chore, C., Nikolay, B., Truscott, J.E., Hollingsworth, T.D., Balabanova, D.,
Griffiths, U.K., Freeman, M.C., Allen, E., Pullan, R.L., Anderson, R.M., 2015a. Inter-
rupting transmission of soil-transmitted helminths: a study protocol for cluster rando-
mised trials evaluating alternative treatment strategies and delivery systems in Kenya.
BMJ Open 5, e008950.
Brooker, S.J., Nikolay, B., Balabanova, D., Pullan, R.L., 2015b. Global feasibility assessment
of interrupting the transmission of soil-transmitted helminths: a statistical modelling
study. Lancet Infect. Dis. 15, 941e950.
Bundy, D.A.P., Medley, G.F., 1992. Immuno-epidemiology of human geohelminthiasis:
ecological and immunological determinants of worm burden. Parasitology 104 (Suppl.),
S105eS119.
Bundy, D.A., Cooper, E.S., 1989. Trichuris and trichuriasis in humans. Adv. Parasitol. 28,
107e173.
Bundy, D.A., Thompson, D.E., Cooper, E.S., Golden, M.H., Anderson, R.M., 1985a.
Population dynamics and chemotherapeutic control of Trichuris trichiura
infection of children in Jamaica and St. Lucia. Trans. R. Soc. Trop. Med. Hyg. 79,
759e764.
Bundy, D.A., Thompson, D.E., Golden, M.H., Cooper, E.S., Anderson, R.M.,
Harland, P.S., 1985b. Population distribution of Trichuris trichiura in a community of
Jamaican children. Trans. R. Soc. Trop. Med. Hyg. 79, 232e237.
Bundy, D.A.P., Cooper, E.S., Thompson, D.E., Didier, J.M., Anderson, R.M.,
Simmons, I., 1987. Predisposition to Trichuris trichiura infection in humans. Epidemiol.
Infect. 98, 65e71.
Callaway, E., 2016. Dogs thwart effort to eradicate Guinea worm. Nature 529, 10e11.
Chan, M.S., 1997. The global burden of intestinal nematode infectionsefifty years on. Para-
sitol. Today 13, 438e443.
Chan, M.S., Bradley, M., Bundy, D.A., 1997. Transmission patterns and the epidemiology of
hookworm infection. Int. J. Epidemiol. 26, 1392e1400.
Chan, M.S., Guyatt, H.L., Bundy, D.A., Medley, G.F., 1994a. The development and
validation of an age-structured model for the evaluation of disease control strategies
for intestinal helminths. Parasitology 109, 389e396.
Chan, M.S., Medley, G.F., Jamison, D., Bundy, D.A., 1994b. The evaluation of potential
global morbidity attributable to intestinal nematode infections. Parasitology 109,
373e387.
192 J.E. Truscott et al.

Cheever, A.W., Kamel, I.A., Elwi, A.M., Mosimann, J.E., Danner, R., 1977. Schistosoma
mansoni and S. haematobium infections in Egypt. II. Quantitative parasitological findings
at necropsy. Am. J. Trop. Med. Hyg. 26, 702e716.
Churcher, T.S., Basan~ez, M.G., 2008. Density dependence and the spread of anthelmintic
resistance. Evolution 62, 528e537.
Colijn, C., Gardy, J., 2014. Phylogenetic tree shapes resolve disease transmission patterns.
Evol. Med. Public Health 2014, 96e108.
Cringoli, G., 2006. FLOTAC, a novel apparatus for a multivalent faecal egg count technique.
Parassitologia 48, 381e384.
Criscione, C.D., Anderson, J.D., Sudimack, D., Subedi, J., Upadhayay, R.P., Jha, B.,
Williams, K.D., Williams-Blangero, S., Anderson, T.J., 2010. Landscape genetics reveals
focal transmission of a human macroparasite. PLoS Negl. Trop. Dis. 4, e665.
Criscione, C.D., Anderson, J.D., Sudimack, D., Peng, W., Jha, B., Williams-Blangero, S.,
Anderson, T.J., 2007. Disentangling hybridization and host colonization in parasitic
roundworms of humans and pigs. Proc. R. Soc. B Biol. Sci. 274, 2669e2677.
Croll, N.A., Anderson, R.M., Gyorkos, T.W., Ghadirian, E., 1982. The population biology
and control of Ascaris lumbricoides in a rural community in Iran. Trans. R. Soc. Trop.
Med. Hyg. 76, 187e197.
Crompton, D.W., 2001. Ascaris and ascariasis. Adv. Parasitol. 48, 285e375.
Cundill, B., Alexander, N., Bethony, J.M., Diemert, D., Pullan, R.L., Brooker, S., 2011.
Rates and intensity of re-infection with human helminths after treatment and the influ-
ence of individual, household, and environmental factors in a Brazilian community.
Parasitology 138, 1406e1416.
Center for Disease Control and Prevention, 2015. Parasites [ONLINE] Available at: http://
www.cdc.gov/parasites/.
Davey, D., Manickam, N., Simms, B.T., Harrison, L.M., Vermeire, J.J., Cappello, M., 2013.
Frequency and intensity of exposure mediate resistance to experimental infection with
the hookworm, Ancylostoma ceylanicum. Exp. Parasitol. 133, 243e249.
Diawara, A., Halpenny, C.M., Churcher, T.S., Mwandawiro, C., Kihara, J., Kaplan, R.M.,
Streit, T.G., Idaghdour, Y., Scott, M.E., Basan~ez, M.G., Prichard, R.K., 2013. Associ-
ation between response to albendazole treatment and b-tubulin genotype frequencies in
soil-transmitted helminths. PLoS Negl. Trop. Dis. 7, e2247.
Duerr, H.P., Dietz, K., B€ uttner, D.W., Schulz-Key, H., 2001. A stochastic model for the
aggregation of Onchocerca volvulus in nodules. Parasitology 123, 193e201.
Easton, A.V., Oliveira, R.G., O’Connell, E.M., Kepha, S., Mwandawiro, C.S.,
Njenga, S.M., Kihara, J.H., Mwatele, C., Odiere, M.R., Brooker, S.J., Webster, J.P.,
Anderson, R.M., Nutman, T.B., 2016. Multi-parallel qPCR provides increased sensi-
tivity and diagnostic breadth for gastrointestinal parasites of humans: field-based infer-
ences on the impact of mass deworming. Parasit. Vectors 9, 38.
Edmunds, W.J., Medley, G.F., Nokes, D.J., 1999. Evaluating the cost-effectiveness of vacci-
nation programmes: a dynamic perspective. Stat. Med. 18, 3263e3282.
Elkins, D.B., Haswell-Elkins, M., Anderson, R.M., 1986. The epidemiology and control of
intestinal helminths in the Pulicat Lake region of Southern India. I. Study design and pre-
and post-treatment observations on Ascaris lumbricoides infection. Trans. R. Soc. Trop.
Med. Hyg. 80, 774e792.
Fiedler, J.L., Chuko, T., 2008. The cost of child health days: a case study of Ethiopia’s
enhanced outreach strategy (EOS). Health Policy Plan. 23, 222e233.
Forrester, J.E., Scott, M.E., 1990. Measurement of Ascaris lumbricoides infection intensity
and the dynamics of expulsion following treatment with mebendazole. Parasitology
100, 303.
Garcia, L.S. (Ed.), 2007. Diagnostic Medical Parasitology, fifth ed. American Society of
Microbiology.
Soil-Transmitted Helminths 193

Geoffard, P.-Y., Philipson, T., 1997. Disease eradication: private versus public vaccination.
Am. Econ. Rev. 87, 222e230.
Gilles, H.M., Williams, E.J., Ball, P.A., 1964. Hookworm infection and anaemia. An epide-
miological, clinical, and laboratory study. Q. J. Med. 33, 1e24.
GiveWell, 2010. Cost-effectiveness Estimates: Inside the Sausage Factory [ONLINE]
Available at: http://blog.givewell.org/2010/03/19/cost-effectiveness-estimates-inside-
the-sausage-factory/.
GiveWell, 2011. Errors in DCP2 Cost-effectiveness Estimate for Deworming [ONLINE]
Available at: http://blog.givewell.org/2011/09/29/errors-in-dcp2-cost-effectiveness-
estimate-for-deworming/.
Glinz, D., Silué, K.D., Knopp, S., Lohourignon, L.K., Yao, K.P., Steinmann, P., Rinaldi, L.,
Cringoli, G., N’Goran, E.K., Utzinger, J., 2010. Comparing diagnostic accuracy of
Kato-Katz, Koga agar plate, ether-concentration, and FLOTAC for Schistosoma mansoni
and soil-transmitted helminths. PLoS Negl. Trop. Dis. 4, e754.
Guyatt, H., 2003. The cost of delivering and sustaining a control programme for schistoso-
miasis and soil-transmitted helminthiasis. Acta Trop. 86, 267e274.
Guyatt, H.L., Bundy, D.A., 1991. Estimating prevalence of community morbidity due to in-
testinal helminths: prevalence of infection as an indicator of the prevalence of disease.
Trans. R. Soc. Trop. Med. Hyg. 85, 778e782.
Guyatt, H.L., Bundy, D.A., Evans, D., 1993. A population dynamic approach to the cost-
effectiveness analysis of mass anthelmintic treatment: effects of treatment frequency on
Ascaris infection. Trans. R. Soc. Trop. Med. Hyg. 87, 570e575.
Guyatt, H.L., Chan, M.S., Medley, G.F., Bundy, D.A., 1995. Control of Ascaris infection by
chemotherapy: which is the most cost-effective option? Trans. R. Soc. Trop. Med. Hyg.
89, 16e20.
Guyatt, H.L., Bundy, D.A., Medley, G.F., Grenfell, B.T., 1990. The relationship between
the frequency distribution of Ascaris lumbricoides and the prevalence and intensity of infec-
tion in human communities. Parasitology 101, 139e143.
Hall, A., Holland, C., 2000. Geographical variation in Ascaris lumbricoides fecundity and its
implications for helminth control. Parasitol. Today 16, 540e544.
Hall, A., Horton, S., De Silva, N., 2009. The costs and cost-effectiveness of mass treatment
for intestinal nematode worm infections using different treatment thresholds. PLoS Negl.
Trop. Dis. 3, e402.
Hall, A., Anwar, K.S., Tomkins, A., Rahman, L., 1999. The distribution of Ascaris lumbri-
coides in human hosts: a study of 1765 people in Bangladesh. Trans. R. Soc. Trop.
Med. Hyg. 93, 503e510.
Hicks, J.H., Kremer, M., Miguel, E., 2015. The case for mass treatment of intestinal hel-
minths in endemic areas. PLoS Negl. Trop. Dis. 9, e0004214.
Holland, C.V., Asaolu, S.O., Crompton, D.W., Stoddart, R.C., Macdonald, R.,
Torimiro, S.E., 1989. The epidemiology of Ascaris lumbricoides and other soil-transmitted
helminths in primary school children from Ile-Ife, Nigeria. Parasitology 99, 275e285.
Holland, C.V., Kennedy, M.W. (Eds.), 2002. The Geohelminths: Ascaris, Trichuris and
Hookworm. Springer, US.
Hotez, P.J., Bundy, D.A.P., Beegle, K., Brooker, S., Drake, L., De Silva, N., Montresor, A.,
Engels, D., Jukes, M., Chitsulo, L., Chow, J., Laxminarayan, R., Michaud, C., Bethony, J.,
Correa-Oliveira, R., Shuhua, X., Fenwick, A., Savioli, L., 2006. Helminth Infections:
Soil-transmitted Helminth Infections and Schistosomiasis. Disease Control Priorities.
Jia, T.W., Melville, S., Utzinger, J., King, C.H., Zhou, X.N., 2012. Soil-transmitted hel-
minth reinfection after drug treatment: a systematic review and meta-analysis. PLoS
Negl. Trop. Dis. 6, e1621.
Keiser, J., Speich, B., Utzinger, J., 2014. Oxantel pamoate-albendazole for Trichuris trichiura
infection. N. Engl. J. Med. 370, 1953e1954.
194 J.E. Truscott et al.

Keiser, J., Utzinger, J., 2008. Efficacy of current drugs against soil-transmitted helminth
infections: systematic review and meta-analysis. JAMA 299, 1937e1948.
Kim, S.Y., Goldie, S.J., 2008. Cost-effectiveness analyses of vaccination programmes: a
focused review of modelling approaches. Pharmacoeconomics 26, 191e215.
Knopp, S., Steinmann, P., Keiser, J., Utzinger, J., 2012. Nematode infections: soil-
transmitted helminths and Trichinella. Infect. Dis. Clin. N. Am. 26, 341e358.
Krauth, S.J., Coulibaly, J.T., Knopp, S., Traoré, M., N’Goran, E.K., Utzinger, J., 2012.
An in-depth analysis of a piece of shit: distribution of schistosoma mansoni and
hookworm eggs in human stool. PLoS Negl. Trop. Dis. 6, e1969.
Lamberton, P.H.L., Jourdan, P.M., 2015. Human ascariasis: diagnostics update. Curr. Trop.
Med. Rep. 2, 189e200.
Lee, B.Y., Bacon, K.M., Bailey, R., Wiringa, A.E., Smith, K.J., 2011. The potential
economic value of a hookworm vaccine. Vaccine 29, 1201e1210.
Lee, B.Y., Bartsch, S.M., Gorham, K.M., 2015. Chapter eight e Economic and financial
evaluation of neglected tropical diseases. In: Anderson, R.M., Basan ~ez, M.G. (Eds.),
Mathematical Models for Neglected Tropical Diseases: Essential Tools for Control and
Elimination, Part A. Adv. Parasitol., 87, pp. 329e417.
Leles, D., Gardner, S.L., Reinhard, K., I~niguez, A., Araujo, A., 2012. Are Ascaris lumbricoides
and Ascaris suum a single species? Parasit. Vectors 5, 42.
Levecke, B., Buttle, D.J., Behnke, J.M., Duce, I.R., Vercruysse, J., 2014. Cysteine protein-
ases from papaya (Carica papaya) in the treatment of experimental Trichuris suis infection
in pigs: two randomized controlled trials. Parasit. Vectors 7, 255.
Leyton, M.K., 1968. Stochastic models in populations of helminthic parasites in the definitive
host, II: sexual mating functions. Math. Biosci. 3, 413e419.
Lima, W.S., 1998. Seasonal infection pattern of gastrointestinal nematodes of beef cattle in
Minas Gerais StateeBrazil. Vet. Parasitol. 74, 203e214.
Lo, N.C., Bogoch, I.I., Blackburn, B.G., Raso, G., N’Goran, E.K., Coulibaly, J.T.,
Becker, S.L., Abrams, H.B., Utzinger, J., Andrews, J.R., 2015. Comparison of
community-wide, integrated mass drug administration strategies for schistosomiasis and
soil-transmitted helminthiasis: a cost-effectiveness modelling study. Lancet Glob. Health
3, e629ee638.
Loukas, A., Prociv, P., 2001. Immune responses in hookworm infections. Clin. Microbiol.
Rev. 14, 689e703.
Lugner, A.K., Mylius, S.D., Wallinga, J., 2010. Dynamic versus static models in cost-
effectiveness analyses of anti-viral drug therapy to mitigate an influenza pandemic.
Health Econ. 19, 518e531.
Lwambo, N.J., Bundy, D.A., Medley, G.F., 1992. A new approach to morbidity risk assess-
ment in hookworm endemic communities. Epidemiol. Infect. 108, 469e481.
Macdonald, G., 1965. The dynamics of helminth infections, with special reference to
schistosomes. Trans. R. Soc. Trop. Med. Hyg. 59, 489e506.
Mansley, E.C., Dunet, D.O., May, D.S., Chattopadhyay, S.K., Mckenna, M.T., 2002.
Variation in average costs among federally sponsored state-organized cancer detection
programs: economies of scale? Med. Decis. Mak. 22, S67eS79.
Martin, J., Keymer, A., Isherwood, R.J., Wainwright, S.M., 1983. The prevalence and
intensity of Ascaris lumbricoides infections in Moslem children from northern
Bangladesh. Trans. R. Soc. Trop. Med. Hyg. 77, 702e706.
Mascarini-Serra, L., 2011. Prevention of soil-transmitted helminth infection. J. Glob. Infect.
Dis. 3, 175e182.
May, R.M., 1977. Togetherness among Schistosomes: its effects on the dynamics of the
infection. Math. Biosci. 35, 301e343.
McManus, C., Do Prado Paim, T., De Melo, C.B., Brasil, B.S., Paiva, S.R., 2014. Selection
methods for resistance to and tolerance of helminths in livestock. Parasite 21, 56.
Soil-Transmitted Helminths 195

Medley, G.F., Guyatt, H.L., Bundy, D.A., 1993. A quantitative framework for evaluating the
effect of community treatment on the morbidity due to ascariasis. Parasitology 106,
211e221.
Medley, G.F., Turner, H.C., Baggaley, R.F., Holland, C., Hollingsworth, T.D., 2016. The
role of more sensitive helminth diagnostics in mass drug administration campaigns: elim-
ination and health impacts. In: Basan~ez, M.G., Anderson, R.M. (Eds.), Mathematical
Models for Neglected Tropical Diseases: Essential Tools for Control and Elimination,
Part B, Adv. Parasitol., 94 (this volume).
Murray, C.J., Vos, T., Lozano, R., Naghavi, M., Flaxman, A.D., Michaud, C., Ezzati, M.,
Shibuya, K., Salomon, J.A., Abdalla, S., Aboyans, V., Abraham, J., Ackerman, I.,
Aggarwal, R., Ahn, S.Y., Ali, M.K., Alvarado, M., Anderson, H.R., Anderson, L.M.,
Andrews, K.G., Atkinson, C., Baddour, L.M., Bahalim, A.N., Barker-Collo, S.,
Barrero, L.H., Bartels, D.H., Basan ~ez, M.G., Baxter, A., Bell, M.L., Benjamin, E.J.,
Bennett, D., Bernabe, E., Bhalla, K., Bhandari, B., Bikbov, B., Bin Abdulhak, A.,
Birbeck, G., Black, J.A., Blencowe, H., Blore, J.D., Blyth, F., Bolliger, I.,
Bonaventure, A., Boufous, S., Bourne, R., Boussinesq, M., Braithwaite, T.,
Brayne, C., Bridgett, L., Brooker, S., Brooks, P., Brugha, T.S., Bryan-Hancock, C.,
Bucello, C., Buchbinder, R., Buckle, G., Budke, C.M., Burch, M., Burney, P.,
Burstein, R., Calabria, B., Campbell, B., Canter, C.E., Carabin, H., Carapetis, J.,
Carmona, L., Cella, C., Charlson, F., Chen, H., Cheng, A.T., Chou, D.,
Chugh, S.S., Coffeng, L.E., Colan, S.D., Colquhoun, S., Colson, K.E., Condon, J.,
Connor, M.D., Cooper, L.T., Corriere, M., Cortinovis, M., de Vaccaro, K.C.,
Couser, W., Cowie, B.C., Criqui, M.H., Cross, M., Dabhadkar, K.C., Dahiya, M.,
Dahodwala, N., Damsere-Derry, J., Danaei, G., Davis, A., De Leo, D.,
Degenhardt, L., Dellavalle, R., Delossantos, A., Denenberg, J., Derrett, S., Des
Jarlais, D.C., Dharmaratne, S.D., Dherani, M., Diaz-Torne, C., Dolk, H.,
Dorsey, E.R., Driscoll, T., Duber, H., Ebel, B., Edmond, K., Elbaz, A., Ali, S.E.,
Erskine, H., Erwin, P.J., Espindola, P., Ewoigbokhan, S.E., Farzadfar, F., Feigin, V.,
Felson, D.T., Ferrari, A., Ferri, C.P., Fevre, E.M., Finucane, M.M., Flaxman, S.,
Flood, L., Foreman, K., Forouzanfar, M.H., Fowkes, F.G., Fransen, M.,
Freeman, M.K., Gabbe, B.J., Gabriel, S.E., Gakidou, E., Ganatra, H.A., Garcia, B.,
Gaspari, F., Gillum, R.F., Gmel, G., Gonzalez-Medina, D., Gosselin, R.,
Grainger, R., Grant, B., Groeger, J., Guillemin, F., Gunnell, D., Gupta, R.,
Haagsma, J., Hagan, H., Halasa, Y.A., Hall, W., Haring, D., Haro, J.M.,
Harrison, J.E., Havmoeller, R., Hay, R.J., Higashi, H., Hill, C., Hoen, B.,
Hoffman, H., Hotez, P.J., Hoy, D., Huang, J.J., Ibeanusi, S.E., Jacobsen, K.H.,
James, S.L., Jarvis, D., Jasrasaria, R., Jayaraman, S., Johns, N., Jonas, J.B.,
Karthikeyan, G., Kassebaum, N., Kawakami, N., Keren, A., Khoo, J.P., King, C.H.,
Knowlton, L.M., Kobusingye, O., Koranteng, A., Krishnamurthi, R., Laden, F.,
Lalloo, R., Laslett, L.L., Lathlean, T., Leasher, J.L., Lee, Y.Y., Leigh, J., Levinson, D.,
Lim, S.S., Limb, E., Lin, J.K., Lipnick, M., Lipshultz, S.E., Liu, W., Loane, M.,
Ohno, S.L., Lyons, R., Mabweijano, J., MacIntyre, M.F., Malekzadeh, R.,
Mallinger, L., Manivannan, S., Marcenes, W., March, L., Margolis, D.J., Marks, G.B.,
Marks, R., Matsumori, A., Matzopoulos, R., Mayosi, B.M., McAnulty, J.H.,
McDermott, M.M., McGill, N., McGrath, J., Medina-Mora, M.E., Meltzer, M.,
Mensah, G.A., Merriman, T.R., Meyer, A.C., Miglioli, V., Miller, M., Miller, T.R.,
Mitchell, P.B., Mock, C., Mocumbi, A.O., Moffitt, T.E., Mokdad, A.A.,
Monasta, L., Montico, M., Moradi-Lakeh, M., Moran, A., Morawska, L., Mori, R.,
Murdoch, M.E., Mwaniki, M.K., Naidoo, K., Nair, M.N., Naldi, L., Narayan, K.M.,
Nelson, P.K., Nelson, R.G., Nevitt, M.C., Newton, C.R., Nolte, S., Norman, P.,
Norman, R., O’Donnell, M., O’Hanlon, S., Olives, C., Omer, S.B., Ortblad, K.,
Osborne, R., Ozgediz, D., Page, A., Pahari, B., Pandian, J.D., Rivero, A.P.,
196 J.E. Truscott et al.

Patten, S.B., Pearce, N., Padilla, R.P., Perez-Ruiz, F., Perico, N., Pesudovs, K.,
Phillips, D., Phillips, M.R., Pierce, K., Pion, S., Polanczyk, G.V., Polinder, S.,
Pope 3rd, C.A., Popova, S., Porrini, E., Pourmalek, F., Prince, M., Pullan, R.L.,
Ramaiah, K.D., Ranganathan, D., Razavi, H., Regan, M., Rehm, J.T., Rein, D.B.,
Remuzzi, G., Richardson, K., Rivara, F.P., Roberts, T., Robinson, C., De
Leon, F.R., Ronfani, L., Room, R., Rosenfeld, L.C., Rushton, L., Sacco, R.L.,
Saha, S., Sampson, U., Sanchez-Riera, L., Sanman, E., Schwebel, D.C., Scott, J.G.,
Segui-Gomez, M., Shahraz, S., Shepard, D.S., Shin, H., Shivakoti, R., Singh, D.,
Singh, G.M., Singh, J.A., Singleton, J., Sleet, D.A., Sliwa, K., Smith, E., Smith, J.L.,
Stapelberg, N.J., Steer, A., Steiner, T., Stolk, W.A., Stovner, L.J., Sudfeld, C.,
Syed, S., Tamburlini, G., Tavakkoli, M., Taylor, H.R., Taylor, J.A., Taylor, W.J.,
Thomas, B., Thomson, W.M., Thurston, G.D., Tleyjeh, I.M., Tonelli, M.,
Towbin, J.A., Truelsen, T., Tsilimbaris, M.K., Ubeda, C., Undurraga, E.A., van der
Werf, M.J., van Os, J., Vavilala, M.S., Venketasubramanian, N., Wang, M.,
Wang, W., Watt, K., Weatherall, D.J., Weinstock, M.A., Weintraub, R.,
Weisskopf, M.G., Weissman, M.M., White, R.A., Whiteford, H., Wiebe, N.,
Wiersma, S.T., Wilkinson, J.D., Williams, H.C., Williams, S.R., Witt, E., Wolfe, F.,
Woolf, A.D., Wulf, S., Yeh, P.H., Zaidi, A.K., Zheng, Z.J., Zonies, D., Lopez, A.D.,
AlMazroa, M.A., Memish, Z.A., 2012. Disability-adjusted life years (DALYs) for 291 dis-
eases and injuries in 21 regions, 1990e2010: a systematic analysis for the Global Burden
of Disease Study 2010. Lancet 380, 2197e2223.
Mutapi, F., Ndhlovu, P.D., Hagan, P., Spicer, J.T., Mduluza,, T., Turner, C.M.,
Chandiwana, S.K., Woolhouse, M.E., 1998. Chemotherapy accelerates the develop-
ment of acquired immune responses to Schistosoma haematobium infection. J. Infect.
Dis. 178, 289e293.
Nawalinski, T., Schad, G.A., Chowdhury, A.B., 1978. Population biology of hookworms in
children in rural West Bengal. I. General parasitological observations. Am. J. Trop. Med.
Hyg. 27, 1152e1161.
Niangaly, H., Djimde, A.A., Traore, B., Sangare, C.P., Guindo, D., Konate, D., Diakite, M.,
Diallo, N., Maïga-Ascofare, O., Sogoba, N., Dabo, A., Doumbo, O.K., 2012. Seasonal
variability of intestinal helminths and Schistosoma haematobium in a rural area of the Sahel
in Mali. Med. Sante Trop. 22, 430e434.
Nejsum, P., Betson, M., Bendall, R.P., Thamsborg, S.M., Stothard, J.R., 2012. Assessing the
zoonotic potential of Ascaris suum and Trichuris suis: looking to the future from an analysis
of the past. J. Helminthol. 86, 148e155.
Nikolay, B., Brooker, S.J., Pullan, R.L., 2014. Sensitivity of diagnostic tests for human soil-
transmitted helminth infections: a meta-analysis in the absence of a true gold standard.
Int. J. Parasitol. 44, 765e774.
Nwosu, A.B., 1978. Desiccation-survival of the eggs and third-stage larvae of hookworms.
Bull. Anim. Health Prod. Afr. 26, 49e53.
Nwosu, C.O., Madu, P.P., Richards, W.S., 2007. Prevalence and seasonal changes in the
population of gastrointestinal nematodes of small ruminants in the semi-arid zone of
north-eastern Nigeria. Vet. Parasitol. 144, 118e124.
PCD, 1998. Cost of school-based drug treatment in Tanzania. The Partnership for Child
development. Health Policy Plan. 13, 384e396.
PCD, 1999. The cost of large-scale school health programmes which deliver anthelmintics to
children in Ghana and Tanzania. The Partnership for Child Development. Acta Trop.
73, 183e204.
Pullan, R.L., Brooker, S.J., 2012. The global limits and population at risk of soil-transmitted
helminth infections in 2010. Parasit. Vectors 5, 81.
Pullan, R.L., Freeman, M.C., Gething, P.W., Brooker, S.J., 2014a. Geographical
inequalities in use of improved drinking water supply and sanitation across Sub-Saharan
Soil-Transmitted Helminths 197

