Professional Documents
Culture Documents
Growth in Bangladesh
Janet Johanna Sanchez Hernandez
Honours Bachelor of Science, Master of International Public Health
BACKGROUND: The dual burden of enteropathogen infection and childhood malnutrition continues
to be a global health concern and a leading cause of morbidity and death among children. Despite
improvements in recent decades, Bangladesh continues to be one of the countries with the highest
burden of growth impairment. In addition, enteropathogen infection continues to be highly prevalent
in the country. An understanding of factors associated with infection and growth could contribute to
public health efforts to reduce this burden through targeted community interventions. This PhD thesis
aimed to examine the association between household factors and enteropathogen infection and
linear child growth in Bangladesh using both national surveys and study-specific data.
RESULTS: In all three analyses, an inverse association was found between increasing age and
LAZ, consistent with previous research. In the DHS and MICS survey data, older children (13-24
months) had more household factors associated with LAZ. Maternal education was positively
associated with linear growth in both national-level datasets, as were improved toilet facilities. In the
MICS analysis, rural households had more factors associated with LAZ. Male sex, dirt flooring
material, and unimproved toilet facility, and animal ownership were negatively associated with LAZ,
while refrigerator ownership and increased maternal education were positively associated. In rural
households in the DHS analysis, an episode of diarrhea in the preceding two weeks, a shared toilet
facility and no maternal education were negatively associated with LAZ. In the LGCM analysis of
children from the MAL-ED Bangladesh site, Campylobacter infection was highly prevalent and had
an effect on growth at specific age intervals. Infection in the preceding 3-month interval was
negatively associated with LAZ between 12 and 18 months of age; similarly, infection in the
preceding 6-month interval was negatively associated with LAZ at between 15 and 21 months of
age. Maternal height and birth order were positively associated with LAZ at birth but were not
associated with infection. Duration of antibiotic use and treated drinking water were negatively
ii
associated with Campylobacter infection, with the strength of the latter effect increasing with
children’s age.
CONCLUSION: The results indicate that growth impairment continues to be a public health concern
in Bangladesh, and that child growth is associated with enteropathogen infection and household-
level factors that may serve as pathogen household reservoirs. Different household factors are
associated with growth in children according to different age groups and their geographical location
(urban versus rural), with older rural children having the most factors associated with growth.
Enteropathogen infection, particularly Campylobacter, was found to have a negative effect on linear
growth, with this effect being greatest at specific age intervals in the second year of life. The results
of these analyses taken together highlight the need for public health interventions to target household
factors involved in enteric pathogen transmission at specific ages of children and specific geographic
locations.
iii
Declaration by author
This thesis is composed of my original work, and contains no material previously published or
written by another person except where due reference has been made in the text. I have clearly
stated the contribution by others to jointly-authored works that I have included in my thesis.
I have clearly stated the contribution of others to my thesis as a whole, including statistical
assistance, survey design, data analysis, significant technical procedures, professional editorial
advice, financial support and any other original research work used or reported in my thesis. The
content of my thesis is the result of work I have carried out since the commencement of my higher
degree by research candidature and does not include a substantial part of work that has been
submitted to qualify for the award of any other degree or diploma in any university or other tertiary
institution. I have clearly stated which parts of my thesis, if any, have been submitted to qualify for
another award.
I acknowledge that an electronic copy of my thesis must be lodged with the University Library and,
subject to the policy and procedures of The University of Queensland, the thesis be made
available for research and study in accordance with the Copyright Act 1968 unless a period of
embargo has been approved by the Dean of the Graduate School.
I acknowledge that copyright of all material contained in my thesis resides with the copyright
holder(s) of that material. Where appropriate I have obtained copyright permission from the
copyright holder to reproduce material in this thesis and have sought permission from co-authors
for any jointly authored works included in the thesis.
J. Johanna Sanchez H.
iv
Publications included in this thesis
Sanchez JJ, Alam MA, Raihan MJ, Haque MA, Das S, Mahfuz M, Stride CB, Sly P, Roth D, Long
KZ, Tahmeed A, Campylobacter infection and household factors are associated with childhood
growth in urban Bangladesh: An analysis of the MAL-ED study. PLoS Negl Trop Dis 2020; 14(5).
v
Submitted manuscripts included in this thesis
Sanchez JJ, Sly PD, Long KZ. Household-Level Risk Factors Associated with Linear Growth
Among Young Children in Bangladesh: An Analysis of the Multiple Indicator Cluster Survey, 2012-
2013.
vi
Other publications during candidature
Oral Presentations
Poster Presentations
1. Sanchez J, Alam MA, Raihan MJ, Haque MA, Das S, Mahfuz M, Stride CB, Sly P, Long
KZ, Tahmeed A, (2018) Campylobacter infection and household-level factors associated
with childhood growth in Mirpur, Bangladesh: An analysis of the MAL-ED Study. Canadian
Conference in Global Health, Toronto, Canada, November 2018.
2. Sanchez J, Alam MA, Raihan MJ, Haque MA, Das S, Mahfuz M, Stride CB, Sly P, Long
KZ, Tahmeed A, (2018) Campylobacter jejuni infection and household-level factors are
associated with childhood growth in Mirpur, Bangladesh: An analysis of the MAL-ED Study.
American Society for Tropical Medicine and Hygiene, New Orleans, USA, October 2018.
3. Hutton EK, Hannah ME, Ross S, Joseph KS, Ohlsson A, Asztalos EV, Willan AR, Allen AC,
Armson BA, Gafni A, Mangoff K, Sanchez JJ, Barrett JF; Urinary incontinence 2 year after
cesarean or vaginal or vaginal birth for twin pregnancy: A multicenter randomized trial
AJOG. 2018 Jan;218(1):S147.
1. Das S*, Sanchez JJ*, Alam MA, Haque MA, Mahfuz M, Long KZ, Ahmed T. Dietary
magnesium, vitamin D and animal protein intake and their association to the linear growth
trajectory of under 2 children: results from MAL-ED birth cohort study conducted in Dhaka,
Bangladesh. Food and Nutrition Bulletin [Accepted October 2019] *Co-first authors
1. Yamamoto JM, Benham JL, Dewey D, Sanchez JJ, Murphy HR, Feig DS, Donovan LE
Neurocognitive and behavioural outcomes in offspring exposed to maternal pre-existing
diabetes: a systematic review and meta-analysis. Diabetologia. 2019 Sep;62(9):1561-1574.
2. Murphy HR, Feig DS, Sanchez JJ, de Portu S, Sale A; CONCEPTT Collaborative Group
Modelling potential cost savings from use of real-time continuous glucose monitoring in
pregnant women with Type 1 diabetes. Diabetic Med. 2019 Dec; 36 (12):1652-1658.
3. Feig DS, Corcoy R, Donovan LE, Murphy KE, Barrett JFR, Sanchez JJ, Ruedy K, Kollman
C, Tomlinson G, Murphy HR; CONCEPTT Collaborative Group. Response to Comment on
Feig et al. Pumps or Multiple Daily Injections in Pregnancy Involving Type 1 Diabetes: A
vii
Prespecified Analysis of the CONCEPTT Randomized Trial. Diabetes Care. 2019
Jun;42(6): e98-e99.
4. Feig D, Corcoy R, Donovan LE, Murphy K, Barrett J, Sanchez JJ, Wysocki T, Ruedy K,
Kollman C, Tomlinson G, Murphy HR, Pumps or Multiple Daily Injections in Pregnancy
Involving Type 1 Diabetes: A Prespecified Analysis of the CONCEPTT Randomized Trial.
Diabetes Care. 2018 Oct; 41(12). doi: 10.2337/dc18-1437
5. Hutton E, Hannah ME, Willan A, Ross S, Allen A, Armson BA, Gafni A, Joseph KS,
Mangoff K, Ohlsson A, Sanchez JJ, Asztalos EV, Barrett J; Urinary stress incontinence
and other maternal outcomes two years after Caesarean or vaginal bith for twin pregnancy:
a multicentre randomised trial. BJOG. 2018 July 125 (13). doi: 10.1111/1471-0528.15407
6. Feig DS, Donovan LE, Corcoy R, Murphy KE, Amiel SA, Hunt KF, Asztalos E, Barrett JFR,
Sanchez JJ, de Leiva A, Hod M, Jovanovic L, Keely E, McManus R, Hutton EK, Meek CL,
Stewart ZA, Wysocki T, O'Brien R, Ruedy K, Kollman C, Tomlinson G, Murphy HR;
CONCEPTT Collaborative Group. Continuous glucose monitoring in pregnant women with
type 1 diabetes (CONCEPTT): a multicentre international randomised controlled trial.
Lancet. 2017 Sep 15. PMID: 28923465
7. Vilder M, Magee LA, von Dadelszen P, Rey E, Ross S, Asztalos E, Murphy KE, Menzies
JM, Sanchez J, Singer J, Gafni A, Gruslin A, Helewa M, Hutton E, Lee SK, Logan AG,
Ganzevoort JW, Welch R, Thornton JG, Moutquin JM, and the CHIPS Study Group.
Women’s views and postpartum follow-up in the CHIPS trial (Control of Hypertension in
Pregnancy Study) Eur J Obstet Gynecol Reprod Biol. 2016 Nov; 206:105-113.
9. Feig DS, Murphy K, Asztalos E, Tomlinson G, Sanchez J, Zinman B, Ohlsson A, Ryan EA,
Fantus IG, Armson AB, Lipscombe LL, Barrett JF; MiTy Collaborative Group. Metformin in
women with type 2 diabetes in pregnancy (MiTy): a multi-center randomized controlled trial.
BMC Pregnancy Childbirth. 2016 Jul 19;16 (1):173. doi: 10.1186/s12884-016-0954-4.
10. Hutton EK, Hannah ME, Ross S, Joseph KS, Ohlsson A, Asztalos E, Willan AR, Allen AC,
Armson BA, Gafni A, Mangoff K, Sanchez JJ, Barrett JF; Twin Birth Study Collaborative
Group. Re: Maternal outcomes at 3 months after planned caesarean section versus
planned vaginal birth for twin pregnancies in the Twin Birth Study: a randomised controlled
trial: Counselling is difficult when outcomes are associated with mode of delivery and not
the plan of mode of delivery. BJOG. 2016 Mar;123(4):644
11. Asztalos EV, Hannah ME, Hutton EK, Willan AR, Allen AC, Armson BA, Gafni A, Joseph
KS, Ohlsson A, Ross S, Sanchez JJ, Mangoff, K and Barrett JFR for Twin Birth Study
Collaborative Group Twin Birth Study: 2-year neurodevelopmental follow-up of the
randomized trial comparing planned cesarean vs planned vaginal delivery for twin
pregnancy. Am J Obstet Gynecol. 2016 Jan; 214(3):371.e - 371.e19.
12. Hutton EK, Hannah ME, Ross S, Joseph KS, Ohlsson A, Asztalos EV, Willan AR, Allen AC,
Armson BA, Gafni A, Mangoff K, Sanchez JJ, Barrett JF; Twin Birth Study Collaborative
Group. Maternal outcomes at 3 months after planned caesarean section versus planned
vaginal birth fortwin pregnancies in the Twin Birth Study: a randomised controlled trial.
BJOG. 2015 Nov;122(12):1653-62.
viii
Contributions by others to the thesis
No contributions by others
Statement of parts of the thesis submitted to qualify for the award of another degree
No works submitted towards another degree have been included in this thesis
The MAL-ED analysis was granted an Exemption to Ethics Review at the University of Queensland,
due to the negligible risk to participants and that it is a secondary analysis. [Clearance Number:
2018000676] An ethics application was also submitted to the regional ethics board in Switzerland
(Ethikkommission Nordwest- und Zentralschweiz” (EKNZ)), on behalf of the Swiss Tropical and
Public Health Institute, the institution of PhD advisor and Project Lead, which granted a Declaration
of No Objection. The DHS and MCIS analyses did not require ethics approval as this was a
secondary analysis of publicly available and de-identified data.
ix
Acknowledgements
I would first like to acknowledge and thank my advisory committee: Dr Kurt Long, Professor Peter
Sly, and Dr Daniel Roth, for their guidance, patience, and continued support throughout my
candidature. Their knowledge and expertise provided valuable input to the development and
completion of this thesis project. I would also like to thank Dr Christopher Stride for his patience and
statistical support in developing the complex longitudinal models. I also thank Dr Anne Bernard for
her statistical support in the MICS and DHS analyses. A special thank you to Dr Kirsten Spann and
Dr Ricardo Soares Magalhaes for participating in my milestone committee and the valuable feedback
they provided during the reviews.
I would also like to acknowledge the Bill & Melinda Gates Foundation, for providing funding for travel
and training in Seattle, USA and London, England, in addition to a site visit to the icddr,b in Dhaka,
Bangladesh for collaborative analysis work and a visit to the study site. I would also like to thank Dr
Karen Moritz and the Children’s Health Research Centre for providing funding to attend major
international conferences to present my research results.
I also thank the icddr,b in Dhaka for providing access to the MAL-ED dataset and for supporting my
analysis work. I would specifically like to thank Dr Tahmeed Ahmed, Dr Mustafa Mahfuz, Dr
Subhasish Das, Ashraful Alam, and Ahshanul Haque, for their collaboration and valuable insight
about the study population and local considerations.
A very special thanks to my family and friends for their support throughout the years, whether it was
by asking how it was going, encouraging me, or by watching my little one so that I had time to write.
A very special thank you to my parents for always pushing the importance of education and for
providing support in every possible way. To my husband Anthony for being a true partner and
supporting my goals, for encouraging me to persevere, and for his constant love. To my little Elena,
who was part of the final year of my PhD journey, for adding new meaning to my life and continuing
to inspire me to work for the health of children. Her existence pushes me to try to be the role model
she deserves.
x
Financial support
Travel funds were provided by the Bill & Melinda Gates Foundation for attendance at a workshop
in Seattle, USA; statistical training in London, England; and a visit to the icddr,b for analysis work
with the study team and a visit to the study site.
Travel funds for presentation of results at the American Society of Tropical Medicine and Hygiene
Annual Meeting in New Orleans and the Canadian Conference for Global Health were provided by
the Children’s Health Research Centre.
xi
Keywords
linear growth, children, Bangladesh, LAZ, enteropathogens, household factors, growth impairment
xii
TABLE OF CONTENTS
ABSTRACT ................................................................................................................................ II
xiii
RESEARCH AIM #1 ................................................................................................................. 39
HOUSEHOLD-LEVEL RISK FACTORS ASSOCIATED WITH LINEAR GROWTH AMONG YOUNG CHILDREN IN
BANGLADESH: AN ANALYSIS OF THE MULTIPLE INDICATOR CLUSTER SURVEY, 2012-2013
7.1 INTRODUCTION................................................................................................................... 85
7.2 METHODS.......................................................................................................................... 85
7.2.1 Study Sample and Site .............................................................................................. 85
7.2.2 Study Data ................................................................................................................ 86
7.2.3 Exploratory Analysis .................................................................................................. 88
7.2 4 Latent Growth Curve Analysis.................................................................................... 89
7.3 EXPLORATORY ANALYSIS RESULTS ...................................................................................... 90
xiv
CHAPTER 8 MAL-ED CAMPYLOBACTER RESULTS ............................................................. 94
MANUSCRIPT ........................................................................................................................... 94
Abstract ............................................................................................................................. 94
Introduction ........................................................................................................................ 95
Methods ............................................................................................................................. 97
Results............................................................................................................................. 101
Discussion ....................................................................................................................... 107
Conclusion ....................................................................................................................... 111
xv
LIST OF FIGURES
xvi
LIST OF TABLES
xvii
LIST OF ABBREVIATIONS
SD Standard deviation
xviii
CHAPTER 1 INTRODUCTION
low- and middle-income countries (LMICs). Linear growth in early childhood is an important
indicator of health and nutritional status and is associated with both short- and long-term
child health and nutritional status; however, it remains one of the countries with the highest
status and inadequate household conditions related to sanitation, water quality and poor
hygienic conditions, all of which contribute to greater exposure of children to infections (3).
impairment, particularly in LMICs due to the high prevalence and frequency of infection (6–8).
The dual burden of enteric infection and childhood undernutrition continues to be a global
health concern and a leading cause of morbidity and death among children under the age of
five. Enteric infections have immediate and long-term impact on the health of children and are
part of a vicious cycle where frequent infections contribute to poor linear growth in young
children, leading to growth impairment, which then increases the risk of further infection (9). It
infection and impaired growth to achieve the greatest impact on childhood health and survival.
Understanding the most important etiological agents that are associated with linear growth
and infection is critical in addressing this important public health problem. The Etiology, Risk
Factors, and Interaction of Enteric Infections and Malnutrition and Consequences for Child
Health (MAL-ED) Study, a longitudinal study, recently determined that the most important
pathogens associated with diarrheal infection in 8 participating sites were norovirus, rotavirus,
and Campylobacter, astrovirus, Cryptosporidium, in the first year of life; and Campylobacter,
1
norovirus, rotavirus, astrovirus, and, Shigella, in the second year of life (10). It is not clear
which of these pathogens contribute most to reduced nutrient absorption and increased
metabolic demands in children and so have the greatest impact on childhood growth.
A clearer understanding of the risk factors for enteric infection and linear growth is essential
in efforts to develop more effective infection prevention efforts. Numerous studies have
identified risk factors associated with enteric disease in different populations such as adequate
water quality, sanitation practices, maternal education, and the presence of animals.
Interventions have been carried out targeting these risk factors, but the results have been
inconsistent. This is partly due to an incomplete understanding of what role such risk factors
whether these factors are associated with growth in all children, or whether their effect varies
by age. Also, household characteristics in an urban versus rural setting may be different and
thus different factors may be important according to region. A greater understanding of how
considerations can lead to greater reductions of diarrhea and impaired childhood growth.
Using data collected in the 2012-2013 Multiple Indicator Cluster Survey (MICS) and the 2014
Demographic and Health Survey (DHS), this study will aim to identify the predictors of linear
growth in children 1-24 months of age. It will also examine whether these predictors are
modified by the age of the child and by the geographic location of the household (urban versus
rural). The second research aim will build upon information gained in the first research aim.
Using longitudinal data from the MAL-ED study Bangladesh site, the association between
urban Bangladesh will be investigated. It will also aim to identify the age intervals in the first
24 months of age where enteropathogen infection had an effect on linear growth. Using
information gained from the MICS and DHS analyses, this study will aim to identify the indirect
2
Figure 1.1 Thesis Outline
Introduction
Literature Review
General Methodology
Results
MAL-ED Study
3
REFERENCES
2010;125(3):e473–80.
et al. Paediatrics and International Child Health Piecing together the stunting puzzle :
a framework for attributable factors of child stunting Paediatr Int Child Health
[Internet]. 2016;9047(October):1–8.
3. Rogawski ET, Liu J, Platts-Mills JA, Kabir F, Lertsethtakarn P, Siguas M, et al. Use of
results from the MAL-ED cohort study. Lancet Glob Heal. 2018;6(12):e1319–28.
2016;53(1):241–67.
5. Islam MM, Sanin KI, Mahfuz M, Ahmed AMS, Mondal D, Haque R, et al. Risk factors
8. Richard SA, Black RE, Gilman RH, Guerrant RL, Kang G, Lanata CF, et al. Catch-Up
9. Jennifer Bryce, Denise Coitinho, Ian Darnton-Hill, David Pelletier PP-A. Maternal and
Child Nutrition- Executive Summary of The Lancet Maternal and Child Nutrition
4
10. Platts-Mills JA, Babji S, Bodhidatta L, Gratz J, Haque R, Havt A, et al. Pathogen-
specific burdens of community diarrhoea in developing countries: A multisite birth cohort
study (MAL-ED). Lancet Glob Heal. 2015;
5
CHAPTER 2 RESEARCH AIMS AND OBJECTIVES
Research Aim #1
To identify the household-level predictors for linear growth impairment in children 1-24
Research Objectives:
a. Determine modification of the predictors for linear growth by age intervals: 1-12
Research Aim #2
Research Objectives:
b. Determine the age intervals where enteropathogen infection had a negative effect
on linear growth
6
CHAPTER 3 LITERATURE REVIEW
Linear growth in children is one of the most important indicators of health and well-being (1).
The first two years of life have been determined to be the most critical for childhood growth
and when linear growth impairment is most likely to take place (2). The period of maximal
postnatal growth velocity takes place between birth and 6 months of age, which has also been
identified as the period critical for long-term cognitive development, while 6 to 24 months
Linear growth impairment, expressed as low height/length for age, is related to the other
measures of childhood malnutrition of wasting (low weight for height), or underweight (low
weight for age) (5,6). Growth impairment is usually defined using height/length for age z-
scores (HAZ/LAZ), with cut-off points for stunting specifically (7,8). A height or length lower
than two standard deviations (SD) below the World Health Organization (WHO) Child Growth
and treat (9–11). Using data from 141 countries, Stevens et al (12) reported that while there
was improvement of nutritional status in children under the age of 5, between 1985 and 2011,
this improvement did not achieve the WHO child growth standards. LAZ improved from -1.86
to -1.16, with prevalence of moderate-severe stunting declining from 47.2% to 29.9% (12). In
2011, 314 million under the age of 5 had a LAZ below -1, and approximately half of these
children had a LAZ below -2 (12). LAZ in children from low- and middle-income countries
(LMICs) are close to the WHO reference standard at birth, then decline rapidly until two years
of age (6,13). Growth impairment is most prevalent in Asian countries, followed by countries
7
Poor nutritional status in early childhood is associated with both short- and long-term negative
outcomes, and metabolic syndromes later in life (6,11,14,15). The economic impact of
undernutrition is also well described. Reduced cognitive and physical development result in a
reduction of productive capacity and therefore less schooling in childhood and poor economic
performance in adulthood (3). Conservative estimates suggest that a stunted person will make
22% less earnings than those not stunted (3). The effects of growth impairment are also
transgenerational. Women who are stunted themselves are more likely to give birth to stunted
Growth references are limited in that they describe the growth of a population in a specific
geographic location and time (16,17). A growth standard, in contrast, describes how a child
should grow in optimal conditions, using a healthy sample of the population as the reference.
The 2006 WHO growth reference standards, based on the Multicentre Growth Reference
Study (MGRS), are widely used in epidemiological research and by national programs
worldwide – 125 countries by April 2011 (18). The multicentre study applied a strict criteria to
define a healthy reference population and examined growth in children in six countries: Brazil,
Ghana, India, Norway, Oman and USA (17). Following the results of this study, the WHO
growth standards demonstrates that healthy breastfed children have similar patterns of growth
across the globe (17). We should therefore expect the same potential for growth in children
from any country (17). The thesis research presented here applies the 2006 WHO Growth
Standards.
