You are on page 1of 34

Effect of Influenza Vaccine Administration: Comparative approach, Interrupted Times Series and

Machine Learning
Chapter 1: Introduction

1.1 Background of the study

Influenza (flu) is a highly contagious respiratory illness caused by influenza viruses that infect the
respiratory system, including the nose, throat, and lungs. Children, adults, and people with chronic
medical conditions are at higher risk of serious flu complications. It is primarily characterized by sudden
fever, cough, sore throat, headache, chills, runny or stuffy nose, body aches, and fatigue (CDC, 2022). It
can cause mild to severe illness and sometimes lead to death. Influenza is a leading cause of morbidity
and mortality worldwide in children under five years. Children under the age of 2 years are at more risk
of developing related severe flu-related complications. (CDC, 2020). There are four main types of
influenza (flu) viruses: Types A, B, C, and D. However, the influenza A and B viruses that routinely
spread in people (human influenza viruses) are responsible for yearly seasonal flu epidemics. Currently,
the best way to reduce the risk of flu infections and their potentially serious complications is by getting
vaccinated each year, as the vaccines are safe and reliable and have been used for more than 60 years, as
outlined by World Health Organization (WHO). Additionally, WHO recommends annual vaccination for
the following group of persons; children aged between 6 months to 5 years; pregnant women at any stage
of pregnancy; elderly individuals (aged more than 65 years); individuals with chronic medical conditions
and healthcare workers(Influenza (Seasonal), n.d.).

Annually, the Centre for Disease Control (CDC) conducts studies to determine how well the influenza
vaccine contributes to the protection against flu. Recent studies reveal that the influenza vaccine reduces
the risks of flu illness by 40.0% to 60.0% among all populations, especially if the flu in circulation is
well-matched with the vaccine administered (Vaccine Effectiveness, 2022). Between 2019 to 2020, the
last flu season of the pre-COVID-19 pandemic period, it was estimated that 7.5 million influenza
illnesses, 3.7 million influenza-associated medical visits, 105,000 influenza-associated hospitalizations,
and 6,300 influenza-associated deaths were prevented by flu vaccination in the United States of America.
A 2022 study to ascertain cause-specific mortality for neonates and children younger than five years from
2000 to 2019 all over the world found that lower respiratory illness contributed to 13.95% of the reported
child deaths. At the same time, vaccine-preventable deaths, such as meningitis, measles, and lower
respiratory illness, among others, were at 21.7% (Perin et al., 2022).
A 2017 Randomized Control Trial (RCT) study among pregnant women showed that influenza
vaccination during pregnancy decreased all-cause Acute Lower Respiratory Illness (ALRI)
hospitalization during the first 90 days of life, which suggested that there was possible protection against
future bacterial infections caused by influenza (Nunes et al., 2017). Three studies conducted between
2013 to 2020 in children under 18 years showed that influenza provided significant protection against
influenza-associated hospitalizations(de Hoog et al., 2019),(Brousseau et al., 2015),(Campbell et al.,
2021). One of the studies found that repeated immunization in children with pre-existing medical
conditions had no negative impact; instead, it could offer long-term protection(de Hoog et al., 2019).
Additionally, a 2021 study found that the live attenuated influenza vaccine (LAIV) and inactivated
influenza vaccine (IIV) was effective in reducing the episodes of influenza and influenza-like illness (ILI)
in children aged between 2 to 16 years. However, the estimated reduction for both types of vaccines
varied substantially among study populations, which made it challenging to predict the precise size
effects of the vaccine in other settings and for different flu seasons (Sambala et al., 2021).

1.1.1 Interrupted Time Series

An interrupted time series (ITS) design involves consistent data collection before and after an
interruption. It entails introducing and withdrawing an intervention or some part of it and observing any
changes in the outcome under assessment within a study population. ITS analysis utilizes statistical
methods to quantify changes in level (slope) and trend from before to after intervention and to assess the
significance of the observed differences. The primary assumption is that observations from the baseline
period predict where the future data points lie in the absence of the introduced intervention. The size of
the effect is expressed in terms of the level of change and slope change. The level change does not
provide all information found in the data, while the change of slope post-intervention period shows the
level of impact (Fretheim & Tomic, 2015). ITS design has increasingly been recommended as a more
robust design where randomization is not possible.