Africa: mapping and spatial analysis of cross-sectional survey data. PLoS Med. 11,
e1001626.
Pullan, R.L., Smith, J.L., Jasrasaria, R., Brooker, S.J., 2014b. Global numbers of infection and
disease burden of soil transmitted helminth infections in 2010. Parasit. Vectors 7, 37.
Schad, G.A., Anderson, R.M., 1985. Predisposition to hookworm infection in humans. Sci-
ence 228, 1537e1540.
Seo, B.S., Cho, S.Y., Chai, J.Y., 1979. Egg discharging patterns of Ascaris lumbricoides in low
worm burden cases. Korean J. Parasitol. 17, 98e104.
Sinniah, B., 1982. Daily egg production of Ascaris lumbricoides: the distribution of eggs in the
faeces and the variability of egg counts. Parasitology 84, 167e175.
Sissay, M.M., Uggla, A., Waller, P.J., 2007. Prevalence and seasonal incidence of nematode
parasites and fluke infections of sheep and goats in eastern Ethiopia. Trop. Anim. Health
Prod. 39, 521e531.
Smith, G., Schad, G.A., 1989. Ancylostoma duodenale and Necator americanus: effect of temper-
ature on egg development and mortality. Parasitology 99, 127e132.
Sonnenberg, F.A., Beck, J.R., 1993. Markov models in medical decision making: a practical
guide. Med. Decis. Mak. 13, 322e338.
Speich, B., Ali, S.M., Ame, S.M., Bogoch, I.I., Alles, R., Huwyler, J., Albonico, M.,
Hattendorf, J., Utzinger, J., Keiser, J., 2015. Efficacy and safety of albendazole
plus ivermectin, albendazole plus mebendazole, albendazole plus oxantel pamoate, and
mebendazole alone against Trichuris trichiura and concomitant soil-transmitted helminth
infections: a four-arm, randomised controlled trial. Lancet Infect. Dis. 15, 277e284.
Strunz, E.C., Addiss, D.G., Stocks, M.E., Ogden, S., Utzinger, J., Freeman, M.C., 2014.
Water, sanitation, hygiene, and soil-transmitted helminth infection: a systematic review
and meta-analysis. PLoS Med. 11, e1001620.
Tallis, G.M., Leyton, M.K., 1966. A stochastic approach to the study of parasite populations.
J. Theor. Biol. 13, 251e260.
Tallis, G.M., Leyton, M.K., 1969. Stochastic models of populations of helminthic parasites in
the definitive host. I. Math. Biosci. 4, 39e48.
Tarafder, M.R., Carabin, H, Joseph, L., Balolong Jr., E., Olveda, R., McGarvey, S.T., 2010.
Estimating the sensitivity and specificity of Kato-Katz stool examination technique for
detection of hookworms, Ascaris lumbricoides and Trichuris trichiura infections in humans
in the absence of a “gold standard.” Int. J. Parasitol. 40, 399e404.
Taylor-Robinson, D.C., Maayan, N., Soares-Weiser, K., Donegan, S., Garner, P., 2015.
Deworming drugs for soil-transmitted intestinal worms in children: effects on nutritional
indicators, haemoglobin, and school performance. Cochrane Database Syst. Rev. 7,
Cd000371.
Truscott, J., Hollingsworth, T.D., Anderson, R., 2014a. Modeling the interruption of the
transmission of soil-transmitted helminths by repeated mass chemotherapy of school-
age children. PLoS Negl. Trop. Dis. 8, e3323.
Truscott, J.E., Hollingsworth, T.D., Brooker, S.J., Anderson, R.M., 2014b. Can chemo-
therapy alone eliminate the transmission of soil transmitted helminths? Parasit. Vectors
7, 266.
Truscott, J.E., Turner, H.C., Anderson, R.M., 2015. What impact will the achievement of the
current World Health Organisation targets for anthelmintic treatment coverage in children
have on the intensity of soil transmitted helminth infections? Parasit. Vectors 8, 551.
Turner, H.C., Truscott, J.E., Fleming, F.M., Hollingsworth, T.D., Brooker, S.J.,
Anderson, R.M., 2016a. Cost-effectiveness of scaling up mass drug administration for
the control of soil-transmitted helminths: a comparison of cost function and constant
costs analyses. Lancet Infect. Dis. 16, 838e846.
Turner, H.C., Truscott, J.E., Bettis, A.A., Hollingsworth, T.D., Brooker, S.J.,
Anderson, R.M., 2016b. Analysis of the population-level impact of co-administering
198 J.E. Truscott et al.

ivermectin with albendazole or mebendazole for the control and elimination of Trichuris
trichiura. Parasite Epidemiol. Control 1, 177e187.
Turner, H.C., Truscott, J.E., Hollingsworth, T.D., Bettis, A.A., Brooker, S.J.,
Anderson, R.M., 2015. Cost and cost-effectiveness of soil-transmitted helminth treat-
ment programmes: systematic review and research needs. Parasit. Vectors 8, 355.
Turner, H.C., Walker, M., Churcher, T.S., Osei-Atweneboana, M.Y., Biritwum, N.-K.,
Hopkins, A., Prichard, R.K., Basan ~ez, M.G., 2014a. Reaching the London Declaration
on Neglected Tropical Diseases goals for onchocerciasis: an economic evaluation of
increasing the frequency of ivermectin treatment in Africa. Clin. Infect. Dis. 59, 923e932.
Turner, H.C., Walker, M., French, M.D., Blake, I.M., Churcher, T.S., Basan ~ez, M.G.,
2014b. Neglected tools for neglected diseases: mathematical models in economic
evaluations. Trends Parasitol. 30, 562e570.
Uniting to Combat NTDs, 2012. The London Declaration on Neglected Tropical Diseases.
Available: http://unitingtocombatntds.org/resource/london-declaration.
Vercruysse, J., Behnke, J.M., Albonico, M., Ame, S.M., Angebault, C., Bethony, J.M.,
Engels, D., Guillard, B., Nguyen, T.V., Kang, G., Kattula, D., Kotze, A.C.,
Mccarthy, J.S., Mekonnen, Z., Montresor, A., Periago, M.V., Sumo, L.,
Tchuente, L.A., Dang, T.C., Zeynudin, A., Levecke, B., 2011. Assessment of the anthel-
mintic efficacy of albendazole in school children in seven countries where soil-
transmitted helminths are endemic. PLoS Negl. Trop. Dis. 5, e948.
Walker, M., Hall, A., Basan ~ez, M.G., 2010a. Trickle or clumped infection process? An
analysis of aggregation in the weights of the parasitic roundworm of humans, Ascaris
lumbricoides. Int J. Parasitol 40, 1373e1380.
Walker, M., Hall, A., Basan ~ez, M.G., 2010b. Trickle or clumped infection process?
A stochastic model for the infection process of the parasitic roundworm of humans,
Ascaris lumbricoides. Int. J. Parasitol 40, 1381e1388.
Walker, M., Hall, A., Anderson, R.M., Basan ~ez, M.G., 2009. Density-dependent effects on
the weight of female Ascaris lumbricoides infections of humans and its impact on patterns of
egg production. Parasit Vectors 2, 11.
WHO, 2002. Prevention and Control of Schistosomiasis and Soil-transmitted Helminthiasis:
Report of a WHO Expert Committee. World Health Organization, Geneva.
WHO, 2006. Preventive Chemotherapy in Human Helminthiasis: Coordinated Use of
Anthelminthic Drugs in Control Interventions: A Manual for Health Professionals and
Programme Managers (Geneva).
WHO, 2012. Accelerating Work to Overcome the Global Impact of Neglected Tropical
Diseases: A Roadmap for Implementation. World Health Organization, Geneva,
Switzerland. Available: http://www.who.int/neglected_diseases/NTD_RoadMap_
2012_Fullversion.pdf.
Williams-Blangero, S., Subedi, J., Upadhayay, R.P., Manral, D.B., Rai, D.R., Jha, B.,
Robinson, E.S., Blangero, J., 1999. Genetic analysis of susceptibility to infection with
Ascaris lumbricoides. Am. J. Trop. Med. Hyg. 60, 921e926.
World Bank, 2003. School deworming at a glance. In: Public Health at a Glance Series.
World Health Organization, 2015. Investing to overcome the global impact of neglected
tropical diseases. In: Third WHO Report on Neglected Tropical Diseases.
Ye, X.P., Wu, Z.X., Sun, F.H., 1994. The population biology and control of Necator
americanus in a village community in south-eastern China. Ann. Trop. Med. Parasitol.
88, 635e643.
CHAPTER FOUR

Studies of the Transmission


Dynamics, Mathematical Model
Development and the Control of
Schistosome Parasites by Mass
Drug Administration in Human
Communities
R.M. Anderson1, H.C. Turner, S.H. Farrell, J.E. Truscott
London Centre for Neglected Tropical Disease Research, London, United Kingdom
1
Corresponding author: E-mail: roy.anderson@imperial.ac.uk

Contents
1. Introduction 200
2. Epidemiological Patterns 205
2.1 Cross-sectional epidemiological surveys of prevalence and intensity of 205
infection
2.2 Morbidity 207
2.3 Distribution of worms per person 208
2.4 Reinfection posttreatment 210
2.5 Predisposition 210
3. Population Processes 211
3.1 Parasite life spans 211
3.2 Human demography 212
3.3 Age-related exposure to infection 213
3.4 Density dependence 213
3.5 Mating probabilities 215
3.6 Basic reproduction number 216
4. Mathematical Models of the Basic Dynamics of Transmission 216
4.1 Simple deterministic model with no age structure 217
4.2 More complex models with age structure 223
4.3 Acquired immunity, heterogeneity in exposure and/or parasite 224
survival within the host
5. Parameter Estimation 229
6. Numerical Studies of Mass Drug Administration Programmes 232
7. Stochastic Models e The Questions of Elimination and Eradication 236
8. Model Validation 239

Advances in Parasitology, Volume 94


© 2016 Elsevier Ltd.
j
ISSN 0065-308X
http://dx.doi.org/10.1016/bs.apar.2016.06.003 All rights reserved. 199
200 R.M. Anderson et al.

9. Discussion 240
Acknowledgements 242
References 243

Abstract
Schistosomiasis is global in extent within developing countries, but more than 90% of
the at-risk population lives in sub-Saharan Africa. In total, 261 million people are esti-
mated to require preventive treatment. However, with increasing drug availability
through donation, the World Health Organization has set a goal of increasing coverage
to 75% of at-risk children in endemic countries and elimination in some regions. In this
chapter, we discuss key biological and epidemiological processes involved in the schis-
tosome transmission cycle and review the history of modelling schistosomiasis and the
impact of mass drug administration, including both deterministic and stochastic ap-
proaches. In particular, we look at the potential impact of the WHO 2020 schistosomi-
asis treatment goals.

List of Abbreviations
CCA Circulating cathodic antigen
epg Eggs per gram of faeces
epml Eggs per milliliter of urine
MDA Mass drug administration
pre-SAC pre-School age children
qPCR Quantitative polymerase chain reaction
R Effective reproduction ratio
R0 Basic reproduction number
SAC School age children
WHO World Health Organization

1. INTRODUCTION
The construction of the theoretical framework that underpins the
study of the transmission dynamics of schistosome parasites was created in
three major phases, based on slightly different approaches. One forerunner
to these phases was a monograph published by V.A. Kositzin in 1934 titled
Symbiose, Parasitisme et Evolution, which described the first deterministic
model of hosteparasite relations based on a differential equation framework
that recorded changes over time in the number of hosts harbouring a given
number of parasites. The model consisted of an infinite number of equa-
tions, representing all possible burdens that could be summed to give the
mean number of parasites per host. This framework is very suitable for
the study of helminth parasites where morbidity is related to worm burden.
However, given its deterministic structure, the distribution of parasites per
host can be shown to be Poisson in form. As we shall see later in this chapter,
this assumption does not match observed patterns, where for schistosomes,
Studies of the Transmission Dynamics, Mathematical Model Development 201

Human host

Populaon of
Mortality
mature worms

Producon
of eggs

Cercarial Miracidial
populaon populaon

Mortality Mortality

Shedding Latent Suscepble


snails snails snails

Mortality Mortality
Mortality

Figure 1 Diagrammatic flow chart of the life cycle of schistosome species defining the
various stages involved and the rate processes determining population size and trans-
mission dynamics.

and indeed all helminths, worm burdens are highly aggregated, meaning that
a few hosts harbour many worms and most harbour few.
The real beginning, and the first phase of the development of the frame-
work with respect to schistosomes, was the pioneering work of Nelson
Hairston and George Macdonald. Each approached the study of the trans-
mission dynamics and population ecology of schistosome parasites using
different conceptual methods. Nelson Hairston was a Professor of Zoology
at Michigan and adopted an approach based on population ecology and life
table analysis. One of his key articles [published in 1962 and 1965 (Hairston,
1962, 1965)] begins with a quote from L.W. Hackett (1937), taken from the
book entitled Malaria in Europe. It is well worth repeating. ‘A closer
collaboration between biometricians and parasitologists, and a better aqua-
intanceship of each with the methods of the other, is one of the most useful
things we can work for today’. This sentiment is as applicable today as it was
in 1937 when Hackett’s book was first published.
Hairston constructed survival tables for each stage in the life cycle of the
schistosome parasites, namely, the adult worms in the human host, the free-
living miracidia, the larval stages in the snail intermediate host and the free-
swimming cercaria that infect humans (Fig. 1). He also calculated fertility,
both for the adult worm and the larval stages in the snail host. He largely
focused on Schistosoma japonicum and the snail host Oncomelania quadrasi for
202 R.M. Anderson et al.

which he had access to very detailed data from the Philippines. However, he
did also attempt calculations for Schistosoma mansoni and Schistosoma haema-
tobium. The data were used to calculate the net reproductive rate (where
births should balance deaths in a stable population state), but he found
that for humans the value was less than unity for S. japonicum (he suggested
that this may have been due to poor data on reservoir mammalian hosts such
as rats and dogs) and greater than unity for S. mansoni and S. haematobium. In
all cases, he ascribed this to poor quantitative information on birth and death
rates of the various life cycle stages. His papers were the first to connect the
study of population ecology with the epidemiological study of infectious
diseases.
George Macdonald was a Professor of Tropical Hygiene at the London
School of Hygiene and Tropical Medicine, and he adopted an approach
more familiar to today’s mathematical epidemiologists (Macdonald, 1965).
His model of transmission was based on differential equations to describe
changes over time in the mean worm burden of parasites in the human
host. Uniquely, at that time he considered the impact of separate sexes on
the dynamics of parasite population growth and decay. He recognized,
that at low parasite population densities, the dioecious nature of schistosome
parasites implied that mating frequency would decline, and this in turn
would influence the production of viable offspring to continue the life cycle.
Concomitant with the work of Macdonald, so-called catalytic differential
equation models were used by Hairston to examine how infection patterns
change with age in both the human and snail hosts, but the models were
based on the assumption that the system was in equilibrium, and they lacked
much detail on the biology of the life cycle of the parasites and the ability to
examine perturbations created, for example, by drug treatment (Anderson
and May, 1982).
The second phases of development of models of transmission by
Anderson and May (1982), built on the simple framework that had been
developed by Macdonald and aimed to add many complexities known
to be present in real life. These more advanced models use partial differen-
tial equations for changes over time and across age classes of the human
population in mean worm burdens. The many observed features of the
known epidemiology and population ecology of the schistosomes
embedded in such models included density dependence in egg production,
overdispersed distributions of parasite numbers per host (Macdonald
assumed a Poisson distribution) in a quasi-stochastic/hybrid framework
and acquired immunity based on past exposure to infection (Anderson
Studies of the Transmission Dynamics, Mathematical Model Development 203

Table 1 Life expectancies of the host populations


and parasite life cycle stages for schistosomes
Population Life span

Human host 50e80 years


Adult parasite in human 3e7 years
host
Snail host 1e6 weeks
Cercaria 8e20 h
Miracidia 4e16 h

and May, 1985a,b; Anderson et al., 1986). This template was also trans-
formed into an individual-based stochastic model to examine both predis-
position to heavy infection (Anderson and Medley, 1985) and various
causes of aggregation in parasite distributions (Anderson and May, 1991).
Parameter estimates for these complex models, which contained many in-
dividual population processes, were derived from data either from human
infections or from animal studies. A major feature of both Macdonald’s
model and those of Anderson and May was the recognition that the
different life cycle stages (miracidium, cercaria and adult parasite in the hu-
man host) have very different timescales of population turnover as detailed
in Table 1. This enables the differential equation models to be collapsed
into one equation for the adult worms that have the longest life span
among the parasite life cycle stages, without any loss of detail in the dy-
namic behaviour of the models (Anderson et al., 2015).
The third phase is the current one involving the most recent publica-
tions, where the focus is on a range of issues. These are the inclusion of
the impact of various control measures such as mass drug administration
(MDA; Truscott et al., 2015; French et al., 2010), individual-based stochas-
tic models to mirror predisposition both to infection and variation in indi-
vidual compliance to drug treatment (Shulford et al., 2016), the fitting of
models to observed epidemiological patterns using various statistical
approaches such as likelihood methods for model validation (Truscott
et al., 2015), understanding the dynamics of immunity and including
them in transmission models (Mitchell, 2011, 2012, 2014) and inclusion
of the treatment of spatial and other forms of heterogeneity in transmission
(Gurarie et al., 2015). Models are increasingly being used to assess the
health economic impact of different treatment programmes (Guyatt &
Chan, 1998; Turner et al., 2014; Lo et al., 2015).
204 R.M. Anderson et al.

This chapter reviews recent approaches with particular attention on the


impact of expanded MDA programmes and on model validation
plus parameter estimation. Criteria are presented for elimination based on
different levels of MDA coverage in school-aged children (SAC), between
the ages of 5 and 14 years, and adults, who are defined as older than 15 years.
The key questions that can be addressed in a quantitative manner by the use
of transmission mathematical models that are examined in this paper can be
summarized as follows (Anderson et al., 2012):
• For a given transmission level, how often should mass or targeted
chemotherapy be administered to sustain infection prevalence and inten-
sity below defined levels?
• As the prevalence and intensity fall after repeated rounds of treatment,
can the interval between treatments increase, and by how much?
• How do the demography of the population and the starting geographical
distribution of infection affect the structure of optimum treatment pro-
grammes when resources are finite?
• What level of infection across a community should trigger mass chemo-
therapy to minimize morbidity?
• Is elimination in a defined area possible by chemotherapy alone?
• How might repeated mass treatment affect the evolution of drug resis-
tance and how can this risk be minimized?
• What should be the target of control programmes?
• What are the best indicators for assessing the impact of control?
• In terms of cost-effectiveness, is it best to target school children, those
predisposed to heavy infection, or the entire community?
Schistosomes, and helminths in general, have much more predictable dy-
namics than most viruses, bacteria and protozoa. After treatment, worm
populations return (bounce back) to precontrol levels in a monotonic
manner, because of tight control created by density-dependent processes
that influence parasite reproduction, infection and mortality (and are partly
related to the buildup of a degree of acquired immunity). The long life
expectancy of established worms in the human host (years rather than
days) is also a factor (Bradley and McCullough, 1973). In the manner high-
lighted by both Hairston way back in the early 1960s (Hairston, 1962) and
Anderson and May in the 1980s (Anderson and May, 1982), and largely
ignored by public health implementers and policy makers, this understand-
ing of population biology can be used both to refine community-based
guidelines and inform policy makers of what to expect from a given inter-
vention programme. This is the main theme of this chapter.
Studies of the Transmission Dynamics, Mathematical Model Development 205

2. EPIDEMIOLOGICAL PATTERNS
Standard epidemiological approaches to the study of schistosome in-
fections in human communities involve the measurement of both the
prevalence of infection (fraction or percentage infected) and the mean inten-
sity of infection (indirectly by eggs per gram (epg) of faeces for S. mansoni or
eggs per millilitre (epml) of urine for S. haematobium).

2.1 Cross-sectional epidemiological surveys of prevalence


and intensity of infection
Such measures are typically performed cross sectionally by age group, and/or
longitudinally over time. In the context of MDA, longitudinal reinfection
studies involve measuring these two epidemiological statistics just before a
round of treatment and over a subsequent period of time. Typical patterns
observed in areas of endemic infection, based on cross-sectional surveys,
are plotted in Fig. 2 for S. mansoni and S. haematobium. These patterns are
convex in form with most infection in the SAC group (4e15 years). Often
in areas of high intensity, the peak and decay rate following this peak are
more marked by comparison with low-transmission areas. Experimental,
field and theoretical studies suggest this trend may be due to a slow buildup
of acquired immunity which is delayed in low-transmission areas (Crombie
and Anderson, 1985; Anderson and Crombie, 1985; Woolhouse, 1998;

Figure 2 Age intensity cross-sectional profiles for Schistosoma haematobium [graph


(A)] and Schistosoma mansoni [graph (B)] (Bradley and McCullough, 1973; Butterworth
et al., 1991). In graph (B) the profiles reflect before (solid line) and after (dotted line) a
round of mass drug administration.
206 R.M. Anderson et al.

350
Kavilinguni
300 Matithini
Kakuyuni
250
Geometric mean epg

Kitengei
200 Misuuni
Miu
150

100

50

0
0 10 20 30 40 50 60 70 80
Mean age (years)

Figure 3 Age intensity profiles for Schistosoma mansoni cross sectional by age in a set
of villages in Kenya (Fulford et al., 1992).

350 100
Intensity of Infection (egg output)

300
Prevalence of Infection

80
250

200 60

150
40
100
20
50

0 0
0.0 20.0 40.0 60.0 80.0 0 20 40 60 80
Age Group (in years) Age Group (in years)
Figure 4 Comparison of age intensity and age prevalence cross-sectional profiles for
Schistosoma haematobium in Tanzania (Bradley and McCullough, 1973).

Anderson, 1987). This is illustrated in Fig. 3 in a series of cross-sectional sur-


veys for S. mansoni in a set of villages in Kenya by Butterworth and col-
leagues (Fulford et al., 1992). By way of a contrast, cross-sectional surveys
of prevalence show a very different pattern as recorded in Fig. 4 for
S. haematobium from the same study where the intensity profile is shown
in Fig. 4A. The explanation for this will be discussed in a later section
that examines the distribution of parasite numbers per host.
As discussed in a seminal paper by Warren in 1973, these convex pat-
terns in intensity with age could be due to the buildup of acquired immu-
nity with age and/or age-dependent exposure to infection (Warren, 1973).
Since Warren raised this issue, its resolution is far from certain today
Studies of the Transmission Dynamics, Mathematical Model Development 207

(Fulford et al., 1992; Sanin et al., 2015). There is abundant evidence of


immunological responses to infection, both humoral and cell mediated,
but their effectiveness in limiting infection remains unclear. The
uncertainty surrounds both laboratory studies in animal models and the
interpretation of field epidemiological studies. What seems certain is that
the host’s genetic background plays an important role in determining the
effectiveness of an immune response in reducing infection postexposure
(Marquet et al., 1996). Later in this paper, we will examine both assump-
tions as explanations of observed epidemiological pattern.

2.2 Morbidity
The schistosome infections create a high burden of morbidity in endemic
regions of the world. Measures of such morbidity are based, for example,
on ultrasound scans of the liver in patients with S. mansoni and on the quan-
tification of blood in the urine for patients with S. haematobium and the asso-
ciated anaemia (Hatz, 2001; King, 2007). In both cases, the severity of
morbidity is believed to be related to worm burden as measured indirectly
by egg output in faeces or urine (Fig. 5). However, this association is not

0.6
Proportion with visual blood, heavy infection

0.5

0.4

0.3

0.2

0.1

0
0 0.2 0.4 0.6 0.8 1
Prevalence
Figure 5 Observed relationship between the prevalence of infection with Schistosoma
haematobium and visual blood (triangles) in the urine and heavy infection (squares).
Data of this type have been used to define the prevalence of infection above which reg-
ular treatment with praziquantel should be employed to protect children from
morbidity.
208 R.M. Anderson et al.

certain and some believe morbidity is more related to the intensity of the
immune response to infection (King, 2007), which may, in turn, be influ-
enced by the genetic background of the host. It is probable, however,
that the severity of morbidity is related to the accumulated burden of infec-
tion, or exposure to infection, over time. Quantifying these relationships has
proved to be difficult in practice, but it is of obvious importance to model
predictions of how MDA might affect morbidity, as well as the more usually
predicted impact of MDA on transmission.

2.3 Distribution of worms per person


The prevalence and mean intensity are summary statistics from the probabil-
ity distribution of worm numbers per person (or the indirect measures of epg
of faeces, or millilitre of urine). In contrast to the soil-transmitted helminths
where worm expulsion post-chemotherapy can be employed to estimate
worm burdens, schistosome burdens cannot be measured directly in humans
except via postmortem studies. The only published study with a reasonable
sample size is that of Cheever in 1968. At present this provides the only
direct information available on the distributions of parasite numbers per per-
son. As recorded in Fig. 6, the distribution for S. mansoni is highly aggregated
in form (variance >> mean), where most hosts harbour none or few worms
and a few harbour many. The pattern is well described by the negative bino-
mial probability model with aggregation parameter k which varies inversely
with the degree of aggregation or clumping. Values of k above 5 indicate

20
18
Frequency of observaon

16
14
12
10
8
6
4
2
0
0 1-5 6-10 11-20 21- 40 41- 80 81- 308
Number of worm pairs
Figure 6 The frequency distribution of the number of worm pairs of Schistosoma man-
soni found at autopsy (Cheever, 1968).
Studies of the Transmission Dynamics, Mathematical Model Development 209

Figure 7 Relationship between the prevalence of infection (fraction infected) and the
mean worm burden for various values of the aggregation parameter k as defined for
the negative binomial distribution (Eq. (1)). When k > 5 the distribution is Poisson in
form (¼ random distribution).

that the worms are randomly distributed in the host population (varian-
ce z mean). For the data recorded in Fig. 6, k is estimated as 0.2, reflecting
a high degree of aggregation (Cheever, 1968).
This distribution predicts the following relationship between the propor-
tion infected, P, and the mean worm burden, M, for a given k value,
  
M k
P ¼1 1þ (1)
k
As displayed in Fig. 7, this relationship helps to explain the patterns
recorded in Figs 2 and 4. The relationship suggests that small differences
in prevalence (particularly where this is high and k is small reflecting a
high degree of worm aggregation) may be associated with considerable
differences in intensity of infection and, therefore, possibly of the prevalence
and severity of clinical disease (Cheever, 1968). As such, the monitoring and
evaluation of MDA programmes must be based on intensity measures, not
just on prevalence, since the latter will not reflect the true magnitude of
the impact on intensity. A relationship between prevalence and mean egg
output can be constructed and used to indirectly estimate k and other param-
eters associated with the worm’s natural history within the host. For different
host populations or subpopulations, values for prevalence and mean number
of eggs per gram (epg) can be calculated. Mean egg output can be related to
210 R.M. Anderson et al.

mean worm burden, which can, in turn, be related to measured prevalence


through Eq. (1). This method has been used to calculate k and worm
fecundity parameters, leading to results comparable to direct estimation
from autopsy studies (Chan et al., 1995).

2.4 Reinfection posttreatment


After a period of continued exposure to infection post-chemotherapeutic
treatment with the most widely used drug, praziquantel, which has on
average a high efficacy (Table 2), people tend to reacquire infection.
Thus, despite abundant evidence of immunological responses to parasite
antigens, continual exposure leads to reinfection, especially in children.
The absence of strong acquired immunity, despite high exposure, is poorly
understood, but of obvious significance to the study of transmission dy-
namics and the impact of MDA. There is limited evidence of a slow buildup
of some acquired immunity in adults (Wilkins et al., 1987; Chandiwana
et al., 1991), although the evidence for S. haematobium infections is possibly
stronger (Butterworth, 1998), but this is insufficient to prevent infection
even in elderly individuals with a long history of exposure. Schistosome
parasites have large genomes and release many secretions that are thought
to modulate the effectiveness of immunological attack by the human host.

2.5 Predisposition
For most human helminth infections including the schistosomes, postche-
motherapy, those heavily infected at the point of treatment tend to reacquire
high burdens of parasites after a period of reinfection. Predisposition for
heavy (or light) infection may be due to a number of influences including

Table 2 Reported drug efficacy figures for praziquantel against the


main human schistosome infections
Parasite Drug efficacy References

Schistosoma mansoni 92.5% Kihara et al. (2007)


Schistosoma haematobium 85.3e100% Ojurongbe et al. (2014),
Tchuente et al. (2004)
S. mansoni 79e90% Utzinger et al. (2000)
S. haematobium 63e85% Utzinger et al. (2000),
Tchuente et al. (2004)
Schistosoma japonicum 80e90% Utzinger et al. (2000)
Schistosoma intercalatum 89% Utzinger et al. (2000)
Studies of the Transmission Dynamics, Mathematical Model Development 211

Table 3 Human helminths where evidence for predisposition to heavy infection has
been recorded
Evidence for
Parasite Country predisposition References

Ascaris lumbricoides India Yes Haswell-Elkins et al. (1987)


A. lumbricoides Burma Yes Hliang (1989)
Enterobius vermicularis India Yes Haswell-Elkins et al. (1987)
Necator americanus India Yes Schad and Anderson (1985),
Haswell-Elkins et al. (1987)
Ancylostoma duodenale India Yes Schad and Anderson (1985)
Trichuris trichiura St Lucia Yes Haswell-Elkins et al. (1987),
Bundy et al. (1985)
Schistosoma mansoni Kenya Yes Wilkins et al. (1987)

host genetic background and social, behavioural or environmental factors.


The precise causes for the observed patterns for human helminths are not
well understood, but behavioural factors are undoubtedly important for
schistosome infections (Bensted-Smith et al., 1987). Table 3 records studies
that have observed predisposition for human helminth infections.

3. POPULATION PROCESSES
Before turning to the development of mathematical models of trans-
mission and control, a brief review of some of the key population processes
provides guidance for model construction and parameterization.

3.1 Parasite life spans


As noted earlier, much simplification can be made in the study of transmis-
sion dynamics by noting the large discrepancies in the average duration of
stay in different compartments in the two-host life cycle of the schistosome
species as recorded in Table 1. The key features of the dynamics can be
captured by simply constructing equations of the population turnover of
the adult worms in the human host, taking due note of human demography
and the influence of the other stages in the snail intermediate host and the
free-living larval stages on the net rate of transmission pertaining to a defined
human community.
The dynamics of the worm population in the human host in response
to perturbations created, for example, by MDA will be largely driven by
the timescale of the adult worm life expectancy. For example, return
212 R.M. Anderson et al.

Table 4 Estimates of the life span of adult worms in the human host
Parasite Life span in years References

Schistosoma mansoni 2.7e4.5, mean 3.3 Anderson and May (1985a)


S. mansoni 5.7e10.5 Fulford et al. (1995)

times to precontrol steady states of the mean worm burden are largely set
by this parameter (Anderson and May, 1985a, 1991). Unfortunately, para-
site life expectancy is difficult to measure in practice. Published attempts
either depend on the duration of egg output in an untreated person who
moves to live in an infection-free environment, or by the use of various
statistical approaches to the analysis of observed epidemiological patterns
or mathematical model fitting procedures, particularly reinfection post-
treatment (Wallerstein, 1949; Berberian et al., 1953; Fulford et al.,
1995). Table 4 summarizes some of these published estimates. The average
value lies in the range 3e5 years. As we shall see later, the precise value has
a strong influence on the predicted dynamics of transmission.