The causes of linear growth impairment are multifactorial and often related to conditions of
poverty, particularly in LMICs, due to inadequate dietary intake and exposure to environmental
8
framework developed by the Lancet in 2013 (21) presents the determinants of malnutrition
and describes stunting as the manifestation of chronic malnutrition (Figure 3.1). Among the
underlying causes of undernutrition are the caregiving category, which includes maternal
education; and the environmental category, which includes the factors associated with
infectious diseases, such as access to safe water, sanitation, and hygienic practices (21).
This research project explores the household-level maternal, environmental and infectious
Infectious disease in children has a negative impact on health and growth, particularly in low-
and middle-income countries (LMICs). It is well established that the relationship between
9
nutritional status, immunity, and infection is cyclical (Figure 3.2) (22–25). A child’s nutritional
status is important for a normal functioning immune system (23). Malnutrition can decrease
immune function, thus making the person more susceptible to infection (22,23). Infection can
then lead to the loss of nutrients and fluid, redistribution of nutrients in the body, and the
reduction of absorptive capacity and so further impacting a child’s nutritional status (26).
structural changes, can also have an impact on nutritional status (23,26,27). Further
deterioration in nutritional status can, in turn, predispose the child to prolonged duration of
Malnutrition
Decreased
Infection
Immunity
children, particularly infants, and is a risk factor for growth impairment. It is thought to lead to
impaired linear growth through reduced food intake, an increase in metabolic demands, and
10
3.5.1 Symptomatic Enteropathogen Infection - Diarrheal Disease
24-hour period followed by at least two free days of diarrhea (30). It has a great impact on
growth due to decreased nutrient absorption, reduced appetite, changing of feeding practices,
and the high prevalence of frequent episodes in LMICs (23,31). The promotion of oral
rehydration therapy has been successful in decreasing mortality through the reduction in
diarrheal-related deaths since the 1980s; however, the burden of diarrhea continues to be
significant (22). In 2015, diarrheal disease was the fourth leading cause of death in children
and was responsible for 499,00 death among children under 5 years old, despite a 20%
Frequent diarrheal episodes, defined as greater than five episodes before 24 months of age
account for a high proportion of stunting at 24 months, and have been identified as being the
primary cause of growth impairment in Asia and Latin America, and the second leading cause
in Africa (23,33). An analysis of nine cohort studies that collected regular diarrhea information
and longitudinal anthropometry determined that the effect of diarrhea on stunting was similar
across studies (33). Each diarrheal episode resulted in an increased odd of stunting at 24
months of age, therefore a higher diarrheal burden was associated with a higher frequency of
stunting (33). Duration of diarrheal episodes have also been determined to be important in
growth, with persistent diarrhea (single episode duration ³14 days) being associated with
The literature often describes enteric infection in its symptomatic form, diarrheal disease, with
many studies examining linear growth in children suffering diarrheal episodes; however,
asymptomatic enteric infections may have a more important role in growth faltering and
malnutrition than has been previously appreciated (31). As described by Petri et al (22), enteric
11
infection, with or without clinical symptoms of diarrhea, can profoundly disrupt intestinal
among children who are malnourished (35). In a Peruvian birth cohort study, it was found that
asymptomatic Campylobacter was associated with reduced weight gain over a three-month
period, but interestingly, was not associated with linear growth reduction (35). In another study
both symptomatic and asymptomatic infection had an impact on linear growth (36,37). They
also found that while symptomatic infection was associated with greater reductions in growth,
asymptomatic Cryptosporidium infection was twice as prevalent and thus had more overall
impact on childhood growth in the community (36,38). A recent analysis of Giardia in the MAL-
ED study found that even in the absence of diarrheal symptoms there was an association with
Some studies have explored the effect of the timing of infection, in relation to children’s age,
on short term and long term linear growth impairment. An analysis conducted by Richard et al
(40) found that diarrhea in the previous 30 days had an effect on weight only, suggesting that
a diarrheal episode in a single month does not have a significant association with length in the
following 1, 2 or 3 months (40). When they explored the lagged effects of diarrheal episodes
on growth over time, they found that the cumulative burden of diarrhea resulted in a
measurable effect on growth (40). They also suggest that if there is adequate time between
diarrheal episodes that the child may experience catch-up growth (40,41). The opportunity for
catch up growth may be limited or missed if periods without diarrhea are short due to repeated
episodes (41). The study by Checkley et al (37) found that the effect of diarrhea on height was
infections determined that infection was associated with reduced growth throughout the 9
12
3.5.4 Etiology of Enteropathogen Infection
(42,43).
Viral
Viral pathogens such as rotavirus, norovirus, adenovirus, and astrovirus, invade the villous
epithelium of the small intestine mucosa (22,44). This results in the loss of mature absorptive
cells and replacement with poorly differentiated crypt cells (44). The extent of the invasion and
lesions will determine the loss of absorptive capacity of the small intestine, and therefore the
Bacterial
and colonize the large intestine (44). They then invade the intestinal mucosa through the
inflammatory reaction that can result in ulceration or synthesis of vasoactive substances (44).
Other bacterial enteropathogens produce cytotoxins that cause cell death and damage (44).
Enterotoxigenic E. coli (ETEC), for example, binds to intestinal epithelial cells and decreases
absorptive capacity through the structural damages to the membrane, while Cholera toxin
binds to specific mucosal receptors to alter salt and water transport (44).
Parasitic
Common parasitic enteropathogens include Giardia lamblia and Cryptosporidium (44). While
the pathogenic mechanism of parasitic enteropathogens is not fully known and understood, it
is thought that they act as physical barriers to absorption by causing injury to the mucosal
lining, creating an immunologic reaction in the host, or altering gastrointestinal motility (44).
13
3.5.5 Important Enteropathogens
The Global Enteric Multicenter Study (GEMS), which enrolled children from four sites in Africa
and three in Asia, determined that, overall, the following pathogens were responsible for most
coli, Cryptosporidium, and Campylobacter (Table 3.1) (45,46). The MAL-ED study enrolled
children from two sites in South America, two in Africa, and four in South Asia. The study
the major pathogens associated with diarrhea in the first year of life while Campylobacter,
norovirus, rotavirus, astrovirus, and Shigella were the major pathogens in the second year of
life (Table 3.1) (30). Both studies identified variability in the most important pathogens between
participating sites.
Table 3.1 Major Pathogens in the GEMS and MAL-ED studies (30,45,46)
GEMS MAL-ED
Enteropathogen Enteropathogen
Enteropathogen Type Type Type
(0-12 months) (13-24 months)
Shigella Bacterial Norovirus Viral Campylobacter Bacterial
Rotavirus Viral Rotavirus Viral Norovirus Viral
Adenovirus Viral Campylobacter Bacterial Rotavirus Viral
ETEC Bacterial Astrovirus Viral Astrovirus Viral
Cryptosporidium Parasitic Cryptosporidium Parasitic Shigella Bacterial
Campylobacter Bacterial
3.6 Household Risk factors for Enteropathogen Infection and Linear Growth
Most of the household risk factors associated with enteropathogen infection are associated
with low socioeconomic status and the associated poor household environmental conditions
(3,14). The majority of enteropathogens are transmitted through contact with feces, both
directly or indirectly through food, water, and contaminated surfaces (23,47). An analysis of
14
data from 137 countries confirmed the importance of environmental factors in not only causing
diarrheal disease but also contributing to growth impairment (19). Following fetal growth
restriction and preterm birth, environmental factors, such as water quality and sanitation, were
the second leading cause of stunting prevalence (19). The negative effect of these factors on
linear growth has been hypothesized to be partially mediated through diarrhea (48).
Food handling and storage practices are important in enteropathogen transmission and are
said to be one of the major contributors to infection and diarrheal disease (23).
Enteropathogen transmission can vary by food type and conditions of storage and handling
(23). Contamination sources can include unclean pots, cooking utensils and baby bottles, with
inadequate washing or storage making them unsafe (23). Cooking and storing food at
inadequate temperature, reheating more often than is recommended and consumption of cold
leftovers has been independently associated with infectious diarrheal disease (23,49).
Animals in the household compound have been found to be a potential risk factor for linear
growth and enteric infection as they can be the reservoirs for many enteropathogens (23,50).
Animals can harbor pathogens in their gastrointestinal tract and then contaminate their
environment by shedding these pathogens through their feces (50). The pathogen may remain
in the environment for long periods of time (50). Contact with domesticated and companion
animals has been associated with Campylobacter infection transmission (51). Chickens in
Maternal Factors
Maternal education has also been identified as an important risk factor for infection and
diarrheal disease. In many studies, higher maternal education has been associated with a
decreased risk of infection (5). In a study in Peru, low maternal education was associated with
15
a 34% increased odds of Shigella infection (52). A study in Ethiopia also found that maternal
primary education was protective against diarrhea in children (53). Maternal height is also a
strong predictor of birth length, as a short mother is more likely to have a child that is small for
gestational age (SGA) (54). This may be due to the physical constraints of a smaller body
Most enteric infections are transmitted via the fecal-oral route (22). As such, improved water,
sanitation and hygiene practices have been commonly suggested as being critical in
decreasing transmission (Table 3.2) (22). Checkley et al (55) found that poor conditions of
water source, water storage and sanitation resulted in children being 1 cm shorter in stature
and suffering 54% more diarrheal episodes than healthy children. Interestingly, they also
found that the effect of water and sanitation on height was independent of the effect on
diarrheal disease (55). They suggested this may be due to the potential presence of
asymptomatic infection, which is known to impact linear growth (55). The consumption of
contaminated water as well as the use of water for cleaning and bathing, are implicated in
children, the greatest reduction in risk of diarrheal infections was associated with better water
storage practices (55). They found that uncovered water containers were associated with
increased diarrheal incidence (55). Also, a lack of access to water can be associated with
inadequate personal hygiene, including hand washing, and an inadequate water reservoir,
one that is stagnant and infrequently cleaned, is also associated with diarrheal infection (56).
16
Table 3.2 JMP Water Source Classification (57)
Improved Unimproved
Piped water Unprotected dug well
Boreholes or tubewells Unprotected spring
Protected dug wells River, dam, lake, pond, stream, canal,
irrigation canal
Protected springs
Rainwater
Packaged or delivered water
Unsafe disposal of human and animal feces, including the practice of open defecation has
been associated with increased diarrheal infection (57). It is estimated that over one billion
people worldwide practice open defecation, which is seen more predominantly in rural areas
(57). Improved sanitation practices through the use of improved toilet facilities (Table 3.3), aim
to prevent enteropathogens from coming into contact with the environment, by preventing the
transmission of pathogens found in feces (57). The literature has consistently identified a
reduction in diarrheal disease with latrine use by 22-51% (58). A recent analysis of the
Campylobacter pathogen in the MAL-ED study found that despite heterogeneity across the
sites, there was a consistency in the reduction of infection with improved latrine facilities (60).
A cohort study in Ethiopia, India, Peru and Vietnam found that access to improved toilets was
Improved Unimproved
Flush/pour flush to piped sewer systems Pit latrine without slab or platform
Septic tanks or pit latrines Hanging latrine
Ventilated improved pit latrines Bucket latrine
Composting toilet or pit latrine with slab Open defecation
17
Behavioural changes can also have an impact on reducing infection transmission. Hand
washing in particular has been determined to be the most cost-effective method for reducing
the incidence of diarrheal diseases (23). An analysis of 17 studies showed a significant pooled
reduction of 42-48% in the risk of diarrheal infection resulting from handwashing (61). Hand
washing with soaps has been found to be more effective than washing with water alone; and
has been found to be effective in contamination of even viruses (23). Unfortunately, global
rates for soap hand washing range from only 0-34% (23).
Recently, the WASH Benefits Bangladesh, WASH Benefits Kenya, and SHINE Zimbabwe
trials aimed to explore the effects of a WASH intervention, a nutrition intervention, and a
combined WASH and nutrition intervention on health and growth outcomes in children (48).
Interestingly, the nutrition intervention significantly increased LAZ in all three interventions, but
the WASH intervention had no effect on LAZ in all three trials (48). Similarly, the WASH
intervention had inconsistent results on incidence of diarrhea. It did not show a reduction in
diarrhea in Kenya or Zimbabwe, but it did show a reduction in the Bangladesh study (48). The
authors acknowledge that this lack of improved in LAZ and the weak diarrhea results are
inconsistent with a large body of literature that report poor household WASH conditions are
strong predictors for poor linear growth in children (48). They also found that despite high
adherence to the study intervention, the children who received WASH interventions continued
to experience very high enteropathogen infection (48). The authors therefore suggest that the
trial interventions may not have been effective enough to elicit a meaningful or large enough
An analysis by the World Health Organization (62) of over 1000 studies conducted between
2012 and 2017 indicated a protective effect of improved sanitation on infectious disease and
outcomes of nutritional status (62). They also acknowledge; however, that there is a low quality
of the evidence, that the health improvements are lower than originally expected, and that
there is a lack of evidence on the role of animals and food contamination in disease
18
transmission pathways. In addition, the review suggest that improved health outcome would
Infant and child feeding practices can have a major impact on both child survival and growth.
Optimal practices include the immediate initiation of breastfeeding, exclusively for the first 6
months, continued until two years of age, and the gradual introduction of safely prepared
nutritional complementary foods that will meet any nutritional requirement gaps left by breast
milk (63). As previously described, the WASH Benefits and SHINE trials in Bangladesh, Kenya
and Zimbabwe found an increased mean LAZ in the nutrition intervention arms, which they
Breastfeeding
during the first 6 months of life, as this prevents exposure to contaminated food or water (9,64).
Additionally, breast milk provides important immunological factors and antimicrobial properties
that can protect the infant as its own immune system develops (9,64). Neonatal mortality is
reduced overall by early or immediate initiation of breastfeeding. Infants that are breastfeed in
the first post-partum days receive the protective antibodies and essential nutrients provided in
colostrum, the milk produced in the first days (65). In the MAL-ED study, exclusive
The same association was not found for non-exclusive breastfeeding (60). In Brazil, children
who were not breastfed were 14 times at greater risk of death from diarrheal disease, and
those that received powdered or cow’s milk in addition to breast milk were 4.2 times at risk of
death from diarrheal disease, compared to those exclusively breastfed (64). Weaning from
breastmilk was associated with increased diarrheal infection, which suggests children’s risk of
19
Complementary feeding
Complementary feeding interventions contribute to small but significant gains in linear growth
in children in LMICs (66). This period is defined as the time when children begin consuming
semi-solid and solid foods in addition to breast milk and typically takes place between 6 and
24 months of age, which is when a significant portion of stunting takes place (4). In LMICs,
children may not meet the minimum dietary quality standard for complementary feeding, even
though it is reported that an estimated 58% of infants between 6-9 months of age receive
complementary foods while they continue to be breastfed (65). Timing may also be an issue,
with some children receiving complementary foods too early or too late, or an inadequate
frequency throughout the day (65). Studies have found that prolonged breastfeeding
Onyango et al (67) found that dietary diversity was positively associated with anthropometric
measurements. An analysis of the data from the Nepal site in the MAL-ED study also found
that inadequate dietary intake impacted growth velocity in the first 2 years of life and growth
The largest number of stunted children resides in South Asia, where approximately 33.3% of
children under the age of 5 are affected (68). Bangladesh has been successful in reducing the
prevalence of stunting, with a sustained reduction of child undernutrition for more than twenty
years, decreasing from 43% in 2007 to 31% in 2017 (69). Suggested reasons for these
improvements are often associated with national-level efforts and programs to improved
education for women and access to prenatal care(70). There was also reduced fertility rates,
reduced open defection rates, and an increase in household assets (70). Despite this
progress, it remains one of the countries with the highest burden of stunting in the world (4).
WASH characteristics continue to be a concern in Bangladesh. While open defection has been
significantly reduced to 4% at the national level, other problems persist in the country (71).
20
Urban areas continue to lack proper human waste management, garbage management, and
continues to experience a fast rate of urbanization (4). Urban settlements are densely
populated and continue to attract the most economically disadvantaged rural population (4).
These conditions of overcrowding and poverty result in a public health challenge, as the
households and their children are more vulnerable to disease (4). Interestingly, despite the
conditions of poverty in the urban settings, a rural-urban disparity is still described in the
literature (72). Rural homes have a lower access to improved sanitation, and lower education
levels (72).
3.9 Summary
The causes of linear growth impairment and enteric infection in young children are
multifactorial. Bangladesh has made impressive improvement in the linear growth status of its
attention. Densely populated urban centres present poor living conditions resulting from
disease, particularly enteric infection, which could have an effect on linear growth in young
children. Impoverished rural communities also experience poor environmental conditions and
lack access to services and education. The first two years of life are critical in preventing and
treating growth impairment, and as such, it is important to understand the associated factors
during this time. This research therefore aims to explore the household-level factors
REFERENCES
1. Foote JM. Optimizing Linear Growth Measurement in Children. J Pediatr Heal Care
21
Paediatr Int Child Health. 2014;34(4):250–65.
3. Budge S, Parker AH, Hutchings PT, Garbutt C. Environmental enteric dysfunction and
4. Islam MM, Sanin KI, Mahfuz M, Ahmed AMS, Mondal D, Haque R, et al. Risk factors
5. Molbak K, Jensen H, Ingholt L, Aaby P. Risk factors for diarrheal disease incidence in
1997;146(3):273–82.
2010;125(3):e473–80.
9. Bhutta ZA, Das JK, Rizvi A, Gaffey MF, Walker N, Horton S, et al. Evidence-based
interventions for improvement of maternal and child nutrition: What can be done and
10. Black RE, Cousens S, Johnson HL, Lawn JE, Rudan I, Bassani DG, et al. Global,
regional, and national causes of child mortality in 2008: a systematic analysis. Lancet
6736(10)60549-1
et al. Paediatrics and International Child Health Piecing together the stunting puzzle :
a framework for attributable factors of child stunting Piecing together the stunting
puzzle : a framework for attributable factors of child stunting. Paediatr Int Child Health
[Internet]. 2016;9047(October):1–8.
22
12. Stevens GA, Finucane MM, Paciorek CJ, Flaxman SR, White RA, Donner AJ, et al.
Trends in mild, moderate, and severe stunting and underweight, and progress
13. Roth DE, Krishna A, Leung M, Shi J, Bassani DG, Barros AJD. Early childhood linear
condition: analysis of 179 Demographic and Health Surveys from 64 countries (1993–
14. Rogawski ET, Liu J, Platts-Mills JA, Kabir F, Lertsethtakarn P, Siguas M, et al. Use of
results from the MAL-ED cohort study. Lancet Glob Heal. 2018;6(12):e1319–28.
2016;53(1):241–67.
enfants. Arch Pédiatrie [Internet]. 2009 Jan 1 [cited 2019 Sep 4];16(1):47–53.
17. De Onis M, Onyango AW, Borghi E, Garza C, Yang H. Comparison of the World
Health Organization (WHO) Child Growth Standards and the National Center for
2012;15(9):1603–10.
19. Danaei G, Andrews KG, Sudfeld CR, Fink G, McCoy DC, Peet E, et al. Risk Factors
20. Schwinger C, Fadnes LT, Shrestha SK, Shrestha PS, Chandyo RK, Shrestha B, et al.
23
Predicting Undernutrition at Age 2 Years with Early Attained Weight and Length
21. Jennifer Bryce, Denise Coitinho, Ian Darnton-Hill, David Pelletier PP-A. Maternal and
Child Nutrition- Executive Summary of The Lancet Maternal and Child Nutrition
22. Petri WA, Miller M, Binder HJ, Levine MM, Dillingham R, Guerrant RL. Review series
Enteric infections , diarrhea , and their impact on function and development. J Clin
Invest. 2008;118(4).
23. Nelson, Konrad ; Williams CM. Infectious Disease Epidemiology. Third. Gartside M,
24. SCRIMSHAW NS, TAYLOR CE, GORDON JE. Interactions of nutrition and infection.
25. DeBoer MD, Scharf RJ, Leite AM, Férrer A, Havt A, Pinkerton R, et al. Systemic
inflammation, growth factors, and linear growth in the setting of infection and
26. Kosek M, Guerrant RL, Kang G, Bhutta Z, Yori PP, Gratz J, et al. Assessment of
27. Katona P, Katona-Apte J. The Interaction between Nutrition and Infection. Clin Infect
28. Guerrant RL, Schorling JB, McAuliffe JF, De Souza MA. Diarrhea as a cause and an
29. Mbuya MNN, Humphrey JH. Preventing environmental enteric dysfunction through
30. Platts-Mills JA, Babji S, Bodhidatta L, Gratz J, Haque R, Havt A, et al. Pathogen-
24
cohort study (MAL-ED). Lancet Glob Heal. 2015;
31. Richard SA, McCormick BJJ, Miller MA, Caulfield LE, Checkley W. Modeling
32. Troeger C, Forouzanfar M, Rao PC, Khalil I, Brown A, Reiner RC, et al. Estimates of
diseases: a systematic analysis for the Global Burden of Disease Study 2015. Lancet
33. Checkley W, Buckley G, Gilman RH, Assis AM, Guerrant RL, Morris SS, et al. Multi-
2008;37(4):816–30.
34. Moore SR, Lima NL, Soares AM, Ori RB, Pinkerton RC, Barrett LJ, et al. Prolonged
episodes of acute diarrhea reduce growth and increase risk of persistent diarrhea in
35. Lee G, Pan W, Peñataro Yori P, Paredes Olortegui M, Tilley D, Gregory M, et al.
36. Checkley W, Gilman RH, Epstein LD, Suarez M, Diaz JF, Cabrera L, et al.
37. Checkley W, Epstein LD, Gilman RH, Cabrera L, Black RE. Effects of acute diarrhea
38. Checkley W, Epstein LD, Gilman RH, Black RE, Cabrera L, Sterling CR. Effects of
39. Rogawski ET, Bartelt LA, Platts-Mills JA, Seidman JC, Samie A, Havt A, et al.
Determinants and Impact of Giardia Infection in the First 2 Years of Life in the MAL-
25
40. Richard SA, Black RE, Gilman RH, Guerrant RL, Kang G, Lanata CF, et al. Diarrhea
in early childhood: Short-Term association with weight and long-Term association with
41. Richard SA, Black RE, Gilman RH, Guerrant RL, Kang G, Lanata CF, et al. Catch-Up
42. Ahs JW, Tao W, Löfgren J, Forsberg BC. Diarrheal Diseases in Low- and Middle-
[Internet]. 2010;4(1):113–24.
43. Guerrant RL, Gilder T Van, Steiner TS, Thielman NM, Slutsker L, Tauxe R V, et al.
Practice Guidelines for the Management of Infectious Diarrhea. Clin Infect Dis.
2001;32:33150.