A 2013 study used ITS to evaluate healthcare quality improvements; it found that it was a simple but
powerful approach for program evaluation. Additionally, it highlighted that it was useful and less
expensive than comparable RCTs conducted to answer similar questions (Penfold & Zhang, 2013). A
similar 2017 study found that it allowed for a more detailed assessment of the longitudinal impact of an
intervention, unlike RCTs. What stood out was how the graphical and numerical presentation of the e
outputs allowed for easy understanding by th ose with little to no knowledge of epidemiological and
statistical methods (Bernal et al., 2017).
1.1.2 Machine Learning

Machine learning (ML) is a branch of computer science and intelligence whose main focus is using data
and algorithms to imitate the learning process of humans and gradually improve the accuracy of learning.
The way it works is in three facets; a decision process, an error function, and a model optimization
process.ML algorithms are used for two main processes; classification or prediction, which is based on
unlabeled or labeled data. The error function is used to evaluate the prediction of the model, especially
where there are known examples that are used for comparison and, eventually, the assessment of
accuracy. The model optimization is carried out in the following fashion; there are continuous weight
adjustments to reduce the discrepancy between known examples and the model to achieve an accuracy
threshold ("What Is Machine Learning (ML)?" 2020).

Recurrent neural networks are machine learning algorithms used for modeling time series data. A
generalization of a feed-forward neural network with internal memory is a recurrent neural network.
RNNs are a specific category of neural networks created specifically to handle sequential input. Text
documents, which may be viewed as a sequence of words; audio files, which can be viewed as a sequence
of sound frequencies over time; and time series (a list of values of some parameters over a specified
period of time) are examples of this type of data. Influenza outbreaks can be prevented, and public health
is protected by foreseeing new and urgent patterns in epidemiological data. The issue has been
extensively researched using both conventional autoregressive (AR) techniques and innovative deep
learning models like Recurrent Neural Networks (RNN). The majority of recent studies concentrate on
influenza's near-term prognosis. Transformer models recently outperformed RNN models in terms of
capturing long-range dependency (Li et al., 2021).

1.2 Statement of the Problem

The ministry of health of Kenya, Washington State University Global Health Kenya, CDC, and other
partners recently implemented influenza vaccination administration among children aged between 6-23
months as a demonstration project in Nakuru and Mombasa counties. The project was conducted in four
sub-counties in Nakuru and Mombasa, which presented diverse geographical and sociocultural
characteristics which were important in assessing vaccine uptake. The program ran from September 2019
to October 2021. The Ministry of Health implemented it through the National Vaccines and
Immunization Program and the Division of Disease Surveillance and Response in collaboration with
Mombasa and
Nakuru counties. In Njoro, Nakuru county, vaccine administration took place between November 26,
2019, to October 31, 2021. In Jomvu, Mombasa County, vaccine administration occurred between
December 16, 2019, to October 31, 2021, and the influenza vaccine was available throughout the year.
However, in Nakuru North, Nakuru county; vaccine administration took place between June 2, 2021, to
September 30, 2021) and in Likoni, Mombasa County; vaccine administration took place between July 9,
2021, October 31, 2021, the vaccine was available for only three months. The two approaches used for
implementation were to assess the coverage, costs, and impact of both the year-round and 3-month
vaccination strategies. Data from this demonstration project would help to inform policy decisions
regarding the design and possible future implementation of a national influenza vaccination program in
Kenya.

In Kenya, low-income communities face significant health challenges, particularly among young children
aged five years and below. One potential solution is the administration of influenza vaccines, which have
been shown to reduce the incidence and severity of influenza among young children. However, there is a
need to rigorously evaluate the impact of influenza vaccine administration in low-income communities in
Kenya to determine its effectiveness.

The aim of this study is to compare the effectiveness of two methods for evaluating the impact of
influenza vaccine administration in low-income communities in Kenya. The first method is an Interrupted
Time Series (ITS) design, a quasi-experimental design that uses data from before and after an
intervention to estimate its effect. The second method is a Recurrent Neural Network (RNN) design,
which is a machine learning method that has not been evaluated in the context of health research but has
shown promise in other applications for the interpretation of time series data.

The study will provide valuable insights into the effectiveness of these two methods for evaluating the
impact of influenza vaccine administration in low-income communities in Kenya. The findings will
inform the development of more effective methods for evaluating the impact of health interventions in
similar settings, which will ultimately contribute to improving the health outcomes of vulnerable
populations.