3.2 Human demography


Since all infected persons contribute to the transmission intensity prevailing
in a defined habitat via water contact by all age groups and passage of faecal
material or urine into snail habitats, human demography with respect to the
age distribution of the affected population will have a significant influence
on net transmission and the impact of MDA targeted at specific age groups
such as SAC. An illustration of this is provided in Fig. 8, where the age dis-
tribution of Uganda in 2011 is plotted. Only 31.2% of the human popula-
tion is in the SAC group (5e14 years). Simple calculations, based on age
intensity profiles (based on epg of faeces) for S. mansoni in two different
countries, Uganda and Brazil, plus the prevailing demography of the human
populations, permit estimates to be made of the percentage of the worms (or
more precisely e egg output) in the SAC group. The figures are 39.7% for
the Ugandan study and 27.4% for the Brazilian study. Such calculations, sim-
ple as they are, are of great importance to the design of MDA programmes. If
only targeted at SAC, a large fraction of the worm population is not exposed
to chemotherapy. This facilitates the persistence of infection within such
communities. However, it can be argued that this reservoir of infection,
to some extent, guards against the evolution of dug resistance by continually
diluting the gene pool of the parasite with offspring not exposed to the se-
lective pressure applied by treatment.
Studies of the Transmission Dynamics, Mathematical Model Development 213

25

Proportion of the population (%)


20

15

10

0
0–4 10–14 20–24 30–34 40–44 50–54 60–64 70–74 80–84 90–94 100+
Age group in years
Figure 8 Age distribution of Uganda in 2011. Note that the proportion of the popula-
tion in the school-aged children (SAC e 5e14 years of age) age classes is 31.2%
(Anderson et al., 2013).

3.3 Age-related exposure to infection


Given that water contact is required for the human host to be exposed to
infection, due to the aquatic lifestyle of the snail intermediate host, age-
related changes in such contact can play an important role in determining
the shape of observed age intensity profiles, independent of any buildup
of acquired immunity with age. Many accounts of age-related water contact
have been published (Chandiwana, 1987; Wilkins et al., 1987; Chan et al.,
2000). One example is presented in Fig. 9, which records observation data
on human behaviour in St Lucia (Jordan, 1972). Note how the severity of
changes with age well mimics the age intensity patterns displayed in Figs
2 and 3, giving strength to the argument that ecology (human behaviour)
is more important than immunology.

3.4 Density dependence


Density dependence in birth and death rates is a well-established principle in
population ecology, where per capita birth rates fall and death rates rise as
population density increases for reasons related to the competition for avail-
able resources. For parasites, these effects may arise from acquired immunity
where the severity of immunological attack rises as parasite burden increases
and/or from resource limitation. In the case of schistosomes, the only well-
documented density-dependent process is that of worm fecundity in the
human host and, indeed, in experimental mammalian hosts, such as mice
214 R.M. Anderson et al.

3500

contact over 8 days (minutes) 3000

2500
Total duraon of water

2000

1500

1000

500

0
0-4 5-9 10-14 15-19 20-29 30-39 40-49 50 +
Age group in years
Figure 9 Total duration of water contact over an 8-day observation period in St Lucia
(Jordan, 1972).

0.9
Eggs/female/gram of faeces

0.8

0.7

0.6
(Thousands)

0.5

0.4

0.3

0.2

0.1

0
0 200 400 600
Female Worm Burden
Figure 10 Density-dependent fecundity in Schistosoma mansoni based on autopsy
data with the best fit exponential decay curve (Cheever, 1968).

and baboons (Medley and Anderson, 1985). Fig. 10 records such a pattern
for S. mansoni in humans from the Cheever (1968) autopsy studies. The
exponential model fits such data adequately where the per female worm
egg output f(M) as a function of worm burden M is given by,
f ðM Þ ¼ aegM (2)
Studies of the Transmission Dynamics, Mathematical Model Development 215

where a and g are parameters estimated from fitting the model to data
(Medley and Anderson, 1985). The term exp(g) is a useful summary
quantity of the severity of density dependence.

3.5 Mating probabilities


Schistosome parasites are dioecious with separate sexes and as noted first by
Macdonald (1965), the need to mate to produce viable offspring creates a
threshold in parasite population density below which mating frequency is
too low to maintain transmission. This threshold is often referred to as the
‘breakpoint’ in transmission. The concept is made more complex by
the parasitic mode of life since male and female worms must be present in
the same host for successful sexual reproduction. Therefore, the probability
distribution of parasites per host greatly influences mating success. Macdon-
ald assumed a Poisson probability distribution, but as shown in and earlier
section, the observed distributions are highly clumped and the negative
binomial model provides a good description of observed pattern. Aggrega-
tion enhances the likelihood of encountering a member of the opposite sex
within the human host and therefore reduces the level of the breakpoint.
Whether or not the parasites are monogamous or polygamous will also in-
fluence the level of the breakpoint as first noted by May (1977), and this
concept is illustrated in Fig. 11, which plots the breakpoint based on the
Probability of being mated ɸ

k=0.01
k=0.1
k=0.3
k=1.0
Random

Mean worm burden


Figure 11 Mating probabilities and the breakpoint in transmission. The plot records the
value of the probability of being mated for different assumptions for the degree of worm
aggregation within the host population (the magnitude of the negative binomial k which
varies inversely with the degree of aggregation) under the assumption of polygamy.
Note: Mathematical models predict that stable equilibria of either a persistent parasite
infection or extinction are separated by an unstable equilibrium. The level of the unstable
point is determined by the value of k and the reproductive biology (monogamous or
polygamous) of the parasite (May, 1977; Anderson and May, 1985a). Values close to
zero are predicted for moderate-to-high worm burdens and low values of k.
216 R.M. Anderson et al.

assumption of monogamy for various degrees of worm aggregation within


the host population (the magnitude of k). Schistosomes that live in pairs
in permanent copulation have been assumed to be monogamous and
mate for life. Robust evidence for this is limited, but in the model develop-
ment described in later sections this assumption is made, although its relax-
ation to a state of polygamy is also examined.

3.6 Basic reproduction number


A central concept in the study of the epidemiology of infectious diseases is
that of the basic reproduction number R0. For macroparasites, R0 is defined
as the average number of female offspring produced by a female worm (or
worm pair) that both infected the definitive human host and survived to
reproductive maturity, in a totally susceptible population. For the parasite
to persist, R0 > 1. In simple terms, R0 can be defined as the product of
reproductive output multiplied by the probability of transmission success,
divided by the mortality terms throughout the life cycle. More will be
said of this in the following mathematical model section, but the measure-
ment of R0 and the impact of MDA on the effective reproduction number,
R, in a defined population is central to an understanding of the transmission
dynamics and control of schistosome infections. High values of R0 reflect
high transmission intensity in a given setting, and concomitantly, reflect
the need for frequent drug treatment to break transmission.

4. MATHEMATICAL MODELS OF THE BASIC


DYNAMICS OF TRANSMISSION
The structure of this section is set to move from simple deterministic
models through increasing degrees of complexity to numerical studies of
individual-based stochastic models. The real world is replete with many
complexities, but simple models shed much light on the key processes influ-
encing transmission and the impact of control measures for helminth para-
sites. In addition, given the relatively simple dynamics of this class of
infectious diseases, they produce predictions that are fairly reliable in a qual-
itative sense. In other words, they are well able to accurately predict the
overall epidemiological patterns which are of importance in setting overall
public health policy for the control of transmission. More complex models
are required to produce quantitative predictions on the precise impact of
defined control measures.
Studies of the Transmission Dynamics, Mathematical Model Development 217

4.1 Simple deterministic model with no age structure


The short life expectancies of the free-living and snail host stages of the life
cycle, relative to the many-year life expectancy of the adult worm in its hu-
man host (Table 1) imply that a sensible discussion of the dynamics can be
based on a single equation for the adult worms. The simplest deterministic
model for the dynamics over time of the mean worm burden in the human,
M(t), is given by a simple differential equation (Anderson and May, 1985a;
Anderson, 1980),
" #
1T T f
dM ðtÞ 2 1 2
¼ m1 M ðtÞ 1 1 (3)
2T2 fM ðtÞ þ 1
dt

Here, ½ arises from an assumed 1:1 sex ratio, 1/m1 is adult worm life
expectancy, and in the terminology of Macdonald (1965), the parameters
T1 and T2 denote transmission from snail to human and human to snail,
respectively, where,
T1 ¼ b1 l2 N2 =½m1 ðb1 N1 þ m5 Þ (4)
and
T2 ¼ f b2 l1 N1 =½m4 ðb2 N2 þ m2 Þ (5)
Here l2 is the rate of cercarial shedding per infected snail, b1 is the trans-
mission coefficient for a cercaria to infect a human, m5 is the death rate of the
cercaria and N1 is human population size. As such the term b1/(b1N1 þ m5)
is the probability that a cercaria infects a human. N2 is the population size of
the snail host, l1 is the rate of egg production per mated female (not assumed
to be density dependent as yet), b2/(b2N2 þ m2) is the probability that an
egg produces a miracidium which infects a susceptible snail, f is the propor-
tion of latent-infected snails that survive to release cercaria, m2 is the death
rate of miracidia and m4 is the death rate of shedding snails.
The function f in Eq. (3) denotes the mating probability. As derived by
May (1977), this function for a dioecious species with a 1:1 sex ratio which is
monogamous and distributed in a negative binomial manner with clumping
parameter k within the human population is given by,
h i Z 2p ð1  cos qÞ
fðM ; kÞ ¼ 1  ð1  aÞð1þkÞ =2p dq (6)
0 ð1 þ a cos qÞð1þkÞ
where a ¼ M/(k þ M). Note that for k small, as it usually is, f / 1 for
moderate-to-high mean worm burdens.
218 R.M. Anderson et al.

For a polygamous parasite the equivalent expression derived by May


(1977) is somewhat simpler (as plotted in Fig. 11),

fðM; kÞ ¼ 1  ½1 þ M =2kð1þkÞ (7)


For the model defined in Eq. (3), the basic reproduction number R0 is
defined as,
1
R0 ¼ T1 T2 f (8)
2
This expression for R0 clearly illustrates how all the various life cycle
population parameters (birth, death and infection rates) influence its
magnitude. Hairston attempted to measure all these individual parameters
with limited success. As illustrated in a later section, a more practical
approach is to estimate R0 directly, from epidemiological data on age in-
tensity of infection profiles and from reinfection data posttreatment, by
fitting the mathematical models to the observed trends.
As a preliminary approach, if we assume that f z 1, this model permits
analytical exploration, to show that two equilibria exist, one of parasite
extinction M* ¼ 0 and another for stable endemic infection M* > 0,
where,
  
 1 1
M ¼ T1 T2 f  1 T2 f (9)
2 2
Put more simply, in terms of R0 this is,
M  ¼ T1 ðR0  1Þ=R0 (10)
This equation confirms that R0 > 1 for parasite persistence.
When the mating probability, f, is less than unity in value, Eq. (3)
cannot be solved analytically, given the complexity of the expression defined
in Eq. (6). However, it can be shown that at equilibrium the system has three
possible solutions, two of which are stable and represent parasite extinction
(M* ¼ 0) and stable endemic infection M* > 0. The two are separated by an
unstable state M*u. This is the transmission breakpoint first identified by
Macdonald. As noted earlier, he assumed a Poisson distribution of parasites
which meant that the unstable point was well above zero in value. For the
observed negative binomial patterns, this unstable mean worm load is very
low and close to zero for k < 0.1. A graphical illustration of this is given in
Fig. 11, which plots the probability mating term, f, as a function of the
value of k and the mean equilibrium worm burden M*, to show that the
Studies of the Transmission Dynamics, Mathematical Model Development 219

value is close to unity for high degrees of aggregation (k ¼ 0.01), even when
the mean worm burden is very low.
One useful result that can be derived from this model with a mating
probability is how the percentage of unfertilized eggs may increase as the
mean worm burden falls due to females not encountering a male, under,
for example, intense MDA (May, 1977; Anderson and May, 1991). In the
case of polygamy (the most pessimistic assumption, since only one male in
the host could fertilize many females, while for monogamy each female
must find its own male), if x is the fraction of people only passing unfertilized
eggs (where y ¼ 1x is the fraction passing fertilized eggs), the relationship
between x, y and the degree of parasite aggregation k given a 1:1 sex ratio in
births is,
n h ik o.
x ¼ 1  y  2ð1  yÞk  1
1
y (11)

As MDA programmes expand in terms of coverage, plotting this rela-


tionship will give an idea of at what level of the prevalence of unfertilized
eggs is the system on the point of transmission cessation. As noted elsewhere,
the predicted relationship well matches observed trends for intestinal hel-
minths in South Korea (Anderson and May, 1991).
The observation of high f values at low k values and low mean worm
burdens raises the question of whether it is worth including the mating
probability term in mathematical models, given the complexity of the
functions, to gain insights into the dynamics of transmission. For general
insights, the answer is probably no. However, for numerical studies of
more complex models that include age structure, age-dependent transmis-
sion and exposure to infection and/or acquired immunity, the answer is
yes. The reason for the latter is the importance of the unstable equilibrium
when assessing the impact of MDA programmes. Once crossed, by
sufficient drug coverage, the parasite extinction state is the attractor in
the system and hence the boundary of drug coverage to place the dynam-
ical system into the region of extinguishing parasite transmission. As shown
later, it is possible to plot the value of the unstable state as a function of the
proportion of different age groups of the human community effectively
treated per annum (or other time interval). In what follows in this subsec-
tion we examine the general analytical insights to be gained when it is
assumed that f ¼ 1 for all values of M* such that no unstable boundary
exists. The transmission boundary between the only two equilibria,
M* ¼ 0 and M* > 1, is R0 < 1.
220 R.M. Anderson et al.

Under this assumption the analytical solution for M* > 0 for the hybrid
deterministic model, given that the distribution of parasite numbers per per-
son is given a probability distribution assumption (negative binomial) but
this distribution is treated in a deterministic manner since k is assumed to
be independent of M(t), is given by,
h 1 i.
M  ¼ k R0kþ1  1 ð1  zÞ (12)

Here z is the density-dependent adult worm fecundity term given by


z ¼ exp(g) (Eq. (2)). This shows that the average worm load at equilib-
rium is simply determined by the magnitude of R0, the degree of worm ag-
gregation, k, and the severity of density dependence on adult worm
fecundity, z. Given the assumption that k is independent of M*, then the
equilibrium prevalence of infection P* is given by Eq. (1).
Eq. (12) can be used to obtain rough estimates of R0, given observations
on the overall mean worm burden, the degree of parasite aggregation and
the severity of density dependence on fecundity (both the latter from the
autopsy study data collected by Cheever, 1968). However, this approach
is not advised given the convex shapes of observed age intensity profiles
for schistosomes which reflect age-dependent exposure and/or acquired
immunity.
The other general insights that can be derived from this simple model
concerns how the system might behave under perturbation from the equi-
librium by, say, MDA. For those who design and implement MDA pro-
grammes, the key questions are what fraction must be treated per unit of
time and what might be an optimal interval between rounds of treatment?
If we assume the system is in endemic equilibrium, then the time, tf, taken
to reach a fraction f of the equilibrium abundance in growth to the equilib-
rium is approximately given by,
lnð1  f Þ
tf ¼  ¼ lnð1  f ÞL (13)
m1
where 1/m1 ¼ L is the adult parasite life expectancy in the human host. This
approximation shows clearly that ‘bounce back’ time is tightly controlled by
the adult worm life expectancy. For the major helminths of humans, which
include onchocerciasis, lymphatic filariasis, schistosomes and soil-transmitted
helminths such as Ascaris lumbricoides, in order, adult worm life expectancies
go from long (10 years or more) to short (around 1 year), with the schis-
tosomes in the middle at around 3.5e5 years (Table 4). Bounce back post a
Studies of the Transmission Dynamics, Mathematical Model Development 221

round of treatment will therefore be fast for Ascaris, medium for schisto-
somes and slow for the filarial worms. The actual time required to return to
the pretreatment equilibrium will depend on the fraction of the total
population treated, but Eq. (13) suggests that for schistosomes, post a round
of treating a high fraction of the population, it will take a number of years to
recover to pretreatment levels. As such, effective control where a high
fraction are treated could be induced by biennial or longer-interval treat-
ments. In other words, the control by MDA of helminth infection would be
easier for filarial worms (if efficacious macrofilaricidal drugs existed) and
schistosomes than for soil-transmitted helminths.
A further insight concerns the faction to be treated to reduce the effective
reproductive number, R, to below unity in value to cross a transmission
threshold. The fraction, gc, that must be treated per unit of time with a
drug of efficacy h (fraction of worms killed) to achieve this is given by,
  
expð1  R0 Þ
gc ¼ 1  h (14)
L
This equation makes clear that the task of breaking transmission (R < 1)
is simply determined by the magnitude of R0 reflecting the intensity of trans-
mission in a defined habitat, adult parasite life expectancy, L (defined in the
same time units as gc), and drug efficacy, h (see Table 2). A plot of the impact
of the treated fraction against prevalence and average intensity is shown in
Fig. 12. As the frequency of treatment rises, mean intensity declines in an
approximate slow exponential manner, but prevalence shows little change
until the fraction treated approaches the critical value (Eq. (14)).

Figure 12 The predicted impact of the fraction treated with chemotherapy, gc, on the
prevalence and mean intensity of infection (Eq. (14)). [R0 ¼ 2, k ¼ 0.024, h ¼ 0.9,
1/m1 ¼ 7 years, z ¼ 0.99].
222 R.M. Anderson et al.

Simple deterministic models can also be employed to examine different


ways to deliver treatment in a population where infection is endemic. So
far, Eq. (14) is based on the assumption that the people treated at each
round of chemotherapy are chosen at random. They could be chosen ac-
cording to worm burden (or egg output), in what is termed selective
chemotherapy (Warren, 1982). Anderson and May (1982) have examined
this problem and derived an expression for the average proportion of the
mean worm burden M* killed, Km, by a single round of selective treat-
ment, where,
(   ðkþ1Þ )
1
Km ¼ ga h 1  zð1  aÞ 1 þ ð1  zÞM
k
,(  k ) (15)
1
1  ð1  aÞ 1 þ ð1  zÞM 
k

Here ga is the average proportion treated, under the assumption of a


negative binomial distribution of worms per person, and its definition de-
pends on the function chosen to define how selective the treatment pro-
gramme is. Anderson and May chose a continuous function, g(i), to
define the probability that a person with i worms gets treated, where,
  
i
gðiÞ ¼ w 1  ð1  aÞexp  (16)
I
Here w and a are constants (both <1) defining the upper and lower
bounds on g(i), respectively. The parameter I characterizes the worm burden
above which treatment is more likely. The derivation of these quantities can
be repeated for the prevalence P or egg output E (Anderson and May, 1982).
The average proportion treated, ga, is then given by,
(   )
ð1  zÞM  k
ga ¼ f 1  ð1  aÞ 1 þ (17)
k

A plot of the function Km for a single round of chemotherapy for various


values of k reveals that highly selective treatment is very beneficial provided
the parasites are highly aggregated in the host population. Since this is the
case in practice, at first sight, selective chemotherapy would seem to be a
very good approach. However, this ignores the costs involved in identifying
those with heavy infection prior to treatment, which could be high.
Studies of the Transmission Dynamics, Mathematical Model Development 223

Predisposition (Table 3) reduces this cost, if it is assumed that those predis-


posed to heavy infection only have to be identified once in a programme
involving multiple rounds of treatment. However, in practice since drugs
are donated free, despite the predicted highly beneficial impact of selective
treatment, treating all in the community may be the better approach.
One final insight that can help in understanding the dynamics of trans-
mission in a defined community concerns the per capita rate, L, of acquiring
infection following a round of treatment. Intuition suggests this rate (defined
as the per person average rate of reacquiring worms, measured indirectly by
egg output) will be related to the magnitude of R0. Under the assumption
that density dependence has a very limited effect after a round of treatment
in the early stages of reinfection, simple models show that this is indeed the
case, with L defined as,
L z ðR0  1Þ=L (18)
where L is adult parasite life expectancy in the human host.

4.2 More complex models with age structure


The discussions in the previous section gave general insights into the factors
determining the main features of transmission and reinfection posttreatment.
Observed epidemiological patterns, however, show clearly that age of the
human host matters with respect to the likelihood of acquiring infection.
This is well illustrated in Fig. 9 which records age-related water contact.
Including age is also of great importance in predicting the impact of
MDA programmes that are often targeted at particular age groups, such as
SAC.
The simple model defined in Eq. (3) can be reformulated as a partial dif-
ferential equation for the mean worm burden in hosts at time t of age a,
M(t,a) (Anderson and May, 1982, 1985a, 1991),
vM ðt; aÞ vM ðt; aÞ
þ ¼ Lðt; aÞ  m1 M ðt; aÞ (19)
vt va
The age-dependent force of infection, L(t, a), is defined as,
(Z ),( Z )
N N
Lðt; aÞ ¼ f1 ða; M ; kÞ lðaÞM ða; tÞf2 ðM ; kÞda lðaÞda
0 0

(20)
224 R.M. Anderson et al.

Here, l(a) is the probability that a person survives to age a, and the func-
tions f1 and f2 are, respectively, the collapsed details of exposure to transmis-
sion stages and parasite establishment ( f1), and parasite reproduction and
fecundity ( f2) within humans; involving the mature worms in humans; den-
sity-dependent constraints on parasite establishment and fecundity, and asso-
ciated mating probabilities.
Evaluation of the behaviour of this model requires numerical work, given
the complexity of the mating functions. In addition, it is possible to add one
further complexity, namely, how each age group of person of age, a,
contributes to the pool of released eggs that contaminate an aquatic habitat
to continue the life cycle of the parasite, and how they also get exposed to
infection. It may be, as reflected in Fig. 9, that both children spend longer
in contact with water and, concomitantly, the eggs they pass contribute
more to sustaining transmission than those passed by adults. This sort of
age-dependent complexity is discussed for soil-transmitted helminths in
Truscott et al. (2015) and for schistosomes in Anderson et al. (2015), along
with equations for model definition, including density dependence in fecun-
dity and mating probabilities. The models described in these papers are deter-
ministic, but with the hybrid assumption of a negative binomial distribution
of parasite numbers per host. They assume that the convexity in changes in
average worm burden with age is due to exposure, not acquired immunity.
They can easily be transformed into individual-based stochastic models,
where the distribution of parasite numbers per person is dynamic and deter-
mined by various assumptions within the model, such as predisposition to
heavy or light infection (Anderson and Medley, 1985; Truscott et al., 2016).

4.3 Acquired immunity, heterogeneity in exposure and/or


parasite survival within the host
Mathematical models can also be developed (with or without age structure)
to mimic the acquisition of acquired immunity (Anderson and May, 1985b;
Feller, 1968; Yang et al., 1997), independent of age-related exposure to
infection. They are also able to generate convex curves in intensity of infec-
tion by age.
The simplest of models would be a parasite infection and parasite death
model of the accumulation of parasites over time t, Pi(t), by the human host,
where i denotes a host population group. Such ‘immigrationedeath’ models
in host population group i (perhaps an age group or a household) take the
form,
dPi ðtÞ
¼ Li  mi Pi ðtÞ (21)
dt
Studies of the Transmission Dynamics, Mathematical Model Development 225

The equivalent stochastic formulation of this immigrationedeath struc-


ture is well known and predicts that the probability distribution of worms
per person is Poisson (¼random) in form with the mean equal to the vari-
ance such that,
EfPi ðtÞg ¼ V fPi ðtÞg ¼ Mi ðtÞ (22)
where
Li
Mi ðtÞ ¼ ½1  expðmi tÞ (23)
mi
The parasite probability distribution in the total host population is given
by a mixture of Poisson distributions, each with a different mean. If these
means are distributed as a gamma distribution, the resultant overall distribu-
tion is negative binomial in form (it is approximately negative binomial for
variation of this general assumption of a gamma distribution of the means).
More formally, the probability generating function (pgf), p (i,z,t) for the dis-
tribution of parasites per person in group i is,
pði; z; tÞ ¼ exp½Mi ðtÞðz  1Þ (24)
And the pgf for the total community is
Xs
Hðz; tÞ ¼ f pði; z; tÞ
i¼1 i
(25)

where fi is the fraction of the total human population in group i. The


variance of this distribution exceeds the mean in value and hence such
simple assumptions of heterogeneity in exposure to infection or parasite
death rates within host groups will in itself generate aggregated worm
distributions. This heterogeneity may be due to environmental, host
behavioural, host genetic or immunological factors. To illustrate this for an
assumption of acquired immunity, if we just consider two groups
representing young children with limited experience of infection who
constitute a fraction f of the total population N with mean worm burden
M1(t), and a group of adults with much experience of infection who
constitute a fraction (1f ) with mean worm burden M2(t), the variance to
mean ratio (V/E) of the Poisson mixture is given by (Anderson and May,
1991),
f ð1  f ÞðM1 ðtÞ  M2 ðtÞÞ2
V =E ¼ 1 þ (26)
f M1 ðtÞ þ ð1  f ÞM2 ðtÞ
and exceeds unity in value (parasites aggregated).
226 R.M. Anderson et al.

This expression shows that acquired immunity, either acting on the rate of
infection or adult parasite mortality (or both), will influence how the value of
the negative binomial aggregation parameter, k, might vary with age [k ¼ E2/
(VE)]. An example is presented graphically in Fig. 13 which shows some
calculations where the groups are set as Infants (0e1 years, a fraction s of
the total population), Pre-school-aged children (Pre-SAC, 2e4 years, a frac-
tion p of the total population), SAC (5e14 years of age, a fraction q of the
population) and Adults (15þ years of age, a fraction [1-(s þ p þ q)] of the
population). Even though parasite distributions within the age groupings
are Poisson in form, given fixed age groupedependent rates of infection
and adult parasite death rates, the overall host population distribution of par-
asites is negative binomial with a k value of around 0.5 as a consequence of the
Poisson mixture across the age groupings [even if more heterogeneity was
introduced within an age grouping, mixtures of negative binomials remain
negative binomial, but with a lower k value (Feller, 1968)]. The parameter
k can vary with age, as is often observed in field studies, in a manner that is
dependent on the age-dependent rates of infection and adult parasite mortal-
ity (hence life span) in the human host. This is a very simple example of a sto-
chastic model generating aggregation in a dynamically changing probability

350

300
Overall mean = 105.2
250 Overall variance = 19818.4
Mean intensity (epg)

Overall k value = 0.5614


200

150

100

50

0
Infants Pre-Sac SAC Adults

Age grouping
Figure 13 Output of a simple model of heterogeneity in exposure and adult worm
(Schistosoma mansoni) life expectancy generating different mean worm burdens in a
set of age groupings [Infants, Pre-school-aged children (Pre-SAC), School-aged children
(SAC) and Adults]. The proportions of the total population in each age group were set at
0.1 (Infants), 0.2 (Pre-SAC), 0.3 (SAC) and 0.4 (Adults), with mean number of eggs per
gram (epg) counts, respectively, of 2, 25, 320 and 10. The distribution of epg output
is Poisson within an age grouping, and negative binomial (Poisson mixture) in the over-
all population with a k value of 0.561.
Studies of the Transmission Dynamics, Mathematical Model Development 227

distribution, by the biological assumption that acquired immunity acts to


decrease rates of infection and increase parasite mortality as hosts build up
experience of infection with age. In other recent descriptions of detailed
life cycle models of schistosome transmission, it has been claimed that the
model described in the publications is the first model to incorporate dynamic
distributions of parasites numbers per host (Gurarie et al., 2010, 2015). An in-
spection of references Anderson and May (1985b) and Anderson and Medley
(1985) reveals much earlier work that has captured dynamically changing dis-
tributions as outlined in this paragraph.
More complex models of the buildup of acquired immunity have been
formulated by Anderson and May, and Woolhouse and colleagues
(Anderson and May, 1985b, Woolhouse, 1991; Woolhouse et al., 1991).
One approach (Anderson and May, 1985b) assumes there is some functional
relationship between past exposure to infection and the current rate of infec-
tion, with immunity acting to decrease the rate as the human host ages. This
requires a partial differential framework to keep track of past exposure via an
integral of worm burden over all past ages up to the present.
Anderson and May used Eq. (18), but with the age-dependent force of
infection defined as,
( Z )
a
Lðt; aÞ ¼ L0 ðtÞ 1  ε M ðt; a0 Þexp½  qða  a0 Þ da0 (27)
0

where L0 is the pristine rate of infection in the absence of past exposure, ε is


a constant that determines the severity of the acquired immunological
response on the rate of infection and the term exp[q(aeaʹ)] describes
immunological memory. Memory is assumed, therefore, to have an average
duration of 1/q time units. In this formulation, acquired immunity acts to
decrease worm establishment in the human host via decreasing the infection
rate, but similar formulations could be constructed with immunity acting on
other rate parameters such as fecundity or adult worm survival (Anderson
and May, 1985b). Similarly, the functional form of the impact of the integral
of past exposure could be changed to represent other assumptions on how
past exposure acts on the rate parameters determining adult worm density
within a host. The model defined by Eq. (18) and Eq. (26) can generate a
variety of shapes for the age intensity profile and include convex curves as
observed. The nature of how immunity is acquired in the model assump-
tions is not at odds with the interpretation of some field studies, especially for
S. haematobium (Wilkins et al., 1987).
228 R.M. Anderson et al.