44. Cohen MB. Etiology and mechanisms of acute infectious diarrhea in infants in the
45. Kotloff KL, Nataro JP, Blackwelder WC, Nasrin D, Farag TH, Panchalingam S, et al.
46. Liu J, Platts-Mills JA, Juma J, Kabir F, Nkeze J, Okoi C, et al. Use of quantitative
47. Lin A, Arnold BF, Afreen S, Goto R, Huda TMN, Haque R, et al. Household
48. Pickering AJ, Null C, Winch PJ, Mangwadu G, Arnold BF, Prendergast AJ, et al. The
weaning food : a major risk factor for diarrhoea and associated malnutrition *.
26
1993;71(1):79–92.
50. Angulo FJ, Steinmuller N, Demma L, Bender JB, Eidson M, Angulo FJ. Outbreaks of
Enteric Disease Associated with Animal Contact: Not Just a Foodborne Problem
51. Kaakoush NO, Castaño-Rodríguez N, Mitchell HM, Man SM. Global epidemiology of
52. Kosek M, Peñataro Yori P, Pan WK, Paredes Olortegui M, GIlman RH, Perez J, et al.
53. Mihrete TS, Alemie GA, Teferra AS. Determinants of childhood diarrhea among
underfive children in Benishangul Gumuz Regional State , North West Ethiopia. BMC
54. Leroy JL, Frongillo EA. Perspective: What Does Stunting Really Mean? A Critical
55. Checkley W, Gilman RH, Black RE, Epstein LD, Cabrera L, Sterling CR. Effect of
56. Pruss A, Kay D, Fewtrell L, Bartram J. Estimating the Burden of Disease from Water,
2002;110(5):537. A
57. Brown J, Cairncross S, Ensink JHJ. Water, sanitation, hygiene and enteric infections
58. Esrey SA, Habicht JP, Casella G. The complementary effect of latrines and increased
1992;135(6):659–66.
27
59. Dearden KA, Schott W, Crookston BT, Humphries DL, Penny ME, Behrman JR.
Children with access to improved sanitation but not improved water are at lower risk
61. Cairncross S, Hunt C, Boisson S, Bostoen K, Curtis V, Fung ICH, et al. Water,
2010;39(SUPPL. 1).
62. WHO/UNICEF. Implications of recent WASH and nutrition studies for WASH policy
63. Caulfield LE, Bose A, Chandyo RK, Nesamvuni C, De Moraes ML, Turab A, et al.
Infant feeding practices, dietary adequacy, and micronutrient status measures in the
Significant Impact on Linear and Ponderal Growth of Children in Low- and Middle-
67. Onyango A, Koski G, Tucker KL. Food diversity versus breastfeeding choice in
68. Kang Y, Aguayo VM, Campbell RK, West KP. Association between stunting and early
childhood development among children aged 36–59 months in South Asia. Matern
69. NIPORT, ICF International. Bangladesh Demographic and Health Survey 2017-18:
70. Headey D. How Did Bangladesh Reduce Stunting So Rapidly? 2014;5–7. Available
28
from: http://www.globalnutritionreport.org/files/2014/11/gnr14_pn4g_09headey.pdf
29
CHAPTER 4 GENERAL METHODOLOGY
This chapter presents general methodological details for this research project. Further details
This research examines linear growth in children 1-24 months of age living in Bangladesh.
With a population of approximately 160 million people, Bangladesh is the most densely
populated country in the world (1). It is also one of the most urbanised countries in South Asia
with approximately 33% of the population living in urban regions (2). To address the aims and
objectives of this research project, different datasets of children in Bangladesh were analysed.
Research Aim #1
To identify the household-level factors associated with linear growth impairment in children 1-
24 months of age in Bangladesh, and then explore by age and geographic location, two
country-level datasets were used. Data from both datasets included children from households
across all seven administrative divisions across Bangladesh (Figure 4.1). This geographically
broad exploration also allowed for exploration of factors by urban vs rural location.
Figure 4.1 Map of Bangladesh and the seven administrative divisions (1)
30
Research Aim #2
To identify the enteropathogen infections and household-level risk factors associated with
childhood growth in urban Bangladesh, the data from the MAL-ED study site in Mirpur, Dhaka,
Bangladesh was used for this project. With a population size of approximately 500,000 people,
the Mirpur community, one of the 21 administrative units of Dhaka, has the conditions
representative of a densely inhabited urban settlement in South Asia, thus increasing the
generalizability of the results of this analysis to other urban centres in the region.
4.2 Data
Several data sources were used to examine linear growth in children in Bangladesh.
Household-level variables selected included maternal factors, child factors, child health
31
Research Aim #1
This analysis included cross-sectional data collected from two major household survey
programs: the 2014 Bangladesh Demographic and Health Survey (DHS) and the 2012-2013
Bangladesh Multiple Indicator Cluster Survey (MICS). The DHS and MICS survey programs
are the primary source for national-level data of child health indicators in LMICs due to the
lack of available local information systems. The DHS is supported by USAID, while the MICS
The 2012-2013 Bangladesh MICS survey was conducted by the Bangladesh Bureau of
Statistics. Data was collected from 51,895 households across the country between December
2012 and April 2013 (3). The survey provides national level estimates of child health indicators
as well as information about access to safe drinking water and improved sanitation. This is the
first Bangladesh MICS survey that assessed the quality of drinking water by collecting
information about the contamination level of arsenic and E. coli (3). This dataset provides
national-level estimates disaggregated by the seven divisions, location, sex, age, education
and wealth quintile (3). A total of 7851 children between the ages of 1 and 24 months were
included in the final analysis. Further information about the study sample is outlined in Chapter
The 2014 Bangladesh DHS survey was conducted as part of a collaborative effort of the
National Institute of Population Research and Training (NIPORT), ICF International, USA and
Mitra & Associates. A total of 17,300 households were interviewed for the survey (1) The DHS
survey is similar to the MICS survey in that it also provides national level estimates of maternal
and child health indicator, as well as some household characteristics, including several WASH
factors. A total of 2633 children between the ages of 1 and 24 months of age were included in
32
the final analysis. Further information about the data can be found in Chapter 6. Table 4.1
Research Aim #2
The MAL-ED Study was a multicentre longitudinal study conducted at 8 sites across the world.
The analysis for this study aim and its objectives focuses on the data collected at the
Bangladesh study site in Mirpur, Dhaka. Children were enrolled between 2009 and 2012
shortly after birth and were followed regularly for 24 months. Data relevant to this analysis
also included household factors, including maternal and WASH characteristics. Table 4.1
outlines the data variables included in the analysis. The main advantages to this analysis are
the longitudinal nature of the study design and the availability of extensive enteropathogen
data collected, which allows for the exploration of the association between enteric infection
and linear growth, and which household factors are associated with both. The study identified
the pathogens found in both symptomatic and asymptomatic stools. Regular collection of
these samples provides a robust dataset from which to explore the temporal relationship
between infection and growth throughout the first 24 months. Further information about the
33
Table 4.1 Summary of Data
Variables
Maternal
Age x x x
Height x X
Education x x x
Child Characteristics
Age x x
Sex x x x
Birth order x
Child health
Breastfeeding x x x
Antibiotic x
Enteric Infection
Diarrhea x x x
Parasite meds x
Pathogen data x
Household
Urban vs rural x x Urban only
Animals x x x
Flooring material
Cooking location x
Refrigerator x x x
# residents/children
WASH
Toilet facility x x x
Shared toilet x x x
Water source x x x
Water treatment x x
Stool disposal methods x x
Handwashing Presence of a station Behaviours
34
4.3 Statistical Methods
The 2006 WHO growth standards were developed from a sample of children from Brazil,
Ghana, India, Norway, Oman, and the United States. The study demonstrated that children in
different parts of the world have similar patterns of linear growth if all their health needs are
reference, accounting for ethnic diversity (4). HAZ/LAZ scores derived from those growth
standards are commonly used in research and as a tool to monitor growth in early childhood.
Changes in absolute height measurements were also considered for this project; however,
they could be a limitation in exploring growth velocity and changes in height in given intervals,
as the literature indicates that smaller children gain less height in a short period (5,6). Using
Research Aim #1
The data were stratified by age group (1-12 months and 13-24 months). Given the large
sample size, there was further stratification of the data by urban versus rural location, for a
total of four subgroups to be analysed. This allowed for the second objective of Aim #2 to be
addressed, which seeks to determine the modification of the association between household
factors and LAZ by geographic location of the household. A linear mixed effects analysis was
then conducted on each of the four subgroups. This method was selected in order to account
for a potential household effect, given that some households had more than one child 1-24
35
Demographic and Health Survey Analysis
As in the MICS analysis, data were also stratified (1-12 months and 13-24 months) and
analysed using multiple regression analysis to identify the household factors significantly
associated with LAZ across the two different age groups. In addition, a moderation and
mediation analysis was conducted to explore the relationships between the household factors
and LAZ; specifically whether there was any mediation via diarrhea incidence and whether
there were any interaction effects between the household factors, the mediator, and LAZ.
Research Aim #2
There are several analytical approaches for longitudinal growth data that allow the analysis of
change through time. Repeated measures analysis of variance is a method that has been
widely used; however, a major limitation is the need for subjects to have the same number of
measurements collected, and for there to be minimal variation in the time of collection.
Generalized estimating equations are another useful approach that allow the calculation of an
average trajectory for the cohort; however, subject-specific effects are not calculated using
this method. Subject-specific effects is important in this analysis in order to allow the
examination of the effect of enteric infection on individual growth outcomes, and how they
result in between-subject differences. Mixed effects models allow for the generation of
individual growth trajectories and, have been widely used to model change in longitudinal data.
Latent growth curve modelling, a less commonly used approach outside of the social sciences,
allows the application of many predictors and determines which exert important effects on the
covariates. This approach was used in the analysis of the MAL-ED data to address aim #2
and associated objectives. In addition to exploring the direct relationship between household
factors and LAZ, the relationship between various enteropathogen infections and LAZ was
explored, as was the association between the household factors and infection, and ultimately
their indirect effect on LAZ via infection. Further information about latent growth curve
modelling and its application in this thesis project is outlined in chapters 7 and 8.
36
4.4 Ethics Approval
Ethical approval was the responsibility of the institutions that administered the surveys. As this
was a secondary analysis of publicly available and de-identified data, additional ethical
MAL-ED
The MAL-ED analysis was granted an Exemption to Ethics Review at the University of
Queensland, due to the negligible risk to participants and that it is a secondary analysis.
[Clearance Number: 2018000676] An ethics application was also submitted to the regional
behalf of the Swiss Tropical and Public Health Institute, the institution of PhD advisor and
REFERENCES
Associates and II. Bangladesh Demongraphic and Health Survey 2014. Rockville,
2012-2013. 2016.
37
5. Voss LD, Mulligan J. Normal growth in the short normal prepubertal child: The
6. Voss LD, Wilkin TJ, Bailey BJR, Betts PR. The reliability of height and height velocity
in the assessment of growth (the Wessex Growth Study). Arch Dis Child.
1991;66(7):833–7.
38
RESEARCH AIM #1
impairment and infection have had inconsistent results. Understanding how differences in
childhood and household characteristics are associated with impaired growth in specific
geographic regions may explain such inconsistencies. This understanding can contribute to
the development of targeted interventions that more effectively reduce impairment. The first
of two research aims examines these factors across Bangladesh using cross-sectional
datasets from two national household level surveys. The analyses of these two large datasets
characteristics of children, mothers, and the household are associated with growth impairment
in children in Bangladesh. The overlap of several data points between the two datasets allows
for the comparison between the results of the two analyses, while different variables contribute
The subsequent two chapters presents the results of these analyses. Chapter 5 presents the
results of the 2012-2013 Bangladesh Multiple Indicator Cluster Survey mixed effects analysis
which has now been submitted for publication. Chapter 6 presents the results of the 2014
Bangladesh Demographic and Health Survey multiple linear regression and mediation-
moderation analyses.
39
CHAPTER 5 MULTIPLE INDICATOR CLUSTER SURVEY
ANALYSIS
5.1 Preamble
This is the first of two analyses addressing the first research aim and its objectives used data
collected in the Bangladesh Multiple Indicator Cluster Survey conducted in 2012-2013. The
large sample size provided the opportunity to conduct a stratified analysis of children to identify
differences in the characteristics of the children, the mothers, and the households as they
The MICS survery has several overlapping data points with the DHS survey; however, the
MICS contains more WASH information, including questions on water quality. In addition,
unlike the DHS, survey respondents were the caregivers of children and not limited to the
biological mother, which allowed for the inclusion of orphans and foster children living in the
household. In this analysis a linear mixed effects approach was applied to account for a
household effect due to the number of households having more than one child enrolled in the
study.
Through the stratification of the data by geographic location and age of the children, this
analysis presents the first of two examinations of the different characteristics of participating
children and household factors as they relate to childhood growth across Bangladesh.
This chapter is in the form of a manuscript which has been submitted for publication.
40
5.2 Household-Level Risk Factors Associated with Linear Growth
Among Young Children in Bangladesh: An Analysis of the Multiple
Indicator Cluster Survey, 2012-2013
Abstract
low- and middle-income countries, including Bangladesh. The further identification of factors
associated with linear growth could contribute to public health efforts to improve childhood
growth.
Methods: This study examined the household-level factors associated with child growth using
data from the 2012-2013 Multiple Indicator Cluster Survey (MICS) in Bangladesh. A total of
7851 children 1-24 months of age were included in the analysis. A linear mixed model analysis
with household as a random effect was used to examine the effect of different household
factors independently associated with length-for-age z-score (LAZ). Significant factors were
included in a stepwise regression to identify the final combination of factors influencing the
then conducted the analysis in four subgroups: 1) Rural children 1-12 months of age; 2) Rural
children 13-24 months of age; 3) Urban children 1-12 months of age, and; 4) Urban children
Results: The age of the child was negatively associated with the outcome, with LAZ scores
decreasing with increasing child age. Differences were identified between the factors
associated with LAZ across the four subgroups. Children 13-24 months of age from rural
households had the highest number of factors associated with LAZ. Maternal education was
important overall and in three of the four subgroups. Male sex was important in both rural
41
subgroups. Unimproved WASH factors such as stool disposal methods were important in rural
settings and in older urban children. Interestingly, there were no factors associated with LAZ
Conclusions: The results indicate that growth impairment continues to be a public health
concern in Bangladesh, and that child growth, as measured by LAZ, is associated with
different household-level factors in different age groups and vary by rural or urban location.
Interventions should therefore not only target these factors but consider the heterogeneity
that exists.
Introduction
Childhood malnutrition continues to be a major global health concern in low- and middle-
income countries (WHO 2006). The prevalence of such indicators as stunting, defined as a
length for age z-scores (LAZ) greater than two standard deviations (SD) below the World
Health Organization (WHO) Child Growth Standards median still remains high. South Asia
has the largest number of stunted children worldwide, with approximately 65 million children
under the age of 5 classified as stunted (1). Despite a sustained reduction of child
undernutrition for more than twenty years, Bangladesh continues to be among the 20 countries
in the world with the highest burden of linear growth impairment (2,3). This challenge is of
critical importance since malnutrition can result in high morbidity and mortality among children
(2). In addition, linear growth impairment has long term health and developmental
(3).
The first 24 months of life are widely recognized as critical for preventing undernutrition. It is
a period of rapid growth of a child and as a result a period when growth faltering predominantly
42
occurs (4). Nutritional interventions can have the greatest impact among children of this age
in reducing growth impairment. Interventions beyond this period have little impact on linear
growth (5,6).
The causes of growth impairment are complex and involve multiple factors. Many of these
factors are associated with low socioeconomic status and poor household environmental
water, hygiene and sanitation and household characteristics, and there is increasing research
on the relationship between socio-economic factors and linear growth (4,7,8). There is a steep
decline in LAZ that occurs between birth and 24 months of age reported in the literature but
this faltering does not occur uniformly in this period (9). As such, it is important to explore the
how variation in household factors between communities and across regions are differentially
associated with growth faltering. In particular, there may be key differences between urban
and rural households that would be important to consider. This understanding would contribute
to efforts develop public health interventions that more effectively reduce growth impairment.
This study therefore aims to investigate the associations of household risk factors with overall
linear growth in the first 24 months of age and explore associations by specific age groups of
Methods
Study Sample
The analyses of child growth in the first 24 months of age was carried out using data from the
sectional survey conducted in Bangladesh. Details of the MICS survey are presented in the
MICS final report, which is publicly available and can be accessed from their website (10). A
total of 8603 children less than 25 months of age were included in the 2012-2013 Bangladesh
43
Survey. Of these children, 457 were excluded from the analysis due to out-of-range or missing
LAZ values and, 257 were excluded due missing household data. A total of 7851 children
8603 children
were less than 25
months of age
7851 children
included in the
analysis
RURAL URBAN
Study Variables
Dependent Variable
The dependent variable in this analysis was LAZ. Per WHO recommendations, children were
Independent Variables
We investigated association between LAZ and various child and clinical characteristics, such
as the occurrence of an episode diarrhea in the past 2 weeks, and breastfeeding. The maternal
characteristic included in the analysis was the attained education level. Household
44
characteristics explored included location of household, numbers of household members, and
number of household members under the age of five, floor material, cooking location, and the
presence of a refrigerator. Water, hygiene, and sanitation (WASH) factors included water
source, water source location, toilet facility, whether the toilet facilities were shared, and the
presence of a handwashing station. Data collected on water source and type of toilet facility
in the household were categorized in accordance with the WHO/UNICEF Joint Monitoring
Program (JMP) classification of improved versus unimproved water source or toilet facilities
(11). An improved water source protects the water from external contamination, particularly
from human excreta. Unimproved water sources include unprotected dug well or spring,
surface water, and tanker truck water. Improved toilet facilities ensure a separation of human
waste from human contact (11). Unimproved toilet facilities include hanging latrines, bucket
latrines, latrines without slabs, or open defecation in fields. In addition, a select number of
household water and water source were tested for levels of Escherichia coli (E. coli). E. coli is
a recommended indicator of feces material. Bangladesh, as many other countries, has set a
standard of no E. coli being present in samples of drinking water, based on a 100 mL sample
of water (12).
Data Analysis
The data were analysed in several stages. Descriptive statistics were first generated for
characteristics of study children and their mothers, as well as select characteristics of the
households. Visualization of the data was achieved using the kernel-weighted local polynomial
smoothing algorithm in Stata, version 14. Independent bivariate associations between LAZ
and child characteristics, maternal characteristics, and the household characteristics were
then investigated using simple linear regression. A number of households had more than one
child included in the survey. Accordingly, a linear mixed effects regression analysis that
included the independent variables found to be significant in the bivariate analysis was carried
out in R, version 3.5.1, to determine whether there was a household effect on LAZ. A stepwise
45
regression method was then applied to confirm the variables to include in the model. Statistical
significance was set at P <0.05, while marginal significance was set as ≤0.20.
To explore whether factors associated with growth were different in younger or older children,
or according to rural/urban location of the household, the children were then subdivided into
four subgroups: rural children 1-12 months of age; rural children 13-24 months of age; urban
children 1-12 months of age; and urban children 13-24 months of age. The same methodology
applied to the overall group was then applied to the four subgroups.
Results
Child Characteristics
The characteristics of the children are presented in Table 5.1. The mean age of children
included in the analysis was 11.98 months. The overall mean LAZ among children 1-24
months of age was -1.45 ±1.50. Mean LAZ was lowest among children 13-24 months of age,
with urban children having a mean LAZ of -1.56±0.06 and rural children -1.86 ±0.02. Children
1-12 months of age living in urban households had the highest mean LAZ scores (-0.98 ±1.50),
followed by the rural children 1-12 months of age (-1.13 ±1.55)). As reported in previous
studies, a negative association was found between increasing age and LAZ, with LAZ
decreasing with increasing child age (Figure 5.2). Information on diarrheal episodes was
collected in the two weeks prior to the survey. A total of 493 children (6.28%) were reported
as having had a recent episode of diarrhea, but diarrhea was not significantly associated with
LAZ. Finally, most children (94.84%) of children were being breastfed at the time of the survey,
including children 13-24 months of age, although exclusive breastfeeding was not
LAZ.
46
Maternal Characteristics
Maternal characteristics and results of the bivariate regression analysis are reported in Table
5.2. As displayed in Figure 5.2, children of mothers with the highest level of education also
had the highest LAZ scores across all age groups. LAZ differences become more evident after
6 months of age, when the LAZ scores of children of mothers with no education is the lowest.
A simple linear regression analysis found a statistically significant relationship between LAZ
and maternal education. Information about maternal height was not collected in the MICS
survey.
Household Characteristics
Household characteristics and results of the bivariate linear regression are reported in Table
5.3. Most households (84.00%) were located in rural areas. There was an association between
the location of the household and LAZ, with higher LAZ scores among children from urban
households (Figure 5.2). The number of children under 5 years of age in the household, the
number of household members, and presence of animals were not associated with LAZ.
Overall, 21.12% of households had an outdoor cooking location, which was independently
associated with lower LAZ scores compared with cooking in the kitchen, both overall and in
rural children 13-24 months of age. Overall, 11.34% of household had a refrigerator; however,
few rural households had one (between 7.5% and 8.5%), while almost 29% of urban
households had a refrigerator. The bivariate analysis identified a positive relationship between
refrigerator presence and LAZ both overall and across all age groups and locations. Most
households overall used earth or sand as the material for flooring (78.95%); however, ceramic
tile and other flooring were positively associated with LAZ scores overall and in older rural and
urban children.