1.3 Study objectives

1.3.1 Overall objective

This study aims to compare the effectiveness of two methods for evaluating the impact of interventions in
healthcare settings: the ITS design, which is a sturdy quasi-experimental design preferred in a healthcare
setting, with RNN, which is an ML algorithm motivated by the fact that there has been a rise of the use of
ML algorithms to accurately predict outcomes in healthcare settings.
1.3.2 Specific objectives

This study aims to utilize two approaches:

i. To determine the impact of the influenza vaccine on the number of medically attended
respiratory illnesses among children under five years old in demonstration sites before, during,
and after the intervention period.
ii. To compare the number of medically attended respiratory illnesses among children under five in
the vaccine demonstration sites (case) with neighboring (control) sub-counties before, during,
and after the intervention period.
iii. To compare the performance of the machine learning (ML) approach with the traditional
Interrupted Time Series Analysis (ITSA) approach for impact evaluation by measuring their
respective Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE).

1.3.3 Research questions

i. Does the number of medically attended respiratory illnesses change post-intervention in the
vaccine demonstration sites?
ii. Does the number of medically attended respiratory illnesses change in the vaccine demonstration
sites compared to the control sites post-intervention?
iii. Which method best highlight the changes post-intervention?

1.4 Justification

Vaccine administration is currently the only recommended way to control influenza virus circulation
among the targeted population. Influenza plays a significant role in the cause of medically attended
respiratory illness, especially in children under five years, and poses a greater risk to children under the
age of 2 years (Dawa et al., 2018),(Emukule et al., 2015). There has been a recommendation from WHO
for countries to carry out year-round vaccination to control the circulation of the virus, but this has yet to
be implemented in Kenya (Recommendations for Influenza Vaccine Composition, n.d.). There were
verbal reports of a reduction in respiratory illness episodes among children who were immunized with the
Influenza vaccine. This study seeks to ascertain whether these reports about the decline of respiratory
illness among vaccinated children, as reported by parents and caregivers, are substantial. Overall, in low-
income countries like Kenya, the burden of influenza is significant, and interventions like influenza vaccine
administration have the potential to reduce its incidence and severity. However, there is a need to evaluate
the impact of such interventions in low-income communities to determine their effectiveness and inform
future policy decisions.

The proposed study aims to address this gap in knowledge by comparing two methods for evaluating the
impact of influenza vaccine administration in low-income communities in Kenya. The first method,
Interrupted Time Series (ITS) design, is a well-established quasi-experimental design that has been used
to evaluate the impact of interventions in health research. The second method, Recurrent Neural Network
(RNN) design, is a machine learning method that has shown promising results in other fields for
analyzing time series data. However, it has not been widely explored in the context of health research.

The proposed study is important for several reasons. First, it will provide valuable insights into the
effectiveness of these two methods for evaluating the impact of influenza vaccine administration in low-
income communities in Kenya. Second, it will contribute to the development of more effective methods
for evaluating the impact of health interventions in similar settings, which will ultimately contribute to
improving the health outcomes of vulnerable populations. Third, it will also contribute to the application
of machine learning methods to the evaluation of health interventions in low-income countries, which
could have broader implications for health research in these settings.

Chapter 2: Literature Review

2.0 Introduction

Public health interventions over the past decades involved the development and introduction of many
vaccines to support preventive care against certain illnesses. Vaccine developers utilize randomized
controlled studies to illustrate that their products meet the safety and efficacy standards required for
licensing. There exists an important need to conduct vaccine assessments to evaluate their performance in
real-world conditions, including effectiveness among population subgroups that were not involved in pre-
licensure trials. The evaluations reveal the impacts of the vaccine on healthcare outcomes excluded from
pre-licensure trials (Verani et al., 2017). The purpose of this dissertation is to establish the extent to
which the administration of influenza vaccine to children five years and below influenced the number of
medically attended respiratory illnesses. Case-control studies that are founded on recorded healthcare
data are essential for generating evidence-based insights about a vaccine's capability to protect recipients
from diseases. These evaluations are crucial for underpinning policy decisions on large-scale vaccine
administration and vaccine implementation program optimization and provide scientific evidence for
sustainable vaccine use and investments from donors and governments.