The first model of Woolhouse (1991) focuses on the prevalence of infec-


tion and assumes that individuals acquire immunity and lose all infection,
remaining in this protected state for a defined average period. This is some-
what removed from what is observed in the field, where following drug
treatment, all age groups reacquire some infection. The second model pub-
lished by Woolhouse et al. (1991) is based on the Anderson and May’s
(1985b) acquired immunity equations (Eqs (18) and (26)), but with different
assumptions concerning the functional form of how past experience influ-
ences parasite establishment, and with comparisons of predictions with field
observations. They conclude that the model is a good descriptor of how im-
munity builds up with exposure and the convex age intensity curves, for
both S. mansoni and S. haematobium. In more recent research, they have
used stochastic models and also compared the predictions of a more detailed
structure including immunological processes, to explore what assumptions
generated model outcomes most consistent with observed patterns (Mitchell
et al., 2012). The main conclusion was that long-lived antibody-mediated
responses acting to increase adult worm death rates gave rise to patterns
most similar to those observed.
Other modelling approaches encompassing much life cycle detail, in a
manner similar to the original approach of Hairston (1962, 1965), are
described by Gurarie, King and colleagues in association with extensive field
studies in Kenya (the SCORE project) to facilitate parameter estimation
(Gurarie et al., 2010). The model template is hybrid in structure, being largely
deterministic but with stochastic terms. It has the T1 and T2 terms defined for
transmission for humans to snails and vice versa, as originally defined by Mac-
donald (1965), but has added complexity including density-dependent fecun-
dity and a number of strata defining different classes of worm burden.
Individuals can move between strata such that the distribution of parasites
per host is dynamic. However, it is unclear what degree of aggregation this
generates and how this relates to observed k values. Parameter estimation
and model fitting and validation are not well developed as yet, with many
parameter assignments made from snail infection data plus changes in intensity
with age in the human host for a variety of communities that show significant
heterogeneity in transmission intensity (Gurarie et al., 2015). Drug treatment
impact is mirrored by individuals moving between parasite number strata. No
formal representation of acquired immunity is included. The desirability of
including the snail host in explicit equations is unclear, given the huge
discrepancy between the life span of an infected snail and that of an adult
worm in the human host (Table 1).
Studies of the Transmission Dynamics, Mathematical Model Development 229

The relative importance of acquired immunity as the key determinant of


the convex age intensity curves, as opposed to age-related exposure (see
Fig. 9), still remains uncertain. From a modelling point of view, both can
act in a somewhat similar manner to restrict parasite establishment in the hu-
man host. However, when examining the population impact of mass
chemotherapy, the two hypotheses will have very different effects. For
example, if the accumulated exposure with age generates immunity, mass
drug treatment of SAC could make them more susceptible to reinfection
by reducing their exposure to parasite antigens. The worms killed by drug
treatment could enhance protection by releasing large amounts of somatic
antigens within the host’s bloodstream as has been argued by Mutapi and
co-workers for S. haematobium (Mutapi et al., 1998). Field evidence for
such an effect is limited to date under the recently expanded MDA
programmes.

5. PARAMETER ESTIMATION
Whichever model is used to describe observed epidemiological pat-
terns, parameter estimation is a key part of the application of such models
to help to develop policies for transmission control. The phrase ‘epidemio-
logical pattern’, may describe cross-sectional age intensity and age preva-
lence profiles, or longitudinal trends stratified by age following a round of
chemotherapy (patterns of reinfection). Maximum likelihood methods are
commonly used to find the best fit parameter combinations to maximize
the likelihood of the model with respect to the data and also gain informa-
tion on parameter uncertainty (Yang et al., 1997; Anderson et al., 2015;
Chan et al., 1995). A problem with this approach is that, in general, no single
source of data is sufficient to resolve all necessary parameter values and it be-
comes necessary to combine a number of different types of data (e.g., worm
fecundity and aggregation data). This can be achieved by taking a Bayesian
approach in which information from different sources are represented as
priors on parameters or parameter combinations.
Likelihood methods can be used to just estimate the age-dependent
infection parameters and R0. Unfortunately, because of the parasitic mode
of life in habitats in humans that do not permit worm expulsion that can
be directly recorded (as is the case for intestinal helminths), much uncer-
tainty surrounds many parameter estimates. This situation is unlikely to
improve in the foreseeable future, so much reliance must be placed on likeli-
hood approaches.
230 R.M. Anderson et al.

We use three cross-sectional age intensity (epg) profiles for S. mansoni


to estimate the parameters of age-dependent exposure and contribution
to egg output into aquatic habitats with intermediate snail hosts, using
some prior values of other parameters as listed in Table 5. Fig. 14 records
three of these model fits to age mean intensity profiles from a study in
Kenya and a second in Uganda, based on four age groupings to derive es-
timates for the infection parameters (Infants, Pre-SAC, SAC and Adults),
and with prespecified estimates for k, density-dependent fecundity and
adult worm life span (Table 5).
Before leaving this short section on parameter estimation, it should be
stressed that this schistosome field of infectious disease epidemiology suffers
greatly from limited information on key parameters, such as estimates of R0
based on detailed age-intensity profiles, estimates of the density-dependent
relationship between egg output and worm burden (which at present relies
on one autopsy based study e that of Cheever, 1968 plus associated studies
in experimental animals) and estimates of adult worm life expectancy.
Good worm life expectancy data could be acquired from records of the
decay in egg output over time (given some knowledge of worm aggrega-
tion within the community the person came from) for individuals who
have left regions where they are exposed to infection. However, if infec-
tion is diagnosed, treatment must follow. As such, better data are unlikely
to become available.
To further complicate matters, the definition of R0 leads immediately to
much correlation between parameter values (in the numerator e the ‘repro-
duction/transmission’ processes e and in the denominator e the mortality
processes). An obvious consequence is that a good model fit to an observed
age intensity profile may be obtained by different parameter combinations
that lead to the same overall R0 value. Although much improvement in
parameter estimates is not to be expected in the near future, two factors
would help to some degree to refine current estimates. The first is better
cross-sectional and longitudinal studies across all age classes of infection in-
tensity and prevalence (the relationship between the two provides informa-
tion on worm aggregation patterns e i.e., the magnitude of the negative
binomial k). Such data provide a basis for estimating R0 in defined commu-
nities. The second is much better data on adherence to treatment, since this
plus pre- and posttreatment (after, say, 1 year) measures of the age intensity
profiles can yield data on which to validate model predictions, which in turn
can help in improving model structural formulation and parameter
estimation.
Studies of the Transmission Dynamics, Mathematical Model Development
Table 5 Parameter estimates for Schistosoma mansoni using Markov Chain Monte Carlo (MCMC) methods on age intensity data from
three different villages in Kenya. The age groupings differ a little between villages to obtain the best fits to the observed trends
Adult worm
life Negative Severity of Age groupings
expectancy binomial density Beta for age Beta for age Beta for age Beta for age (years) from groupings
R0 (years) k dependence grouping 1 grouping 2 grouping 3 grouping 4 1e4
Village
Bulidha 2.0 7.154 0.24 0.006 0.18 1.50 1.00 0.01 0e4,5e9,10e15, 16þ
Matithini 1.9 7.154 0.24 0.006 0.01 0.61 1 0.12 0e4, 5e11, 12e19,
20þ
Iietune 1.68 7.154 0.24 0.007 0.03 0.16 1 0.06 0e4,5e9,10e15, 16þ
From Butterworth, A.E., Sturrock, R.F., Ouma, J.H., Mbugua, G.G., Fulford, A.J., Kariuki, H.C., Koech, D. 1991. Comparison of different chemotherapy strategies
against Schistosoma mansoni in Machakos District, Kenya: effects on human infection and morbidity. Parasitology 103, 339e355 and Fulford, A.J., Butterworth, A.E.,
Sturrock, R.F. & Ouma, J.H. 1992. On the use of age-intensity data to detect immunity to parasitic infections, with special reference to Schistosoma mansoni in Kenya.
Parasitology 105, 219e227.

231
232 R.M. Anderson et al.

Figure 14 MCMC model fit to an observed age intensity profile for Schistosoma mansoni
in Kenya (Butterworth et al., 1991), based on four age groupings [Infants, Pre-school-aged
children (Pre-SAC), School-aged children (SAC) and Adults]. Solid line denotes model fit
and open circles are observed mean number of eggs per gram (epg) values. The fit is
good in all cases given the variation in overall transmission intensity, but could be
made better if a finer structure for age groupings were to be used (more than four clas-
ses). The parameter values estimated for each fit are listed in Table 5.

6. NUMERICAL STUDIES OF MASS DRUG


ADMINISTRATION PROGRAMMES
As noted in the introduction, the main thrust of this paper is on
models to mimic the impact of MDA to help in optimal programme design.
For this purpose, we focus attention on a deterministic model with age
Studies of the Transmission Dynamics, Mathematical Model Development 233

structure and age-dependent exposure to infection (no acquired immunity),


with a hybrid structure where the aggregation parameter of the probability
distribution of adult worms per host is fixed (constant k value) as described in
earlier publications (Anderson et al., 2015; Truscott et al., 2015). The gen-
eral conclusions derived from these models are then checked for robustness
against a fully individual-based stochastic model with dynamic probability
distributions permitting the inclusion of various complexities such as predis-
position to heavy infection and to taking up (or not) drug treatment when
offered to individuals (discussed in the following section). The sexual mating
probability term is included under the assumption of monogamy, since if
transmission interruption is a target for MDA, the unstable point becomes
important when mean worm burdens are low, especially within stochastic
simulations of times to extinction under a defined treatment programme.
Surface plots of the breakpoint in transmission below which the system
moves to parasite extinction have been presented in a previous publication
(Anderson et al., 2015). The general conclusion from this earlier study of the
properties of the deterministic age-structured model is that the current treat-
ment coverage trends in SAC and adults, extrapolated to the WHO 2020
goal of 75% coverage in SAC, are able to greatly reduce host worm burdens
in low-to-moderate transmission intensity settings. Low levels of adult
coverage have a significant impact on worm burden in all settings. Model
validation against one reinfection survey demonstrates that the age-
structured model is able to match posttreatment data well in terms of egg
output, but that some details of reinfection among school children and
young adults are not currently well represented.
Expanding on this analysis, Table 5 records parameter estimates for three
different villages in the Machakos District of Kenya of S. mansoni from a
study by Butterworth et al. (1991) derived from Markov Chain Monte
Carlo (MCMC) fits to age intensity profiles (mean epg in various age group-
ings). These fits to the observed data for the three village settings are
recorded in Fig. 14, from which it can be seen that the model characterizes
observed trends well, even given the crude assumption of four age groupings
(Infants, Pre-SAC, SAC and Adults) for the estimation of age-dependent
rates of infection. Table 5 records the variation in parameter estimates
obtained from these fits, with life expectancy, k, and density dependence
held approximately constant (note life expectancy of the adult worm set
at roughly 7 years e which is at the longer end of current estimates; see
Table 4). The variation in R0 reflects moderate transmission levels for
S. mansoni with values lying between 1.6 and 2.0.
234 R.M. Anderson et al.

Predictions of the time to extinction in years in three different Kenyan


villages (Butterworth et al., 1991), to match the model fits presented in
Fig. 14, given various effective drug treatment levels (as proportions treated)
in SAC and adults, are recorded in Table 6. Treatment coverage levels that
result in crossing the transmission breakpoint do not immediately result in
parasite extinction (where Mi * ¼ 0 in all age groups i). This takes a little
longer to be achieved, with numbers decaying over time postcrossing the
breakpoint. Infants and Pre-SAC cannot be treated with praziquantel. NA
in Table 6 denotes coverage levels which are not predicted to break trans-
mission. Note that the time to extinction is very much reduced if coverage
in adults is moderate to high. In the case of a high transmission setting

Table 6 Deterministic model predictions of time to extinction in years (crossing the


transmission breakpoint) in three different Kenyan villages (Butterworth et al., 1991)
to match the model fits presented in Fig. 14, given various effective drug treatment
levels (as proportions treated) in School-aged children (SAC) and Adults. Infants and
Pre-school-aged children (Pre-SAC) cannot be treated with praziquantel
Village: Bulidha
Adult proportion treated
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

SAC proportion 0 NA NA NA NA NA NA NA NA NA NA NA
treated 0.1 NA NA NA NA NA NA NA NA NA NA NA
0.2 NA NA NA NA NA NA NA NA NA NA NA
0.3 NA NA NA NA NA NA NA NA NA NA NA
0.4 NA NA NA NA 25 25 25 25 24 24 24
0.5 19 18 18 17 17 17 17 17 17 16 16
0.6 14 13 13 13 13 12 12 12 12 12 12
0.7 11 11 10 10 10 10 10 9 9 9 9
0.8 10 9 8 8 8 8 8 8 7 7 7
0.9 8 7 7 7 6 6 6 6 6 6 6
1 8 6 6 5 5 5 5 5 5 5 5
Village: Matithini
Adult proportion treated
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

SAC proportion 0 NA NA NA NA NA NA NA NA NA NA NA
treated 0.1 NA NA NA NA NA NA NA NA NA NA NA
0.2 NA NA NA NA NA NA NA NA 25 24 23
0.3 NA NA NA NA 24 21 20 18 18 17 17
0.4 NA NA NA 24 19 17 15 14 14 13 13
Studies of the Transmission Dynamics, Mathematical Model Development 235

Table 6 Deterministic model predictions of time to extinction in years (crossing the


transmission breakpoint) in three different Kenyan villages (Butterworth et al., 1991)
to match the model fits presented in Fig. 14, given various effective drug treatment
levels (as proportions treated) in School-aged children (SAC) and Adults. Infants and
Pre-school-aged children (Pre-SAC) cannot be treated with praziquanteldcont'd
Village: Matithini
Adult proportion treated
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0.5 NA NA NA 20 16 14 12 12 11 11 10
0.6 NA NA 25 19 15 12 11 10 9 9 8
0.7 NA NA 24 18 14 11 10 9 8 7 7
0.8 NA NA 23 17 13 11 9 8 7 6 6
0.9 NA NA 22 16 13 10 8 7 6 6 5
1 NA NA 22 16 12 10 8 7 6 5 5
Village: Iietune
Adult proportion treated
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

SAC proportion 0 NA NA NA NA NA NA NA NA NA NA NA
treated 0.1 NA NA NA NA NA NA NA NA NA NA NA
0.2 NA NA NA NA NA NA NA NA NA NA NA
0.3 NA NA NA NA NA NA NA NA 25 24 23
0.4 NA NA 23 21 20 18 18 17 17 16 16
0.5 NA 20 17 15 14 14 13 13 12 12 12
0.6 22 17 14 12 11 11 10 10 10 9 9
0.7 20 15 12 10 9 9 8 8 8 7 7
0.8 19 14 11 9 8 7 7 7 6 6 6
0.9 18 13 10 8 7 6 5 5 5 5 5
1 17 12 10 8 7 6 5 5 4 4 4
NA denotes coverage levels will not break transmission.

R0 > 2), adult coverage is essential to interrupt transmission. For the age in-
tensity profiles from which model fits were obtained, note that the age-
dependent rate of infection also greatly influences whether or not adult
treatment is essential. If this rate is high, even if R0 is moderate overall, adult
treatment is required (Table 6). It has been assumed that the convex nature
of the age intensity profile is entirely set by age-dependent exposure. If ac-
quired immunity plays a role, the estimated R0 values would be higher, but
the conclusions about the need for some treatment of adults are unlikely to
change. Interestingly, the only large-scale field trial of different MDA
strategies to control S. mansoni morbidity and transmission is that of
236 R.M. Anderson et al.

Butterworth et al. (1991) in Kenya. They concluded that, in areas of low


morbidity, chemotherapy of school children only at intervals of up to
3 years, is a satisfactory way of producing a long-term reduction in both in-
tensity of infection and morbidity.

7. STOCHASTIC MODELS e THE QUESTIONS OF


ELIMINATION AND ERADICATION
Much current attention is focused on the question of whether or not
MDA alone can break transmission, if coverage is high and the duration of
MDA is many decades (Anderson et al., 2015; Truscott et al., 2015). Table 6
well illustrates that long durations of MDA are required for parasite elimina-
tion in a defined habitat. Deterministic predictions of these durations will
give a rough guide to the expected extinction times, but stochastic models
are required to give some idea of the probability distribution of such times.
An associated issue is the important, and very general issue, of how should
‘elimination’ be defined. From an infectious disease epidemiological context
this is easy to do e it is when the effective reproduction number, R, in a
defined region is less than unity in value (R < 1), even given the immigra-
tion of infected people from other spatial locations at some defined rate.
Once R falls below unity in value (the transmission threshold is breached),
it will still take some time for the overall mean worm burden to fall to zero.
Roughly, this time will scale according to the life expectancy of the adult
worm in the human host. For a possibly monogamous dioecious species,
this concept is modified by the ‘breakpoint’ in transmission concept outlined
earlier. In this case, elimination is achieved when the unstable breakpoint is
crossed and the attractor is the state of parasite extinction (Fig. 11). When
mating functions are taken into account, the effective reproductive number,
below which extinction is the attractor, may be in excess of unity in value
and even as high as close to 2 if the parasite is truly monogamous.
WHO uses the 1998 definitions proposed by Dowdle (1998). He
defined control as a reduction in the incidence, prevalence, morbidity or
mortality of an infectious disease to a locally acceptable level; elimination
as reduction to zero of the incidence of disease or infection in a defined
geographical area; and eradication as permanent reduction to zero of the
worldwide incidence of infection (Dowdle, 1998). His definitions were
largely framed for microparasitic infections, where prevalence is the major
epidemiological measure and incidence is defined as new infections (change
from susceptible to infected) per unit of time. For the macroparasitic
Studies of the Transmission Dynamics, Mathematical Model Development 237

helminths, where parasite burden determines morbidity, and incidence can


be defined as the acquisition of new individual parasites, these definitions still
hold with some minor modification. Elimination is again simply the reduc-
tion of any new parasite infections in a defined locality to zero. However,
note that existing adult worms post transmission elimination may live for
many years and release transmission stages. Elimination requires that none
of these successfully mature in a new host. A number of case studies for
both S. mansoni and S. haematobium have been published, that record long
life spans of these parasites, often many decades, after moving to areas
with no transmission (Arnon, 1990). Thus elimination of transmission
may not equate to the overall mean parasite burden being zero. With this
notion in mind, examination of the predictions of individual-based stochas-
tic models is informative. The development of these models is described
elsewhere (Anderson and Medley, 1985; Farrell et al. (unpublished data)),
but the basic template and biological assumptions are as defined for the
age-structured deterministic model discussed in the previous section. The
dynamic parasite negative binomial distribution is generated by heterogene-
ity in exposure/parasite establishment between and within age groups
(essentially predisposition to light or heavy infection, with a parameter
determining this set at birth of the human host).
Some predictions for S. mansoni, using the data sets defined in Fig. 14,
from one of these stochastic formulations (Farrell et al. (unpublished
data)), are displayed in Fig. 15. This records changes in the overall mean
worm burden over years in the population (defined as a village of 500 peo-
ple, with an age distribution and age-specific mortality plus birth rates set to
that of rural Kenya in 2003) for five replicated simulations. Chemotherapy is
initiated in year 10, with 75% coverage in SAC and 27.4% coverage in
adults and continued for either 5 or 10 years. The deterministic model pre-
dicts that 5 years is insufficient to break transmission, but 10 years is suffi-
cient. The stochastic simulations confirm the former, but show that for
10 years of treatment, three out of the five replicate runs show bounce
back to precontrol levels of infection once treatment ceases. Fig. 16 shows
the high variance in individual parasite loads by plotting the burdens in
five people from a 10-year treatment run in which parasite elimination
did occur in the individual and the population.
Many improvements are required in the current generation of stochastic
models. Two of the most important are migration in and out of study com-
munities. When detailed studies match the full demography with the sampled
demography of those who gave faecal or urine samples, or who received drug
238 R.M. Anderson et al.

Graph (A) Graph (B)


Mean intensity of infecon (worm burden)

300 250

250
200

200
150
150

100
100

50 50

0
0
5 25 45
5 15 25 35 45

Time in years Time in years


Figure 15 Stochastic simulations of mass drug administration at 75% coverage of
School-aged children (SAC) and 27.4% coverage of Adults starting in year 10 and
continuing for either 10 years [graph (A)] or 5 years [graph (B)]. Five replicates are shown
and the plots record the mean number of eggs per gram (epg) in the total human pop-
ulation. Note that in graph (A) three replicates bounce back out of the five runs. Two rep-
licates record the elimination of transmission and parasite extinction. Parameter values
are as for Iietune village in Kenya (Fig. 14, R0 ¼ 1.68). In graph (B) none reach elimination
after 5 years of mass drug administration (MDA) and all bounce back once treatment
stops. Note the long duration of bounce back to precontrol levels which is set by adult
worm life expectancy.

250
Intensity of infecon (worm burden)

200

150

100

50

0
5 10 15 20 25
Time in years
Figure 16 One of the Iietune (in Kenya) village simulations in which the transmission of
Schistosoma mansoni is eliminated by 10 yearly rounds of treatment; model outputs are
the time courses of parasite burden [as reflected by eggs per gram (epg)] in five people.
Note the high variance between individuals.
Studies of the Transmission Dynamics, Mathematical Model Development 239

treatment, it is common to see a discrepancy in the young adult male age clas-
ses in those sampled or treated. This is often due to migration to urban centres
for work, but may also be due to a reluctance to give samples. In assessing the
impact of MDA, these migrant workers must be taken into account since they
often return to their home villages and therefore can be a source of immigrant
infections. The second much broader issue is that of social, behavioural and
environmental change due to economic growth. China provides a good
example, since rapid economic growth over the past two decades has intro-
duced clean water and sanitation to some rural areas which, in turn, impacts
the underlying transmission intensity in a positive manner. Changes in the
pristine R0 must be taken into account when mimicking the impact of
MDA imposed on top of economic development.

8. MODEL VALIDATION
A number of key conclusions can be drawn from these analyses, espe-
cially the stochastic simulations, but the policy maker is right to ask to what
extent can they be relied on, given they derive from mathematical models
that only mirror the real world. One obvious response is that in the very
complex nonlinear real world, intuition alone cannot be relied on and
some sort of quantitative experimental framework must be used. This is
the case, for example, in all areas of engineering and physics. But a reasoned
response should be based on attempts to validate the model before making
predictions. Few attempts have been made to do this (Anderson et al., 2015;
French et al., 2015), and both are based on the Anderson and May age-
structured deterministic framework. One of these relied on fitting to pre-
and posttreatment patterns to estimate the key parameters. For example,
in French et al. (2010) models were fitted to a longitudinal cohort followed
up across successive rounds of annual treatment for 4 years. This is far from
ideal since the complex underlying model with many parameters can be
manipulated to fit many different patterns.
The ideal test is to fit to pretreatment age intensity patterns and then pre-
dict what happens post-MDA given data on the drug coverage levels realized.
One attempt has been made, but in the absence of detailed information on
who received treatment. Despite this, but based on the assumption all were
treated in the village, overall the predictions well mirrored observed reinfec-
tion patterns except in the very young children (Anderson et al., 2015). The
need in the future is to repeat this exercise on epidemiological patterns for
240 R.M. Anderson et al.

S. mansoni, S. haematobium and S. japonicum, in varied settings with different


prevailing transmission intensities, where good drug coverage data on who
was treated stratified by age have been collected.

9. DISCUSSION
Three broad sets of conclusions can be drawn from the research
reviewed and reported in this paper. The first set concerns the mathematical
models themselves. Ideally, some common agreement is needed on the rela-
tive importance of acquired immunity versus age-related exposure. This
may differ between the different schistosome species, where, for example,
acquired immunity is believed to be of high importance in urinary schisto-
somiasis. Models that incorporate both factors need to be developed and
various procedures adopted to assess goodness of fit to observed epidemio-
logical trends, not just to age intensity profiles, but more importantly to
reinfection patterns.
Predisposition to both infection and the uptake of treatment (adherence)
is of great importance and rather little work has been published on model
development and the associated predictions for control measure impact.
Current limitations in analysis and model development are largely associated
with the limited data available (Shulford et al., 2016).
Some argue that snail control is an important addition to MDA and that
seasonality in transmission, due to its importance to the population biology
of the snail host, will impact the optimal timing of treatment in terms of
interrupting transmission (King et al., 2015). Investigations of this notion
are limited to date but should be carried out, ideally based on individual-
based stochastic models. It is likely, however, that the long average timescale
set by adult worm life expectancy (5e7 years) will negate a strong influence
of changes on an annual basis.
The second set of conclusions concern data and parameter estimates.
Many studies report intensity and prevalence profiles for schistosome infec-
tions in human communities and a smaller subset records reinfection data
over time post a round of MDA. Most of these studies, however, are not
accompanied by work on the estimation of the key parameters such as the
basic reproduction number and the force of infection. Such estimates are
key to understanding the potential impact of MDA programmes and the
coverage and frequency of drug treatment required to eliminate transmission
in a defined setting. Helminth epidemiology lags behind many other fields
Studies of the Transmission Dynamics, Mathematical Model Development 241

of infectious disease epidemiology in the use of statistical and computational


tools to estimate these key parameters, despite the availability of software
packages for methods such as MCMC. There are other data needed to sup-
port the use of mathematical models as tools in public health control policy
design. As mentioned earlier, good compliance data to treatment, collected
at multiple rounds and recorded at an individual level, are of obvious impor-
tance. Sadly, this is rarely done (King et al., 2015). Other key parameters are
parasite life expectancy in the human host and the precise nature of density-
dependent fecundity. Data accumulation for these processes is difficult and
dependent on autopsy studies. Little improvement in understanding is likely
in the near future. A further complexity relates to the accuracy of faecal or
urine examination for parasite eggs as diagnostic tools for measuring parasite
intensity and prevalence. As drug coverage in MDA programmes expands in
the coming decade, detecting infection in low-transmission areas becomes
more important if transmission elimination is the goal. New technologies
based on qPCR methods offer much improvement in sensitivity (Easton
et al., 2016). These need to be applied in future epidemiological studies
of age intensity and prevalence profiles (Carabin et al., 2000).
The third set of conclusions concerns the lessons from studies based on
deterministic and stochastic models of the use of MDA to eliminate transmis-
sion by drug treatment alone. Deterministic models point to the value of SAC
treatment in areas of low-to-medium transmission, provided coverage is very
high (75% at least). But they also point to the value of treating adults as well,
to shorten the time to elimination, which can be many years even when
coverage is sufficient to drive the system below the transmission breakpoint.
Given the observed age profiles of schistosome infection, little is to be gained
by treating pre-SAC with respect to transmission control. However, benefit is
likely in terms of morbidity control in the young and vulnerable.
Given the ambitions of WHO to achieve schistosome transmission elim-
ination in defined regions in the coming decade, stochastic models point to
the long durations of parasite persistence in some communities despite high
drug treatment coverage at levels above those required to break transmis-
sion. The expected outcome predicted by stochastic models is, in general,
in good agreement with deterministic predictions, but the average predicted
behaviour does not preclude parasite persistence in some communities. This
observation has major implications for any planned clinical trials [as in the
case of STH control (Brooker et al., 2015)] of the ability of MDA alone
to interrupt schistosome transmission and lead to parasite elimination in
defined communities or areas. Stochastic individual-based models can be
242 R.M. Anderson et al.

used to simulate various trial design to examine what is the most appropriate
option for testing such hypotheses. Classical cluster randomized clinical trial
designs may or may not be appropriate. The main need will be for sufficient
replication of villages at a defined coverage level to compensate for the
stochastic effects operating on transmission elimination, especially in small
populations. Stochastic simulation models will be of great value in
determining what level of replication is required in practice to test for trans-
mission elimination, and what pitfalls may arise for a given trial design.
A final issue that requires careful consideration is the quality of the cur-
rent diagnostic tools based on egg count methods. New tools are emerging
such as qPCR (quantitative polymerase chain reaction) and CCA (circu-
lating cathodic antigen) assays which suggest that in low-prevalence com-
munities these new tests are much better at detecting low-intensity
infections. If MDA progresses towards elimination in some regions, these
new tests will be of great value in assessing transmission interruption.
However, to date many uncertainties remain concerning these new methods
and much calibration is required on how qPCR methods and CCA assays
correlate with epg scores. What is not in doubt is the need for better moni-
toring and evaluation as MDA coverage rises in the coming decade and it is
to be hoped that such monitoring could be fully cross sectional and not just
targeted at the SAC for both intensity and prevalence. In addition, much
greater attention needs to be placed on recording drug coverage and individ-
ual adherence to repeated treatment community by community.
Competing interests
Roy M. Anderson is a Non-Executive Director of GlaxoSmithKline
(GSK). GlaxoSmithKline played no role in study design, data collection
and analysis, decision to publish, or preparation of the manuscript.
Authors’ contributions
Roy M. Anderson prepared the manuscript and helped to design models
and simulations. H.C. Turner participated in the design of the study. S.H.
Farrell ran model simulations, helped in the design of the study, and prepared
the manuscript. J.E. Truscott helped in the design and ran model simulations,
carried out parameter estimation procedures and prepared the manuscript. All
authors read and approved the final version of the manuscript.

ACKNOWLEDGEMENTS
Research for this article was funded by the Children’s Investment Fund Foundation (UK)
(‘CIFF’) and the Bill and Melinda Gates Foundation (BMGF). We thank Arminder Deol
for access to data from Uganda, collected as part of a CIFF-funded operational research grant
Studies of the Transmission Dynamics, Mathematical Model Development 243

to the Schistosomiasis Control Initiative. The views, opinions, assumptions or any other in-
formation set out in this paper are solely those of the authors and should not be attributed to
BMGF and CIFF or any person connected with these foundations.