The bivariate linear regression results for WASH factors are reported in Table 5.4a. A
majority of children included in the survey lived in households with an improved source of
water (95.83%). These numbers were similar between urban and rural households. Most
47
households had their water source located outside the dwelling. Less than 2% of rural
households had a water source in the dwelling, while 6.4-7.1% of urban households had their
water source in the dwelling. The remaining households had it in the yard or elsewhere. The
location of the water source was found to be significantly associated with LAZ, with source of
water located outside the house negatively associated overall, among all urban children and
among older rural children. Few households overall treated their water (5.46%), with more
urban households reporting treating water (15.0-15.57%) while very few rural household
treated water (3.53%-3.64%). Water treatment was found to be positively associated with LAZ
overall among older urban and rural children, in the bivariate analysis. Overall, 27.84% of
households used unimproved toilets facilities, which was negatively associated with LAZ
overall, among all rural children, and older urban children. Shared toilet facilities were reported
in 32.26% of households overall, with similar percentages reported in rural and urban
household. Sharing a toilet facility was also negatively associated with LAZ scores overall,
among all rural children and older urban children. Finally, the presence of a handwashing
Finally, water samples for Water Quality Survey were collected on a subset of households
(one household per 20 households per cluster), which includes 371 children under 25 months
of age included in the analysis. Results are presented in table 5.4b.A total of 61.8% of children
lived in a household with detectable E. coli in the household water sample, while 41.14%
children lived in households with detectable E. coli in the water source. Interestingly, no
statistically significant relationship was found between LAZ and presence of E.coli in the
48
Table 5.1: MICS Children Characteristics
Variable Overall Rural Urban
N=7851 N = 3433 N= 3162 N = 620 N = 636
1-24 months Estimate (SE) p-value 1-12 months Estimate (SE) p-value 13-24 months Estimate p-value 1-12 months Estimate p-value 13-24 months Estimate p-value
(SE) (SE) (SE)
Mean LAZ -1.45 ± 1.50 -0.05 (0.002) <0.001 -1.13 ± 1.55 -0.03 (0.007) <0.001 -1.86 ± 0.02 -0.06 (0.01) <0.001 -0.98 ± 0.06 -0.001 0.949 -1.56 ± 0.06 -0.03 (0.02) 0.054
(0.16)
Sex
Female 3875 (49.43%) ref 1711 (49.84%) ref 1512 (47.82%) ref 324 (52.26%) ref 328 (51.57%) ref
Male 3976 (50.64%) -0.13 (0.03) <0.001 1722 (50.16%) -0.12 (0.05) 0.028 1650 (52.18%) 0.16 (0.05) 0.001 296 (47.74%) -0.11 (0.12) 0.348 308 (48.43%) 0.13 (0.12) 0.285
Diarrhea 493 (6.28%) -0.07 (0.07) 0.342 173 (5.04%) -0.11 (0.12) 0.378 232 (7.34%) -0.01 (0.09 0.907 38 (6.13%) 0.33 (0.24) 0.164 50 (7.85%) 0.13 (0.22) 0.556
Currently 7394 (94.88%) 0.06 (0.08) 0.415 3342 (97.98%) 0.03 (0.19) 0.87 2899 (92.30%) -0.16 (0.09) 0.079 591 (96.57%) 0.001 (0.32) 0.997 562 (89.35%) -0.32 (0.19) 0.096
breastfeeding
49
Table 5.4a: MICS Water, Hygiene and Sanitation
Variable Overall Rural Urban
N=7851 N = 3433 N= 3162 N = 620 N = 636
1-24 months Estimate p-value 1-12 months Estimate p-value 13-24 months Estimate p-value 1-12 months Estimate p-value 13-24 months Estimate (SE) p-value
(SE) (SE) (SE) (SE)
Water source
Improved 7524 (95.83%) ref 3298 (96.07%) ref 3006 (95.07%) ref 603 (97.26%) ref 617 (97.01%) ref
Unimproved 327 (4.17%) -0.13 (0.08) 0.114 135 (3.93%) -0.04 (0.14) 0.743 156 (4.93%) -0.11 (0.11) 0.318 17 (2.74%) 0.14 (0.35) 0.684 19 (2.99%) -0.34 (0.35) 0.326
Water source location
In dwelling 111 (2.45%) ref 57 (1.67%) ref 55 (1.75%) ref 40 (7.09%) ref 36 (6.36%) ref
In yard/plot 5017 (65.26%) -0.42 (0.11) <0.001 2243 (65.74%) -0.098 (0.21) 0.637 2056 (65.35%) -0.35 (0.18) 0.057 354 (62.77%) -0.50 (0.24) 0.037 364 (64.31%) -0.65 (0.25) 0.011
Elsewhere 2483 (32.30%) -0.58 (0.11) <0.001 1112 (32.59%) -0.193 (0.21) 0.358 1035 (32.90%) -0.53 (0.19) 0.004 170 (30.14%) -0.75 (0.25) 0.003 166 (29.33%) -0.96 (0.27) <0.001
Water treatment
Treated 428 (5.46%) ref 121 (3.53%) ref 115 (3.64%) ref 93 (15.00%) ref 99 (15.57%) ref
Untreated 7418 (94.54%) -0.32 (0.75) <0.001 3308 (96.47%) -0.169 (0.14) 0.237 3046 (96.36%) -0.29 (0.13) 0.024 527 (85.00%) -0.15 (0.16) 0.344 537 (84.43%) -0.52 (0.16) 0.001
Toilet Facility
Improved 5665 (72.16%) ref 2434 (70.90%) ref 2225 (70.37%) ref 493 (79.52%) ref 513 (80.66%) ref
Unimproved 2186 (27.84%) -0.31 (0.04) <0.001 999 (29.10%) -0.233 (0.06) <0.001 937 (29.63%) -0.35 (0.05) <0.001 127 (20.48%) -0.22 (0.14) 0.119 123 (19.34%) -0.39 (0.15) 0.009
Shared toilet facility 2405 (32.26%) 0.21 (0.04) <0.001 1026 (31.50%) -0.166 (0.06) 0.004 973 (32.63%) -0.164 (0.05) 0.002 199 (33.00%) -0.076 (0.12) 0.543 207 (33.77%) -0.64 (0.12) <0.001
Total households sharing
facility
<10 2274 (97.14%) 986 (98.40%) ref 939 (98.32%) ref 168 (90.32%) ref 181 (91.41%) ref
>10 67 (2.86%) -0.42 (0.18) 0.017 16 (1.60%) -0.801 (0.37) 0.029 16 (1.68%) -0.56 (0.32) 0.084 18 (9.68%) -0.82 (0.35) 0.53 17 (66.23%) -0.32 (0.36) 0.372
Stool disposal
Improved 2456 (31.50%) ref 776 (22.80%) ref 1063 (33.74%) ref 268 (43.79%) ref 349 (55.22%) ref
Unimproved 5342 (68.50%) 0.30 (0.04) <0.001 2627 (77.20%) -0.370 (0.06) <0.001 2088 (66.26%) -0.25 (0.05) <0.001 344 (56.21%) -0.50 (0.11) <0.001 283 (44.78%) -0.81 (0.11) <0.001
Handwashing place
Not in dwelling/plot/yard 1487 (19.14%) ref 1646 (19.03%) ref 626 (20.03%) ref 106 (17.21%) ref 109 (17.25%) ref
Observed 6282 (80.86%) 0.17 (0.04) <0.001 2749 (80.97%) 0.129 (0.07) 0.055 2500 (79.97%) 0.135 (0.06) 0.025 510 (82.79%) 0.25 (0.15) 0.10 523 (82.75%) 0.27 (0.16) 0.079
50
Linear Mixed Effects Models
The overall final model of the linear mixed effects analysis contained five significant variables
associated with LAZ (Table 5.5). Having a mother who completed primary school (b=0.20
p=0.006), some secondary school (b=0.30, p=<0.001) and secondary school completed
(b=0.57, p=<0.001) were associated with improved LAZ scores, when compared to children
of women with no education. Sharing the toilet facility with other households was negatively
associated with LAZ (b=-0.098, p=0.016). Flooring material was also important, with earth and
sand flooring have a negative association (b=-0.20, p=<0.001). The presence of a refrigerator
Table 5.5: Linear mixed effects model results for all participating children
Variables Estimate (SE) p-value
51
Rural Children 1-12 Months
Among children 1-12 months of age living in rural households, three factors were associated
with LAZ (Table 5.6). Being male was negatively associated with LAZ (b=-0.14, p=0.011).
Having a mother with some secondary school (b=0.27, p=<0.001) or secondary school
completed (b=0.39, p=<0.001) was associated with improved LAZ scores, when compared to
children of women with no education. Poor stool disposal methods were negatively associated
Table 5.6: Linear mixed effects model results for children 1-12 months of age from rural households
Variables Estimate (SE) p-value
Six factors were associated with LAZ among children 13-24 months of age from rural
households, (Table 5.7). As with the younger rural children, being male was also negatively
associated with LAZ (b=-0.11, p=0.033). Maternal education was also important among older
rural children with completed primary school (b=0.23 p=0.006), some secondary school
increased LAZ scores. The presence of an animal in the household was positively associated
with LAZ (b=0.14, p=0.017). Flooring material was also important, with earth and sand flooring
have a negative association (b=-0.23, p=0.007). Finally, an unimproved toilet facility and not
having a refrigerator in the home were both negatively associated with LAZ (b=-0.121,
52
Table 5.7: Linear mixed effects model results for children 13-24 months of age from rural households
Variables Estimate (SE) p-value
There were no factors associated with LAZ in urban children 0-12 months of age (Table 5.8);
however, unimproved stool disposal methods were marginally associated. In addition, the
relationship of maternal education was positive, with increasing magnitude by education level.
Earth and other flooring material, and unimproved toilets both had a negative relationship.
Table 5.8: Linear mixed effects model results for children 1-12 months of age from rural households
Variables Estimate (SE) p-value
53
Urban Children 13-24 Months
Among children 13-24 months of age from urban households, two factors were associated
with LAZ (Table 5.9). Maternal education was again important among this age group living in
urban households. Having a mother secondary school completed (b=0.66, p=0.016) was
associated with improved LAZ scores, when compared to children of women with no
education. Poor child stool disposal methods were negatively associated with LAZ (b=-0.39,
p=0.009).
Table 5.9: Linear mixed effects model results for children 13-24 months of age
from urban households 13-24 months of age
Variables Estimate (SE) p-value
54
Figure 5.2 Summary of key overall characteristics and mixed effect analysis results
-1
-1
7851 CHILDREN INCLUDED IN ANALYSIS DIARRHEA
-1.5
HAZ
HAZ
-1.5
49.4%
-2
FEMALE MEAN LAZ
-2
-1.45
-2.5
94.9% NO DIARRHEA DIARRHEA
0 5 10 15
Age (months)
20 25 0 5 10 15
Age (Months)
20 25
-.5
-1
-1
EDUCATION
-1
-1
-1.5
-1.5
HAZ
HAZ
HAZ
HAZ
-1.5
-1.5
-2
-2
-2
-2
-2.5
-2.5
-2.5
-2.5
Urban Rural Refrigerator No refrigerator Improved Unimproved Toilet shared Toilet not shared
NO EDUCATION 20.89%
-1
EDUCATION
PRIMARY INCOMPLETE 13.76% MATERNAL STOOL DISPOSAL
EDUCATION
-2
55
Discussion
In this study, we examined early childhood linear growth among children 1 to 24 months of
age in Bangladesh in an effort to identify household-level risk factors associated with growth.
While it is well understood that the causes for growth impairment are multifactorial, we wanted
to explore the differences in select household factors by both age and location of the
residence. Our analysis identified an overall low mean LAZ in the children included in the
analysis, which was expected given the high prevalence of growth impairment in Bangladesh,
as is widely discussed in the literature (3). Also consistent with the literature is the negative
relationship between age and LAZ, with a pattern of decreasing LAZ as the children get older
(4). We also identified differences in the household-level factors associated with LAZ
This study highlights some notable differences in household and WASH characteristics
between urban and rural households across Bangladesh. Only approximately 8% of rural
households had a refrigerator, while approximately 28% of urban households had one present.
Most participating rural households had dirt as the flooring material, whereas approximately
half of urban households had a dirt floor. A higher proportion of urban homes treated their
water, had improved toilet facilities and improved toilet disposal methods compared to rural
households.
An important contribution made by the study is to identify that factors associated with LAZ
differ across the age groups and by location. Younger children in both rural and urban
households had fewer factors associated with LAZ than the older children in the same location.
This may be due to their lower exposure to the factors due to limited mobility in the first year
of life. Children in the second year of life interact with their environment more, as a result of
hypothesized that breastfeeding could also have a protective effect on children, through the
between breastfeeding and LAZ. Children 13 to 24 months of age living in rural households
56
had the highest number of factors associated with LAZ out of all four subgroups. Household
and WASH characteristics of rural households demonstrate poorer living conditions than
urban households, as described above. These living conditions along with the increase
environmental exposure of older and more mobile children may explain the high number of
WASH factors negatively associated with LAZ were unimproved stool disposal methods
among younger rural children and older urban children, and unimproved toilet facilities among
older rural children. Both the type of toilet facility and stool disposal methods were associated
with LAZ in the overall model. Poor WASH characteristics have been found to increase the
risk of enteropathogen infection, which can then have a negative effect on linear growth (14).
A recent large international study found that despite heterogeneity across the participating
sites, there was a consistency in the reduction of infection with improved toilet facilities (15).
A multicentre cohort study found that access to improved sanitation was associated with
Maternal education was an important factor overall and in three of the four groups, with
increased education being positively associated with LAZ when compared to no education.
There is a clear disparity between the education levels of urban versus rural women. A higher
number of rural women had no education, while more women from urban areas completed
secondary or higher education. The linkage between maternal education and child health has
been widely reported in the literature with children of more educated women less likely to be
malnourished and growth impaired (16). Literacy skills enhance a mother’s ability to both
recognize and seek treatment for illness in her children, and also more likely to be receptive
to modern treatments (16). There is also the socioeconomic benefit of education, with more
educated women being more likely to have employment and higher income, which have an
influence on child health and growth, due to increased access to resources (16). A study
examining linear growth in Bangladesh and Nepal describe the important role of maternal
education, and also the disparity that exists between rural and urban Bangladesh (17). This
disparity was found in this analysis as well as in other studies despite national initiatives to
57
improve school enrolment of women in Bangladesh in the past few decades, particularly rural
women, (17).
The male sex was negatively associated with LAZ in both younger and older rural groups. This
same association has been identified in other studies, which suggested that this may be due
to different feeding practices for male compared to female infants (18,19). Male infants may
be more likely to receive complementary feeding at an earlier age and therefore exclusively
breastfed for a shorter duration than females (19). Previous studies have found that such
differences may result from beliefs around the nutritional needs of male infants, as they were
Most of these characteristics have been collected in surveys such as MICS and the DHS
survey as durable assets and used to calculate an asset index, or wealth index, to determine
the socio-economic position of the households included in the survey (13). Impaired child
growth is often the result of poor living conditions and low socioeconomic status; however, we
also aimed to explored how these predictors of socio-economic status were associated with
childhood growth, such as through the transmission of infection. Enteric infections have a
negative effect on linear growth and is more commonly seen in households and communities
living in poor conditions, particularly poor sanitary conditions. WASH factors such as
untreated water, unimproved toilet facilities, and unimproved stool disposal methods, for
example, could serve as pathogen reservoirs and be part of the infection transmission
One of the limitations of the study was the cross-sectional nature of the data. As such, we can
determine heterogeneities across the age groups and by rural or urban location on child
growth, but it does not allow us to appropriately identify growth faltering in the children. Instead
we aimed to identify potential predictors of negative LAZ, which could then result in growth
faltering in children. The cross-sectional nature also prevents causal inferences from being
made. WASH variables included in this analysis were potentially associated with
enteropathogen infection; however, this dataset lacked the pathogen data required to explore
58
this further. However, one of the strengths of the study was the large sample size which
Conclusion
Despite global efforts to improve nutritional status in children, growth impairment continues to
be a global health challenge (20–22). Using the 2012-2013 MICS Bangladesh survey data,
we examined household-level factors associated with linear growth in children 1-24 months of
age. We specifically aimed to explore the rural-urban and age-specific differences in these
associated factors. The results from the analyses indicate that growth impairment continues
to be a public health concern in Bangladesh, and that child growth, as measured by LAZ, is
associated with several household-level factors. The study identified heterogeneity in the
factors that were associated with linear growth according to the child’s age and location and
residence, with children 13-24 months of age living in rural Bangladesh having the most
characteristics associated with LAZ. This highlights the importance of considering the age of
children and the location of the household when designing and implementing public health
REFERENCES
1. Shekar M, Dayton Eberwein J, Kakietek J. The costs of stunting in South Asia and the
from: http://www.globalnutritionreport.org/files/2014/11/gnr14_pn4g_09headey.pdf
3. Islam MM, Sanin KI, Mahfuz M, Ahmed AMS, Mondal D, Haque R, et al. Risk factors
2016;53(1):241–67.
5. Leroy JL, Ruel M, Habicht J-P, Frongillo EA. Linear Growth Deficit Continues to
Accumulate beyond the First 1000 Days in Low- and Middle-Income Countries: Global
Lancet. 2013;382(9890):427–51.
7. Danaei G, Andrews KG, Sudfeld CR, Fink G, McCoy DC, Peet E, et al. Risk Factors
10. Bangladesh Bureau of Statistics (BBS); Unicef. Multiple Indicator Cluster Survey
https://www.unicef.org/bangladesh/en/reports/multiple-indicator-cluster-survey-2012-
2013
11. World Health Organization. Progress on Drinking Water, Sanitation and Hygiene.
12. Unicef; Bangladesh Bureau of Statistics. Bangladesh MICS 2012-2013: Water quality
2012;41(3):871–86.
14. Dearden KA, Schott W, Crookston BT, Humphries DL, Penny ME, Behrman JR.
Children with access to improved sanitation but not improved water are at lower risk
60
Settings : Results From the MAL-ED Study. 2016;63:1171–9.
18. Wamani H, Åstrøm AN, Peterson S, Tumwine JK, Tylleskär T. Boys are more stunted
19. Bork KA, Diallo A. Boys Are More Stunted than Girls from Early Infancy to 3 Years of
20. Bhutta ZA, Das JK, Rizvi A, Gaffey MF, Walker N, Horton S, et al. Evidence-based
interventions for improvement of maternal and child nutrition: What can be done and
21. Black RE, Cousens S, Johnson HL, Lawn JE, Rudan I, Bassani DG, et al. Global,
regional, and national causes of child mortality in 2008: a systematic analysis. Lancet
[Internet]. 2010;375(9730):1969–87.
et al. Piecing together the stunting puzzle : a framework for attributable factors of child
61
CHAPTER 6 BANGLADESH 2014 DEMOGRAPHIC AND
HEALTH SURVEY ANALYSIS RESULTS
6.1 Preamble
The MICS analysis identified differences in the household factors associated with linear
growth according to age and geographic location of the household through a mixed effects
analysis. The results presented in this chapter continue to address the first research aim and
its objectives with the application of different statistical methods, using data collected from the
Bangladesh Demographic and Health Survey conducted in 2014. This analysis adds the
exploration of the relationships between the variables through the examination of direct and
indirect effects via mediation, while also exploring the moderating effects of other variables.
This analysis applied two statistical methods. A simple multiple linear regression model was
first applied and explored associations between household factors and LAZ overall and by
relationship between diarrhea and linear growth provided the opportunity to develop a
mediation analysis to explore whether the household factors had an effect on LAZ that was
children allowed for the examination of heterogeneity by age and location. The groupings of
children were the same as those used for the MICS analysis, with an additional analysis also
carried out from children 1-12 months and 13-24 months of age.
The mediation analysis presents the first exploration of direct and indirect effects of household
62
6.2 Methods
Data from the 2014 Bangladesh Demographic and Health Survey (DHS) were used for this
analysis. To focus on the critical period for child growth, the first two years of life, only children
24 months of age and younger were included in the analyses. A total of 3283 children were
less than 25 months of age in the 2014 survey. Of these children, 120 had died, 351 were non
dejure residents, and 179 had out of range or missing data. In addition, children with length
for age (LAZ) z-scores less than -5 or great than 5 were also excluded. The remaining 2633
120 died
351 non dejure
residents excluded
179 excluded
due to out of
range or
missing data
2633 children
included in the
analysis
Dependent Variable
The dependent variable in this analysis was height for age (LAZ) z-score. Length for age z-
scores were calculated using the World Health Organization (WHO) 2006 international growth
63
Independent Variables
This analysis investigated the associations between LAZ and various household-level factors.
Child characteristics included the occurrence of a diarrheal episode in the preceding two
weeks, taking medication for parasites, and breastfeeding. Maternal characteristics included
maternal age, height, and education level. Household characteristics explored included
unimproved), type of toilet facility (improved versus unimproved), whether the toilet facility was
shared, the number of household members, and the number of household members under
the age of five. Data collected on source of household water was categorized in accordance
with the WHO/UNICEF Joint Monitoring Program (JMP) classification. ‘Improved’ water
sources include piped water, boreholes, and protected springs and wells (1). Other water
sources, which included unprotected wells, unprotected springs, natural water sources, rain
water, cart with a small tank were categorized as ‘unimproved’ (1). Data collected on type of
toilet facility in the household was also categorized in accordance with the WHO/UNICEF JMP
classification. ‘Improved’ toilet facilities include a connection to a sewer system, septic tanks,
pour-flush toilets, ventilated pit latrines and pit latrines with a slab. Other toilet facilities are
The first stage of the analysis involved generating descriptive characteristics for the children
included in the analysis, the mothers of these children, and their households. The exploratory
stage involved investigating the independent bivariate associations between LAZ and each
independent variable using simple linear regression. In addition, simple logistic regression
analysis was undertaken to explore the association between diarrhea and selected risk
factors. To examine whether the age of the child or the location of the household has an effect
on the factors associated with LAZ the exploratory analysis was also conducted on age-
stratified data (0-12 months, 13-24 months) and by location (rural vs urban). Statistical
significance was set at <0.05, while marginal significance was at P=0.2 or less. Factors
determined to be significantly associated with LAZ in the exploratory phase were then included
64
in multiple linear regression models. Several models were generated: 1) An overall model that
included all children; 2) Children 1-12 months of age; 3) Children 13-24 months of age; 4)
Rural children; 5) Urban Children. Stata version 14.0 was used for these analyses.
Visualization of the data was achieved using the kernel-weighted local polynomial smoothing
algorithm in Stata.
Finally, using a structural equation modelling (SEM) approach, a moderation and mediation
analysis was conducted. This analysis explored whether the effect of household risk factors
according to the age of the child and the location of the household: 1) All participating children;
2) Children 1-12 months; 3) Children 13-24 months 4) Urban children 1-12 months; 5) Rural
children 13-24 months; 6) Urban children 1-12 months; 7) Rural children 13-24 months. Mplus
The Demographic and Health Surveys (DHS) received ethical approval in participating
by ORC Macro’s institutional review board. As this was a secondary analysis of anonymized
6.3 Results
Child Characteristics
Descriptive statistics and exploratory analysis results are presented in Table 6.1. The mean
age of children included in the analysis was 12.66 months, with an even distribution of children
across the various age categories. The average LAZ score of included children was -1.23,
which can also be described that, on average, this sample of children is 1.23 standard
deviations shorter for their age than the WHO reference population of healthy breastfed
65
children. A total of 783 (29.74%) children were stunted, with a mean LAZ score of -2.83 among
those children. Consistent with what is described in the literature, LAZ was inversely
associated with age, with scores decreasing with increasing age. LAZ scores were also lower
in rural children (-1.30 ± 1.40) when compared to urban children (-1.10 ± 1.49). A total of 170
children (6.46%) were reported as having diarrhea while 428 (16.6%) children received
medication for intestinal parasites within the past 6 months. Both diarrhea and use of
medication for parasites were negatively associated with LAZ. When exploring by age group
and by location diarrhea was significantly associated with LAZ in older children (13-24 months)
and in children from rural household. Most children (95.33%) were being breastfed at the time
of the survey, with high rates reported even in the highest age category – 22-24 months
Maternal Characteristics
Descriptive statistics and exploratory analysis results are presented in Table 6.2. Overall, the
mean of age of mothers in the analysis was 24.27 years, and their mean height was 152.35
cm. Maternal age was not associated with LAZ; however, increased maternal height was
found to be positively associated with child LAZ both overall and in the different groupings (by
age and location). A total of 362 (13.75%) women did not have any formal education, 719
(27.31%) had primary, 1254 (47.63%) had secondary, and 298 (11.32%) had higher
education. Overall, the relationship between maternal education and LAZ was positively
associated compared to no education. All education levels were positively associated with
LAZ in children 13-24 months of age and in children from rural households. As displayed in
Figure 6.2, children of mothers with the highest level of education also had the highest LAZ
scores across all age groups. LAZ differences become more evident after 6 months of age,
when the LAZ scores of children of mothers with no education are the lowest.
66
Household Characteristics
Descriptive statistics and exploratory analysis results are presented in Table 6.3. Most
participating households (68.06%) were located in rural areas. Most children lived in
households with an improved source of water (96.85%) and it was not found to be associated
with LAZ overall, but was significant in children 1-12 months and in children from rural
households. Unimproved toilet facilities (31.67%) were associated with lower LAZ scores
overall and in most of the groupings. Shared toilet facilities were also negatively associated
with LAZ. The number of children in the household and the total number of household
67
Table 6.1: DHS Children Characteristics
Variable Overall Age Categories Location
(N=2633) 1-12 Months (N = 1312) 13-24 Months (N= 1321) Rural (N = 1792) Urban (N = 841)
Estimate (SE) p-value Estimate (SE) p-value Estimate (SE) p-value Estimate p-value Estimate p-value
(SE) (SE)
Mean LAZ -1.23 ± 1.44 -0.77 ± 1.42 -0.110 <0.001 -1.71 ± 1.30 -0.063 <0.001 -1.30 ± 1.40 -0.08 <0.001 -1.10 ± 1.49 -0.07 <0.001
Sex
Female 1279 (48.48%) 0.07 (0.06) 0.240 635 (48.40%) 0.06 (0.08) 0.473 644 (48.75%) 0.08 (0.07) 0.251 874 (48.77) 0.05 (0.07) 0.437 405 (48.16%) 0.10 (0.10) 0.334
Diarrhea 170 (6.46%) -0.35 (0.11) 0.002 82 (6.25%) -0.18 (0.16) 0.278 88 (6.67%) -0.49 (0.14) 0.001 113 (6.31) -0.38 (0.14) 0.005 57 (6.78%) -0.30 (0.20) 0.146
Medication for parasites 428 (16.26%) -0.40 (0.08) <0.001 45 (3.43%) -0.04 (0.04) 0.842 383 (29.02%) 0.07 (0.08) 0.403 276 (15.42) -0.42 (0.09) <0.001 152 (18.12%) -0.39 (0.13) 0.003
Currently breastfeeding 2510 (95.33%) 0.14 (0.13) 0.289 1280 (97.56%) 0.19 (0.25) 0.446 1230 (93.11%) -0.20 (0.14) 0.152 1714 (95.65) 0.12 (0.16) 0.450 796 (94.65%) 0.20 (0.22) 0.384
68
6.3.2 Multiple Linear Regression Analysis
Variables identified as having a significant independent relationship with LAZ overall (0-24
months of age) were included in multiple linear regression models. To explore whether age
had an effect on the factors associated with linear growth, the children were then subdivided
into two groups: 0-12 months and 13-24 months. To explore the effect of location, separate
multiple linear regression models were generated for rural children and urban children.
All children
Results for all children are presented in Table 6.4. Occurrence of diarrhea in the past two
weeks (b=-0.33, p=0.003), rural location of the household (b=-0.12, p=0.035), shared toilet
facilities (b=-0.13, p=0.032), and the use of medication for parasites in the past six months
(b=-0.39, p=<0.001) were negatively associated with LAZ. While higher level maternal
education (Secondary b=0.32, p=<0.001; Higher b=0.72, p=<0.001) and increased maternal
height (b=-0.003, p=<0.001) were associated with improved LAZ. Unimproved toilet facilities
No formal education
Primary education 0.12 (-0.06, 0.30) 0.203
Secondary education 0.32 (0.15, 0.49) <0.001
Higher education 0.72 (0.49, 0.94) <0.001
No diarrhea
Diarrhea -0.33 (-0.56, -0.11) 0.003
No parasite meds
Parasite meds given -0.39 (-0.53, -0.23) <0.001
Urban location
Rural location -0.12 (-0.25, -0.009) 0.035
69
Children 1-12 Months
Results are presented in Table 6.5. Maternal factors were important predictors for LAZ in
younger children. The highest level of maternal education, when compared to the reference
group of no education, was associated with increased LAZ (b=0.54, p=0.001), as were
maternal age and height. Having an unimproved water source was marginally associated with
LAZ (b=-0.44, p=0.054). The type of toilet facility and whether it was shared did not have an
Urban location
Rural location -0.12 (-0.29, -0.04) 0.149
Results are presented in Table 6.6. Diarrhea occurrence in the past two weeks (b=-0.46,
p=0.001) and unimproved toilet facility (b=-0.22, p=0.005) were negatively associated with
LAZ, while maternal education (Secondary b=0.41, p=<0.001; Higher b=0.70, p=<0.001) and
70
Table 6.6 Multiple Regression Model - Children 13-24 Months of Age
No diarrhea
Diarrhea -0.46 (-0.72, -0.18) 0.001
Urban location
Rural location -0.10 (-0.25, 0.046) 0.179
Rural Children
Results are presented in Table 6.7. Diarrhea occurrence in the past two weeks (b=-0.34,
p=0.013) and the use of medication for parasites in the past 6 months (b=-0.37, p=<0.001)
were negatively associated with LAZ, while maternal education (Secondary b=0.40, p=<0.001;
Higher b=0.75, p=<0.001) had a positive effect on LAZ. Having an unimproved water source
and unimproved toilet facility was marginally negatively associated with LAZ (b=-0.34,
p=0.059)
Rural
Variable ß (95% CI) P value
No formal education
Primary education 0.25 (0.04, 0.46) 0.019
Secondary education 0.40 (0.20, 0.60) <0.001
Higher education 0.75 (0.46, 1.06) <0.001
No diarrhea
Diarrhea -0.34 (-0.61, -0.07) 0.013
No parasite meds
Parasite meds given -0.37 (-0.55, -0.19) <0.001
71
Urban Children
Results are presented in Table 6.8. The use of medication for parasites in the past 6 months
(b=-0.44, p=0.001) was negatively associated with LAZ, while higher maternal education
Urban
Variable ß (95% CI) P value
No formal education
Primary education -0.25 (-0.62, 0.12) 0.185
Secondary education 0.14 (-0.21, 0.48) 0.431
Higher education 0.58 (0.19, 0.97) 0.004
No diarrhea
Diarrhea -0.21 (-0.60, 0.19) 0.306
No parasite meds
Parasite meds given -0.44 (-0.70, -0.19) 0.001
72
Figure 6.2 Summary of key overall characteristics and multiple regression analysis results
CHILDREN CHARACTERISTICS (OVERALL) HAZ Score by Child Age HAZ Score by Diarrhea
1
2633 CHILDREN INCLUDED IN ANALYSIS
-.5
0
DIARRHEA PARASITE
MEDS
HAZ
-1
HAZ
48.5%
-1
-1.5
FEMALE MEAN LAZ
-2
-1.23
-2
95.3%
-3
NO DIARRHEA NO MEDS
0 5 10 15 20 25
Age (Months) 0 5 10 15 20 25
Age (months)
DIARRHEA
BREASTFED MEDS TAKEN 95% CI lpoly smooth
kernel = epanechnikov, degree = 0, bandwidth = 1.98, pwidth = 2.96
No Diarrhea Diarrhea
URBAN RURAL IMPROVED UNIMPROVED NOT SHARED SHARED MATERNAL MATERNAL MATERNAL
OVERALL
MATERNAL MATERNAL AGE TOILET FACILITY
EDUCATION HEIGHT EDUCATION
HEIGHT
HAZ by Location HAZ Scores by Toilet Facility HAZ scores by shared toilets
0
0
0
-.5
-.5
-.5
HAZ
-1
HAZ
HAZ
-1
-1
MATERNAL
-1.5
-1.5
-2
-2
0 5 10 15 20 25 0 5 10 15 20 25 0 5 10 15 20 25
Age (months) Age (months) Age (months)
Urban Rural Improved Toilet Facility Unimproved Toilet Facility Not sharing toilet Shared toilet
NO EDUCATION 13.75%
HAZ
-1.5
PRIMARY 27.31%
-2
47.63%
EDUCATION FACILITY
SECONDARY 0 5 10 15
Age (Months)
20 25
No Education Primary
73
6.3.3 Mediation Analysis
All Children
directly or via diarrhea was generated for all participating children (Figure 6.3). Maternal
education (b=0.306, p=<0.001) and maternal height (b=0.003, p=<0.001) were positively
associated with LAZ, while diarrhea (b=-0.319, p=0.005) and the use of parasite medication
in the past six months (b=-0.377, p=<0.001) were negatively associated. Shared toilet facilities
were not significantly associated with LAZ or diarrhea but the relationship with LAZ was
source (b=1.266, p=0.027) was positively associated with diarrhea, but was not associated
with LAZ directly. When testing for indirect effects of the household factors, there no significant
effects identified.
Diarrhea
-0.
3 19
(0.
0 05
4)
0.49 )
94 (
-0.0
Maternal
education
)
99 0.306
0.1 (0.000
6 3( )
Meds for 0.2
7)
parasites 02 -0.377 (0.000)
0.
6(
26
1.
Water 0.091 (0.757)
LAZ
source
)
0.003 (0.000
)
37
Maternal
5
(0.
height
26
.027)
1
(0
-0.154
-0.
Education
x toiletshr
74
Children 1-12 Months
directly or via diarrhea was then generated for all children 1-12 months of age (Figure 6.4). As
in the overall model, maternal education (b=0.236, p=0.001) and maternal height (b=0.003,
p=0.021) were positively associated with LAZ. Diarrhea (b=-0.149, p=0.3955) was not
significantly associated with LAZ. Shared toilet facilities were not significantly associated with
LAZ (b=0.214, p=0.224); however, they were marginally significantly associated with diarrhea
(b=-1.105, p=0.050) and the relationship with LAZ was significantly moderated by maternal
education (b=-0.233, p=0.025). When testing for indirect effects of the household factors, there
1
Figure 6.4 Mediation Model – Children 1-12 Months
Diarrhea
-0.
1 49
(0.
3 95
)
)
.2 37
5 8(0
2
Maternal -0.
0) 0.236(0.001)
education 05
0.
0 5(
.1
-1
Shared 0.214 (0.224)
LAZ
toilet
)
0.003 (0.021
)
Maternal
.0 67
height
(0
02
.025)
-0.233 (0
0.6
Education
x toiletshr
In the mediation-moderation model for children 13-24 months of age living in rural households
(Figure 6.5) maternal education (b=0.352, p=0.004) and height (b=0.058, p=<0.001) were the
only factors significantly associated with LAZ. As with the rural model, there were no significant
indirect effects.
75
1-
Diarrhea
0.4
34
(0.
1 83
)
)
59
0.2
2 0(
4
Maternal -0.
education 0.352(0.004)
A mediation-moderation model was generated for children 1-12 months of age living in rural
households (Figure 6.6). Maternal height (b=0.002, p=0.045) was the only factor significantly
Diarrhea
-0.
3 72
(0.
3 06
8)
0.002 (0.045)
Maternal
LAZ
height
In a mediation-moderation model for all participating children ages 13-24 months (Figure 6.7),
maternal education (b=0.343, p=<0.001) and maternal height (b=0.004, p=<0.001) were
positively associated with LAZ, while diarrhea (b=-0.459, p=0.001) was negatively associated.
An unimproved water source (b=1.803, p=0.022) was positively associated with diarrhea, but
76
was not associated with LAZ directly. Shared toilet facilities were not significantly associated
with LAZ (b=-0.017, p=0.905); however, they were significantly associated with diarrhea
(b=0.979, p=0.024). This relationship was significantly moderated by maternal education (b=-
Diarrhea
-0.
4 59
(0.
5) 0 01
0.70 )
70 (
-0.0
Maternal
education
)
24 0.343
0.0 (0.000
7 9( )
Shared 0.9
)
toilet .0 22 -0.017 (0.905)
(0
03
1.8
Water 0.432 (0.210)
LAZ
source
)
0.004 (0.000
)
14
Maternal
.0
height
(0
77
.392)
-0.072 (0
.6
-0
Education
x toiletshr
In a mediation-moderation model for all urban children ages 13-24 months (Figure 6.8),
maternal education (b=0.414, p=<0.001) and maternal height (b=0.057, p=<0.001) were
positively associated with LAZ, while diarrhea (b=-0.465, p=0.038) was negatively associated.
Shared toilet facilities were not significantly associated with LAZ or diarrhea; however, the
77
>
Figure 6.8 Mediation Model – Urban Children 13-24 Months
Diarrhea
-0.
4 65
(0.
0
85) 38
0 (0.6 )
0.13
Maternal
education
0.414
(0.000
)
Maternal
0.057 (0.000)
height
)
45
LAZ
.2
(0
36
.5
)
-0.378 (0.031
-0
Education
x toiletfac
In a mediation-moderation model for all rural children ages 13-24 months (Figure 6.9),
maternal education (b=0.222, p=0.001) and maternal height (b=0.003, p=0.001) were
positively associated with LAZ, while diarrhea (b=-0.474, p=0.004) and an unimproved toilet
facility (b=-0.362, p=0.023) were negatively associated. Shared toilet facilities were not
significantly associated with LAZ or diarrhea; however, the effect on diarrhea was significantly
p=0.003) was positively associated with diarrhea, but was not associated with LAZ directly;
however, a significant indirect effect on LAZ via diarrhea was identified. In addition, the effect
78
Figure 6.9 Mediation Model – Rural Children 13-24 Months
>1
Maternal
education
f2
Diarrhea
f1 -0.
4 74
8) (0.
Water 2.27 8 (0.00 0 04
)
source Maternal
education
e1
)
9 94 0.407
(0. (0.27
02 2 ) g1
0.0 g2
Maternal
education 5) 0.222 (0.001
12 )
0.
5 1(
.8
-0
Toilet -0.362 (0.023)
facility LAZ
)
0.003 (0.001
24
Maternal
.0
height
(0
73
.8
-0
.569)
Education 0.059 (0
x toiletshr
f1 * e1 = -1.080 (0.050)
e1 * (f1 + f2) = -0.341 (0.321)
f1 * (e1 + g2) = -1.864 (0.051)
79
Table 6.9 Mediation Model Results
0-12 13-24
Diarrhea
Maternal education -0.094 0.138 0.494 -0.258 0.218 0.237 -0.157 0.284 0.581 -0.42 0.372 0.259 0.07 0.184 0.705 0.002 0.234 0.994 0.13 0.32 0.685
Type of toilet facility 0.072 0.329 0.828 0.56 0.514 0.275 0.77 0.624 0.218 0.083 1.088 0.939 -0.349 0.44 0.428 -0.851 0.555 0.125 1.068 0.756 0.158
Shared toilet 0.098 0.327 0.764 -1.105 0.563 0.050 1.016 0.759 0.181 -1.103 0.932 0.237 0.979 0.434 0.024 0.922 0.544 0.09 0.899 0.801 0.262
Maternal education x shared toilet -0.126 0.204 0.537 0.602 0.328 0.067 0.451 0.457 0.324 0.810 0.508 0.111 -0.677 0.275 0.014 -0.873 0.388 0.024 -0.434 0.427 0.309
Breastfeeding -0.39 0.331 0.238 9.092 64.159 0.887 8.09 45.612 0.859 8.03 72.863 0.912 -0.674 0.346 0.052 -0.798 0.438 0.068 -0.592 0.575 0.303
Parasite medication 0.263 0.205 0.199 0.66 0.637 0.301 0.90 0.66 0.173 -7.702 85.056 0.928 0.168 0.234 0.471 0.366 0.298 0.22 -0.134 0.385 0.727
Maternal education x toilet type 0.037 0.207 0.859 -0.084 0.314 0.789 -0.08 0.377 0.832 -0.303 0.726 0.676 0.173 0.275 0.53 0.487 0.356 0.172 -0.536 0.461 0.245
Water source 1.266 0.574 0.027 0.768 0.867 0.376 0.927 1.073 0.387 0.512 1.689 0.762 1.803 0.786 0.022 2.278 0.852 0.008 -7.785 44.57 0.861
Maternal education x water source -0.579 0.43 0.178 0.123 0.54 0.819 0.064 0.678 0.925 0.068 1.045 0.948 -1.48 0.807 0.067 -1.559 0.822 0.058 0.331 NA NA
LAZ
Diarrhea -0.319 0.113 0.005 -0.149 0.175 0.395 -0.372 0.204 0.068 0.434 0.326 0.183 -0.459 0.133 0.001 -0.474 0.164 0.004 -0.465 0.224 0.038
Maternal education 0.306 0.047 0.000 0.236 0.069 0.001 0.111 0.086 0.195 0.352 0.123 0.004 0.343 0.057 0.00 0.222 0.068 0.001 0.414 0.109 0.00
Type of toilet facility -0.038 0.12 0.749 0.031 0.181 0.866 -0.233 0.212 0.273 0.577 0.351 0.10 -0.155 0.143 0.277 -0.362 0.159 0.023 0.35 0.33 0.288
Shared toilet 0.105 0.119 0.376 0.214 0.176 0.224 0.046 0.223 0.836 0.559 0.314 0.075 -0.017 0.143 0.905 -0.15 0.165 0.365 0.125 0.294 0.669
Maternal education x shared toilet -0.154 0.07 0.027 -0.233 0.104 0.025 -0.186 0.135 0.169 -0.314 0.171 0.066 -0.072 0.094 0.392 0.059 0.104 0.569 -0.212 0.151 0.162
-
Breastfeeding 0.121 0.929 0.353 0.34 0.267 0.202 0.294 0.323 0.362 0.363 0.459 0.428 -0.146 0.134 0.275 -0.195 1.199 0.23 -0.174 0.228 0.447
Parasite medication -0.377 0.074 0.000 0.261 0.269 0.332 0.334 0.304 0.273 0.17 0.543 0.754 0.048 0.074 0.514 0.045 0.089 0.61 0.015 0.131 0.907
Maternal education x toilet type -0.046 0.074 0.525 -0.003 0.111 0.982 0.135 0.131 0.304 -0.216 0.216 0.318 -0.042 0.086 0.623 0.127 0.101 0.207 -0.378 0.175 0.031
Maternal height 0.003 0.001 0.000 0.003 0.001 0.021 0.002 0.001 0.045 0.058 0.015 0.00 0.004 0.001 0.00 0.003 0.001 0.001 0.057 0.012 0.00
Water source 0.091 0.294 0.757 -0.239 0.451 0.596 -0.552 0.604 0.360 -0.163 0.695 0.815 0.432 0.345 0.21 0.407 0.371 0.272 0.001 0.867 0.999
Maternal education x water source -0.215 0.176 0.223 -0.042 0.265 0.875 0.053 0.358 0.881 -0.003 0.395 0.995 -0.352 0.211 0.095 -0.344 0.20 0.117 -0.013 0.664 0.985
80
6.4 Discussion
This study used another nationally representative survey to explore the household predictors
of linear growth in children 0-24 months of age in Bangladesh and determine whether they
were modified by the age of the children and the location of the household. Two different
statistical methodologies were applied. The multiple linear regression models identified the
different factors associated with LAZ by age groups and by location. The SEM analysis further
explored this heterogeneity with the added groupings by both age and location and exploring
whether the association between the factors and LAZ were mediated by diarrhea and
Consistent with the MICS analysis is the difference between characteristics of children from
rural versus urban locations. In children from urban households overall, mean LAZ score was
higher and the mothers had a higher level of education; however, WASH characteristics were
better in rural households. A higher proportion of households had improved toilet facilities and
water source, and few households shared toilet facilities. Also consistent is the finding that
more factors were associated with LAZ in children older than 12 months. A unique finding
was that an episode of diarrhea in the preceding two weeks and the use of medication for
parasites in the last six months were significantly and negatively associated with LAZ overall.
Diarrhea was associated with LAZ in children 13 months or older, both in urban and rural
households; however, it wasn’t significant in younger children. Again, this could be due to the
increased mobility of older children and thus greater exposure to enteric pathogens in their
environment. Children in this age group may be more exposed to pathogen reservoirs, such
Unimproved water sources were positively associated with diarrhea overall and in children 13
months and older in both rural and urban settings. The household water source may be a
pathogen reservoir responsible for causing diarrheal infection. It may not have been
associated in younger children due to a more limited exposure to water in the first year of life,
with older children expected to be consuming water and thus being exposed to enteric
pathogens.
81
Given the relationship between diarrheal infection and linear growth, the mediation-
moderation analysis examined the relationships between the household factor and LAZ and
whether they were mediated by a diarrheal episode in the preceding two weeks. Water source
had an indirect effect on LAZ through diarrhea although it did not have a direct effect. This
suggests that household water may be an important pathogen reservoir involved in diarrheal
Overall, several maternal factors, such as higher maternal education and increased height
were positively associated with LAZ. Maternal education has been identified as an important
risk factor for child growth (2). An analysis of DHS data across 56 countries found that the
positive association between maternal education and child LAZ increases with age, thus there
being a potential cumulative impact of education on child growth (3). Education was
associated with LAZ in several of the mediation models and moderated the indirect effect of
toilet sharing on LAZ. Maternal height has also been found to be associated with child LAZ,
with children of tall mothers being having higher stature (3). In this analysis it was found to be
Several limitations need to be considered in the interpretation of these results. As with the
MICS analysis, a major limitation of this analysis is the cross-sectional nature of the data,
which prevents causal inferences from being made. While growth faltering in children cannot
be properly identified, we can determine the differences that exist across age groups and
geographic locations and identify the potential predictors of linear growth faltering. Again, the
lack of pathogen data did not allow us to explore the relationship between enteric infection
and linear growth. The diarrhea variable was used as a proxy for enteric infection; however,
this variable is limited in that it only represents diarrheal episodes occurring in the preceding
two weeks of the survey. While the sample size of the dataset is not as large as the MICS
82
6.5 Conclusion
The results of these analyses further emphasize the need to consider the age of children and
location of the households in the development of public health programs. Using data from the
DHS survey conducted in Bangladesh, there was a consistency in the findings that household
characteristics differ between rural and urban settings and thus different factors may be further
contributing to growth faltering in children. Children of different ages are impacted by their
environment differently due to changing mobility and dietary intake that comes with age.