REFERENCES
Anderson, R., Hollingsworth, T.D., Truscott, J., Brooker, S., 2012. Optimisation of mass
chemotherapy to control soil-transmitted helminth infection. Lancet 379, 289e290.
Anderson, R., Turner, H., Farrell, S., Yang, J., Truscott, J., 2015. What is required in terms
of mass drug administration to interrupt the transmission of schistosome parasites in re-
gions of endemic infection? Parasites Vectors 8, 553.
Anderson, R.M., 1980. The dynamics and control of direct life cycle helminth parasites. In:
Barigozzi, C. (Ed.), Vito Volterra Symposium on Mathematical Models in Biology.
Springer, Berlin Heidelberg.
Anderson, R.M., 1987. Determinants of infection in human schistosomiasis. In: Mahmoud, A.F.
(Ed.), Balliere’s Clinical Tropical Medicine, second ed. Balliere Tindall, London.
Anderson, R.M., Crombie, J.A., May, R.M., 1986. Predisposition to helminth infection in
man (reply). Nature 320, 195e196.
Anderson, R.M., Crombie, J.C., 1985. Experimental studies of age intensity and age preva-
lence profiles of infection: Schistosoma mansoni in snails and mice. In: Rollinson, D.,
Anderson, R.M. (Eds.), Ecology and Genetics of Host-parasite Interactions. Academic
Press, London.
Anderson, R.M., May, R.M., 1982. Population dynamics of human helminth infections:
control by chemotherapy. Nature 297, 557e563.
Anderson, R.M., May, R.M., 1985a. Helminth infections of humans: mathematical models,
population dynamics, and control. Adv. Parasitol. 24, 1e101.
Anderson, R.M., May, R.M., 1985b. Herd immunity to helminth infection and implications
for parasite control. Nature 315, 493e496.
Anderson, R.M., May, R.M., 1991. Infectious Diseases of Humans: Dynamics and Control.
Oxford University Press, Oxford.
Anderson, R.M., Medley, G.F., 1985. Community control of helminth infections of man by
mass and selective chemotherapy. Parasitology 90, 629e660.
Anderson, R.M., Truscott, J.E., Pullan, R.L., Brooker, S.J., Hollingsworth, T.D., 2013.
How effective is school-based deworming for the community-wide control of soil-
transmitted helminths? PLoS Negl. Trop. Dis. 7, e2027.
Arnon, R., 1990. Life span of parasite in schistosomiasis patients. Isr. J. Med. Sci. 26, 404e405.
Bensted-Smith, R., Anderson, R.M., Butterworth, A.E., Dalton, P.R., Kariuki, H.C.,
Koech, D., Mugambi, M., Ouma, J.H., Arap Siongok, T.K., Sturrock, R.F., 1987. Ev-
idence for predisposition of individual patients to reinfection with Schistosoma mansoni af-
ter treatment. Trans. R. Soc. Trop. Med. Hyg. 81, 651e654.
Berberian, D.A., Paquin Jr., H.O., Fantauzzi, A., 1953. Longevity of Schistosoma hematobium
and Schistosoma mansoni: observations based on a case. J. Parasitol. 39, 517e519.
Bradley, D.J., Mccullough, F.S., 1973. Egg output stability and the epidemiology of Schisto-
soma haematobium. II. An analysis of the epidemiology of endemic S. haematobium. Trans.
R. Soc. Trop. Med. Hyg. 67, 491e500.
Brooker, S.J., Mwandawiro, C.S., Halliday, K.E., Njenga, S.M., Mcharo, C., Gichuki, P.M.,
Wasunna, B., Kihara, J.H., Njomo, D., Alusala, D., Chiguzo, A., Turner, H.C., Teti, C.,
Gwayi-Chore, C., Nikolay, B., Truscott, J.E., Hollingsworth, T.D., Balabanova, D.,
Griffiths, U.K., Freeman, M.C., Allen, E., Pullan, R.L., Anderson, R.M., 2015. Inter-
rupting transmission of soil-transmitted helminths: a study protocol for cluster rando-
mised trials evaluating alternative treatment strategies and delivery systems in Kenya.
BMJ Open 5, e008950.
244 R.M. Anderson et al.

Bundy, D.A., Thompson, D.E., Cooper, E.S., Golden, M.H., Anderson, R.M., 1985. Pop-
ulation dynamics and chemotherapeutic control of Trichuris trichiura infection of children
in Jamaica and St. Lucia. Trans. R. Soc. Trop. Med. Hyg. 79, 759e764.
Butterworth, A.E., 1998. Immunological aspects of human schistosomiasis. Br. Med. Bull.
54, 357e368.
Butterworth, A.E., Sturrock, R.F., Ouma, J.H., Mbugua, G.G., Fulford, A.J., Kariuki, H.C.,
Koech, D., 1991. Comparison of different chemotherapy strategies against Schistosoma
mansoni in Machakos District, Kenya: effects on human infection and morbidity.
Parasitology.
Carabin, H., Chan, M.S., Guyatt, H.L., 2000. A population dynamic approach to evaluating
the impact of school attendance on the unit cost and effectiveness of school-based schis-
tosomiasis chemotherapy programmes. Parasitology 121, 171e183.
Chan, M.S., Guyatt, H.L., Bundy, D.A.P., Booth, M., Fulford, A.J., Medley, G.F., 1995.
The development of an age structured model for schistosomiasis transmission dynamics
and control and its validation for Schistosoma mansoni. Epidemiol. Infect. 115, 325e344.
Chan, M.S., Mutapi, F., Woolhouse, M.E., Isham, V.S., 2000. Stochastic simulation and the
detection of immunity to schistosome infections. Parasitology 120, 161e169.
Chandiwana, S.K., 1987. Community water-contact patterns and the transmission of Schis-
tosoma haematobium in the highyield region of Zimbabwe. Soc. Sci. Med. 25, 495e505.
Chandiwana, S.K., Woolhouse, M.E.J., Bradley, M., 1991. Factors affecting the intensity of
reinfection with Schistosoma haematobium following treatment with praziquantel. Parasi-
tology 102, 73e83.
Cheever, A.W., 1968. A quantitative post-mortem study of Schistosomiasis mansoni in man.
Am. J. Trop. Med. Hyg. 17, 38e64.
Crombie, J.A., Anderson, R.M., 1985. Population dynamics of Schistosoma mansoni in mice
repeatedly exposed to infection. Nature 315, 491e493.
Dowdle, W.R., 1998. The principles of disease elimination and eradication. Bull. World
Health Organ. 76, 22e25.
Easton, A.V., Oliveira, R.G., O’Connell, E.M., Kepha, S., Mwandawiro, C.S.,
Njenga, S.M., Kihara, J.H., Mwatele, C., Odiere, M.R., Brooker, S.J., Webster, J.P.,
Anderson, R.M., Nutman, T.B., 2016. Multi-parallel qPCR provides increased sensi-
tivity and diagnostic breadth for gastrointestinal parasites of humans: field-based infer-
ences on the impact of mass deworming. Parasites Vectors 9, 38.
Feller, W., 1968. An Introduction to Probability Theory and Its Applications, vol. 1. John
Wiley, New York.
French, M.D., Churcher, T.S., Gambhir, M., Fenwick, A., Webster, J.P., Kabatereine, N.B.,
~ez, M.G., 2010. Observed reductions in Schistosoma mansoni transmission from
Basan
large-scale administration of praziquantel in Uganda: a mathematical modelling study.
PLoS Negl. Trop. Dis. 4, e897.
French, M.D., Churcher, T.S., Webster, J.P., Fleming, F.M., Fenwick, A.,
Kabatereine, N.B., Sacko, M., Garba, A., Toure, S., Nyandindi, U., Mwansa, J.,
Blair, L., Bosqué-Oliva, E., Basan ~ez, M.G., 2015. Estimation of changes in the force
of infection for intestinal and urogenital schistosomiasis in countries with Schistosomiasis
Control Initiative-assisted programmes. Parasit. Vectors 8, 558.
Fulford, A.J., Butterworth, A.E., Ouma, J.H., Sturrock, R.F., 1995. A statistical approach to
schistosome population dynamics and estimation of the life-span of Schistosoma mansoni in
man. Parasitology 110, 307e316.
Fulford, A.J., Butterworth, A.E., Sturrock, R.F., Ouma, J.H., 1992. On the use of age-in-
tensity data to detect immunity to parasitic infections, with special reference to Schisto-
soma mansoni in Kenya. Parasitology 105, 219e227.
Gurarie, D., King, C.H., Wang, X., 2010. A new approach to modelling schistosomiasis
transmission based on stratified worm burden. Parasitology 137, 1951e1965.
Studies of the Transmission Dynamics, Mathematical Model Development 245

Gurarie, D., Yoon, N., Li, E., Ndeffo-Mbah, M., Durham, D., Phillips, A.E., Aurelio, H.O.,
Ferro, J., Galvani, A.P., King, C.H., 2015. Modelling control of Schistosoma haematobium
infection: predictions of the long-term impact of mass drug administration in Africa. Par-
asites Vectors 8, 529.
Guyatt, H.L., Chan, M.S., 1998. An investigation into the interaction between drug efficacy
and drug price of praziquantel in determining the cost-effectiveness of school-targeted
treatment for Schistosoma mansoni using a population dynamic model. Trop. Med. Int.
Health 3, 425e435.
Hackett, L.W., 1937. Malaria in Europe. Oxford University Press, Oxford.
Hairston, N.G., 1962. Population ecology and epidemiological problems. In: Ciba Founda-
tion Symposium e Bilharziasis. John Wiley & Sons, Ltd.
Hairston, N.G., 1965. On the mathematical analysis of schistosome populations. Bull. World
Health Organ. 33, 45e62.
Haswell-Elkins, M.R., Elkins, D.B., Anderson, R.M., 1987. Evidence for predisposition in
humans to infection with Ascaris, hookworm, Enterobius and Trichuris in a South Indian
fishing community. Parasitology 95, 323e337.
Hatz, C.F., 2001. The use of ultrasound in schistosomiasis. Adv. Parasitol. 48, 225e284.
Hliang, T., 1989. Epidemiological Basis of Survey Design, Methodology and Data Analysis
for Ascaris. Ascariasis and Its Prevention and Control. Taylor and Francis, London.
Jordan, P., 1972. Epidemiology and control of schistosomiasis. Br. Med. Bull. 28, 55e59.
Kihara, J.H., Muhoho, N., Njomo, D., Mwobobia, I.K., Josyline, K., Mitsui, Y., Awazawa, T.,
Amano, T., Mwandawiro, C., 2007. Drug efficacy of praziquantel and albendazole in
school children in Mwea Division, Central Province, Kenya. Acta Trop. 102, 165e171.
King, C.H., 2007. Lifting the burden of schistosomiasisedefining elements of infection-associated
disease and the benefits of antiparasite treatment. J. Infect. Dis. 196, 653e655.
King, C.H., Sutherland, L.J., Bertsch, D., 2015. Systematic review and meta-analysis of the
impact of chemical-based mollusciciding for control of Schistosoma mansoni and S. haema-
tobium transmission. PLoS Negl. Trop. Dis. 9, e0004290.
Kostizin, V.A., 1934. Symbiose, parasitisme et evolution. Hermann, Paris.
Lo, N.C., Bogoch, I.I., Blackburn, B.G., Raso, G., N’Goran, E.K., Coulibaly, J.T.,
Becke, S.L., Abrams, H.B., Utzinger, J., Andrews, J.R, 2015. Comparison of commu-
nity-wide, integrated mass drug administration strategies for schistosomiasis and soil-
transmitted helminthiasis: a cost-effectiveness modelling study. Lancet Glob. Health 3,
e629ee638.
Macdonald, G., 1965. The dynamics of helminth infections, with special reference to
schistosomes. Trans. R. Soc. Trop. Med. Hyg. 59, 489e506.
Marquet, S., Abel, L., Hillaire, D., Dessein, H., Kalil, J., Feingold, J., Weissenbach, J.,
Dessein, A.J., 1996. Genetic localization of a locus controlling the intensity of infection
by Schistosoma mansoni on chromosome 5q31-q33. Nat. Genet. 14, 181e184.
May, R.M., 1977. Togetherness among schistosomes: its effects on the dynamics of the
infection. Math. Biosci. 35, 301e343.
Medley, G., Anderson, R.M., 1985. Density-dependent fecundity in Schistosoma mansoni in-
fections in man. Trans. R. Soc. Trop. Med. Hyg. 79, 532e534.
Mitchell, K.M., Mutapi, F., Savill, N.J., Woolhouse, M.E., 2011. Explaining observed infec-
tion and antibody age-profiles in populations with urogenital schistosomiasis. PLoS
Comput. Biol. 7, e1002237.
Mitchell, K.M., Mutapi, F., Savill, N.J., Woolhouse, M.E., 2012. Protective immunity to
Schistosoma haematobium infection is primarily an anti-fecundity response stimulated by
the death of adult worms. Proc. Natl. Acad. Sci. U.S.A. 109, 13347e13352.
Mitchell, K.M., Mutapi, F., Mduluza, T., Midzi, N., Savill, N.J., Woolhouse, M.E., 2014. Pre-
dicted impact of mass drug administration on the development of protective immunity
against Schistosoma haematobium. PLoS Negl. Trop. Dis. 8, e3059.
246 R.M. Anderson et al.

Mutapi, F., Ndhlovu, P.D., Hagan, P., Spicer, J.T., Mduluza, T., Turner, C.M.,
Chandiwana, S.K., Woolhouse, M.E., 1998. Chemotherapy accelerates the develop-
ment of acquired immune responses to Schistosoma haematobium infection. J. Infect.
Dis. 178, 289e293.
Ojurongbe, O., Sina-Agbaje, O.R., Busari, A., Okorie, P.N., Ojurongbe, T.A.,
Akindele, A.A., 2014. Efficacy of praziquantel in the treatment of Schistosoma haema-
tobium infection among school-age children in rural communities of Abeokuta,
Nigeria. Infect. Dis. Poverty 3, 30.
Sanin, D.E., Prendergast, C.T., Bourke, C.D., Mountford, A.P., 2015. Helminth infection
and commensal microbiota drive early IL-10 production in the skin by CD4þ T cells
that are functionally suppressive. PLoS Pathog. 11, e1004841.
Schad, G.A., Anderson, R.M., 1985. Predisposition to hookworm infection in humans.
Science 228, 1537e1540.
Shulford, K.V., Turner, H.C., Anderson, R.M., 2016. Compliance with anthelmintic treat-
ment in the neglected tropical diseases control programmes: a systematic review. Parasites
Vectors 9, 29.
Tchuente, L.A., Shaw, D.J., Polla, L., Cioli, D., Vercruysse, J., 2004. Efficacy of
praziquantel against Schistosoma haematobium infection in children. Am. J. Trop. Med.
Hyg. 71, 778e782.
Truscott, J.E., Turner, H.C., Anderson, R.M., 2015. What impact will the achievement of the
current World Health Organisation targets for anthelmintic treatment coverage in children
have on the intensity of soil-transmitted helminth infections? Parasites Vectors 8, 551.
Truscott, J.E., Turner, H.C., Farrell, S.,H., Anderson, R.M., 2016. Soil-transmitted
helminths: mathematical models of transmission, the impact of mass drug administration
~ez, M.G., Anderson, R.M. (Eds.), Math-
and transmission elimination criteria. In: Basan
ematical Models for Neglected Tropical Diseases: Essential Tools for Control and Elim-
ination, Part B. Adv. Parasitol. 94 (in this volume).
Turner, H.C., Walker, M., French, M.D., Blake, I.M., Churcher, T.S., Basan ~ez, M.G., 2014.
Neglected tools for neglected diseases: mathematical models in economic evaluations.
Trends Parasitol. 30, 562e570.
Utzinger, J., N’goran, E.K., N’dri, A., Lengeler, C., Tanner, M., 2000. Efficacy of praziquan-
tel against Schistosoma mansoni with particular consideration for intensity of infection.
Trop. Med. Int. Health 5, 771e778.
Wallerstein, R.S., 1949. Longevity of Schistosoma mansoni; observations based on a case. Am.
J. Trop. Med. Hyg. 29, 717e722.
Warren, K.S., 1973. Regulation of the prevalence and intensity of schistosomiasis in man:
immunology or ecology? J. Infect. Dis. 127, 595e609.
Warren, K.S., 1982. Selective primary health care: strategies for control of disease in the
developing world. I. Schistosomiasis. Rev. Infect. Dis. 4, 715e726.
Wilkins, H.A., Blumenthal, U.J., Hagan, P., Hayes, R.J., Tulloch, S., 1987. Resistance to
reinfection after treatment of urinary schistosomiasis. Trans. R. Soc. Trop. Med. Hyg.
81, 29e35.
Woolhouse, M.E., 1991. On the application of mathematical models of schistosome trans-
mission dynamics. I. Natural transmission. Acta Trop. 49, 241e270.
Woolhouse, M.E., 1998. Patterns in parasite epidemiology: the peak shift. Parasitol. Today
14, 428e434.
Woolhouse, M.E., Taylor, P., Matanhire, D., Chandiwana, S.K., 1991. Acquired immunity
and epidemiology of Schistosoma haematobium. Nature 351, 757e759.
Yang, H.M., Coutinho, F.A.B., Massad, E., 1997. Acquired immunity on a schistosomiasis
transmission model d fitting the data. J. Theor. Biol. 188, 495e506.
CHAPTER FIVE

River Blindness: Mathematical


Models for Control and
Elimination
M.G. Basan~ ez*, 1, a, M. Walker*, a, H.C. Turner*, L.E. Coffengx,
S.J. de Vlasx, W.A. Stolkx
*Imperial College London, London, United Kingdom
x
University Medical Center, Rotterdam, Rotterdam, The Netherlands
1
Corresponding author: E-mail: m.basanez@imperial.ac.uk

Contents
1. Introduction 250
1.1 Life cycle of Onchocerca volvulus 250
1.2 Control and elimination of onchocerciasis and the role of mathematical 251
modelling
1.2.1 The Onchocerciasis Control Programme in West Africa (OCP, 1974e2002) 252
1.2.2 The African Programme for Onchocerciasis Control (APOC, 1995e2015) 255
1.2.3 The Onchocerciasis Elimination Program for the Americas (OEPA, 1993epresent) 256
1.2.4 Accelerating the control and elimination of onchocerciasis 257
1.3 Alternative treatment strategies 257
1.3.1 Increasing coverage, adherence and frequency of ivermectin treatment 257
1.3.2 Other microfilaricidal therapies (moxidectin) 258
1.3.3 Macrofilaricidal therapies (doxycycline) 258
1.3.4 Vector control 259
1.4 The NTD Modelling Consortium and review aims 259
2. The Models 260
2.1 EPIONCHO 260
2.1.1 Parasite population regulation in humans 261
2.1.2 Parasite population regulation in vectors 269
2.1.3 Pretreatment parasite dynamics 270
2.1.4 Posttreatment parasite dynamics 272
2.1.5 Model outputs 280
2.2 ONCHOSIM 281
2.2.1 Human population demography 281
2.2.2 Parasite population regulation in humans 282
2.2.3 Parasite population regulation in vectors 286
2.3 Comparison between EPIONCHO and ONCHOSIM 290

a
These authors contributed equally to the work.
Advances in Parasitology, Volume 94
© 2016 Elsevier Ltd.
j
ISSN 0065-308X
http://dx.doi.org/10.1016/bs.apar.2016.08.003 All rights reserved. 247
248 ~ez et al.
M.G. Basan

3. Model Validation 302


3.1 EPIONCHO 302
3.2 ONCHOSIM 303
4. Modelling Current Treatment Strategies 304
4.1 Defining elimination endpoints 305
4.2 Treatment with ivermectin 306
4.2.1 Epidemiological and programmatic variables 307
5. Modelling Alternative Intervention Strategies 309
5.1 Moxidectin treatment 312
5.2 Macrofilaricidal treatment 312
5.3 Vector control 314
5.4 An onchocerciasis vaccine 315
6. Economic Evaluations 315
6.1 Burden of disease 316
6.1.1 Blindness and visual impairment 316
6.1.2 Skin disease 317
6.1.3 Excess mortality 317
6.2 Costing intervention strategies 318
6.3 The economics of elimination and eradication 318
7. Challenges and Future Directions 319
7.1 Relationship between operational endpoints, transmission breakpoints and 320
stochastic fade-out
7.2 Estimating basic and effective reproduction ratios 321
7.3 Modelling the diagnostic performance of the skin snip method and serological 324
assays in near-elimination scenarios
7.4 Modelling hypoendemic onchocerciasis 326
7.5 Spatial models of onchocerciasis transmission 327
8. Conclusions 328
Acknowledgements 329
References 329

Abstract
Human onchocerciasis (river blindness) is one of the few neglected tropical diseases
(NTDs) whose control strategies have been informed by mathematical modelling.
With the change in focus from elimination of the disease burden to elimination
of Onchocerca volvulus, much remains to be done to refine, calibrate and validate
existing models. Under the impetus of the NTD Modelling Consortium, the teams
that developed EPIONCHO and ONCHOSIM have joined forces to compare and
improve these frameworks to better assist ongoing elimination efforts. We review
their current versions and describe how they are being used to address two key ques-
tions: (1) where can onchocerciasis be eliminated with current intervention strategies
River Blindness 249

by 2020/2025? and (2) what alternative/complementary strategies could help to


accelerate elimination where (1) cannot be achieved? The control and elimination
of onchocerciasis from the African continent is at a crucial crossroad. The African Pro-
gramme for Onchocerciasis Control closed at the end of 2015, and although a new
platform for support and integration of NTD control has been launched, the
disease will have to compete with a myriad of other national health priorities at a
pivotal time in the road to elimination. However, never before had onchocerciasis
control a better arsenal of intervention strategies as well as diagnostics. It is, therefore,
timely to present two models of different geneses and modelling traditions as they
come together to produce robust decision-support tools. We start by describing
the structural and parametric assumptions of EPIONCHO and ONCHOSIM; we
continue by summarizing the modelling of current treatment strategies with annual
(or biannual) mass ivermectin distribution and introduce a number of alternative stra-
tegies, including other microfilaricidal therapies (such as moxidectin), macrofilaricidal
(anti-wolbachial) treatments, focal vector control and the possibility of an onchocer-
ciasis vaccine. We conclude by discussing challenges, opportunities and future
directions.

List of Abbreviations
ABR Annual biting rate
APOC African Programme for Onchocerciasis Control
ATP Anniual transmission potential
CDTI Community-directed treatment with ivermectin
CMFL Community microfilarial load
DALY disability-adjusted life years
ESPEN Expanded Special Project for Elimination of Neglected Tropical Diseases
MDA Mass drug administration
mf Microfilariae
mg Milligram
MTP Monthly transmission potential
NBD Negative binomial distribution
NTD Neglected tropical disease
OCP Onchocerciasis Control Programme in West Aftica
OEPA Onchocerciasis Elimination Program for the Americas
PCT Preventative chemotherapy
pOTTIS Provisional Operational Thresholds for Treatment Interruption and commence-
ment of Surveillance
REMO Rapid epidemiological mapping of onchocerciasis
R0 Basic reproduction number
SAE Severe adverse event
s.l. Sensu lato
s.s. Sensu stricto
ss Skin snip
TOVA The Onchocerciasis Vaccine for Africa Initiative
WHO World Health Organization.
250 ~ez et al.
M.G. Basan

1. INTRODUCTION
Human onchocerciasis, a neglected tropical disease (NTD) also known
as ‘river blindness’, is a parasitic infection caused by the filarial nematode
Onchocerca volvulus Leuckart (Filarioidea: Onchocercidae). Transmission
among hosts occurs via the bites of Simulium Latreille flies (Diptera: Simulii-
dae). The name river blindness illustrates that the worst sequela of the infec-
tion is the loss of vision and that the vectors breed in fast flowing water.
Onchocerciasis has been endemic in 27 sub-Saharan African countries, the
Yemen and 6 Latin American countries, with an estimated 37 million people
infected and 90 million at risk, 99% of them in Africa (APOC, 2005). For a
map of the global distribution and status of control as of 2006 see Basan ~ez
et al. (2006). Due to the efforts of intervention programmes such as the
Onchocerciasis Control Programme in West Africa (OCP), the African Pro-
gramme for Onchocerciasis Control (APOC) and the Onchocerciasis Elim-
ination Program for the Americas (OEPA) (Richards et al., 2001),
onchocerciasis has been successfully eliminated in some foci of Mali, Senegal
(Diawara et al., 2009; Traore et al., 2012) and Nigeria (Tekle et al., 2012), in
Mexico (Rodríguez-Pérez et al., 2015), in the northern foci of Venezuela
(Convit et al., 2013), in Colombia (West et al., 2013) and Ecuador (Lovato
et al., 2014). From the early years of the OCP until present, mathematical
modelling has played an important role to support planning, evaluation
and decision making in large-scale control programmes (Remme, 2004a;
~ez and Ricardez-Esquinca, 2001). In this chapter we will review the
Basan
contribution that mathematical models have made to our understanding of
the epidemiology, population biology, transmission dynamics, control and
feasibility of elimination of onchocerciasis with current and novel interven-
tion tools and strategies.

1.1 Life cycle of Onchocerca volvulus


In this section we describe some of the salient features of the parasite’s life
cycle that determine the dynamics of the infection. The adult worms
(macrofilariae) are dioecious (separate sexes) and, therefore, reproduction
requires mating of male and female worms. (In modelling terms, a mating
probability e the probability that female worms are mated e needs to be
included.) Macrofilariae live in subcutaneous palpable nodules called
onchocercomata and in deeper, inaccessible worm bundles, for an average
of 10  3 years (Duerr et al., 2003; Plaisier et al., 1991a). This is the
River Blindness 251

longest-lived parasite stage and mostly responsible for the protracted dynamics
characteristic of onchocerciasis. Females are fertilized by males during distinct
reproductive cycles, with three to four such cycles occurring each year
(Schulz-Key and Karam, 1986). Therefore, unlike other nematodes, female
O. volvulus oscillates between being nonfertile and being fertile, producing
thousands of embryos (microfilariae) throughout their fertile cycles and
millions during their lives. A heavily infected person may harbour 50e
200 million of them. Microfilariae live for about 12e15 months in the skin
(Duke, 1993). If they are ingested by simuliid species competent for O.
volvulus, they will develop into infective L3 larvae during the extrinsic incu-
bation period, whose duration e in the poikilotherm insect vectors e is tem-
perature dependent, with a mean of 8  3 days (Cheke et al., 2015). L3 larvae
enter another human via the bite wound made when the blackfly next feeds
and mature into reproductive adult stages during the intrinsic incubation
period within 12e18 months (Duke, 1991; Nelson, 1991). The prepatent
period from infection to production of a sizeable microfilarial population,
detectable by the typical skin snip diagnostic method, can be as long as 3 years
(Prost, 1980). For a pictorial representation of the life cycle see Fig. 1.
Those microfilariae not taken up by the vectors die within skin and eye
tissues, provoking an immune response to the released somatic antigens,
which leads to the inflammatory lesions that underlie onchocerciasis-
associated pathology. In addition, proinflammatory cytokines released from
Wolbachia bacteria that live in mutualistic symbiosis with the worms generate
inflammatory responses that are also involved in the cutaneous and ocular
pathogenesis of the disease (Brattig, 2004; Saint André et al., 2002).

1.2 Control and elimination of onchocerciasis and the role of


mathematical modelling
Onchocerciasis is an exemplary case among human helminthiases in which
mathematical models have been used to inform control policy (e.g., inter-
vention strategy, programme duration, frequency of application, health
impact) since the early stages of their implementation (Basan ~ez et al.,
2012), due to a long-lasting and fruitful partnership between research and
control activities (Remme et al., 1995; Remme, 2004a). In the next sections
we review the three major onchocerciasis control and elimination initiatives
that have been put in place worldwide and highlight the questions addressed
by models within these programmes. We proceed to discuss how best can
models work together to tackle remaining challenges within the current
impetus for transmission interruption and parasite elimination at a global
252 ~ez et al.
M.G. Basan

Figure 1 Life cycle of Onchocerca volvulus. Mean dimensions of parasite stages are:
adult females, 35e70 cm  400 mm; adult males, 2e4 cm  150e200 mm; microfilar-
iae, 250e360  5e9 mm; L1 larvae, 200 mm  12 mm (front) and 20 mm (rear); L3,
440e700  20 mm. L1 larvae moult into L2, preinfective larvae, and L2 into L3, infective
larvae. (Illustration: Giovanni Maki, derived from a CDC image at http://www.dpd.cdc.
gov/dpdx/HTML/Filariasis.htm) (Basan ~ez et al., 2006).

scale (World Health Organization, 2012; African Programme for


Onchocerciasis Control, 2012) and particularly in Africa (World Health
Organization/African Programme for Onchocerciasis Control, 2015).