Diarrheal infection may be part of the pathway of the association between household factors
and growth, with water source indirectly affecting linear growth through diarrhea, thus
suggesting that enteric infection may have an important role in growth impairment. Maternal
education is an important factor across age groups and geographic location, further
suggesting the importance of education in preventing growth faltering in children. Public health
linear growth in children should consider the heterogeneity identified in these analyses and
develop tailored interventions instead of a ‘one size fits all’ approach. Further analyses
integrating pathogen data could help identify the contributing role of enteric infection in the
linear growth.
REFERENCES
1. Brown J, Cairncross S, Ensink JHJ. Water, sanitation, hygiene and enteric infections
2016;53(1):241–67.
83
RESEARCH AIM #2
The MICS and DHS analyses presented the important factors associated with linear growth
and how they differ by age and location. While understanding this heterogeneity provides
important information for addressing growth impairment, the absence of pathogen data does
not allow for the role of infection to be examined. Building upon the information generated
from the first research aim, the MAL-ED analyses explored the factors identified as important
for growth with the addition of enteropathogen data, using a longitudinal dataset. Conducting
an analysis using a longitudinal dataset provides a basis for inferring causality between risk
factors, enteric infections and linear growth, which is not possible in cross-sectional studies.
It also allows the opportunity to explore the importance of asymptomatic infection. This
method, to MAL-ED data in an effort to address the second research aim. LGCM has had
The subsequent chapters present the results of the LGCM analysis. Chapter 7 presents
methods of the LGCM analysis and the results of the exploratory analysis, which examined
the bivariate relationships between the selected household-level predictors, growth, and
infection. The selection of predictors for inclusion in the exploratory analysis were guided by
the results of the first research aim of the project. Chapter 8 is a manuscript submitted for
publication which presents the results of the Campylobacter pathogen latent growth curve
model. Chapter 9 presents the other pathogen-specific latent growth curve models.
84
CHAPTER 7 MAL-ED ANALYSIS METHODS AND
EXPLORATORY ANALYSIS RESULTS
7.1 Introduction
The ‘Etiology, Risk Factors, Interactions of Enteric Infections and Malnutrition, and the
Consequences for Child Health and Development Study’ (MAL-ED study) was an international
multi-centre longitudinal research project that examined the effect of enteric infection and a
variety of risk factors on child growth up from birth to 24 months of age. A total of 8 sites across
South Asia, Sub-Saharan Africa, and South America participated in the study (1).
The study involved extensive collection of child health data, including diarrheal and non-
diarrheal stool samples. Samples were screened for a wide variety of enteric pathogens to
As previously described, using the data collected from the children participating in the Dhaka,
Bangladesh MAL-ED site, advanced longitudinal statistical methods were used to address the
second aim of this thesis: To investigate the association between enteropathogen infections
and household-level predictors and linear growth in children in urban Bangladesh. This data
will allow the determination of enteropathogens associations with linear growth; the age
intervals where a specific enteropathogen infection had a negative effect on linear growth and;
the indirect and direct effects of household-level factors on enteropathogen infection and linear
growth
7.2 Methods
The data of 265 children enrolled in the Dhaka, Bangladesh MAL-ED study site were included
in the analyses. Newborn children (within 17 days of birth) were enrolled between 2009 and
2012 from the urban community of Mirpur (Figure 7.1, Figure 7.2), one of the 21 administrative
85
units of Dhaka. Inclusion criteria included: a birthweight or enrolment weight of greater than
1500g, a mother aged at least 16 years, and no other siblings enrolled (2). Further
The results of the MICS and DHS analysis in research aim #1 provided guidance in the
selection of factors to include in this exploratory analysis. Relevant data includes household
characteristics, child health status, maternal characteristics, pathogens isolated from collected
stool samples, and child anthropometry (Table 7.1). Stool samples were collected from
children every month for the first 12 months of age, and every 3 months between 12 and 24
months of age. Stool samples were also collected during diarrheal episodes. Care-givers were
interviewed every 6 months about WASH characteristics, such as water source, treatment of
drinking water, toilet facility type, whether toilets were shared, and handwashing behaviours.
Data was also collected on household assets, including the presence of a refrigerator and
animal ownership. Care-giver data included years of education, age, and anthropometric
Table 7.1 Data Collection Timeline for the MAL-ED Study (3)
86
Figure 7.1 Mirpur, Dhaka, Bangladesh (2)
87
7.2.3 Exploratory Analysis
latent growth curve models. Pathogens were individually examined for a significant or
marginal independent relationship with LAZ, while household factors were also examine for
an association with LAZ and pathogen infection. Statistical significance was set at <0.05,
while marginal significance was set at <0.2. Given their relationship with enteric infection,
duration of antibiotic use in children and duration of exclusive breastfeeding were included in
the latent growth curve models and not included in the exploratory analysis.
Linear regression was first used to explore the relationship between the most prevalent
pathogens identified at the Bangladesh MAL-ED site and LAZ. Data was analysed at 3-month
intervals (0, 3, 6, 9, 12, 15, 18, 21, 24) to maintain consistent spacing between data collection
points and to examine changing growth rates, A total of nine pathogens were included in the
coli (EPEC), Enterotoxigenic Escherichia coli (ETEC), Giardia lamblia, and Rotavirus.
The independent associations between household-level factors at baseline and LAZ were also
explored. These included the WASH factors previously described. A dichotomous water
treatment variable was constructed, with improved versus unimproved treatment methods.
improved, such as boiling, filtering, bleaching and chlorination. Letting water stand and settle,
and cloth straining were considered inadequate or unimproved methods. Water source was
not included in the analysis as all participating households had improved water sources, either
piped into dwelling or piped to yard/plot or pubic tap/standpipe (4). The presence of a
refrigerator was also included in the analysis, as well as ownership of an animal, which
included chickens, which have been identified as a pathogen reservoir, particularly for
88
Campylobacter (5). Child characteristics examined were the sex of the child and birth order,
which other studies have found to be associated with child growth (6). Maternal factors
included in the analysis were maternal education included as total years of formal education,
height (cm) and age (at child’s birth), which were all continuous variables.
Using logistic regression, we also explored the association between these household factors
and positive Campylobacter infection across all 3-month intervals. All exploratory bivariate
Latent growth curve modelling (LGCM), a form of longitudinal analysis that uses the structural
equation modelling (SEM) framework, allows for the modelling of change in an outcome
variable and identify the factors that predict the trajectory of change, as well as have time-
specific effects. This modelling framework was used to estimate children’s growth trajectory.
Change in the outcome variable, LAZ, is modelled through the estimation of a latent variable
for the interception, or the baseline/starting LAZ, and a latent variable for the slope, or the
trajectory of change of LAZ. These are estimated using the observed repeated measure of
Time invariant covariates can be included in the model as predictors of the intercept and the
trajectory of change. In this model, these were the covariates not expected to change within
the 24 months of participation of the children, which includes the maternal, household, and
WASH characteristics. Covariates found to have an independent association with LAZ in the
To explore the changes at the individual and population level, we added enteric infection as a
time-varying covariate, as infection was expected to vary across time intervals. Separate
models were developed for the enteropathogens associated with LAZ, which included each
enteric infection as a dichotomous variable. As we expect it would take a while for enteric
89
infection to have an effect on growth, we lagged these effects by a 3- and 6-month lag. The
lagged effects of enteric infection were tested by regressing infection at a preceding time
interval with LAZ at a subsequent interval. The mediation model was then developed to test
whether the pathogen reservoirs have a direct effect on the latent growth variable for childhood
growth or was mediated by the lagged latent growth variable for enteric infections. (Note:
of the latent growth curve models.) Mplus software was used for this analysis.
Enteropathogens
Six enteric pathogens were found to be significantly or marginally significantly associated with
LAZ in the Bangladesh study site (Table 7.3). These were: Campylobacter species,
Campylobacter species was the most prevalent, with prevalence increasing with age, from
18% children having positive stool samples at 3 months of age to 69% children at 24 months
Campylobacter 20 (8.00) 41 (17.75) 41 (17.75) 106 (49.30) 124 (61.39) 120 (59.11) 145 (73.23) 142 (72.08) 133 (68.91)
Cryptosporidium 16 (6.40) 2 (0.87) 3 (1.38) 5 (2.33) 7(3.47) 4 (1.97) 10 (5.05) 9 (4.47) 4 (2.07)
Astrovirus 6 (2.40) 4 (1.73) 10 (4.61) 7 (3.26) 6 (2.97) 2 (0.99) 7 (3.54) 7 (3.55) 11 (5.70)
EPEC 2 (0.8) 3 (1.30) 15 (6.91) 28 (13.02) 17 (8.42) 14 (6.90) 8 (4.04) 5 (2.54) 4 (2.07)
90
Figure 7.3 Enteropathogen prevalence graph
Household-level Factors
All 265 children enrolled at the Bangladesh study site were included in the exploratory bivariate
association was found between increasing age and LAZ, with LAZ decreasing with increasing
child age, which was also described in the MICS and DHS analyses as well as other research
(7). Visualization of the data also suggested this decline in LAZ over the study period.
Maternal height and education were found to be significantly and positively associated with
LAZ.
Data for other household characteristics were available for 242 children. WASH factors found
to be positively associated with LAZ were treatment of drinking water, improved toilet facility,
and not sharing a toilet. The presence of a refrigerator was found to be statistically and
positively associated with LAZ. Birth order was found to be marginally significant (p=>0.05),
animal.
91
A logistic regression analysis identified the household and maternal factors associated with
the various enteropathogen infections (Table 7.3). All factors had a significant or marginally
REFERENCES
2. Ahmed T, Mahfuz M, Islam M, Mondal D, Hossain I, Ahmed AMS, et al. The MAL-ED
206.
92
Settings : Results From the MAL-ED Study. 2016;63:1171–9.
6. Wells JCK, Hallal PC, Reichert FF, Dumith SC, Menezes AM, Victora CG.
Associations of birth order with early growth and adolescent height, body composition,
2011;174(9):1028–35.
2016;53(1):241–67.
93
CHAPTER 8 CAMPYLOBACTER ANALYSIS RESULTS
6\8
Campylobacter infection and household factors are associated with
childhood growth in urban Bangladesh: An analysis of the MAL-ED
study
J.Johanna Sanchez, Md. Ashraful Alam, Christopher B. Stride, Md. Ahshanul Haque,
Subhasish Das, Mustafa Mahfuz2, Daniel E. Roth5, Peter D. Sly1, Kurt Z. Long1,3*, and
Tahmeed Ahmed
Abstract
Background
The dual burden of enteric infection and childhood malnutrition continues to be a global health
concern and a leading cause of morbidity and death among children. Campylobacter infection,
We examined longitudinal data to evaluate the trajectories of change in child growth, and to
Methodology/Principal Findings
The study analysed data from 265 children participating in the MAL-ED Study in Mirpur,
Bangladesh. We applied latent growth curve modelling to evaluate the trajectories of change
in children’s height, as measured by length-for-age z-score (LAZ), from age 0-24 months.
lagged time-varying covariates, while household risk factors were included as time-invariant
covariates. Maternal height and birth order were positively associated with LAZ at birth. An
inverse association was found between increasing age and LAZ. Campylobacter infection
prevalence increased with age, with over 70% of children 18-24 months of age testing positive
for infection. In the final model, Campylobacter infection in the preceding 3-month interval was
negatively associated with LAZ at 12, 15, and 18 months of age; similarly, infection in the
preceding 6-month interval was negatively associated with LAZ at 15, 18, and 21 months of
age. Duration of antibiotic use and access to treated drinking water were negatively associated
94
with Campylobacter infection, with the strength of the latter effect increasing with children’s
age.
Conclusion
Campylobacter infection had a significant negative effect on child’s growth and this effect was
most powerful between 12 and 21 months. The treatment of drinking water and increased
antibiotic use have a positive indirect effect on linear child growth trajectory, acting via their
Introduction
low- and middle-income countries (LMICs). Linear growth in early childhood is an important
indicator of nutritional status and is associated with both short- and long-term negative
height/length for age is partly a consequence of fetal growth restriction, preterm birth, and
intergenerational effects. It is also the result of low socioeconomic status which leads to
inadequate dietary intake and household sanitation, water quality and poor hygienic
conditions; all of which contribute to greater exposure of children to infections and clinical
disease (2,5–7). Enteric infection has been recognized as an important cause of child growth
impairment since enteric infection rates are negatively associated with child growth,
particularly in LMICs (8). This may be due to decreased nutrient absorption, inflammatory
responses, reduced appetite, changing feeding practices, and the high prevalence of frequent
The ‘Etiology, Risk Factors, Interactions of Enteric Infections and Malnutrition, and the
Consequences for Child Health and Development Study’ (MAL-ED) was a multi-centre
longitudinal research project designed to examine the effect of enteric infection and risk factors
on child growth. The study screened stools collected from both diarrheal episodes and non-
diarrheal stools for a wide variety of pathogens to estimate the pathogen-specific burden of
95
diarrhea in children 0-24 months of age. Campylobacter was identified as one of the most
prevalent enteric pathogens among children who participated in the study and was negatively
MAL-ED study participants were negatively associated with linear growth, a finding consistent
with other studies that suggest that asymptomatic enteric infections may play a larger role in
growth faltering and malnutrition than has previously been appreciated (2,13,14). A cohort
study in Peru also identified a marginal association between both symptomatic and
asymptomatic Campylobacter infection and linear growth impairment in the 9 months following
among children who are malnourished (17). In a recent MAL-ED analysis Rouhani et al
reported that Campylobacter infection was associated with disruptions in the gut microbiota,
which may explain the effects of asymptomatic infection on linear growth (18).
Bangladesh has one of the highest burden of growth impairment in the world despite declines
in recent decades (19). Dhaka, Bangladesh’s capital city, has become one of the world’s most
densely populated urban areas, and urbanization continues to increase (12). Urban
in crowded housing provision with poor environmental conditions (19). This presents a public
health challenge, as the households and their children are more vulnerable to disease.
Studying the factors associated with enteric infection and childhood linear growth in such an
Previous studies have explored the drivers of growth impairment in children; including
household-level factors such as water, sanitation, and hygiene (WASH), child characteristics,
and maternal factors. However; few have linked these factors to both specific enteric pathogen
data and longitudinal growth trajectories. Latent growth curve modelling, a less commonly
used approach outside of the social sciences, allows the application of many predictors and
determines which exert important effects on the rate of change in growth, including both time-
96
varying and time-invariant predictors or covariates. In addition, this method allows the
opportunity to also explore the indirect effect of household factors on LAZ via infection. The
application of such a method could contribute additional detail to the growing body of
Therefore, this analysis aims to investigate, for children in the Dhaka, Bangladesh MAL-ED
study site: 1) whether these household-level factors and Campylobacter infection are
associated with changes in LAZ across the first 24 months of life; 2) whether the effect of
Campylobacter infection on changes in LAZ is consistent across the first 24 months of a child’s
age or whether children are more susceptible to this effect at certain ages; and, 3) whether
the effects of household-level factors on LAZ operated directly, or operated indirectly via their
Methods
Study Sample
A total of 265 children were enrolled within 17 days of birth between 2010 and 2012 from the
Bauniabadh area of Mirpur, one of the 21 administrative units of Dhaka. Inclusion criteria
included a birthweight or enrolment weight of greater than 1500g, a mother aged at least 16
years, singletons, and no other siblings being enrolled. Children with congenital or severe
Length and weight measures of children were taken every month following their enrollment,
using standard scales (Seca GmbH & Co. KG., Hamburg, Germany). LAZ scores (length-for-
age z-score), were then calculated by mapping the individual child length, standardized by
age and length/height, onto the WHO reference population distribution (21). Non-diarrheal
stool samples were collected from children every month for the first 12 months of age, and
every 3 months from children between 12 and 24 months of age. Diarrheal stool samples were
collected during a diarrheal episode or up to one day after. Diarrhea was defined as the
97
maternal report of three or more loose stools in 24 hours (12). Our analysis included the
enteropathogen data collected from the non-diarrheal stool samples. Laboratory analyses of
manufacturers procedures (12,22). The ELISA method was the preferred analysis method as
it was found to have higher sensitivity than culture (23). Campylobacter species was the most
prevalent enteropathogen in the MAL-ED Bangladesh site, with prevalence increasing with
age, from 18% children having positive stool samples at 3 months of age to 69% children at
Household sanitation, and hygiene (WASH) characteristics were collected from caregiver
interviews at study baseline. These included water source, treatment of drinking water, toilet
facility type, whether toilets were shared, handwashing behaviours, and the presence of a
refrigerator. Maternal data, such as years of education, age, and anthropometric measures,
including height, were also collected at study baseline. Antibiotic duration was included in the
models as total duration of use throughout the 24 months. The breastfeeding variable was
dichotomized as exclusive breastfeeding yes/no in the first 3 months of life and in the first 6
months of life as none of the participating children were exclusively breastfed beyond the 6
months.
Statistical Analysis
Latent growth curve modelling (LGCM) was used to assess the overall shape of growth in
LAZ, variation in this pattern between the children studied, and antecedents of that variation.
LGCM, a form of longitudinal analysis that utilizes a structural equation modelling (SEM)
framework, models change in an outcome through estimating latent variables for the intercept
(starting point) and slope (trajectory) of change, using the observed values of the variable
across time as indicators of these underlying ‘true’ latent variables. Potential predictors of the
intercept and trajectory of change can then be included in the model, both as subject-level
98
predictors (i.e., time invariant constructs) and time-specific effects. (24)
Given the exploratory nature of part of our analyses - we did not have a clear hypothesis
regarding the shape of the change in LAZ and the temporal pattern of the relationship between
Campylobacter infection and LAZ - we split the data into random halves, generated through
the statistical software, with one half of the data used to build our model, and the other used
for testing the fit of the resulting model and the statistical significance of its parameters. This
was necessary to ensure robustness when evaluating our final model: a model built and then
tested on the same subjects would naturally be likely to achieve a higher level of fit than when
tested on any other random sample from the population to which we wish to generalize. The
longitudinal design of the study gained us the power required to conduct the analysis on halves
of the data.
Model building was initiated with the development of an ‘unconditional’ growth curve model,
which identified the trajectory of LAZ without the inclusion of any covariates. In this model LAZ
measurements at 3-month intervals (0, 3, 6, 9, 12, 15, 18, 21, 24 months) were used as
indicators of the intercept and slope latent variables, with the intercept mean describing the
population-average baseline LAZ, the (linear) slope mean representing the population
average rate of change of LAZ over time (per 3 month period), and intercept and slope
variances respectively describing how starting level and change in LAZ varied amongst our
variable representing quadratic change was added to the model. We used the model chi-
square statistics, Comparative Fit Index (CFI), Tucker-Lewis Index (TLI), Root Mean Square
Error Approximation (RMSEA) and the Standardized Root Mean Square Residual (SRMR)
Root to assess model fit (25). A CFI value of above 0.95, a TLI value above 0.95, an SRMR
value of below 0.08, and an RMSEA below 0.06 are all indicative of a good model fit (26).
Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) were also used to
Campylobacter infection, defined as a positive stool sample collected from children regardless
99
of clinical status, was then included as a time-varying covariate of LAZ. As such, our 3-monthly
observations of LAZ (as opposed to the latent variables representing intercept and slope for
presence/absence. Given that the effect of infection on linear growth is delayed by several
months, we lagged the effect of Campylobacter infection on LAZ by 3 and 6 month periods
(e.g. LAZ at 9 months was regressed upon Campylobacter infection at both 3 and 6 months;
LAZ at 12 months was regressed upon Campylobacter infection at 6 and 9 months, etc)
first fitting a model in which the 3-month lag and 6-month lag effects were both free to vary
across all time points; having examined the pattern of coefficients from this model, we then
fitted a series of simpler models based on these patterns, with fixings in place to model the
child’s development.
Finally, the subject-level demographic, family, and household factors collected at study
baseline were considered as potential predictors of the latent variables for intercept, slope and
quadratic change (i.e., having a direct effect upon LAZ), and of Campylobacter infection at
each time point (therefore enabling an indirect effect on LAZ via Campylobacter infection).
Specific household factors considered were exposure to treated drinking water, type of toilet
facility, shared toilet facilities, and presence of a refrigerator; demographic factors were sex of
the child, birth order, and maternal education. While they were not included in the exploratory
analysis phase, a decision was made to include duration of antibiotic use and presence of
exclusive breastfeeding in the analysis as time-varying covariates in the model given their
importance in infection and growth (8). We examined whether the effects of these
time or showed variation by first fitting a model in which they were free to vary, and then
checking whether there was loss in model fit when applying temporal fixings to these
relationships: specifically, whether they were each constant across time, or whether they
100
throughout the 24 months. Again, models were compared using the fit indices listed above.
The model emerging from this building process was then tested on the other half of the data,
to obtain a robust evaluation of model fit and the statistical significance of model parameters.
Mplus version 8.0 software was used for all LGCM modelling (29). Models were estimated
Ethics Statement
The MAL-ED study was approved in Bangladesh by the icddr,b review committee. Informed
consent was obtained from the parent or guardian of every participating child in the study.
The project was granted an Exemption to Ethics Review at the University of Queensland, as
a secondary data analysis. An ethics application was also submitted to the regional ethics
the Swiss Tropical and Public Health Institute, which granted a Declaration of No Objection.
Results
Characteristics of study participants included in the final model are show in Table 8.1.
Our exploratory (model building) analyses on one half of the data involved the development
of an unconditional model, which estimated the starting point and change in the outcome
variable LAZ without the inclusion of any covariates. The initial model included only the
intercept and slope as latent variables, with their means and variances freely estimated (AIC
= 1292.534, BIC = 1335.78, CFI = 0.918, TLI = 0.924, SRMR = 0.092, RMSEA = 0.162). To
investigate whether a curvilinear trajectory of LAZ offered a better fit to the data, a quadratic
term was added to the model, resulting in a substantial improvement in all model fit indices
(AIC = 1237.20, BIC = 1291.98, CFI = 0.95, TLI = 0.95, SRMR = 0.078, RMSEA = 0.128).
101
Table 8.1 Characteristics of study participants in the final model
Birth order
1 108 40.8
2 94 35.5
3+ 63 23.77
Mean SD
Duration of antibiotic use (mean days) 109.6 56.7
Duration of exclusive breastfeeding (mean days) 98.64 57.4
We then explored temporal variation in the relationship between Campylobacter infection and
LAZ. A model in which these regression coefficients were free to differ across time suggested
that the effect was strongest for children over 12 months old but diminished as the children
approached 24 months of age. Therefore, we compared the free model to one in which the
effects on LAZ at 12, 15, 18, and 21 months were fixed to be equal. This was done separately
for the 3- and 6-month lag Campylobacter infection to LAZ paths to account for potential
difference in time length effect. We then separately fixed the effects on LAZ at 3, 6, 9 and 24
months to be equal. This latter model (i.e. the one with two sets of fixed equal paths (12-21
Finally, we explored the direct and indirect effect of household factors. The latent variables for
the intercept, slope and quadratic term, and Campylobacter infection at each time point were
regressed on child, maternal and household characteristics. Maternal age was not found to
improve the model and therefore completely removed. Paths from toilet facility, antibiotic use,
102
and exclusive breastfeeding to LAZ intercept, slope and quadratic terms, were likewise
removed. The relationships with Campylobacter infection were fixed across all intervals as per
exploratory results, instead of being allowed to vary freely. LAZ at each time point was
regressed on the time-lagged Campylobacter infection variables, again with the equality
We then tested our proposed final LGCM on the other half of the data. This model explained
69% of the variance in LAZ observed scores, through both the growth curve structure and the
level predictors explained 17% of the variance in the intercept factor (i.e. initial level) of LAZ,
15% of the variance in the linear slope factor for change in LAZ, and 13% of the variance in
the quadratic slope factor for LAZ. The percentage of within-subject variance in LAZ uniquely
explained by Campylobacter was 7.1%. Model results are presented in Table 8.2 and Figure
8.1.
103
Table 8.2 Final LGCM results for LAZ change over time and association with Campylobacter
Time-fixed covariates
Campylobacter 1- Sex of child (1=male, 2=female) 0.055 0.189 0.770
24mos
Maternal education (years) -0.011 0.030 0.722
Water treated (y/n) -0.430 0.193 0.026
Improved toilet facility (y/n) 0.141 0.208 0.499
Shared toilet (y/n) 0.135 0.240 0.574
Antibiotic duration (days) -0.048 0.009 0.000
Time-Varying covariates
Campylobacter 3mos Exclusive breastfeeding - 3 months -0.647 0.602 0.283
(days)
Campylobacter 6mos Exclusive breastfeeding - 3 months (y/n) -0.294 0.500 0.557
Exclusive breastfeeding - 6 months (y/n) 1.607 0.860 0.062
Campylobacter 9mos Exclusive breastfeeding - 6 months (y/n) -0.402 0.843 0.634
104
Figure 8.1 Final Latent Growth Curve Model
Water
Antibiotics treatment
duration
6) -0
26)
-0.4
-0
-0.0 .4
-0.048 (0.000)
02
.04
48 6) 30
0.
0.0
0.02
30(
(0.0 ( (0
)
8(
00
00) 30( 30 .0
30(
0.0
-0.4 .4 26
0.0
0.02
-0 )
00
-0.4
8(
)
.04
-0.0
6)
26) 26) 48 ( -0.048
-0
-0 -0.098 -0
.1
-0.0
98 ( -0
.1
-0.0
98 (
.1 (0.0 15 0.00 15 0.00
15 01) 1)
(< (< (< 1)
0. 0. 0.
00 00 00
05 05 05
) ) )
i s q
)
25
(0.0
74 0.
04
0.2 1
(0
.0
37
)
i Intercept
s Slope
q Quadratic term
Time invariant covariate had statistically significant association with latent variable
Time varying covariate had statistically significant association with outcome variable
105
Campylobacter infection in the previous 3-month interval was negatively associated with LAZ
at 12, 15, and 18 months (b=-0.115, p=<0.001) (Fig. 1 and Fig. 2). Infection in the previous
6-month interval was negatively associated with LAZ at 15, 18, and 21 months, (b=-0.098,
Figure 8.2 Relationship between Campylobacter infection and LAZ across time intervals
Coefficient (b)
Of the demographic and household predictors, maternal height (b=0.042, p=0.033) was
unsurprisingly positively associated with the intercept factor (i.e. the initial level of LAZ). Birth
order was also significantly associated with initial LAZ (b=0.274, p=0.025). Increased duration
of antibiotic use was negatively associated with Campylobacter infection (b= -0.048,
p=<0.001). Water treatment was negatively associated with Campylobacter infection, with this
effect equally fixed across time (b=-0.430, p=0.026). Type of toilet facility, having a shared
facility, and the presence of a refrigerator were not associated with Campylobacter infection
Significant indirect effects, both operating via Campylobacter infection between the 12th and
21st months of the child’s life, were found between water treatment and LAZ (total indirect
effect for the 3 and 6 month Campylobacter infection-LAZ lags, indirect effect = 0.092, SE =
0.046, p = 0.047) and between antibiotic use and LAZ (total indirect effect for the 3 and 6
106
month Campylobacter infection-LAZ lags, indirect effect = 0.010, SE = 0.003, p = 0.001).
Indirect effects of water treatment and antibiotic use outside of the 12- to 21-month age range,
and of improved toilet facility and toilet sharing at any age, were not statistically significant.
Discussion
To our knowledge, this study describes the first application of LGCM to household-level risk
factors associated with enteric infection and childhood linear growth. The study population is
The decrease of LAZ across the 24 months is consistent with other study findings (4,19).
Previous studies have found that such decreases may be due to increased exposure to a poor
hygienic environment and pathogen reservoirs as the child develops and mobility increases
(31,32). In addition, children may experience potential nutritional challenges and exposure to
(33).
community. increasing throughout the first 24 months to reach a peak prevalence at 18 months
of age (Chapter 7, Table 7.2). Again, supporting prior research, we found that Campylobacter
infection has a negative association with linear growth; however, this study extends these
observations by identifying the age intervals in the first 24 months of life where infection is
most strongly associated with growth. Our analysis specifically identified that children between
12 and 21 months old are most likely to show a negative relationship between having a
Campylobacter episode 3 or 6 months previously and their age adjusted length (Fig. 2).
We explored the impact of Campylobacter infection on age intervals instead of total attained
height at the end of the study. This approach was applied to specifically examine the effect of
107
infection on short term growth faltering and to identify the period most critical for the
association between infection and growth impairment. A previous MAL-ED analysis across
the 8 study sites described the second year of life as being the most critical period for
Campylobacter-associated growth impairment. (8) Our findings provide further support that
the effect of exposure to potential Campylobacter pathogen reservoirs and resulting infection
on growth may have been limited in the first year of child’s life, given that we did not find an
association between infection and growth during this period. Interestingly, our analysis did not
find an association between preceding infection and LAZ in the latest time interval, specifically
at 24 months of age. This may suggest that the increasing and initial exposure to
Campylobacter has the most significant effect on the growth trajectory of children. Most
of age, with few new incidents of infection occurring after this period; however, further analyses
would need to confirm this. In addition, children may develop protective immunity following
early infection, which may explain the reduction in infection rates in addition to the high number
of asymptomatic cases (34). As pathogen data used for this analysis was collected from
asymptomatic stool samples, this analysis further highlights the importance of asymptomatic
Campylobacter infection on child growth. Additional analyses should also compare the effect
Treatment of drinking water and increased duration of antibiotic use were important negative
for Campylobacter infection, may therefore be indirectly impacting positively upon growth.
These findings have important implications for the development of public health interventions
designed to reduce childhood growth failure in urban communities. Approximately 2.2 million
people in Bangladesh, and 87,000 in the Mirpur study area, live in such communities and the
numbers are expected to grow, with increasing urbanization (19,20,35). All participating
households received their water from an improved piped water source as defined by World
Health Organization Joint Monitoring Programme (36); however, treatment of drinking water
was positively associated with LAZ. Interestingly, the WASH Benefits Bangladesh study did
108
not find a benefit in water treatment to the reduction of reported diarrhea. In addition, the
integration of WASH interventions to nutrition did not have a benefit to child LAZ. (37,38). This
36% reduction in diarrhea in children from the water chlorination plus safe storage arm when
compared to controls (38). Another study reported that children from households with better
conditions had higher LAZ scores compared with household with poorer hygienic conditions
(39). This highlights a potential challenge with the quality of the water provided to households
and so provides an opportunity for central infrastructure changes that could have an impact
on Campylobacter infection and linear growth. While household-based methods are cheaper,
centralized municipal treatment of piped water provide wide health benefits for the general
population, particularly in urban and densely populated areas (40). Our analysis does identify
important relationships between water treatment and Campylobacter infection, which needs
to be explored further.
Antibiotic use was prevalent in this study population. A previous study of the MAL-ED data
reported that the Bangladesh site had one of the highest rates of antibiotic use, second to the
Pakistan site (2). They also found that antibiotic use in the previous month was associated
with a reduced risk of detection of Campylobacter in the stool samples (2). While it was
about its long-term effects beyond the first 24 months of age, given the current global health
challenge of the increasing rate of antibiotic resistance and potential negative effects on the
Previous research has suggested the exposure to chickens in the home as a risk factor for
childhood impaired growth with LAZ; however, our analysis did not identify an association
between chicken ownership and LAZ (8). Interestingly, the Bangladesh site had one of the
lowest rates of reported chicken ownership of all participating MAL-ED sites, with few
households reporting the presence of chickens in the home (1.3%). However, reported
chicken and fowl ownership may not accurately reflect children’s exposure to such zoonosis
109
as Campylobacter since chickens from other households wander freely throughout the
Increased maternal education in previous research has also as being associated with
improved linear growth in young children (17,42,43). In our preliminary models, maternal
education was significantly associated with the rate of change of growth; however, after
adjusting for other covariates in our final model, the relationship was no longer present. An
analysis of the MAL-ED cohort did not find an association with breastfeeding practices and
growth in the first 24 months, and commented that duration of exclusive breastfeeding was
low among the participants (14). Similarly, we did not identify an association between
breastfeeding and LAZ and Campylobacter infection among the Bangladesh participants.
Research on the implication of birth order on child growth is inconsistent. Several studies have
reported that firstborns have lower birth weight but that they become significantly taller in later
childhood and in adulthood, while others reported that they remained shorter (44,45). In this
analysis, increased birth order was positively associated with baseline LAZ of the children;
however, it was not associated with the rate of change in LAZ. Maternal height was also
associated with the intercept, or the birth length, of the participating children. However, they
were not associated with the slope or rate of change of LAZ in the first 24 months of age.
Limitations
The small sample size limited our ability to include additional variables in this pathogen-
specific analysis, given our focus on exploring the various household predictors for growth
and Campylobacter infection. The relationship between co-infections and linear growth was
not explored in these specific models and should be considered in future analyses. Dietary
intake has widely been identified as having an effect on linear child growth and so future
statistical models should consider including data representing dietary intake in addition to
exclusive breastfeeding. Our team is currently developing such models. Also, our analysis
uses pathogen data collected from regularly-collected non-diarrheal samples; however, future
110
analyses should also consider the pathogen information collected from diarrheal samples.
Finally, the urban slum population may result in a generalizability limitation, with findings being
specific to a densely populated urban settlement. However, given rapid rate of urbanization
on the global scale, particularly in low- and middle-income countries, child health in urban
Conclusion
Linear growth impairment in children is a complex public health challenge resulting from a
variety of factors associated with poor socioeconomic conditions. At the same time,
particularly in the Bangladesh MAL-ED population. To our knowledge, this is the first
application of the latent growth curve modelling statistical method to explore Campylobacter
infection, childhood linear growth, and the association with household factors. The negative
association between Campylobacter infection and LAZ at 12, 15, 18 and 21 months of age
household-level model also highlights the need to reduce the prevalence of infection by
addressing the factors associated with it. While water sources in the households of
participating children at the Bangladesh site were considered to be “improved” piped sources,
treatment of drinking water was identified as being negatively associated with Campylobacter
infection, highlighting an issue with the quality of the water provided to households in the
region. There is a need for government to not only provide access to improved water sources,
but to ensure access to high quality drinking water is provided to all households, eliminating
Conflicts of interest
111
Acknowledgements
The authors would like to thank the staff and participants of the MAL-ED network for their
Sweden, and the UK for providing core/unrestricted support to icddr,b. This research protocol
(MAL-ED birth cohort study) was funded by University of Virginia (UVA) with support from
Fogarty International Centre (FIC) with overall support from the Bill & Melinda Gates
Author Contributions
Ahshanul Haque, Subhasish Das, Mustafa Mahfuz, Daniel E. Roth, Peter D. Sly, Kurt Z.
Data curation: Md. Ashraful Alam, Md. Ahshanul Haque, Subhasish Das, Mustafa Mahfuz,
Tahmeed Ahmed
Formal analysis: J.Johanna Sanchez, Md. Ashraful Alam, Christopher B. Stride, Md.
Ahshanul Haque, Subhasish Das, Mustafa Mahfuz, Kurt Z. Long, and Tahmeed Ahmed
Investigation: J.Johanna Sanchez, Md. Ashraful Alam, Christopher B. Stride, Md. Ahshanul
Haque, Subhasish Das, Mustafa Mahfuz, Kurt Z. Long, and Tahmeed Ahmed
Methodology: J.Johanna Sanchez, Md. Ashraful Alam, Christopher B. Stride, Md. Ahshanul
Haque, Subhasish Das, Mustafa Mahfuz, Daniel E. Roth, Peter D. Sly, Kurt Z. Long, and
Tahmeed Ahmed
Project administration: Md. Ashraful Alam, Md. Ahshanul Haque, Subhasish Das, Mustafa
112
Validation: J.Johanna Sanchez, Md. Ashraful Alam, Christopher B. Stride, Md. Ahshanul
Visualization: J.Johanna Sanchez, Md. Ashraful Alam, Christopher B. Stride, Md. Ahshanul
Writing – first draft: J.Johanna Sanchez, Md. Ashraful Alam, Ahshanul Haque, Subhasish
Das,
Writing – review & editing: J.Johanna Sanchez, Md. Ashraful Alam, Christopher B. Stride,
Md. Ahshanul Haque, Subhasish Das, Mustafa Mahfuz, Daniel E. Roth, Peter D. Sly, Kurt
REFERENCES
et al. Paediatrics and International Child Health Piecing together the stunting puzzle :
a framework for attributable factors of child stunting Piecing together the stunting
puzzle : a framework for attributable factors of child stunting. Paediatr Int Child Health
http://dx.doi.org/10.1080/20469047.2016.1230952
2. Rogawski ET, Liu J, Platts-Mills JA, Kabir F, Lertsethtakarn P, Siguas M, et al. Use of
results from the MAL-ED cohort study. Lancet Glob Heal. 2018;6(12):e1319–28.
2010;125(3):e473–80.
2016;53(1):241–67.
5. Budge S, Parker AH, Hutchings PT, Garbutt C. Environmental enteric dysfunction and
113
child stunting. Nutr Rev. 2019;77(4):240–53.
6. Danaei G, Andrews KG, Sudfeld CR, Fink G, McCoy DC, Peet E, et al. Risk Factors
7. Schwinger C, Fadnes LT, Shrestha SK, Shrestha PS, Chandyo RK, Shrestha B, et al.
Predicting Undernutrition at Age 2 Years with Early Attained Weight and Length
Compared with Weight and Length Velocity. J Pediatr [Internet]. 2016;1–7. Available
from: http://linkinghub.elsevier.com/retrieve/pii/S0022347616312410
9. Richard SA, Barrett LJ, Guerrant RL, Checkley W, Miller MA. Disease surveillance
methods used in the 8-site MAL-ED cohort study. Clin Infect Dis. 2014;59(Suppl
4):S220–4.
10. Guerrant RL, Leite AM, Pinkerton R, Medeiros PHQS, Cavalcante PA, DeBoer M, et
11. Kosek MN, Lee GO, Guerrant RL, Haque R, Kang G, Ahmed T, et al. Age and Sex
Enteropathy in Infancy and Early Childhood: Results from the MAL-ED Study. J
12. Platts-Mills JA, Babji S, Bodhidatta L, Gratz J, Haque R, Havt A, et al. Pathogen-
13. Richard SA, McCormick BJJ, Miller MA, Caulfield LE, Checkley W. Modeling
14. Caulfield LE. Relationship between growth and illness, enteropathogens and dietary
intakes in the first 2 years of life: findings from the MAL-ED birth cohort study. BMJ
114
Glob Heal [Internet]. 2017;2(4):e000370. Available from:
http://gh.bmj.com/lookup/doi/10.1136/bmjgh-2017-000370
15. Riddle MS, Guerry P. Status of vaccine research and development for Campylobacter
http://dx.doi.org/10.1016/j.vaccine.2016.02.080
16. Lee GO, Richard SA, Kang G, Houpt ER, Seidman JC, Pendergast LL, et al. A
17. Lee G, Pan W, Peñataro Yori P, Paredes Olortegui M, Tilley D, Gregory M, et al.
18. Rouhani S, Griffin NW, Yori PP, Olortegui MP, Siguas Salas M, Rengifo Trigoso D, et
al. Gut Microbiota Features Associated With Campylobacter Burden and Postnatal
Linear Growth Deficits in a Peruvian Birth Cohort. Clin Infect Dis. 2019;(Xx Xxxx):1–8.
19. Islam MM, Sanin KI, Mahfuz M, Ahmed AMS, Mondal D, Haque R, et al. Risk factors
20. Ahmed T, Mahfuz M, Islam M, Mondal D, Hossain I, Ahmed AMS, et al. The MAL-ED
21. Roth DE, Krishna A, Leung M, Shi J, Bassani DG, Barros AJD. Early childhood linear
condition: analysis of 179 Demographic and Health Surveys from 64 countries (1993–
22. Fahim SM, Das S, Gazi MA, Mahfuz M, Ahmed T. Association of intestinal pathogens
children in the first 2 years of life in Bangladesh. Trop Med Int Heal.
2018;23(11):1242–50.
23. Houpt E, Gratz J, Kosek M, Zaidi AKM, Qureshi S, Kang G, et al. Microbiologic
115
24. Park I, Schutz RW. An Introduction to Latent Growth Model. Res Q Exerc Sport
[Internet]. 2005;76(2):176–92. 9
25. Curran PJ, Obeidat K, Losardo D. Twelve Frequently Asked Questions About Growth
26. Felt JM, Depaoli S, Tiemensma J. Latent growth curve models for biomarkers of the
27. Checkley W, Epstein LD, Gilman RH, Cabrera L, Black RE. Effects of acute diarrhea
28. Checkley W, Gilman RH, Epstein LD, Suarez M, Diaz JF, Cabrera L, et al.
29. Muthen LK, Muthén BO. MPlus User’s Guide. Eighth. Los Angeles, CA: Muthen &
Muthen;
30. Akram R, Sultana M, Ali N, Sheikh N, Sarker AR. Prevalence and Determinants of
31. Perin J, Thomas A, Oldja L, Ahmed S, Parvin T, Bhuyian SI, et al. Geophagy Is
32. George CM, Oldja L, Biswas S, Perin J, Lee GO, Kosek M, et al. Geophagy is
33. Onyango AW, Borghi E, Onis M De, Casanovas C, Garza C. Complementary feeding
and attained linear growth among 6 – 23-month-old children. Public Health Nutr.
2013;17(9):1975–83.
34. Kaakoush NO, Castaño-Rodríguez N, Mitchell HM, Man SM. Global epidemiology of
35. Bangladesh Bureau of Statistics (BBS). Census of Slum Areas and Floating
116
36. World Health Organization. Progress on Drinking Water, Sanitation and Hygiene.
37. Luby SP, Rahman M, Arnold BF, Unicomb L, Ashraf S, Winch PJ, et al. Effects of
child growth in rural Bangladesh: a cluster randomised controlled trial. Lancet Glob
38. Ercumen A, Naser AM, Unicomb L, Arnold BF, Colford JM, Luby SP. Effects of
39. Lin A, Arnold BF, Afreen S, Goto R, Huda TMN, Haque R, et al. Household
40. Baker KK, Sow SO, Kotloff KL, Nataro JP, Farag TH, Tamboura B, et al. Quality of
piped and stored water in households with children under five years of age enrolled in
the Mali site of the Global Enteric Multi-Center Study (GEMS). Am J Trop Med Hyg.
2013;89(2):214–22.
42. Miller LC, Joshi N, Lohani M, Rogers B, Mahato S, Ghosh S, et al. Women’s
wealth, child diet, and child growth in rural Nepal. Int J Equity Health. 2017;16(1):1–
17.
44. Wells JCK, Hallal PC, Reichert FF, Dumith SC, Menezes AM, Victora CG.
Associations of birth order with early growth and adolescent height, body composition,
2011;174(9):1028–35.
117
45. Wells JCK, Devakumar D, Manandhar DS, Saville N, Chaube SS, Costello A, et al.
years of age: longitudinal cohort analysis from lowland Nepal. Eur J Clin Nutr
118
CHAPTER 9 EPEC AND GIARDIA RESULTS
9.1 Introduction
significant association with LAZ in the Bangladesh MAL-ED site (Table 7.2) in the exploratory
EPEC, Giardia lamblia, and Ascaris lumbricoides. Results of the Campylobacter analysis, the
most prevalent pathogen, are presented in Chapter 8. This chapter presents the results of the
EPEC and Giardia latent growth curve modelling analysis. Due to the low prevalence of
Cryptosporidium, Astrovirus, and Ascaris we did not have sufficient information to develop
EPEC has been described to be one of the leading causes of diarrheal disease in low-income
countries (1); however, there is limited literature on the relationship with linear growth in
children. A study by Santos et al (2) found a decrease in LAZ scores in children infected with
EPEC compared to those not infected. Kotloff et al (3) described the association between
EPEC and acute and persistent diarrhea, which leads to nutritional faltering. The Global
Enteric Multicentre Study (GEMS) identified EPEC as being significantly associated with
moderate to severe diarrhea in participating children but did not find an association between
diarrhea and Giardia, which was interestingly identified more frequently in control cases (3).
The MAL-ED study; however, found that, overall, Giardia was one of the most common
pathogens isolated in the diarrheal stools of participating children (4). Results of the
examination of the association between Giardia infection and linear growth are inconsistent.
Some studies have identified an effect on linear growth delay, while others have found no
association (5). A recent analysis of the MAL-ED data found that EPEC and Giardia infection
resulted in a tendency towards lower growth velocity; however, they did not find a consistent
and long term relationship with child growth, although this may have been due to the low
119
As with the Campylobacter analysis, this analysis aims to investigate, for children in the
Dhaka, Bangladesh MAL-ED study site: 1) whether specific household-level factors and EPEC
and Giardia infection are associated with changes in LAZ across the first 24 months of life; 2)
whether the effect of infection on changes in LAZ is consistent across the first 24 months of a
child’ age or whether children are more susceptible to this effect at certain ages; and, 3)
whether the effects of household-level factors on LAZ operated directly, or operated indirectly
9.2 Methods
As with the Campylobacter model we did not have a clear hypothesis regarding the temporal
pattern of the relationship between EPEC, Giardia and LAZ; therefore, the data was also split
into random halves, the first half to build a model for relationships between these variables,
and the second half to test the fit of the model and the statistical results of the parameters. As
is described in Chapter 8, this splitting of the data ensures robustness in the evaluation of the
final model.