1.2.1 The Onchocerciasis Control Programme in West Africa (OCP,


1974e2002)
The OCP started as a vector control programme aiming to eliminate
the public health burden of blinding onchocerciasis in savannah areas of
River Blindness 253

West Africa. From a core, original area of 654,000 km2 in nine countries, it
was successively extended to 1,300,000 km2 in 11 countries to prevent rein-
vasion by flies from uncontrolled areas (Boatin, 2008; Boatin and Richards,
2006). O’Hanlon et al. (2016) have used model-based geostatistics to pro-
duce maps of the initial microfilarial prevalence in the OCP. The OCP
initially aimed at interrupting transmission by reducing simuliid populations
(via weekly aerial distribution of larviciding insecticides in the blackfly
breeding sites) for long enough to curtail acquisition of new infections
and let the adult worm population die of natural attrition.
Towards the end of the 1980s, the microfilaricidal drug ivermectin
(MectizanÔ ) was licensed for human use after a series of clinical trials, which
have been metaanalysed using mathematical modelling by Basan ~ez et al.
(2008). A standard dose of ivermectin (150 mg/kg of body weight, orally)
causes a 98e99% reduction in skin microfilarial density within 1e2 months
after treatment (due to the microfilaricidal effect of the drug), with new
microfilariae gradually repopulating the skin (as female worms resume the
production of live microfilariae) from the third month posttreatment on-
wards. In addition to reducing transmission from humans to vectors, the
microfilaricidal effect of ivermectin helps to prevent the morbidity associated
with onchocerciasis, for which microfilariae are essentially responsible.
Therefore, ivermectin is a drug in the arsenal of the so-called preventive
chemotherapy treatment (PCT), a strategy endorsed by the World Health
Organization (WHO) to tackle some of its prioritized NTDs (World Health
Organization, 2006). The OCP conducted community trials of mass iver-
mectin distribution in Ghana, showing that it was feasible and safe to admin-
istrate treatment at a large scale (Remme et al., 1989, 1990b; Alley et al.,
1994). In 1987, Merck & Co. Inc. announced the donation of ivermectin
for as long as necessary to combat river blindness (Colatrella, 2008). The
OCP adopted an annual treatment strategy by 1989, which was used to
complement vector control in some areas and as the only intervention in
most of the extension areas (Boatin, 2008). In some of these (extension)
foci, mass ivermectin treatment was provided biannually (6-monthly)
(Diawara et al., 2009). Ivermectin mass treatment was initially delivered
by mobile teams and nongovernmental organizations and later via commu-
nity-based distribution, overseen by trained nurses and/or technicians.
Finally, this evolved over several years into community-directed treatment
with ivermectin (CDTI), the preferred mode of drug delivery e a strategy
pioneered by APOC to enhance its long-term sustainability (Boatin,
2008). In 1995 Kim and Benton (1995) estimated that between 1974 and
254 ~ez et al.
M.G. Basan

2002, the overall operations of the OCP would have averted 593,440 cases
of blindness.
During the earlier (vector control) stages of the OCP, mathematical
models were used to determine: (1) the threshold biting rates of the savannah
members of the Simulium damnosum sensu lato (s.l.) complex, below which
endemic onchocerciasis would not be able to persist (Dietz, 1982), (2) the
dynamics of recolonization by blackflies of breeding sites after spraying of
the larvicides (Birley et al., 1983) and (3) the duration of vector control
that would be necessary to interrupt transmission and prevent recrudescence
(Plaisier et al., 1991b). It was estimated that at least 300e700 bites per person
per year by S. damnosum sensu stricto/S. sirbanum (depending on whether the
local vector population was more or less anthropophagic) would be neces-
sary for the basic reproduction ratio (Section 7.2), R0, of the parasite to be
one or greater (Dietz, 1982). These figures were confirmed by subsequent
modelling studies using a precursor of the EPIONCHO model (described
in Section 2.1) (Basan~ez and Boussinesq, 1999) and a refinement of the Dietz
(1982) model (Duerr and Eichner, 2010). In the absence of immigration of
infected humans or invasion by infected flies, the microsimulation model
ONCHOSIM (developed in the early 1990s by Plaisier et al., 1990;
described in Section 2.2) predicted that 14 years of full-scale vector control
would be required to reduce the risk of recrudescence to less than 1%
(Plaisier et al., 1991b). These projections were confirmed by the epidemio-
logical trends presented by Hougard et al. (2001).
As the OCP integrated vector control with ivermectin distribution,
ONCHOSIM was used to estimate the reduction in programme duration
that could be achieved by combining vector control with ivermectin mass
treatment (Plaisier et al., 1997), as well as to assess the feasibility of elimi-
nating onchocerciasis with ivermectin treatment alone (Winnen et al.,
2002). It was concluded that 20e25 years of treatment at a high coverage
would be required in foci with very high precontrol endemicity levels to
achieve elimination. These projections were confirmed by subsequent
modelling studies (Turner et al., 2013a; Coffeng et al., 2014a; Stolk
et al., 2015). Presently, although onchocerciasis-associated morbidity has,
by and large, been eliminated in most of the former OCP area, transmission
remains in some foci and countries (due to conflict interrupting control
operations, incomplete geographical coverage, and suboptimal responses
to ivermectin treatment, among other factors) (Frempong et al., 2016;
Hodges et al., 2011; Lamberton et al., 2015; Lloyd et al., 2015; Osei-
Atweneboana et al., 2007).
River Blindness 255

1.2.2 The African Programme for Onchocerciasis Control (APOC,


1995e2015)
In view of the potential of ivermectin for onchocerciasis control if high
levels of geographic and therapeutic coverage could be achieved, APOC
was launched in 1995 (Remme, 1995) under the auspices of the WHO to
cover the remaining endemic African countries outside the OCP. Countries
and areas in need of treatment were mapped through rapid epidemiological
mapping of onchocerciasis (REMO) (Ngoumou and Walsh, 1993; Noma
et al., 2002). Zouré et al. (2014) have used model-based geostatistics to
map the prevalence of onchocercal nodules (the basis of REMO) in the
APOC area. APOC’s original objective was to establish, within an initial
period of 12 years, effective and self-sustainable CDTI projects in order to
control morbidity from onchocerciasis (World Health Organization, 1996;
Amazigo et al., 2002). CDTI was implemented in 16 of 20 mapped
countries, covering endemic areas inhabited by roughly 71.5 million people
in 1995 (Coffeng et al., 2013a). APOC delivered its first treatments through
CDTI in 1997 (80,000 treatments). The programme was progressively
scaled up to reach an overall therapeutic coverage of about 73% in 2010
(75.8 million treatments). By the end of 2013, around 100.7 million people
in 132,919 communities were receiving ivermectin (World Health
Organization/African Programme for Onchocerciasis Control, 2014). In
2014, the number of people receiving ivermectin had increased to around
112.5 million people, despite the fact that no treatments occurred in Liberia
and Sierra Leone, due to the Ebola outbreak, with an estimated coverage of
78% by 2015, the final year of the programme (World Health Organization/
African Programme for Onchocerciasis Control, 2015).
In 2010, and motivated by the successful elimination of the infection in
some foci of Mali and Senegal (Diawara et al., 2009) e within the western
extension of the OCP and without vector control e APOC shifted its goals
from the control of onchocerciasis morbidity to elimination of the parasite
reservoir, where feasible, with annual (or biannual) ivermectin MDA
(African Programme for Onchocerciasis Control, 2010). Other examples
of onchocerciasis elimination in APOC projects provided further evidence
to support the feasibility of this change in programme objectives (Tekle
et al., 2012; Higazi et al., 2013).
ONCHOSIM was used to model the impact, on achieving elimination
in APOC settings, of increasing the frequency of ivermectin mass treatment
(Coffeng et al., 2014a). Doubling the frequency of treatment from yearly to
6-monthly or 3-monthly was predicted to reduce remaining programme
256 ~ez et al.
M.G. Basan

duration by about 40% or 60%, respectively. Using data on precontrol


prevalence of infection and MDA coverage, trends in infection, visual
impairment and blindness (eye disease) and severe itch (skin disease) between
1995 and 2010 were simulated, estimating that during that period
8.2 million disability-adjusted life years (DALYs) due to onchocerciasis
were averted in APOC areas, at a nominal cost of about US$257 million.
The ONCHOSIM model was also used to project health impact and cost
for the final period of 2011e2015, anticipating that APOC will have averted
another 9.2 million DALYs at a nominal cost of US$221 million during its
final phase (Coffeng et al., 2013a).

1.2.3 The Onchocerciasis Elimination Program for the Americas


(OEPA, 1993epresent)
Since its inception in 1993 OEPA was conceived as an elimination
programme (both of the morbidity associated with onchocerciasis and of
the transmission and reservoir of infection) in its six constituent Latin Amer-
ican countries (Blanks et al., 1998; Sauerbrey, 2008). This approach was
based on the perceived strength and visibility that a cohesive regional
onchocerciasis initiative would have (for otherwise a low-priority disease
in the continent). Biannual (6-monthly) ivermectin mass treatment was
adopted as the main strategy, since epidemiological and entomological
studies conducted in areas where the vector was S. ochraceum s.l. indicated
that twice-yearly treatment could lead to interruption of transmission
despite high vector density and initial endemicity, provided high coverage
levels could be achieved (Collins et al., 1992; Cupp et al., 1992). In some
communities with high precontrol onchocerciasis endemicity and
transmission intensity, a regimen of 3-monthly (quarterly) ivermectin treat-
ments was implemented, helping to accelerate interruption of transmission
(Rodríguez-Pérez et al., 2008).
The SIMON simulation model, originally developed for forest oncho-
cerciasis in Sierra Leone (Davies, 1993) was adapted for the Americas
(SIMON-A), and although not formally published, has been used to inform
OEPA’s regional strategy. All versions of the SIMON-A model indicate that
at the end of the treatment programme, there will be a period of 3e5 years
during which some members of a community will continue to be infected at
a low parasite level, giving an indication of the period necessary for post-
treatment surveillance (Rodríguez-Pérez et al., 2011). A precursor of the
deterministic EPIONCHO model was used to investigate the role of regular
nodulectomy campaigns (excision of palpable onchocercomata) in addition
River Blindness 257

~ez
to that of ivermectin for the control of onchocerciasis in Mexico (Basan
and Ricardez-Esquinca, 2001).
Presently, onchocerciasis transmission has been eliminated from
Colombia (West et al., 2013); Ecuador (Lovato et al., 2014), Mexico
(Rodriguez-Pérez et al., 2015) and northern Venezuela (Convit et al.,
2013), but it remains in the hard-to-reach indigenous Yanomami popula-
tion inhabiting the Amazon rainforest straddling Venezuela and Brazil
(Botto et al., 2016).

1.2.4 Accelerating the control and elimination of onchocerciasis


In January 2012 Dr. Margaret Chan, Director General of the WHO, pre-
sented a roadmap for accelerating work to overcome the global impact of
NTDs. This announcement inspired the launching of the London Declara-
tion on NTDs and set targets and milestones for elimination during the period
2015e2020 (Uniting to Combat Neglected Tropical Diseases, 2012; World
Health Organization, 2012). Regarding onchocerciasis, regional elimination
in the Americas was anticipated for 2015 (achieved in most foci with the
exception of the hard-to-reach Yanomami population of the Amazonian
focus (World Health Organization, 2014; Botto et al., 2016). For Africa,
the roadmap set goals of elimination by 2020 in selected African countries
and by 2025 in 80% of African countries (African Programme for Onchocer-
ciasis Control, 2012). However, modelling work has indicated that in some
areas with high precontrol endemicity (and hence with conditions propitious
to intense transmission), annual CDTI e the predominant control strategy in
Africa e may not be sufficient for the achievement of the elimination goals in
the proposed time frameworks (Turner et al., 2013a; Coffeng et al., 2014a;
Stolk et al., 2015). For these areas, as well as for others, lagging behind the
elimination goals, implementation of alternative or complementary strategies
may be necessary (World Health Organization/African Programme for
Onchocerciasis Control, 2015). In the next section we describe some of these
strategies, which will be revisited later in the chapter within the context of our
current modelling work to provide rigorous and quantitative decision-
making support to the elimination endeavour.

1.3 Alternative treatment strategies


1.3.1 Increasing coverage, adherence and frequency of ivermectin
treatment
If programmatic difficulties have slowed down progress, an increased effort
to improve treatment coverage and adherence may be necessary. For areas
258 ~ez et al.
M.G. Basan

starting to implement CDTI or needing to switch strategies, the most


obvious option may be increasing the frequency of ivermectin treatment
(to twice or four times per year if possible), as highlighted by the Latin
American experience. In Section 6, we discuss the effectiveness and cost
effectiveness of these strategies as explored by combining epidemiological
and cost data with mathematical models (Coffeng et al., 2014a; Turner
et al., 2013b; Turner et al., 2014c).

1.3.2 Other microfilaricidal therapies (moxidectin)


Moxidectin (a macrocyclic lactone derivative of the milbemycins and not of
the avermectins, as is the case for ivermectin) has not yet been licensed for
human use. However, phase II and phase III clinical trials have been con-
ducted in Africa to assess its safety and efficacy for the treatment of human
onchocerciasis (Awadzi et al., 2014). Motivated by these studies, modelling
has been used to investigate its potential use in MDA programmes for the
control and elimination of O. volvulus infection in comparison with iver-
mectin (Turner et al., 2015a). Salient results will be discussed in Section 5.1.

1.3.3 Macrofilaricidal therapies (doxycycline)


If attainment of high levels of treatment coverage and adherence has been
hindered by coendemicity of onchocerciasis and loiasis (caused by the filarial
nematode Loa loa), as is the case in some forest foci of Cameroon (Wanji
et al., 2015a,b), microfilaricidal MDA treatment with ivermectin (or mox-
idectin) may be contraindicated in some epidemiological settings. This is
because severe adverse events (SAEs), such as encephalopathy, have been
reported following ivermectin treatment in a proportion of individuals
with high L. loa microfilaraemia (Mackenzie et al., 2007). Maps of loiasis
prevalence distribution have been developed to aid understanding of the
true extension of the problem (Zouré et al., 2011). In some of these areas
doxycycline may be deployed on a test-and-treat basis. Doxycycline is effec-
tive against adult O. volvulus by depleting the worms from their Wolbachia
endosymbiotic bacteria, essential for parasite reproduction and survival
(Taylor et al., 2014). Although not microfilaricidal, O. volvulus microfilariae
from treated individuals do not develop at their normal rate in their simuliid
vectors, potentially impacting on transmission (Albers et al., 2012). Besides,
L. loa does not contain Wolbachia, rendering doxycycline safe for treatment
of coinfected patients. However, key barriers to using doxycycline as an
MDA strategy for widespread community-based control are the logistics
of a relatively lengthy course of treatment (4e6 weeks) and
River Blindness 259

contraindications in children under eight years and in pregnancy. Mathe-


matical modelling has been used to quantify the macrofilaricidal efficacy
of doxycycline (Walker et al., 2015) and will be discussed in Section 5.2.

1.3.4 Vector control


Large-scale vector control, by weekly spraying of blackfly breeding sites
with larvicidal insecticides, as it was successfully implemented by the OCP
(Hougard et al., 2001), has been modelled using ONCHOSIM as an instan-
taneous reduction in the level of blackfly biting followed by immediate
recolonization (and resumption of blackfly biting) following cessation of
control (Plaisier et al., 1997; Alley et al., 2001). Under these assumptions,
large-scale vector control has been shown to be highly complementary to
annual ivermectin MDA, such that combined strategies e even in highly
endemic settings e can achieve elimination with a lowered risk of recrudes-
cence in little over 10 years (Plaisier et al., 1997).

1.4 The NTD Modelling Consortium and review aims


Both the WHO roadmap and the London Declaration on NTDs provided
the impetus for the establishment, in 2014, of the Bill & Melinda Gates
Foundation-supported NTD Modelling Consortium (http://www.
ntdmodelling.org/). The motivation of the consortium lies in the recogni-
tion that many urgent policy issues concerning the control and elimination
of NTDs can only be answered through the use of quantitative tools, and
that this can only be truly achieved through strong collaborations between
modellers, epidemiologists, policy makers and field epidemiologists (Basan ~ez
and Anderson, 2015).
The issues posed by the NTD Modelling Consortium regarding oncho-
cerciasis elimination can be distilled into two main questions: (1) where (and
under which epidemiological scenarios) do models predict that onchocerci-
asis can be eliminated with current strategies by the timelines proposed in the
WHO (2012) roadmap and (2) in those scenarios where elimination cannot
be achieved using the current strategy, which (and where) alternative and
complementary intervention strategies should be deployed to facilitate/
accelerate progress towards elimination. These questions, and the modelling
work necessary to address them, are particularly poignant at a time when
onchocerciasis control and elimination activities are being devolved to the
national health programmes and systems of endemic countries in Africa
following APOC’s closure at the end of 2015.
260 ~ez et al.
M.G. Basan

Therefore, this chapter aims to review the progress made towards


addressing these questions and supporting the WHO (2012) goals, and to
highlight areas of future work. We commence by reviewing the structure
of the two main onchocerciasis models that have been included under the
umbrella of the NTD Modelling Consortium. We do this in order to pro-
vide, in a single document, the formal description of both models in their
current phase of development. We continue by comparing some of their
salient features in the context of providing reliable elimination projections
and by describing the efforts that have been made thus far to validate such
models. We proceed by reviewing the modelling of current interventions
as well as of alternative treatment strategies. Finally, we discuss the challenges
faced in addressing the questions posed by the NTD Modelling Consortium
and indicate future research directions that are essential to refine our models
in order to best serve onchocerciasis control and elimination efforts in areas
of ongoing transmission.

2. THE MODELS
2.1 EPIONCHO
EPIONCHO is a deterministic, population-based model that uses
partial differential equations to describe, with respect to time and host
age, the rate of change of the mean numbers of nonfertile, N, and fertile,
F, adult female worms per host; of microfilariae per milligram of skin, M,
and of infective (L3) larvae, L, per blackfly vector. A model prototype
(without explicit age structure) was presented by Basan ~ez and Boussinesq
(1999) and extended to incorporate host age and sex to account for varying
age- and sex-specific microfilarial profiles in endemic areas of northern
Cameroon, central Guatemala and southern Venezuela (Filipe et al.,
2005). Fig. 2 presents these profiles and the corresponding model fit. The
main blackfly vectors in each of these areas are, respectively, S. damnosum
sensu stricto (s.s.)/S. sirbanum (in African savannah), S. ochraceum s.l.
(in Mesoamerica) and S. guianense s.l. (in the Amazonian focus). For this
chapter we focus on the parameterization of EPIONCHO with data on
human demography, parasitology and entomology of northern Cameroon
(representing areas of Sudan-type savannah in Africa). Detailed equations
are given in Section 2.1.3.1. Table 1 lists state variables and parameters in
EPIONCHO, and provides their definitions, notation, values and sources.
River Blindness 261

Figure 2 Observed and EPIONCHO-predicted age profiles of microfilarial load (mf/mg)


by sex for three different ecological settings at endemic equilibrium. The circles and error
bars are, respectively, the mean and standard error of observed mf/mg within w10-year
age groups (open circles denote women; solid circles denote men) and the lines are the
model fit (to individual) data from (A) northern Cameroon (w5000 individuals); (B) central
Guatemala (w900 individuals) and (C) southern Venezuela (w1000 individuals) (dashed
lines represent women and solid lines represent men) (Filipe et al., 2005).

2.1.1 Parasite population regulation in humans


2.1.1.1 Parasite establishment
Following Dietz (1982), the probability of L3 larvae establishing (and devel-
oping into adult worms) within the human host, dH, is assumed to be
described by a decreasing function of the rate at which L3 larvae are acquired
when the host is bitten (which itself depends on the vector biting rate, m b,
and the mean number of L3 larvae per fly, L), the so-called annual transmis-
sion potential (ATP, No. L3/person/year). The parameters of this (negative)
density-dependent parasite establishment rate were estimated by fitting the
relationship between the (arithmetic) age- and sex-standardized mean
microfilarial load in the community, M, and the ATP (m b L, in yeare1)
measured for each community for a number of villages with associated
parasitological and entomological data, as described in Basan ~ez and
Boussinesq (1999) and Basan~ez et al. (2002),
Table 1 Definition and values of variables and parameters for EPIONCHO

262
Expression,
average value
Symbol Definition of variables and parameters and units Sources
Parasite in human host
Ns,d(t,a) Mean number of nonfertile female adult worms Eq. (20) Turner et al. (2013a)
per person at time (t) and age (a); subscript s
denotes host sex and d denotes treatment
adherence category
Fs,d(t,a) Mean number of fertile female adult worms per Eq. (21)
person at time (t) and age (a); subscripts s and
d as above
Ms,d(t,a) Mean number of microfilariae (mf ) per Eq. (22)
milligram of skin at time (t) and age (a);
subscripts s and d as above
dH[L(t)] Proportion of L3 larvae developing to adult dH0 þdHN cH mb LðtÞ ~ez and Boussinesq
Basan
1þcH mb LðtÞ
worms within the human host as a function of (1999)
the number of infective larvae received per
unit time
d H0 Proportion of L3 larvae developing to adult 0.0712e0.0854 ~ez et al. (2002) and
Basan
worms within the human host when Filipe et al. (2005)
mbL(t) / 0
d HN Proportion of L3 larvae developing to adult 0.00299

M.G. Basan
worms within the human host when
mbL(t) / N
5.86  103 year per L3 larva

~ez et al.
cH Severity of transmission intensity-dependent
parasite establishment within the human host
p Prepatent period (from infection with L3 larvae 2 year Filipe et al. (2005)
to presence of detectable microfilariae in the
skin)
River Blindness
mH Per capita death rate of human hosts 0.02 year1
0.04 year1 ~ez and Boussinesq
Basan
(1999) and Filipe et al.
(2005)
sW Per capita death rate of adult worms 0.1 year1 Plaisier et al. (1991a)
sM0 Per capita death rate of microfilariae in the 0.8 year1 Duke (1993)
absence of ivermectin
6 Per capita rate at which untreated, 0.59 year1 ~ez et al. (2008)
Basan
nonreproducing female worms become fertile
l0 Per capita rate at which untreated fertile female 0.33 year1
worms become nonfertile in the absence of
treatment
ε Per capita rate of production of microfilariae per 1.1538 year1 Turner et al. (2013a)
fertile female worm scaled by the total weight
(in milligrams) of microfilariae-bearing skina
am Maximum recorded human age in the reference 80 year Filipe et al. (2005)
population of northern Cameroon
r(a) Probability density function of host age a (using a mH expðmH a Þ Filipe et al. (2005)
1expðmH am Þ
truncated exponential distribution of survival
times)
hd Proportion of the host population in adherence e Turner et al. (2013a)
group d
qs Proportion of the host population in sex group s qF ¼ 0.45; qM ¼ 0.55 Filipe et al. (2005)
Parasite in simuliid vector
L(t) Mean number of infective larvae per fly Eq. (24) Turner et al. (2013a)
at time (t)

263
m Vector to host ratio 609, for ABR ¼ 19,000 bites For 70% prevalence
person1 year1
(Continued)
Table 1 Definition and values of variables and parameters for EPIONCHOdcont'd

264
Expression,
average value
Symbol Definition of variables and parameters and units Sources
b Biting rate per fly on humans assuming a human 31.2 year1 ~ez and Boussinesq
Basan
blood index ¼ 0.3 (1999)
dV 0
dV[Ms,d(t,a)] Proportion of microfilariae developing to the ½1þcV Ms;d ðt;aÞ
infective stage within the vector, per bite  
sL[Ms,d(t,a)] Per capita net rate of loss of L3 larvae from aH
þ sL þ mV þ aV Ms;d ðt; aÞ
g
vectors
d V0 Proportion of microfilariae developing to L3 0.0207
within vectors when Ms,d(t,a) / 0, per bite
cV Severity of density-dependent limitation of 0.0148
larval development within (savannah) vectors
aH Proportion of infective, L3 larvae shed per bite 0.54e0.8 Renz (1987) and Duke
(1973)
g Average duration between consecutive blood- 0.0096 year ~ez and Boussinesq
Basan
meals (1999)
sL Per capita death rate of L3 larvae within the 52e104 year1
vector
m V0 Per capita death rate of uninfected blackflies 26e52 year1
aV Parasite-induced death rate of infected blackflies 0.597
 year1 per microfilaria ~ez et al. (2002)
Basan
0
U s(a) Age- and sex-specific measure of exposure to Es gs E0 ; a < a Filipe et al. (2005)

M.G. Basan
0 0
vectors Es gs exp½as ða a Þ; a > a
Es Sex-specific exposure to vector bites EF ¼ 0.90; EM ¼ 1.08
E0 Fraction of exposure at age 0 in relation to that at 0.10; in Cameroon a0 was

~ez et al.
age a’ from which exposure changes with age estimated to be ¼ 0
River Blindness
gs Normalization factors to ensure that the gF ¼ 0.548; gM ¼ 1.154 Filipe et al. (2005)
distribution of bites among age groups
sums to 1
as Age-specific change in contact rate with vectors aF ¼ 0.023; aM ¼ 0.007 Filipe et al. (2005)
for human hosts of sex s
Mating probability and parasite prevalence
f[Ws,d(t,a), kW] Mating probability at time t, Eq. (25) Turner et al. (2013a)
age a, sex s and treatment adherence group d
Ws,d(t,a) Mean number of female adult worms per person Ns,d(t,a) þ Fs,d(t,a) Turner et al. (2013a)
at time (t) and age (a), s denotes sex and
d denotes treatment adherence category
kW Inverse measure of degree of overdispersion in 0.35 Bottomley et al. (2008)
the distribution of worms among hosts
pd(t) Microfilarial prevalence at time t in adherence Eq. (39) Turner et al. (2013a)
group d
kM[Md(t)] Inverse measure of the degree of overdispersion Eq. (41) Turner et al. (2014a)
in the distribution of skin microfilariae among
hosts of adherence group d, as a function of
the mean microfilarial load
k0 Parameters of the relationship between kM and 0.013 Turner et al. (2014a)
skin microfilarial load
k1 0.025
a
Note that ε is different from ε0 the scaled per capita rate of microfilarial production by a (mated) female worm (regardless of fertility status), whose value is 0.667 year1
~ez and Boussinesq, 1999).
(Basan

265
266 ~ez et al.
M.G. Basan

ε0 dH ðLÞmbL
M ðmbLÞ ¼ ; (1)
2ðsW þ mH ÞðsM0þ mH Þ
dH0 þ dHN cH mbL
dH ðLÞ ¼ ; (2)
1 þ cH mbL
where m is the vector to human ratio (V/H); b is the biting rate per fly on
humans; L is the average number of infective (L3) larvae per fly; ε0 is the per
capita rate of microfilarial production per (mated) female worm, scaled by
the total weight (in milligrams) of microfilariae-bearing skin; sW is the per
capita mortality rate of adult worms; sM0 the per capita mortality rate of
microfilariae (in the absence of microfilaricidal treatment); mH the per capita
death rate of human hosts; dH0 and dHN are, respectively, the maximum and
minimum establishment probabilities of L3 larvae (L) within humans
(as transmission rate tends to zero or becomes infinitely large), and measures
the severity of transmission rate-dependent constraints upon L3 establish-
ment within humans. Parameter b is the product of multiplying the biting
rate per fly (the reciprocal of the mean duration between two consecutive
blood meals, g, taken as 3.5 days) times the proportion of blood meals taken
on human hosts, h. (i.e., b ¼ h/g). Previous runs of EPIONCHO had
assumed that one-third of the blood meals are of human origin (h ¼ 0.33) in
Cameroon, based on Disney and Boreham (1969), but a more recent and
extensive study of S. damnosum host choice in Ghana (Lamberton et al.,
2012, 2016) indicates that two-thirds of blood meals may be taken on hu-
man hosts by S. damnosum s.s./S. sirbanum (h ¼ 0.67). Also, note that the
estimated value of dHN (0.003) is very similar to the success ratio (sr) of
0.0031 used by ONCHOSIM, as described in Section 2.2.2.2 (Tables 1, 3
and 5).
Other authors (Duerr et al., 2003) have assumed, instead, that the parasite
establishment rate is an increasing (positive density-dependent) function of
the number of adult worms already established, describing a phenomenon
of immunosuppression that would explain profiles of microfilarial load
that increase with host age. In EPIONCHO this phenomenon is explained
by age- (and sex)-dependent exposure to blackfly bites as described in Filipe
et al. (2005). The contact rate per human with vectors is m b Us(a), with m
and b as above and,

Es gs E0 ; a < a0
Us ðaÞ (3)
Es gs exp½ as ða  a Þ; a  a0
0
River Blindness 267

is an age (a)- and sex (s)-specific measure of exposure to vectors with


parameters
P E0, Es and as, and population average equal to 1 (ensured by
Es qs ¼ 1 and factors gs, where qs is the proportion of the population in
s
each sex). Relative exposure (of males with respect to females) is defined by
Q ¼ EM/EF. After an increase in exposure, approximated by a step function
(E0 < 1), during childhood period a0 (Dietz, 1982; Remme et al., 1986),
contacts can increase (as < 0), decrease (as > 0) or remain constant (as ¼ 0)
with age. The value of a0 was estimated to be equal to zero for Cameroon
(Filipe et al., 2005).
2.1.1.2 Mating probability of female adult worms
Following May (1977), the probability that a female worm is mated depends
on: (1) the sex ratio (assumed to be 1:1 in Onchocerca volvulus, Schulz-Key,
1990); (2) the sexual system (assumed to be polygamous, Schulz-Key and
Karam, 1986) and (3) the distribution of adult worms among the human
host population (assuming that males and females are distributed together
following a negative binomial with overdispersion parameter kW, Duerr
et al., 2004). The mating probability, f(W,kW), is an increasing (positive
density-dependent) function of mean female (nonfertile, N plus fertile, F)
worm burden, W ¼ N þ F, whose shape and rate of increase (from 0 to
1) is inversely influenced by the value of kW (the smaller the value, the stron-
ger the degree of parasite overdispersion and the faster the mating probability
increases with mean worm burden) (Anderson and May, 1985),
 
W ðkW þ1Þ
f½W ; kW  ¼ 1  1 þ : (4)
kW

2.1.1.3 Excess human mortality


In addition to excess mortality due to onchocerciasis-associated blindness
(Kirkwood et al., 1983), there is a density-dependent relationship between
microfilarial load and relative risk of mortality (Little et al., 2004b; Walker
et al., 2012b). Some versions of EPIONCHO have incorporated the
relationship between microfilarial load and blindness incidence (Little
et al., 2004a), and between microfilarial load and human excess mortality
(Little et al., 2004b; Walker et al., 2012b) in order to link the core infection
model with a disease model for the estimation of disease burden and cost-
effectiveness of control interventions (Turner et al., 2014a, 2014b). Both
relationships are parameterized using: (1) the microfilarial load lagged by
2 years (reflecting that blindness and excess mortality are associated with
268 ~ez et al.
M.G. Basan

past rather than current infection) and (2) the mean number of microfilariae
per skin snip (as opposed to per mg of skin), which is designated M 0
(as opposed to M ) and is derived from the modelled microfilarial load
(per mg of skin) assuming an arithmetic mean skin snip sample weight of
1.7 mg estimated using data from Collins et al. (1992). The relative risk of
blindness associated with infection is given by a simple log-linear function
of M 0 (lagged by 2 years, i.e., M 0 ðt 2Þ, but with the time dependence
omitted for brevity). By contrast, relative risk of mortality associated with
(past) infection is nonlinearly related to M 0 and host age, a, by the expression
exp½ f ðM 0 Þaw  where,
b1 M 0b3
f ðM 0 Þ ¼ ; (5)
ð1 þ b2 M 0b3 Þ
and parameters b ¼ {b1,b2,b3} and w were estimated by fitting to the
longitudinal OCP dataset collected from 1974 through 2001 (Walker et al.,
2012b). Results indicate that for a given microfilarial load the relative risk of
mortality is statistically significantly greater in those aged less than 20 years
(Fig. 3).

Figure 3 Observed and fitted relative mortality risk with Onchocerca volvulus
microfilarial skin load according to age group. Individuals <20 years old (open squares
and solid line, respectively); individuals 20 years old (open circles and dashed line,
respectively). The fitted sigmoid doseeresponse model is adjusted to the average
age of the respective age groups. Shaded (grey) areas around the fitted lines represent
95% Bayesian credible intervals; error bars represent 95% confidence intervals around
observations. Inset permits visual inspection of the mortality relative risk at parasite
loads 40 microfilariae per skin snip (Walker et al., 2012b).
River Blindness 269

2.1.2 Parasite population regulation in vectors


2.1.2.1 Parasite establishment
Following Basan ~ez et al. (1994, 1995, 2009) and Soumbey-Alley et al.
(2004), the probability of microfilariae (ingested by the vector) developing
into L3 larvae, dV(M), is a decreasing (negative density-dependent) function
of microfilarial load in African savannah vectors such as S. damnosum s.s./
S. sirbanum. The parameters of this function have been estimated by fitting
the relationship between thoracic microfilariae (L1 uptake) and microfilarial
load, L(M) ¼ dV(M)M, assuming that once established within the thoracic
muscles, L1 larvae will develop to the infective stage, with,
dV0
dV ðM Þ ¼ ; (6)
ð1þ cV M Þ
dV0 M
LðM Þ ¼ ; (7)
1 þ cV M
where dV0 is the initial slope or maximum establishment probability (as
microfilarial load tends to zero), and cV the severity of density-dependent
constraints acting on microfilariae (M). The maximum number of larvae
establishing (as microfilarial load becomes infinitely large) is given by dV0 =cV
and, with the parameter values given in Table 1 (0.0207/0.0148), is equal to
1.4. (In ONCHOSIM, this maximum value is 1.2, as described in Section
2.2.3.1; Tables 3 and 5.)

2.1.2.2 Excess vector mortality


Following Basan ~ez et al. (1996, 2009), there is a density-dependent relation-
ship between microfilarial intake (linearly related to skin microfilarial load,
~ez et al., 1994; Demanou et al., 2003) and the rate of vector mortality,
Basan
such that the life expectancy of infected flies decreases with the number of
ingested microfilariae. Blackfly mortality as a function of microfilarial load,
mV(M), is given by,
mV ðM Þ ¼ mV0 þ aV M ; (8)
where mV0 is the background per capita mortality rate taken as the reciprocal
of the life expectancy of uninfected flies (Basan ~ez et al., 1996) and aV the
excess rate of mortality per microfilaria (Basan~ez and Boussinesq, 1999). The
overall per capita rate of loss of infective larvae, sL(M) is, therefore, deter-
mined by their own background death rate, sL, the vector mortality rate,
270 ~ez et al.
M.G. Basan

mV(M) and the rate at which L3 larvae are shed when flies bite (on human or
nonhuman blood hosts, aH/g),
sL ðMÞ ¼ sL þ mV0 þ aV M þ ðaH =gÞ; (9)
where aH is the proportion of L3 larvae shed per bite, varying between 0.54
(Renz, 1987) and 0.8 (Duke, 1973) and g the average duration between two
consecutive blood meals, also known as the length of the gonotrophic cycle,
taken as stated above to be 3.5 days (Crosskey, 1990).
The EPIONCHO precursors (Basanez and Boussinesq, 1999; Basanez
and Ricardez-Esquinca, 2001), prior to the introduction of human host
age structure (Filipe et al., 2005), have incorporated the full set of density-
dependent processes described here, with the model used in Basan ~ez et al.
(2007) also including the rates of progression through the larval L1, L2,
and L3 stages within the simuliid vector. Age-structured versions such as
Filipe et al. (2005) have subsumed the regulatory processes within the vector
into a single (linearly increasing) relationship between microfilarial load and
excess fly mortality; accordingly, different parameter values have been used.

2.1.3 Pretreatment parasite dynamics


2.1.3.1 Endemic equilibrium situation
Filipe et al. (2005) have presented the EPIONCHO model for the situation
of endemic equilibrium, i.e., assuming that the parasite population is in equi-
librium with respect to time before the inception of control interventions.
This assumption was justified by comparing and confirming stability of
microfilarial prevalence and intensity reported in the 1970s by Anderson
et al. (1974) and in the late 1980s by Boussinesq (Boussinesq, 1991; Basan ~ez
and Boussinesq, 1999) in the same area of northern Cameroon. We provide
the equations describing mean numbers of worms with respect to host age
here to facilitate access by readers and to highlight modifications that were
introduced by Turner et al. (2013a),
dNs ðaÞ 1
¼ m b Us ða pÞdH ðL  ÞL  expð mH pÞ þ l0 Fs ðaÞ
da 2
 ð6þ sW ÞNs ðaÞ (10)

dFðaÞs
¼ 6 Ns ðaÞ  ½l0 þ sW Fs ðaÞ; (11)
da
such that adding Eqs (10) and (11) gives the total mean number of adult
female worms (W ¼ N þ F) per host,
River Blindness 271

dWs ðaÞ 1 
¼ m b Us ða pÞdH ðL  Þ L  expð mH pÞ  sW Ws ðaÞ; (12)
da 2
dMs ðaÞ
¼ f½Ws ðaÞ; kW εFs ðaÞ  sM0 Ms ðaÞ; (13)
da
dLs ðaÞ
¼ b Us ðaÞdV ½Ms ðaÞM ðaÞ  sL ½Ms ðaÞLs ðaÞ; (14)
da
X Z

L ¼ qs rðaÞUs ðaÞLs ðaÞda: (15)
s
a

In addition to the parameters described above, ε is the per capita rate of


microfilarial production per fertile female worm scaled by the total weight (in
milligrams) of microfilariae-bearing skin (notice that this parameter is
different from ε0 , defined in Section 2.1.1.1); p is the prepatent period
(the period between infection with infective larvae and microfilariae being
detectable in the skin (Prost, 1980); l0 is the per capita rate at which un-
treated fertile female worms become nonfertile in the absence of treatment;
6 is the per capita rate at which untreated, nonreproducing female worms
become fertile; qs is the proportion of the human population in sex group
(women, men) s and r(a) is the proportion of individuals alive at age a.

2.1.3.2 Transmission seasonality


In some foci, the breeding sites of the simuliid vectors dry up and biting rates
dwindle to zero during the dry months of the year, transmission being
confined to the rainy months when the vectors have repopulated the
breeding sites. This is for instance the case of S. sirbanum in the western exten-
sion of the OCP, potentially decreasing the effectiveness of ivermectin treat-
ment if it is not timed to ensure minimal skin microfilarial levels when biting
rates are highest (Diawara et al., 2009). Some versions of EPIONCHO
(Turner et al., 2015b) have been modified to permit the instantaneous biting
rate of blackfly vectors (the number of bites received per person per unit time)
BR ¼ m b to vary within the year according to a sinusoidal functional that has
been used to model seasonality of malaria transmission (Griffin et al., 2010),
BRðtÞ ¼ X  fc þ ð1 cÞ ½1=2 þ 1=2  cos½2p ðt uÞk g: (16)
Here, X is the instantaneous peak biting rate within the year, c defines the
minimum biting (¼c  X, with (0  c  1), u defines the position of the
transmission peak and k is a shape parameter, with k > 0. In foci with
272 ~ez et al.
M.G. Basan

seasonal transmission, the endemic equilibrium conditions given in Section


2.1.3.1 correspond to the annual biting rate (ABR), the instantaneous biting
rate averaged over the year,
Z1
ABR ¼ BRðtÞdt (17)
0

2.1.4 Posttreatment parasite dynamics


2.1.4.1 Ivermectin
In EPIONCHO, the dynamics of skin microfilarial load and of the propor-
tion of adult female worms producing live microfilariae following iver-
mectin treatment with the standard dose of 150 mg/kg are modelled
according to the results of the systematic review, metaanalysis and modelling
presented by Basan~ez et al. (2008). Briefly, microfilaridermia is reduced by
half after 24 h, by 85% after 72 h, by 94% after 1 week and by 98%e99%
after 1e2 months (microfilaricidal effect), the latter also corresponding to
the time when the fraction of females harbouring live microfilariae is at its
lowest (embryostatic effect), reduced by around 70% from its original value.
After the first 2 months following treatment, microfilariae gradually reap-
pear in the skin following the resumption of microfilarial production by
the female worms. The microfilaricidal effect (excess mortality of microfilar-
iae due to treatment), sM1 ðsÞ, is modelled as,
sM1 ðsÞ ¼ ðs þ vÞu ; (18)
where s is time after treatment, v is a constant added to time after treatment
to allow for a very large, yet finite, microfilaricidal effect at the point of
treatment and u is a shape parameter for the per capita death rate of
microfilariae following treatment. The embryostatic effect (the treatment-
induced per capita rate at which fertile females become nonfertile), l1(s), is
modelled as,
l1 ðsÞ ¼ lMAX
1 expð4sÞ; (19)
where lMAX
1 is the maximum rate of treatment-induced sterility, and 4 is the
rate of decay of this effect with time after treatment.
At present, EPIONCHO does not assume that standard dose ivermectin
treatment has a direct macrofilaricidal (killing of the adult worms) effect, but
an antimacrofilarial effect (a cumulative reduction in the rate of microfilarial
production by adult females of 7% per dose) has been assumed in more
River Blindness 273

recent versions (Turner et al., 2014c) to account for results presented by


Gardon et al. (2002) and Cupp and Cupp (2005). The modelling of this cu-
mulative, antimacrofilarial effect is described in Section 2.1.5.

2.1.4.2 Moxidectin
The temporal dynamics of skin microfilarial loads from the ivermectin treat-
ment arm in the phase II moxidectin study (Awadzi et al., 2014) were within
the range observed by Basan~ez et al. (2008) (Fig. 4A). Moxidectin treatment
was assumed to exert the same types of effects on the parasite as ivermectin.
Therefore, moxidectin’s effects were parameterized by fitting the functions
used by Basan~ez et al. (2008) (Eqs (18) and (19) above) to the percentage
reduction in skin microfilarial densities from pretreatment, measured
8 days, 1, 2, 3, 6, 12 and 18 months after a single dose of 8 mg moxidectin
(91e186 mg/kg) (Fig. 4B).

Figure 4 The dynamic effect of a single dose of ivermectin (A) and moxidectin (B) on
skin microfilarial load. The data points are derived from skin microfilarial loads (the
mean of four microfilarial counts) collected from (A) 45 individuals treated with iver-
mectin and (B) 38 individuals treated with moxidectin, as part of the Phase II clinical
safety trial of moxidectin for the treatment of onchocerciasis (Awadzi et al., 2014).
The effect of a single dose of ivermectin previously fitted to microfilarial load data
by Basan~ez et al. (2008) is shown as a solid line in (A). The microfilarial dynamics induced
by ivermectin are not re-estimated here and hence provide a validation of the previous
parameterization. The dynamical effect of moxidectin was fitted to the trial data on
microfilarial loads from treated participants using the same approach as in Basan ~ez
et al. (2008) and is shown as a solid red line in (B). Error bars are the 95% confidence
intervals which in some circumstances were narrower than the plotted data point
and so are not discernible (Turner et al., 2015b).
274 ~ez et al.
M.G. Basan

2.1.4.3 Coverage and adherence


Whilst (therapeutic) coverage describes the proportion of the population
treated at a particular treatment round, adherence (compliance) describes
the degree to which individuals adhere correctly to the treatment schedule.
In EPIONCHO, the human host population (and subsequently the parasite
population) is partitioned into different treatment groups according to how
regularly they receive ivermectin treatment: (1) a fully compliant group who
takes treatment every round; (2) two semicompliant groups who take treat-
ment every other round alternately, and (3) a systematically noncompliant,
fourth, group who never takes treatment.

2.1.4.4 Infection dynamics


The following set of partial differential equations (omitting time and age
dependencies on the left hand side for simplicity) describes the dynamics
of infection intensity in human and vector hosts under treatment with iver-
mectin (or moxidectin), with subscript s denoting host sex and d denoting
treatment adherence category,
vNs;d vNs;d 1
þ ¼ m b Us ða pÞdH ½Lðt pÞLðt pÞexpð mH pÞ
vt va 2 (20)
þ ½l0 þ l1 ðsÞFs;d ðt; aÞ  ð6þ sW ÞNs;d ðt; aÞ

vFs;d vFs;d
þ ¼ 6 Ns;d ðt; aÞ  ½l0 þ l1 ðsÞ þ sW Fs;d ðt; aÞ (21)
vt va
vMs;d vMs;d  
þ ¼ f Ws;d ðt; aÞ; kW εjd ðtÞFs;d ðt; aÞ
vt va (22)
 ½sM0 þ sM1 ðsÞMs;d ðt; aÞ

vLs;d vLs;d    
þ ¼ b Us ðaÞdV Ms;d ðt; aÞ Ms;d ðt; aÞ  sL Ms;d ðt; aÞ Ls;d ðt; aÞ
vt va
(23)
XX Z
LðtÞ ¼ qs hd rðaÞUs ðaÞLs;d ðt; aÞda (24)
s d a
Function l1(s) denotes the excess per capita rate at which fertile females
become nonfertile following treatment (embryostatic effect), with s being
the time since last treatment (Eq. (19) above); f[Ws,d(t,a),kW] the mating
probability as described in Eq. (4) and Section 2.1.4.5; sM1 ðsÞ is the excess
per capita death rate of microfilariae following ivermectin or moxidectin
River Blindness 275

~ez et al., 2008, Eq. (18)), and jd(t)


treatment (microfilaricidal effect, Basan
the average value of the factor modifying (decreasing) female worm fertility
in adherence group d when a cumulative effect of treatment is considered
(Section 2.1.5).

2.1.4.5 Mating probability


It is assumed that the distribution of adult worms among hosts of the same
adherence group is adequately described by a negative binomial distribution
(NBD) with mean (female) worm load, Ws,d(t,a), and overdispersion param-
eter, kW. Assuming that: (1) male and female worms are distributed together,
(2) they are polygamous, i.e., a single male has the potential to fertilize all
females within a host (Schulz-Key and Karam, 1986; Hildebrandt et al.,
2012) and (3) they have a balanced (1:1) worm sex ratio, the probability
that a female worm is mated (May (1977) is,
 
  Ws;d ðt; aÞ ðkW þ1Þ
f Ws;d ðt; aÞ; kW ¼ 1  1 þ : (25)
kW
Note that the degree of overdispersion of the adult worm population
(inversely measured by the value of kW) is assumed to be unaffected by
treatment.

2.1.4.6 Cumulative effect of treatment on female worm fertility


At any time after the start of a simulated treatment programme, the worm
population in adherence group d comprises worms previously exposed to
different numbers of ivermectin (or moxidectin) treatments. This is because
(1) worms continually infect hosts throughout the treatment programme
and (2) hosts in different adherence groups receive different numbers of treat-
ments at different times. If ivermectin or moxidectin are assumed to suppress
cumulatively the fertility of female O. volvulus, then the average reduction in
fertility of the worm population will critically depend on the fraction of
worms exposed to different numbers of treatments. To this end, n was defined
as the maximum number of previous exposures to ivermectin, and n þ 1 sub-
models were formulated to track worm populations acquired during discrete
time intervals throughout the course of a simulated treatment programme.
Note that n varies among adherence groups (for example, for systematic non-
compliers n ¼ 0), and exposure group (number of treatments to which
worms have been exposed, j), as some worms, acquired after the final treat-
ment, will be unexposed to treatment ( j ¼ 0). (The possibility of unexposed
worms gives rise to the n þ 1 (as opposed to n) submodels.)
276 ~ez et al.
M.G. Basan

Consider a treatment programme starting at time s0 (that is, the first dose
of ivermectin or moxidectin is administered at time t ¼ s0 ). Worms exposed
to all n treatments ( j ¼ n) are those that were acquired at time t < s0 . By
redefining the rate of establishment of female adult worms as (first term of
the right hand side of Eq. (20)),
1
Ls ðt; aÞ ¼ m b Us ða pÞdH ½Lðt pÞLðt pÞexpð mH pÞ; (26)
2
the rate of establishment of adult female worms that will be exposed to all n
treatments (i.e., those acquired before the commencement of treatment) in
adherence group d can be expressed as,

Ls ðt; aÞ for 0 < t < s0
Ls;d;j¼n ðt; aÞ ¼ (27)
0 otherwise:
By contrast, unexposed female worms ( j ¼ 0) are those acquired after
the last treatment which, if the n treatments were administered at frequency
f (where f ¼ 1 represents annual treatment and f ¼ 2 represents biannual
treatment), indicates that infection occurred at t > s0 þ (n  1)/f. (In this
chapter we explore an annual (ivermectin, moxidectin) or a 6-monthly
(ivermectin) treatment frequency.) That is,

Ls ðt; aÞ for s0 þ ðn  1Þ=f < t < N
Ls;d;j¼0 ðt; aÞ ¼ : (28)
0 otherwise:
It follows that the rate of establishment of adult worms exposed to the
intervening numbers of treatments j ¼ 1, 2,., n  1 is given by,

Ls ðt; aÞ for s0 þ ðn  1  jÞ=f < t < s0 þ ðn  jÞ=f
Ls;d;j¼0 ðt; aÞ ¼
0 otherwise:
(29)
These conditions are used to define partial differential equations for the
mean number of female adult worms, Ws,d,j(t, a), in each exposure group
j ¼ 0, 1,., n,
vWs;d;j ðt; aÞ vWs;d;j ðt; aÞ
þ ¼ Ls;d;j ðt; aÞ  sW Ws;d;j ðt; aÞ: (30)
vt va
Note that for the purposes of tracking adult worms exposed to different
numbers of treatments, the fertility status (fertile/nonfertile) of female
worms is not distinguished. Taking the expectation of Ws,d,j(t,a) with respect
to host age a and sex s yields,
River Blindness 277

X Z
Ws;d ðtÞ ¼ q rðaÞWs;d;j ðt; aÞda;
s s
(31)
a

where r(a), the probability density function of host age, a, is


mH expðmH aÞ
rðaÞ ¼ : (32)
1  expðmH am Þ
Summing over treatment exposure groups gives the mean number of
worms per host in adherence group d,
X
j¼n
Wd ðtÞ ¼ Wd;j ðtÞ: (33)
j¼0

The fraction of the total female worm population in treatment exposure


group j, denoted ud,j(t), is now given by,
Wd;j ðtÞ
ud;j ðtÞ ¼ : (34)
Wd ðtÞ
Each subsequent exposure to ivermectin or moxidectin (after the first
exposure) was assumed to cause a 7% reduction in female worm fertility
(varied in the sensitivity analysis between 1% e weak cumulative effect e
and 30% e strong cumulative effect), such that the fertility of female worms
exposed to j treatments, Jj, is given by,

1 for j ¼ 0
Jj ¼ (35)
ð1  zÞj1 for j > 0
with parameter z ¼ 0.01 (minimum value), 0.07 (nominal value) or 0.3
(maximum value) (Table 2). The maximum value was motivated by the
findings of the modelling study conducted by Plaisier et al. (1995),
fitting to data from Alley et al. (1994) on the first community trial of
annual ivermectin treatment in the then highly hyperendemic focus of
Asubende in Ghana using ONCHOSIM. Note that for j ¼ 0 (and for
j ¼ 1), Jj ¼ 1 indicates that EPIONCHO assumes that worms previously
unexposed to treatment ( j ¼ 0), or exposed to a single, first dose ( j ¼ 1)
have, respectively, full fertility, or the potential to regain full fertility.
Subsequent treatments may cause a cumulative reduction of female worm
fertility.
The average value of the factor modifying the per capita microfilarial
production rate of (fertile) female worms in adherence group d, jd(t), is
Table 2 Definition and values of parameters and variables for ivermectin or moxidectin treatment effects in EPIONCHO

278
Symbol Definition of variables and parameters Expression, average value and units Sources

n Maximum number of previous 0 for those hosts never taking treatment to Turner et al. (2013a)
exposures to ivermectin by worms in 15 (annual) or 30 (biannual) for those
a given adherence group taking all treatments
f Frequency of treatment Ivermectin: Annual or biannual Turner et al. (2013a)
Moxidectin: Annual
s Time since last ivermectin (or years e
moxidectin) treatment
l1(s) Excess per capita rate at which fertile lMAX
1 expð4 sÞ year1 ~ez et al. (2008)
Basan
females become non-fertile
following treatment (embryostatic
effect)
lMAX
1 Maximum rate of treatment-induced Ivermectin: 32.4 year1 Turner et al. (2015b)
sterility Moxidectin: 462 year1
4 Rate of decay of treatment-induced Ivermectin: 19.6 year1 Turner et al. (2015b)
sterility with time after treatment Moxidectin: 4.83 year1
sM1 ðsÞ Excess per capita death rate of (s þ n)u year1 ~ez et al. (2008)
Basan
microfilariae following ivermectin
treatment (microfilaricidal effect)
n Constant added to time after treatment Ivermectin: 0.0096 years Turner et al. (2015b)
to allow for a very large, yet finite, Moxidectin: 0.04 years
microfilaricidal effect at the point of

M.G. Basan
treatment
u Shape parameter for the per capita Ivermectin: 1.25 Turner et al. (2015b)
death rate of microfilariae following Moxidectin: 1.82

~ez et al.
treatment
River Blindness
s0 Time at which treatment programme e e
starts
Ls,d,j(t,a) The rate of establishment of female Eqs (27)e(29) Turner et al. (2013a)
adult worms at time t in hosts of age
a, sex s, treatment adherence group
d and exposure group (number of
treatments to which worms have
been exposed) j
Ws,d,j(t,a) Mean number of female adult worms at Eqs (30) and (31) Turner et al. (2013a)
time (t) in hosts of age (a); sex s,
treatment adherence group d and
treatment exposure group j
ud,j(t) The fraction of the total worm Eq. (34) Turner et al. (2013a)
population in treatment exposure
group j
Jj The fertility of adult worms in Eq. (35) Turner et al. (2013a)
treatment exposure group j
z The per dose reduction in fertility 0.01, 0.07, 0.30 Turner et al. (2015b)
caused by treatment when a
cumulative effect is assumed
jd(t) The average value of the factor Eq. (36) Turner et al. (2013a)
modifying female worm fertility in
adherence group d

279
280 ~ez et al.
M.G. Basan

calculated using the fraction of the total worm population in each treatment
exposure group ud,j(t) (Eq. (32)) and Jj (Eq. (33)),
X
j¼n
jd ðtÞ ¼ Jj ud;j ðtÞ: (36)
j¼0

Definitions and values of parameters for treatment effects in EPION-


CHO are given in Table 2.

2.1.5 Model outputs


2.1.5.1 Intensity of infection
In this chapter infection intensity refers to microfilarial load in those aged
20 years (to facilitate comparison with ONCHOSIM results, which use
the community microfilarial load [CMFL]). However, CMFL (Remme
et al., 1986) refers to a geometric mean microfilarial load per skin snip in
those aged 20 years, rather than to an arithmetic mean microfilarial load
per mg of skin, as used in EPIONCHO. Microfilarial load in those aged
20 years is calculated from Eq. (22) above by integrating over age (from
a ¼ 20 to a ¼ am, the maximum human age of 80 years recorded in the
Cameroonian datasets) and summing over sex s and adherence group d,

XX Z m
a¼a

M ðtÞ20 ¼ qs hd r0ðaÞMs;d ðt; aÞda; (37)


s d
a¼20

where r0 (a)
is the probability density function of host age between 20 and
am ¼ 80 years,
mH expð mH aÞ
r0 ðaÞ ¼ ; (38)
½expð mH 20Þ  expðmH am Þ
and mH is the per capita death rate of humans.

2.1.5.2 Prevalence of infection


According to the current version of EPIONCHO, overall microfilarial
prevalence in adherence group d (pd(t)) is derived by using a relationship
between prevalence and microfilarial load at the community level in
Cameroon described in Basan ~ez and Boussinesq (1999) and reparameterized
by Turner et al. (2014a). This relationship assumes that skin microfilarial load
per person is distributed according to an NBD with mean Md(t) and
overdispersion parameter kM. The best fit to the microfilarial prevalence
River Blindness 281

vs. intensity relationship was obtained when kM was allowed to be a function


of the mean. Assuming that the degree of microfilarial overdispersion does
not depend on adherence group, pd(t) is given by,
 kM ½Md ðtÞ
Md ðtÞ
pd ðtÞ ¼ 1  1 þ (39)
kM ½Md ðtÞ
where Md(t) is given by,
X Z
Md ðtÞ ¼ qs rðaÞMs;d ðt; aÞda; (40)
s
a

and kM is given by,


k0 Md ðtÞ
kM ½Md ðtÞ ¼ : (41)
1 þ k1 Md ðtÞ
The overall population prevalence at time t was obtained by summing
pd(t) across adherence groups,
X
pðtÞ ¼ hd pd ðtÞ: (42)
d

2.2 ONCHOSIM
ONCHOSIM is a stochastic, individual-based model for the transmission
and control of onchocerciasis. This model describes a dynamic human pop-
ulation, consisting of a discrete number of individuals. The computer pro-
gram tracks change in the composition of the human population and in
the infection status of each individual in the population over time (t, in 1-
month time steps) and age (a). The transmission of infection between indi-
viduals is captured by a deterministic submodel, accounting for the Simulium
fly population dynamics and the fate of the parasite in the fly. The model was
developed in close collaboration with the OCP (Plaisier et al., 1990). A
formal description of the model, presented previously by Habbema et al.
(1996a) and Coffeng et al. (2014a), is included here to facilitate access to
the readers and to allow a direct comparison with EPIONCHO. Table 3 lists
ONCHOSIM’s parameters, notation, values and sources.

2.2.1 Human population demography


The human population dynamics is governed by birth and death processes.
We define F(a) as the probability to survive to age a (apart from excess
282 ~ez et al.
M.G. Basan

mortality due to onchocerciasis-associated blindness). The values used are as


follows:

age (a) 0 5 10 15 20 30 50 90
F(a) 1.000 0.804 0.772 0.760 0.740 0.686 0.509 0.000

Survival at intermediate ages is obtained by linear interpolation.


The expected number of births (per year) at a given moment t is
given by,
X
na
Rb ðtÞ ¼ Nf ðk; tÞ$rb ðkÞ; (43)
k¼1

where Nf (k,t) is the no. of women in age group k at time t, rb(k) is the annual
birth rate in age-group k: 0.109 babies per year for women between 15 and
20 years; 0.300 between 20 and 30 years; 0.119 between 30 and 50 years;
0.0 for all other ages and na is the number of age groups considered.
Each month, Rb(t) is adapted according to the number of women and
their age-distribution. Depending on the size of the initial population and
birth and death rates, the human population may increase. The program
allows the specification of a ‘maximum population size’, in order to keep
the size population representative for the type of community being simu-
lated and limit computation time. As soon as the simulated population ex-
ceeds this maximum, a random fraction of the population is sampled and
removed from the population. This can be thought of as emigration. In
most published model applications, the model is used to simulate a village
population with a maximum size of 440. This value is well within the ranges
(23e828, with a median of 171), recorded in the OCP database (O’Hanlon
et al., 2016). The population distribution resulting from the aforementioned
parameters closely follows the age distribution in Sub-Saharan Africa as
shown elsewhere (Coffeng et al., 2014a).