EPEC and Giardia infection were defined as a positive stool sample regardless of clinical
the respective pathogen models. LAZ every 3 months was then regressed upon this infection
covariate. As we would also expect a lagged effect of infection on linear growth, we lagged
the effects by 3- and 6-month intervals, as with the Campylobacter models. The temporal
variation in the effect of each infection on LAZ were explored by first fitting a model that
allowed the 3- and 6-month lag effects to vary freely across time points. The coefficients were
then examined to determine whether there was a pattern that could be fitted in a simpler
model, for example, whether EPEC or Giardia had a stronger or weaker effect for particular
periods in the first 24 months, or even if the effects remained constant, as opposed to varying
each month.
120
Household factors determined to be independently associated with EPEC and Giardia (Table
7.3) were included in the model. To examine whether the effects of these demographic and
household variables on infection were consistent across time or showed variation, they were
first allowed to vary freely. Coefficients were then examined for patterns and the model was
then reviewed when temporal fixings were applied to the relationships, either constant across
time or to have an increasingly or decreasingly important effect on infection. Using fit indices
described in Chapter 8, we check whether there was a loss in model fit after these fixings were
applied.
The final model was then tested on the second half of the data. Mplus version 8.0 software
was used for all LGCM modelling (7). Models were estimated using Monte Carlo integration.
9.3 Results
EPEC
The final model included LAZ intervals starting at 6 months as there were very few children
with EPEC infection at 1 month (2; 0.8%) or 3 months of age (3; 1.30%: see Chapter 7, Table
7.2). The exploration of temporal variation in the relationship between EPEC infection and
LAZ, did not reveal a pattern in the regression coefficients when they were free to differ across
time. This suggested that there wasn’t a specific period where the effect of infection was the
strongest. As it was expected that there could be a difference in time length effect, the 3- and
6-month lag of EPEC infection to LAZ paths were separately fixed to be equal across time.
The model with two sets of fixed equal paths (3-month lag and 6-month lag) provided the best
fit.
The direct and indirect effects of household factors on LAZ were then explored. The latent
variables for the intercept, slope and quadratic term, and EPEC infection at each time point
were regressed on child, maternal, and household characteristics. Characteristics not found
to improve the model were completely removed. Birth order, maternal height, water treatment,
improved toilet facility and shared toilet facility remained in the model and were regressed on
the latent variables. EPEC infections across time points, starting at 6 months, were regressed
121
on sex of the child, maternal education, water treatment, shared toilet, antibiotic use, and
exclusive breastfeeding. The relationships with EPEC infection were fixed across all intervals
as per exploratory results, instead of being allowed to vary freely. LAZ at each time point was
regressed on the time-lagged EPEC infection variables, again with the equality restrictions
The proposed final LGCM (see figure 9.1) was then tested on the other half of the data. As in
the Campylobacter model, maternal height (b=0.052, p=0.005) was positively associated with
the intercept factor (the baseline value of LAZ). There were no other variables associated with
the latent variables. EPEC infection was not associated with LAZ at any of the time intervals.
MAL-
None of the child, maternal, and household characteristics were significantly associated with
i s q
122
Giardia
Giardia infection prevalence increased with age, with very few children presenting with Giardia
and 6 (2.79%) at 9 months (see Chapter 7, Table 7.2). Therefore, the final model explored the
impacts of infection on LAZ from 12 to 24 months of age and excluded LAZ intervals before
12 months. The exploration of temporal variation in the relationship between Giardia infection
and LAZ, did not identify a pattern in the regression coefficients when they were free to differ
across time. This suggested that there wasn’t a specific period where the effect of infection
was the strongest. As with the EPEC and Campylobacter models, the 3- and 6-month lags
The latent variables for the intercept, slope and quadratic term, and Giardia infection at each
time point were then regressed on child, maternal and household characteristics.
Characteristics not found to improve the model were completely removed. Birth order,
maternal education, maternal height, water treatment, shared toilet facility and presence of a
refrigerator remained in the model and were regressed on the latent variables. Giardia
infections across time points, starting at 12 months, were regressed on maternal education,
water treatment, shared toilet, shared toilet, and exclusive breastfeeding. The regression
coefficients of maternal education, water treatment, and shared toilet followed a pattern of
increasing value, thus the relationship between those household factors and Giardia were
fixed with increasing value to reflect that pattern instead of being allowed to vary freely or to
being fixed equally across time. LAZ at each time point was regressed on the time-lagged
Giardia infection variables, again with the equality restrictions across time suggested by our
The final proposed LGCM model was then tested on the other half of the data. Male sex
(b=0.340, p=0.049) and maternal height (b=0.050, p=0.003) were positively associated with
the intercept factor. Exclusive breastfeeding for 6 months was paradoxically negatively related
123
of age. Maternal education was positively associated with Giardia infection at 12 months of
age (b=0.139, p=0.030) and negatively associated in later time intervals – 18 months (b=-
MA
0.161, p=0.012) and 21 months (b=-0.311, p=<0.001). Gi
i s q
9.4 Discussion
This analysis applied a novel statistical method to explore the relationship between EPEC
infection and Giardia infection and linear growth (LAZ) using the longitudinal dataset from the
MAL-ED Bangladesh site. It also examined the household predictors of linear growth and
infection. EPEC and Giardia lamblia infection were among the most prevalent pathogens
identified in the MAL-ED study. EPEC was also marginally significantly associated with LAZ
124
EPEC infection increased during the first few months of life peaking in prevalence at 9 months
of age and then decreasing to a few cases by 24 months of age (Table 7.2). The low
prevalence in the first few months results in the exclusion of the data from the 1 to 6-month
time intervals. There was no association between infection and linear growth (LAZ) in the final
model. Maternal height was associated with the intercept, or the starting length, of the
participating children. However, it was not associated with the slope (i.e. rate of change) in
LAZ between 6 and 24 months of age. This association is consistent with the Campylobacter
analysis, DHS analysis, and other studies. Those subjects with short maternal height were
more likely to have had a smaller birth size and childhood stunting (8). This relationship may
be due to the physical constraints of a smaller frame during the fetal growth period limiting the
Few children had Giardia infection in the first 12 months. Prevalence increased throughout the
first 24 months, with the highest prevalence at 24 months of age. This analysis only included
data from the 12- to 24-month time intervals, due to the low prevalence in the first years. While
there was no association found between Giardia infection and LAZ, the analysis identified a
few predictors of the growth curve and infection. Again, maternal height was associated with
the starting length of the child, as was the sex of the child. Maternal education was negatively
associated with Giardia infection at 18 and 21 months, but interestingly was positively
associated with Giardia at 12 months. The relationship between breastfeeding and LAZ was
against enteropathogen infection, thus indirectly promoting healthy growth (9). In addition,
longer exclusive breastfeeding would be expected to provide the nutritional benefits required
for infants, and therefore directly promoting adequate linear growth (9,10).
An important limitation to consider in the interpretation of these results is the sample size. The
low prevalence of EPEC and Giardia made it harder to achieve a robust examination of the
relationship between infection and linear growth. Other limitations of this MAL-ED latent
125
9.5 Conclusions
income countries, including Bangladesh. The latent growth curve modelling methodology
allows for a flexible approach to exploring longitudinal growth and infection data. This method
has not been frequently applied to epidemiology research but, through the ease of estimating,
and then regressing latent variables for rate and shape of change upon potential covariates,
could provide additional insight into the temporal relationship between enteropathogen
infection and linear growth while also allowing the addition of direct and indirect predictors.
Studies examining the effect of Giardia infection on linear growth in children have provided
inconsistent results. While the exploratory bivariate analysis found a statistically significant
relationship between Giardia and LAZ, the final LGCM model did not present a significant
association. Similarly, there were no significant associations between EPEC and growth, or
any predictors for infection identified. Studies exploring the effect of EPEC on growth are very
limited. It is unclear whether these insignificant findings in both the EPEC and Giardia models
were due to a true lack of association or due to insufficient power to properly address this
question. Future studies are needed to continue to explore the role of EPEC and Giardia on
child growth faltering, as well as the predictors of infection. This information is critical for the
provides the opportunity to explore this; however, larger sample sizes are essential for a more
robust analysis.
REFERENCES
2004;53(11):1137–44.
2. Santos AKS, De Medeiros PHQS, Bona MD, Prata MMG, Amaral MSMG, Veras HN,
126
outcomes of enteropathogenic Escherichia coli infections in children from the Brazilian
3. Kotloff KL, Nataro JP, Blackwelder WC, Nasrin D, Farag TH, Panchalingam S, et al.
5. Donowitz JR, Alam M, Kabir M, Ma JZ, Nazib F, Platts-Mills JA, et al. A Prospective
Longitudinal Cohort to Investigate the Effects of Early Life Giardiasis on Growth and
6. Caulfield LE. Relationship between growth and illness, enteropathogens and dietary
intakes in the first 2 years of life: findings from the MAL-ED birth cohort study. BMJ
http://gh.bmj.com/lookup/doi/10.1136/bmjgh-2017-000370
7. Muthen LK, Muthén BO. MPlus User’s Guide. Eighth. Los Angeles, CA: Muthen &
Muthen;
8. Addo OY, Stein AD, Fall CH, Gigante DP, Guntupalli AM, Horta BL, et al. Maternal
10. WHO. Diarrhoeal disease. WHO Factsheet [Internet]. 2013;1–4. Available from:
http://www.who.int/mediacentre/factsheets/fs330/en/
127
CHAPTER 10 DISCUSSION AND CONCLUSIONS
10.1 Introduction
country of relatively small geographic size and one of the densest populations in the world has
made child health a priority in recent years. Investments in health and education have resulted
in marked improved in the health status and linear growth in children (2). As a result, stunting
rates have been on the decline for several decades, in contrast to the neighbouring countries
in South Asia. Despite these impressive improvements in child growth, Bangladesh continues
to be among the countries with the highest burden of growth impairment in children in the
world.
children in Bangladesh. The first aim presented an across country examination of the
household-level factors associated with childhood growth, with a particular focus on WASH
factors, given their association with enteric infection. As previously discussed, the burden of
infection in low- and middle- income countries remains high. Enteric infection in particular is
involved in a vicious cycle with undernutrition. The second aim of this research examined the
associations of most prevalent enteropathogens associated with growth in the MAL-ED study.
Studies often focus on the effect of diarrheal infection on growth; however, due to the
increasing evidence of the role of asymptomatic infection on growth, this research did not
distinguish between types of infection and instead considered all positive cases of infection.
Household-level predictors were incorporated into the models, to determine their effects on
128
10.2 Summary of findings
Research Aim #1
The first research aim investigated the household-level predictors for linear growth impairment
in Bangladesh among children 0-24 months of age. This was achieved through the analysis
of two national surveys: the 2012-2013 Multiple Indicator Cluster Survey and the 2014
Demographic and Health survey using different methodological approaches. In the MICS
analysis, a mixed effects analysis accounted for a potential household effect, as this dataset
included many households with more than one enrolled child. In the DHS analysis, a multiple
linear regression analysis found an association between diarrheal infection and LAZ. A
mediation analysis was then carried out to determine whether the effects of household
predictors of LAZ were both direct and indirect, mediated through the presence of diarrhea. In
addition, the models explored the moderating effects of factors on each other. Both survey
analyses were carried out overall and then by age interval and geographic location, to address
Both analyses identified a negative association between age of the child and LAZ. In addition,
they also identified differences in the household characteristics in the different groups with
LAZ. In the rural setting, a higher proportion of households had unimproved water source,
unimproved toilet facility, and a lower level of maternal education. The MICS analysis also
identified a higher proportion of unimproved stool disposal methods in rural households, and
a higher amount of untreated water. Breastfeeding was consistently high across the different
groups.
The MICS analysis identified differences in the factors associated with LAZ across four
subgroups defined by age and setting – rural children 0-12 months; rural children 13-24
months; urban children 0-12 months, and; urban children 13-24 months. Interestingly, there
were no factors associated with growth in younger (0-12 months) children from urban
households. Older children (13-24 months) from rural households had the highest number of
129
factors associated with growth. Maternal education was an important factor overall, and in
three of the four subgroups. Finally, boys from rural households had significantly lower LAZ
than girls.
In the DHS analysis, the same groupings as the MICS analysis were created in the mediation
analysis. As in MICS, older rural children had the most factors associated with LAZ. An
episode of diarrhea in the past two weeks was negatively associated with LAZ in children
overall, but not in younger children from both rural and urban households. Increased maternal
height and education were positively associated with LAZ in most groups. In older rural
children, toilet facility was associated with higher LAZ, while unimproved water source had no
direct effect. It was positively associated with diarrhea and therefore had an indirect effect on
LAZ.
In summary, the two analyses identified heterogeneity not only in the characteristics of the
children, mothers and households between urban and rural population, but also in the
predictors of linear growth. Additionally, heterogeneity was not only regional, but by age
groups, with older children having more predictors associated with growth. Both analyses
determined that more factors were associated with LAZ in older children from rural households
Research Aim #2
The second research aim investigated the associations between enteropathogen infection,
household-level predictors and linear growth in children in urban Bangladesh. This was
achieved through the analyses of the data from the Mirpur, Dhaka site of the longitudinal MAL-
ED study. Latent growth curve modelling, an advanced statistical modelling approach, was
applied to determine the slope of change of linear growth in participating children during the
first 24 months of life, and determine the predictors of this growth curve, including the positive
130
As in research aim #1, an inverse association between age and LAZ was identified.
Campylobacter was the most prevalent pathogen among participating children. Maternal
height and birth order were positively associated with the baseline length or birth length.
Campylobacter infection had a negative and lagged effect on LAZ. Infection in the preceding
3-month interval had a negative effect on growth between 12 and 18 months of age, while
infection in the preceding 6-month interval had a negative effect between 15 and 21 months
of age. Antibiotic use and the treatment of water were negatively associated with infection,
with the strength of the effect of drinking treated water increasing with age. In the EPEC model,
the only statistically significant association was between maternal height and birth LAZ. In the
Giardia model, maternal height and male sex were positively associated with birth length, and
maternal education was negatively associated with Giardia infection in older children. The low
prevalence of these two pathogens may have limited the strength of associations found in
these models.
The results of the Campylobacter analysis complement a mounting body of research on the
association between this pathogen and early childhood growth. In addition to a consistent
finding of increased effect of infection in the second year of life, the latent growth curve model
presents the age intervals most affected by a preceding episode of infection. In addition, it
highlights the importance of water quality issues by identifying a negative relationship between
treated water and infection that strengthens with age, and an indirect effect on growth.
The results of this research presents information for consideration in the design of health
Heterogeneity
Characteristics between urban and rural households were found to be different, as were the
predictors for linear growth. As previously described, rural households had poorer
131
children from rural households had the highest number of predictors of associated growth,
likely due to a combination of exposure to poorer environmental conditions and the increase
in mobility and access to their surroundings. The disparities between rural and urban
communities are not unique to Bangladesh and are well documented in the literature (3).
Access to basic services such as water and sanitation in rural communities in LMICs are
generally lower than in urban communities (3). Interventions may therefore achieve more
approaches.
Age of Children
Older children (13-24 months) may benefit more from environmental interventions, particularly
WASH interventions, given their increased interaction with their environment and resulting
higher prevalence of growth impairment in the second year of life. Fewer household predictors
were associated with growth in children during the first year of life. In addition, enteric infection
by several pathogens had the highest prevalence in older children. Campylobacter infection,
in particular, peaked in prevalence between 18 and 24 months of age and was negatively
Maternal Factors
association with growth identified in both survey analyses and in the MAL-ED exploratory
analysis. These findings are consistent with a large body of research (4–6). Maternal height
was also an important predictor of length in children. In the DHS analysis increased maternal
height was positively associated with length, while in the MAL-ED analysis it was associated
with birth length. This association suggests the need to address intergenerational effects of
growth impairment.
132
Enteric Infection Reduction
Campylobacter infection given its high prevalence in an urban sample and its negative
association with linear growth. The lagged approach of the analysis identified when positive
infection would have an effect on growth in the subsequent interval. As discussed above,
interventions and strategies may achieve more success or be more cost-effective by targeting
the second year of life for infection reduction, given the impact on growth and the prevalence
A number of WASH factors were important predictors for linear growth consistent with a large
body of research. Special consideration should be given to the quality of water provided to
indirectly impact growth and overall health. Household-based treatment of water was found to
water may have a wider impact in the urban population given most households have access
to piped water and not all households were treating their water.
As previously discussed; the recent WASH Benefits trial only demonstrated improved linear
growth in children in the nutrition arm with no improvements found among children in the
WASH intervention arms (7). These findings contrasted with a number studies suggesting that
improving WASH factors is associated with improved linear growth. While it is the intent of
most research and public health interventions to see an improvement in linear growth, it is
possible to positively impact the health and well-being of children without linear growth
improvement. For example, in a mass nutrition campaign in Bangladesh, there was marked
improvement in complementary feeding practices but not in linear growth, which can have a
positive long-term health impact on the children (1). Similarly, a breastfeeding and
supplementation program had a positive impact on health through a reduction in morbidity and
mortality, but it did not improve linear growth (1). The elimination of linear growth impairment
133
in populations is one component that contributes to the health of children (1). Improvements
in the area of WASH could have an impact on infection and other health indicators that may
This research explored linear growth through the relationship of predictors with linear growth,
as measured by LAZ, instead of stunting. Stunting is widely used in the literature to describe
growth impairment; and while there are benefits to comparing stunting prevalence between
populations, there are several limitations to its use as the primary measure for growth
impairment (1). Stunting is defined by a cut-off of a HAZ/LAZ of less than -2 SD, yet this is an
arbitrary cut-off without a biological basis (1,8). In addition, examining the number of stunted
since it has been demonstrated that in populations with a stunting prevalence greater than
2.5%, the entire distribution of LAZ is shifted (1). As discussed in Chapter 4, changes in
absolute height measurements were also considered for this project; however, a significant
limitation when exploring changes in height in given intervals is that smaller children gain less
This thesis includes analyses using three different data sources. Cross-country analyses
examined the heterogeneity of predictors of growth between rural and urban households and
by age group using two survey datasets. Similar data variables between the two datasets
allowed for a comparison between the results, while some differences provided added
information. The use of a longitudinal dataset containing pathogen data provided the
opportunity to explore the temporal relationships of infection and growth, using information
gained about important household factors associated with growth in the analyses of the survey
datasets. This study thereby provides both a general overview of the predictors of growth in
134
children in Bangladesh, how they are modified by location and age, and then specifically
Given the highly multifactorial nature of infant growth, there are limitations to regression
modelling. Latent growth curve modelling has had limited application in epidemiological
research may provide an approach that addresses this limitation. The models developed in
the MAL-ED analysis provide an alternative approach to identifying the time intervals most
affected by Campylobacter infection, while also identifying the predictors of the change in
growth either directly, or indirectly via infection. These results complement other recent studies
on Campylobacter infection.
10.5 Limitations
There are several limitations of this research project that need to be considered when
Enteric pathogen data was limited to the MAL-ED dataset, which represents an urban sample
of children from Dhaka. The national level survey data did not contain pathogen data, and
instead included proxies for enteric infection, such as recent diarrheal episode or use of
parasite medication in the past six months. The limited pathogen data did not allow for the
examination of the relationship between infection, growth, and associated household factors,
Cross-sectional Data
As previously discussed, the MICS and DHS analyses were limited by the cross-sectional
nature of the datasets. This prevents the proper identification of growth faltering in children or
135
allow causal inferences from being made; however, the large sample size allowed us to
The variables included in the models developed in the MICS and DHS analyses as part of the
first research aim were inconsistent due to different data collection points between the two
surveys. This did not allow for a direct comparison between the results of the two analyses;
however, they both contributed information that guided the development of models for the
MAL-ED dataset.
The small sample size of the Bangladesh MAL-ED site did not allow the development of
models for some of the pathogens associated with LAZ, given the low prevalence of infection.
Despite sample size limitations, interesting findings related to Campylobacter infection have
Single pathogen latent growth curve models were developed and did not account for co-
infection by multiple enteric pathogens. As discussed below in Section 10.6, future models
should account for multiple pathogens given the expected high prevalence.
Nutrition Data
information was not included. Given the important role of nutrition and linear growth, future
analyses of household-level factors may benefit from the inclusion of dietary information.
136
10.6 Future Research Direction
Future research in the environmental predictors of growth and infection should consist of or
predictors and linear growth. A larger sample size would provide the opportunity to
apply advanced statistical methods for the examination of many enteric pathogens.
The collection of enteric pathogen data with a broader sampling across several
communities within a country would capture urban and rural participants and provide
• Models Incorporating Nutrition Data: Given the important role of nutrition in linear
growth, latent growth models using nutrition covariates are currently under
development. Additional models will integrate the pathogen and nutrition models.
10.7 Conclusions
This study provides an understanding of the household predictors associated with linear
sectional national survey datasets and a longitudinal dataset containing pathogen information
to identify these predictors and also explore some of the interrelations. The survey data
analysis identified the associated factors overall and by subgroups using a large sample size
of children from across Bangladesh. The results highlight the necessity to recognize the
heterogeneity of factors associated with growth between rural and urban regions, and by the
age of the children, in Bangladesh. These identified differences provide opportunities for the
design of interventions that take this into consideration. The advanced statistical methods
137
applied to the MAL-ED urban Bangladesh data provided a new approach for exploring
Campylobacter and growth. The specific intervals affected where growth is affected were
identified and suggest the age children may benefit most from infection reducing interventions.
The identification of the treatment of water as an important factor for predicting infection and
growth, despite all households using an improved water source, highlights the need for
government to improve the quality of water, not just the water source.
multifactorial nature of this public health problem. The factors identified in this research
project, both at the national level and in the smaller urban sample, provide information that
can contribute to addressing this problem. The methodological approach applied to the MAL-
rarely been applied to epidemiological research. The application of a mediation analysis and
the latent growth curve modelling provide some insight into the relationships between the
factors and growth, through direct and indirect effects. An improved understanding of the
environmental conditions that contribute to growth impairment and their complex relationships
provides an opportunity to address some of the causes of linear growth impairment in targeted
REFERENCES
1. Leroy JL, Frongillo EA. Perspective: What Does Stunting Really Mean? A Critical
https://dhsprogram.com/pubs/pdf/PR104/PR104.pdf
3. Cumming O, Cairncross S. Can water, sanitation and hygiene help eliminate stunting?
138
Symptomatic and Asymptomatic Campylobacter Infections Associated with Reduced
wealth, child diet, and child growth in rural Nepal. Int J Equity Health. 2017;16(1):1–
17.
7. Pickering AJ, Null C, Winch PJ, Mangwadu G, Arnold BF, Prendergast AJ, et al. The
8. Perumal N, Bassani DG, Roth DE. Use and misuse of stunting as a measure of child
9. Voss LD, Mulligan J. Normal growth in the short normal prepubertal child: The
10. Voss LD, Wilkin TJ, Bailey BJR, Betts PR. The reliability of height and height velocity
in the assessment of growth (the Wessex Growth Study). Arch Dis Child.
1991;66(7):833–7.
139