2.2.2 Parasite population regulation in humans


2.2.2.1 Exposure to fly bites
The number of bites mbri(t) a person i gets in month t (in the absence of vec-
tor control) is given by,
mbri ðtÞ ¼ MbrðtÞ Exi (44)
River Blindness 283

where Mbr(t) is the number of bites in month t (Jan., Feb., .) for a person
with relative exposure equal to 1. The relative exposure Exi is calculated as,
Exi ¼ Exaðai ; si Þ$Exii ; (45)
where Exa(ai,si) is the relative exposure of a person of age a and sex s,
assumed to be zero at birth, to increase linearly with age between the ages of
0 and 20 years until a maximum of 1.0 for men and 0.7 for women, and to
remain constant from 20 years onwards, and,
Exii w Gamma(1.0,aExi) is the exposure index of person i. Exii is
assumed to follow a gamma distribution with mean 1.0 and shape and
rate equal to aExi. The exposure index of a person remains constant
throughout their lifetime. For selected West African villages (within
the OCP), estimated aExi values vary between 1.6 and 12.7 (for the sim-
ulations illustrated in this chapter values were 1.0 or 3.5; see Table 3 and
Coffeng et al., 2014a).
Mbr(t) values were obtained from six years of blackfly collections near the
village of Asubende (Ghana) conducted between 1978 and 1985. In this site
with perennial transmission, monthly biting rates of, on average, 2570 bites
per person, varying from 1500 in March to 3750 in November had been
found. For the actual biting rates (Mbr(t)) inside the village Asubende, these
figures were multiplied by a factor (called the relative biting rate) of 0.95.
(Since we have no measurements of biting rates actually experienced by vil-
lagers, we have arbitrarily defined a relative biting rate of 1.0, i.e., a mean
Mbr ¼ 2750 as the biting rate that results in a geometric mean number of
microfilariae (mf) per skin snip (ss) of 100 in a hypothetical village where
all its inhabitants are permanently characterized by a relative exposure of
1.0.) Assuming the same seasonal pattern for other villages, relative biting
rates have been estimated to vary from 0.4 to 0.9.

2.2.2.2 Parasite establishment


The monthly transmission potential (MTP) in ONCHOSIM is defined as
mbr(t)  lr(t). When accumulated over a year, we get the ATP. If during a
blood meal by a simuliid fly in month t, lr infective larvae are released on
average, the force-of-infection foii(t), defined as the expected number of
new adult parasites acquired by person i in month t, is calculated as,
foii ðtÞ ¼ mbri ðtÞ lrðtÞ sr; (46)
where sr is the success ratio, namely, the fraction of inoculated L3 larvae
succeeding in developing to adult male or female worms, with value
284 ~ez et al.
M.G. Basan

sr ¼ 0.0031. An average male:female sex ratio of 1:1 is assumed (Schulz-


Key, 1990).
In month t, a person i is assumed to become infected according to a Pois-
son process with rate foii(t). ONCHOSIM assumes that there is no density
dependence in this process.

2.2.2.3 Mating probability of female adult worms


Chance processes determine the number of male and female worms present
in each human individual. The degree of parasite overdispersion in the hu-
man population depends on exposure heterogeneity assumptions. The
reproductive lifespan of male and female parasites is a random variable,
Tl w Weibull(muTl,aTl), with mean muTl ¼ 10 years and shape aTl ¼ 3.8.
(For readers used to other commonly used parameterizations of the Weibull
distribution in terms of shape k and scale l, shape k is aTl (as described in this
document) and scale is l ¼ muTl/G(1 þ 1/aTl).) The microfilarial produc-
tivity r(a,t) of a female worm of age a in month t is calculated as,
rða; tÞ ¼ RðaÞ fmðtÞ; (47)
where
R(a) is the potential microfilarial productivity of a female worm of age a
(in years):
R(a) ¼ 0 for 0  a < 1;
R(a) ¼ 1 for 1  a < 6;
R(a) ¼ 1  ((a  6)/15) for 6  a < 21;
R(a) ¼ 0 for a > 21 and
fm(t) mating factor at time t.
Quantifying R(a) ¼ 0 for 0  a < 1 is equivalent to assuming an intrinsic
latency period (i.e., the time needed for a parasite to become mature and be
able to reproduce) of exactly 1 year for all male and female worms. This
duration is user-specified in ONCHOSIM; other values can be given and
it can also be specified as a continuous probability distribution. Similarly
other values can be specified for the potential microfilarial production by
female worms once the intrinsic latency period has elapsed. However, the
intrinsic latency period is not necessarily equal to the prepatent period, as
the latter is defined as the time from the moment of infection until patency
(microfilarial infection) can be detected in the skin.
The mating factor is defined as follows. To continue microfilarial
production, a female worm must be inseminated each rc months (rc ¼ repro-
ductive cycle ¼ 3 per year). If insemination takes place less than rc months
River Blindness 285

ago, then fm(t) ¼ 1. Otherwise, the probability of insemination or reinsemi-


nation Pins(t) in month t is given by,

Pins ðtÞ ¼ Wm ðtÞ Wf ðtÞ if Wm < Wf


; (48)
Pins ðtÞ ¼ 1 if otherwise
where W(t) is the number of male (Wm) or female (Wf) parasites in the human
host at time t. If no insemination takes place then fm(t) ¼ 0 and the female
worm has a new opportunity in month t þ 1. If insemination occurs in
month ti then fm(t) ¼ 1 during ti  t < ti þ rc. In ONCHOSIM there is one
additional parameter influencing the mating probability Pins. This parameter is
called male potential and is multiplied with the male:female worm sex ratio.
Assigning a high value to this male potential (e.g., 100) implies that mating (if
required) will always take place if there is at least one adult male worm.
The skin microfilarial density sl(t) at time t is calculated by accumulating
the microfilarial production of all female parasites over the past Tm months,
slðtÞ ¼ cw$elðtÞ; (49)

1 Xni X
Tm
elðtÞ ¼ rj aj  x; t  x ; (50)
Tm j¼1 x¼1

where
rj(aj,t) is the microfilarial productivity of a female worm j of age aj in
month t
el(t) is the effective parasite load at time t. This intermediate variable can be
interpreted as the total number of female worms that have contributed to
skin microfilarial counts at time t, weighted for their average microfilarial
productivity over the past Tm months,
cw is the average contribution of an inseminated female worm at peak
fecundity (R ¼ 1) to the skin microfilarial density: cw ¼ 7.6 microfilar-
iae/worm. (Instead of a linear relationship between sl and el, other func-
tional relationships can be chosen, e.g., a saturating function.)
Tm is the (fixed) microfilarial lifespan, with an assumed value
Tm ¼ 9 months (Plaisier et al., 1995), and
ni is the number of parasites alive during at least one of the months
t  1,.,t  Tm.

2.2.2.4 Microfilarial counts in skin snips


The expected number of microfilariae in a skin snip (of 2 mg) is given by,

cw Xni X
Tm
ssðtÞ ¼ dj rj aj  x; t  x ; (51)
Tm j¼1 x¼1
286 ~ez et al.
M.G. Basan

where, dj is the dispersal factor of female parasite j. This is a random variable


drawn for every ‘newborn’ worm and accounts for differences in the
contribution of female worms to the microfilarial density at the standard site
of the body where snips are taken (hips in Africa). We assume that dj follows
an exponential distribution, dj w Expo(1.0).
The actual (observed) number of microfilariae per skin snip (mf/ss) at
examination time t, ssobs(t) follows a Poisson distribution, ssobs(t) w Pois-
son(ss(t)) e although other discrete probability functions (e.g., geometric)
can be used. At each epidemiological survey two snips (or any other number
as appropriate) are taken from all simulated persons. The results of such a sur-
vey are postprocessed to arrive at age- and sex-specific and standardized
microfilarial prevalences.

2.2.2.5 Blindness and excess human mortality


The event of a person going blind at age a (in months) depends on the
cumulative parasite load (elc) of a person,
X
a
elcðaÞ ¼ elðxÞ: (52)
x¼0

Each person has a threshold level elc (denoted as Elc) at which a person
goes blind. Elc follows a Weibull probability distribution,
Elc w Weibull(muElc,aElc), with mean muElc ¼ 10,000 and shape aElc ¼ 2.0.
Person i goes blind at age a when,
elci ðaÞ  Elci > elci ða  1Þ: (53)
At that moment the remaining lifespan at age a is reduced by a factor rl,
which follows a uniform distribution on [0,1] (hence on average rl ¼ 0.5).
(Any other probability distribution defined on [0,1] can be used, e.g., a
beta distribution.)

2.2.3 Parasite population regulation in vectors


2.2.3.1 Parasite establishment
On the basis of fly-feeding experiments conducted in the OCP (analyzed by
~ez et al., 1995 and summarized by Soumbey-Alley et al., 2004), the
Basan
following expression for the relation between L1 uptake (lu) and skin micro-
filarial density in humans (sl) was derived by Plaisier et al. (1991b),

lu ¼ a$ 1 eb$sl $ 1þ ec$sl ; (54)
with a ¼ 1.2, b ¼ 0.0213 and c ¼ 0.0861 (the initial slope of this relationship
equals 2ab and, therefore, the maximum probability of parasite establishment
is 0.0511 per microfilaria in a skin snip; the maximum number of larvae
River Blindness 287

establishing within the thoracic muscles of the fly is 1.2). Other functional


relationships can also be defined. For instance, we can set c ¼ 1.0 to simulate a
situation with less pronounced negative density dependence in transmission
(which may be reflective of the situation in forest areas). This alternative
parameter value results in a less concave shape of the function, while the slope
in the origin (which equals 2 ab) and the final saturation level (a) remain the
same. In terms of L1 uptake, this means that uptake is up to 40% lower for
skin microfilarial densities <10 mf/ss, and nearly unchanged for skin mf
densities >40 mf/ss. Increases in c beyond 1.0 do not affect the shape of the
function by much. The choice of setting the value c ¼ 1.0 was arbitrary and
does not necessarily represent forest vectoreparasite complexes.
The mean L1 uptake in the blackfly population per fly bite in month t is
calculated as,
, NðtÞ
X
NðtÞ X
luðtÞ ¼ ðExi lui Þ Exi ; (55)
i¼1 i¼1

where N(t) is the number of persons bitten in month t.


It is assumed that a fixed proportion of the L1 (thoracic) larvae will
develop to the L3 stage and be released during subsequent bites,
lrðtÞ ¼ n luðtÞ; (56)
where
lr(t) is the mean number of L3 larvae released per bite in month t and
v is the transmission probability from vector to humans, defined as the
average probability that an L1 larva completes the extrinsic incubation
period within the blackfly vector and is released as an infective, L3 larva.
In calculating v we take into account the life history of the fly starting
from her first blood meal. We assume that blood meals are taken at fixed
hours during daytime, so that we can use 1 day time steps. Although we
take into account differences in the length of the gonotrophic cycle between
flies (the cycle may take between 2 and 5 days with a mean of 3.5 days), in
the model we assume that a particular fly always has the same cycle length
(which equals the time between two successive blood meals). We further
explicitly account for variation in the duration of development from L1
to L3 (which is mainly determined by environmental temperature; Cheke
et al., 2015). The basic assumption underlying the use of a fixed proportion
v is that at any moment the fly population has a stable age distribution and
that the number of bites per person is large enough to disregard the age of
the biting flies.
288 ~ez et al.
M.G. Basan

2.2.3.2 Transmission probability


For the current version of ONCHOSIM, transmission probability v has to
be calculated outside the model and given as a parameter. This section de-
scribes the necessary calculations.
Assume that a fly engorges one L1 larva (microfilariae are essentially L1
larvae as there is no moult between the microfilarial and the L1 stage) at her
mth blood meal, then the probability to release an L3 larva n blood meals
later is given by,
Prel ðnji; j; mÞ ¼ PL1/L3 $ð1 PL3/ Þi $PL3/L3
i
$PL3/ $Sðm; n$jÞ; (57)
where
Prel(nji,j,m) is the probability to release one L3 larva at the (m þ n)th
blood meal if one L1 larva has been ingested during the mth blood
meal, given that
• a gonotrophic cycle takes j days,
• between blood meals m and m þ n there have been i potentially
infective blood meals (i.e., blood meals at which the L1 larvae
have already developed to the L3 stage),
PL1/L3 is the probability that an L1 larva develops to the L3 stage, given
survival of the fly, with PL1/L3 ¼ 0.85 (Collins et al., 1977),
PL3/L3 is the probability that an L3 larva, which is not released at a given
blood meal survives to a next blood meal, given survival of the fly:
PL3/L3 ¼ 0.90,
PL3/ is the probability that an L3 larva is released at a blood meal:
PL3/ ¼ 0.65 (Duke, 1973; Renz, 1987), and
S(m,t) is the probability that a fly survives for t days until blood meal m.
In order to arrive at a general solution for all possible values of i, we use
the probability distribution of the number of potentially infective blood
meals since the meal during which microfilariae (L1 larvae) were ingested
and before the blood meal during which the infective L3 larvae are
released,
X
n1
Prel ðnjj; mÞ ¼ ½Prel ðnji; j; mÞ$Pib ðijn; jÞ; (58)
i¼0

Pib ðijn; jÞ ¼ FdL1/L3 ðjðn  iÞÞ  FdL1/L3 ðjðn  i  1ÞÞ; (59)


where
Pib(ijn,j) is the probability that before the nth blood meal since microfi-
larial intake, i blood meals have been potentially infective (L1 has
become L3), given a cycle length of j days, and
River Blindness 289

FdL1/L3(t) is the probability that the duration of development of L1 to


L3 is equal to or less than t days (FdL1/L3(t) ¼ 0.0 for t  5; 0.07 for
t ¼ 6; 0.86 for t ¼ 7; 1.0 for t  8 days).
A general solution for all possible values of m can be obtained by incor-
porating the probability that a fly takes her mth blood meal,
X
mmax
Prel ðnjjÞ ¼ ½Prel ðnjj; mÞ$Pb ðmjjÞ; (60)
m¼1
,
X
mmax
Pb ðmjjÞ ¼ Lðjðm  1ÞÞ ðjðm  1ÞÞ; (61)
m¼1

where
Pb(mjj) is the probability that a feeding fly takes her mth blood meal at a
cycle length of j days and
L(t) is the probability that a fly lives for at least t days. At present we
assume an age- (and microfilarial load)-independent daily survival of
0.78. This is in rough agreement with a probability of daily survival of
0.81 at 25 C for S. damnosum s.s (Cheke et al., 2015).
Generalizing for j can be achieved by summation, weighted by the prob-
ability distribution of the duration of the gonotrophic cycle,
X
jmax
 
PðnÞ ¼ Prel ðnjjÞ$Pgc ðjÞ ; (62)
j¼jmin

where Pgc( j) is the probability that a gonotrophic cycle takes j days (i.e., j
days between successive blood meals; Pgc( j) ¼ 0.0 for j  2; 0.2 for j ¼ 3;
0.6 for j ¼ 4; 0.2 for j ¼ 5; 0.0 for j  6 days).
Using the following equality,
Sðm; n$jÞ ¼ Lðjðmþ n  1ÞÞ=Lðjðm 1ÞÞ; (63)
the average probability that an L1 larva taken from a human will develop to
the L3 stage and be released to another human is given by,
Prel ¼ PL1/L3 $PL3/ $
8 " # ( 9
> X
mmax
1 X
nmax X
n1
i >
>
> P ðjÞ$ P Lðjðm þ n  1ÞÞ$ ½ð1  P Þ$P  >
>
>
> gc m $ L3/ L3/L3 $ >
>
> m¼1 Lðjðm  1ÞÞ >
max
> >
max < =
m¼1 n¼1 i¼0
jP
:
>
j¼jmin > ) >
>
>
> >
>
> P
> n1 >
>
>
: ½ð1  PL3/ Þ$PL3/L3 i $½FdL1/L3 ðjðn  iÞÞ  FdL1/L3 ðjðn  i  1ÞÞ >
;
i¼0

(64)
290 ~ez et al.
M.G. Basan

In Eqs (58)e(61)
amax
mmax ¼ þ 1; truncated to integer
j
; (65)
amax  ðm$jÞ
nmax ¼ þ 1; truncated to integer
j
where amax is the maximum attainable age of the fly (i.e., age at which L(T)
approaches zero).
The transmission probability from vectors to humans v is now given by,
v ¼ Prel $ð1  zÞ; (66)
where z is the fraction of fly bites taken on nonhuman blood host (zoo-
phagy). This value is highly dependent on local circumstances (e.g., human
and nonhuman blood host density, blackfly density and blackfly species
(Lamberton et al., 2016). In ONCHOSIM the value used is z ¼ 0.04,
meaning that 96% of the blood meals are assumed to be taken on human
hosts. This is close to the zoophagy index, z ¼ 0.08 recorded in the Beffa
form of S. soubrense, in Ghana (Lamberton et al., 2016).
Using the indicated quantifications, we have calculated a value for v of
0.073 released infective L3 larvae per L1 larva resulting from a given micro-
filarial uptake. Note that Eq (64) reduces to a much simpler form if we
assume that each day a fraction S of the flies survive, that the gonotrophic
cycle has a fixed duration of dgc days and that the number of blood meals
needed to complete the development of L1 to L3 is fixed to n1 / 3,
S nl/3$dgc
Prel ¼ PL1/L3 $PL3/ $ (67)
1  S dgc $ð1  PL3/ Þ$PL3/L3

2.2.3.3 Excess vector mortality


Excess mortality of infected flies is not considered in ONCHOSIM.

2.3 Comparison between EPIONCHO and ONCHOSIM


The EPIONCHO and ONCHOSIM models were developed indepen-
dently by research teams (respectively based at Imperial College London,
London, and Erasmus Medical Center, Rotterdam) using distinct modelling
approaches, on the basis of diverse datasets and with different initial pur-
poses. EPIONCHO was built following the modelling tradition and
methodology of Anderson and May (1982, 1985, 1991), and Dietz (1976,
1982) for infectious diseases in general and helminthic infections in
Table 3 Definition and values of variables and parameters for ONCHOSIM

River Blindness
Parameter Values and Units Sources
Human demography
Human life table, F(a) See Section 2.2.1 Human population United Nations (2013)
demography
Human fertility, R(t) See Section 2.2.1 Human population United Nations (2013)
demography
Exposure to simuliid vectors
Interindividual variation in exposure to Gamma distribution with mean 1.0 and Plaisier (1996); unpublished OCP data
fly bites (Exi) shape and rate equal to 1.0 or 3.5
Variation in exposure to fly bites by age Eq. (45) Plaisier (1996)
and sex (Exa)
Seasonal variation in exposure to fly 104%, 91%, 58%, 75%, 75%, 66%, 102%, Alley et al. (1994)
bites (mbr) 133%, 117%, 128%, 146%, and 105%
times the average monthly biting rate
(JanuaryeDecember)
Life history and microfilarial productivity of the parasite in the human host
Worm longevity (Tl ) Weibull distribution with mean 10 and Plaisier et al. (1991a)
shape 3.8 (year)
Pre-patent period 1 year Duke (1980) and Prost (1980)
Age (a)-dependent potential R(a) ¼ 0 for 0  a < 1 year Albiez (1985) and Karam et al. (1987)
microfilarial production, R(a) R(a) ¼ 1 for 1  a < 6 years
R(a) ¼ 1  ((a  6)/15) for
6  a < 21 years
R(a) ¼ 0 for a > 21 years

291
(Continued)
Table 3 Definition and values of variables and parameters for ONCHOSIMdcont'd

292
Parameter Values and Units Sources
Longevity of microfilariae (Tm) 9 months Plaisier et al. (1995)
Worm contribution to the skin 7.6 mf/worm Plaisier (1996)
microfilarial load (cw)
Variability in microfilariae per skin snip Poisson distribution with mean ss(t), Habbema et al. (1996a)
(2 mg) Eq. (51)
Dispersal factor for worm contribution Exponential distribution with mean 1 Habbema et al. (1996a)
to skin snip (d )
Mating cycle (rc) 3 months Schulz-Key (1990) and Schulz-Key
and Karam (1986)
Male potential 100 female worms Habbema et al. (1996a)
Vision loss
Blindness threshold (Elc) Weibull distribution with mean 10,000 Coffeng et al. (2013a)
and shape 2.0
Reduction in remaining life 50% Coffeng et al. (2013a), Kirkwood et al.
expectancy due to blindness (1983) and Plaisier et al. (1990)
Parasite in simuliid vector
Fly survival, L(t) 0.78 day1 Habbema et al. (1996a) and Cheke
et al. (2015)
Probability of gonotrophic cycle Pgc( j ) ¼ 0.0 for j  2 days Habbema et al. (1996a); expert opinion

M.G. Basan
duration, Pgc( j ) Pgc( j ) ¼ 0.2 for j ¼ 3 days (OCP entomologists)
Pgc( j ) ¼ 0.6 for j ¼ 4 days
Pgc( j ) ¼ 0.2 for j ¼ 5 days

~ez et al.
Pgc( j ) ¼ 0.0 for j  6 days
River Blindness
Zoophagy index, proportion of blood 0.04 Habbema et al. (1996a); expert opinion
meals taken on nonhumans blood (OCP entomologists)
hosts (z)
Microfilarial uptake, lu Eq. (54); a ¼ 1.2, b ¼ 0.0213, and Plaisier et al. (1991b)
c ¼ 0.0861 (main analysis); a ¼ 1.2,
b ¼ 0.0213, and c ¼ 1.0 (sensitivity
analysis)
Probability of duration of larval FdL1/L3(t) ¼ 0 for t  5 days Habbema et al. (1996a); expert opinion
development (from L1 to L3), FdL1/L3(t) ¼ 0.07 for t ¼ 6 days (OCP entomologists)
FdL1/L3(t) FdL1/L3(t) ¼ 0.86 for t ¼ 7 days
FdL1/L3(t) ¼ 1.0 for t  8 days
Larval survival (L1 / L3) 0.85
L3 survival (L3 / L3) 0.90
Larval release (L3) 0.65
Success ratio (sr) 0.0031 Plaisier et al. (1996)
Mass treatment coverage and adherence
Coverage, Cw User-defined
Age- and sex-specific adherence cr(k,s) See page s13 of Supplementary File S1 Unpublished OCP data
Text of Coffeng et al. (2014a)
Individual adherence index (co) Uniform distribution [0,1] Habbema et al. (1996a)
Ivermectin
Microfilaricidal efficacy (assumption 100% Plaisier et al. (1995)
sets 1 and 2)
Assumption set 1
Relative effectiveness (v) Weibull distribution with mean 1 and Plaisier et al. (1995)

293
shape 2
(Continued)
Table 3 Definition and values of variables and parameters for ONCHOSIMdcont'd

294
Parameter Values and Units Sources
Duration of embryostatic 11 months
effect (Tr, s)
Per dose (cumulative) reduction in 35%
worm fecundity, d
Macrofilaricidal efficacy 0%
Assumption set 2
Embryostatic effect (tau) Exponential distribution with mean 3.5 Coffeng et al. (2014a)
(years)
Macrofilaricidal efficacy (on male Beta distribution with mean 12.3% and
worms) sample size 50
Macrofilaricidal efficacy (on female Beta distribution with mean 6% and
worms) sample size 50
Larviciding
Timing User-defined Plaisier et al. (1997)
Coverage User-defined

M.G. Basan
~ez et al.
River Blindness 295

particular. The precursors of EPIONCHO (Basan ~ez and Boussinesq, 1999;


Basan~ez and Ricardez Esquinca, 2001; Basan
~ez et al., 2002, 2007), based on
(deterministic) differential equations, were developed with the primary
objective of understanding the impact of parasite population regulatory pro-
cesses on the population biology of O. volvulus and the transmission dy-
namics of onchocerciasis. These precursors were subsequently extended to
explore the influence of parasite overdispersion (Churcher et al., 2005,
2006), to include human host age- and sex-structure (Filipe et al., 2005),
to account for parasite population genetic structure with the aim to investi-
gate the spread of anthelmintic resistance (Churcher and Basan ~ez, 2008,
2009) and to add realism regarding treatment coverage and adherence pat-
terns (Turner et al., 2013a), thus evolving into a full-transmission and con-
trol model capable of supporting decision-making by intervention
programmes. ONCHOSIM is an individual-based model for simulating
onchocerciasis transmission and control in a dynamic human population,
based on the technique of stochastic microsimulation (Alley, 1992; Hab-
bema et al., 1996b). The underlying generalized modelling framework has
formed the basis of similar models for other helminthic diseases (De Vlas
et al., 1996; Plaisier et al., 1998). ONCHOSIM was conceived from the start
with the purpose of informing control strategies (Remme, 2004a), formerly
during the OCP (Habbema et al., 1996a; Plaisier et al., 1990,b, 1997;
Remme et al., 1986; Remme et al., 1990a), and more recently to support
APOC’s elimination efforts and to quantify its overall health impact (Cof-
feng et al., 2013a, 2014a).
A formal comparison of EPIONCHO and ONCHOSIM regarding
the required duration of mass ivermectin treatment for onchocerciasis
elimination in Africa has been presented by Stolk et al. (2015). In this pa-
per, the two models were ‘docked’ as much as possible in terms of param-
eter values (see Table 2 of Stolk et al., 2015), and the focus was on
comparing, contrasting and understanding the similarities and differences
in projected elimination outcomes (under annual or biannual ivermectin
distribution) for a number of epidemiological scenarios (defined in terms
of initial endemicity, microfilarial prevalence, CMFL and vector biting
rates), ranging from mesoendemic to holoendemic onchocerciasis, and a
number of programmatic scenarios (defined by therapeutic coverage and
treatment adherence). This comparison revealed several differences in
model predictions, despite harmonization of key parameters. The
remainder of this section discusses some of the convergences and diver-
gences identified.
296 ~ez et al.
M.G. Basan

EPIONCHO is a deterministic, population-based model; ONCHOSIM


is a stochastic individual-based model. Although there are some similarities,
the models also differ in important aspects, e.g., on the extent to which het-
erogeneities in the human population (e.g., in exposure to blackfly bites) and
density dependencies in various processes are captured (e.g., in parasite
establishment rate within humans and excess mortality of infected flies, as
described above). More specifically, and in the context of the ability of
both models to predict elimination, EPIONCHO does not account for
the possibility of chance elimination of the parasite population (stochastic
fadeout), which becomes increasingly likely at very low intensities of infec-
tion, especially for small settings (villages) with a couple of hundred inhab-
itants (as assumed by ONCHOSIM). Secondly, the models differ with
respect to assumptions about density dependence in the various processes
involved in transmission dynamics (Table 4), which may also be important
for elimination prospects (Duerr et al., 2005, 2011). In particular, EPION-
CHO includes a (negative) density-dependent relationship between the
annual transmission potential and the parasite establishment rate (Section
2.1.1.1); ONCHOSIM does not capture this mechanism, which makes
the model more optimistic regarding elimination prospects. Thirdly, the
assumed distribution of adult worm and microfilarial survival times and as-
sumptions regarding microfilarial productivity in relation to worm age
may play a role. The current version of EPIONCHO assumes an exponen-
tial distribution of worm survival times with a long right tail, implying that
worm mortality rates are independent of worm age (an implicit assumption
of the exponential model). ONCHOSIM assumes a Weibull distribution
(Plaisier et al., 1991a), a more symmetrical distribution with the same
mean survival time but a shorter right tail, implying age dependency of
worm mortality rates (Section 2.2.2.3). Therefore, it takes considerably
longer for the parasite population to die out naturally in EPIONCHO
than in ONCHOSIM. In addition to this, ONCHOSIM assumes that the
microfilarial production rate declines in older worms (Section 2.2.2.3), so
that the relatively old worm population remaining after long-term iver-
mectin mass treatment has a relatively low microfilarial production. Such
a process is not considered by EPIONCHO. EPIONCHO models micro-
filarial prevalence as a function of mean microfilarial density assuming an un-
derlying negative binomial distribution whose overdispersion parameter is a
function of the mean (such that the distribution becomes increasingly aggre-
gated as the intensity of infection decreases). In ONCHOSIM individual
outputs are aggregated to obtain information on the microfilarial prevalence
River Blindness 297

Table 4 Overview of the main characteristics of the EPIONCHO and ONCHOSIM


models
Characteristics EPIONCHO ONCHOSIM
Basic model structure
Modelling approach Deterministic, Stochastic, individual-
population mean- based (excepting the
based vector component)
Number and type of Single population Single population
spatial locations
modelled
Way of representing Mean density in Presence and density at
infection in hosts population subgroups individual level
(e.g., age, sex,
treatment adherence
group). Prevalence as
a function of mean
density assuming an
underlying negative
binomial distribution
Interventions Mass treatment Mass treatment,
considered in selective treatment
previous publications (test-and-treat),
vector control
Features included in the model
Human population Birth and death rate; age Birth and death rate; age
demographics and sex composition and sex composition
Heterogeneities in the Age, sex, life Age, sex, life
human population expectancy, level of expectancy, level of
exposure to blackflies, exposure to blackflies,
adherence with MDA adherence with
MDA, treatment
efficacy
Blackfly population Fixed input as annual Fixed input as annual
density biting rate (ABR); biting rate (ABR);
seasonality in biting seasonal monthly
rates can be included biting rates
Exposure to blackfly Heterogeneous Heterogeneous
vectors (dependent on age (dependent on age,
and sex) sex, personal
attractiveness to
blackflies)
(Continued)
298 ~ez et al.
M.G. Basan

Table 4 Overview of the main characteristics of the EPIONCHO and ONCHOSIM


modelsdcont'd
Characteristics EPIONCHO ONCHOSIM
Uptake of infection by Varying non-linearly Varying non-linearly
blackfly vectors (de