This chapter presents findings from peer-reviewed scholarly literature that provide a basis for this
dissertation. The review identified influenza vaccine effectiveness and influenza-related disease burden
as the major themes. The first section of the review presents general literature on efficacy, effectiveness,
and impact in relation to vaccines. The second section provides a review of literature relevant to the
themes and highlights the research gap identified in every source. The final section focuses on reviewing
scholarly literature that covers conceptual methods for assessing vaccine effectiveness and impact.

2.1 Effectiveness, Efficacy, and Impact

Terms like effectiveness, efficacy, and impact have distinctly different meanings in the context of vaccine
studies. Efficacy refers to the percentage by which the influenza vaccine reduces the target disease rate
among the vaccinated individuals compared to the rate of target disease prevalence among the
unvaccinated. The purpose of efficacy is to establish vaccine performance under optimal conditions.
Effectiveness measures the disease rate percent reduction in the context of real-world vaccine use.
Effectiveness may be the same as efficacy, but its use of the vaccinated population in real-world
conditions with multiple influences makes it differ in magnitude. On the other hand, impact refers to the
quantified disease reduction at the population level after vaccine introduction. The metric can be
expressed as the absolute change or percentage decline in the disease rate. The impact is determined by
the combined contribution of vaccine coverage, vaccine effectiveness, and herd effects, including
lowered transmission rates, which in turn lowers the disease risk in the population (Verani et al., 2017).

One of the pioneering studies on vaccine effectiveness (Clemens et al., 1996) showed that conventional
evaluation sequences for new vaccines targeting emerging economies involving the assessment of safety,
immunogenicity, and clinical protection in the first, second, and third phases are scientifically powerful
and rigorous. However, policymakers in developing countries targeted by vaccines need to get
information regarding vaccine suitability for local populations. The lack of insights about effectiveness,
efficacy, and impact results in indecision that is illustrated by ambivalence about the introduction of new
vaccines. Ambivalence about vaccines is mainly linked with the unknown effectiveness perspectives
whose biological outcomes are constrained to study populations considered for pre-license trials. It
evaluates vaccine effectiveness through studies that balance its beneficial and detrimental outcomes after
administration. The use of case-control studies to measure vaccine effectiveness is particularly applicable
to cases where the vaccine has been considered protective and safe in pre-license trials, but its cost-
benefit profile in public health practice is still unknown (Clemens et al., 1996).

2.2 Influenza Vaccine Effectiveness

The ability of influenza vaccines to protect subjects against respiratory illnesses varies between regions
and seasons. Vaccine protection can also vary depending on the individual targeted by the vaccination.
According to a 2016 study (Belongia et al., 2016), the two factors associated with the effectiveness of a
vaccine in protecting individuals from medically attended respiratory illnesses are the recipient’s
characteristics, including health status and age, and the vaccine’s match with the type of influenza virus
spreading illnesses in the target community. Vaccines that do not match the characteristics of the
circulating influenza viruses are likely to provide limited protection from respiratory illnesses. Influenza
vaccination provides significant healthcare benefits by preventing vulnerable populations from acquiring
flu infections when a perfect match exists between the circulating viruses and the vaccine. Although the
researchers provided seminal findings about influenza vaccine effectiveness by subtype, they did not
assess the vaccines’ impact on preventing critical healthcare outcomes like inpatient admission. The
proposed study will include evidence-based findings on the effects of the influenza vaccine on inpatient
and outpatient statistics at the study site. Additionally, it was recommended to focus on age-stratified
estimations of the influenza vaccine's effects on pediatrics, which is the focus of this study.

Influenza vaccination is the recommended strategy for reducing virus spread and medically
attended respiratory illnesses. However, opportunities exist to explore the effectiveness of the vaccine in
population subgroups, including children under five years. The effectiveness of the vaccine in children
varies, complicating the approaches for estimating its efficacy. Influenza vaccine administration requires
annual immunization, which could affect its effectiveness among the unvaccinated, newly vaccinated,
and those with prior immunization records in previous years. Children under five years rely on maternal
immunization for respiratory illnesses protection due to the lack of influenza vaccine for the age group.
Since the seemingly reliable adjuvanted, recombinant, and high-dose influenza vaccines have not been
approved universally for the pediatric population, an urgent need exists to evaluate the impacts of current
influenza vaccination (Mameli et al., 2019). The study will focus on a time series approach to evaluate
the number of medically attended respiratory illnesses to report on the relevance of influenza
immunization at the study sites.

Their recent study exploring the effectiveness of the influenza vaccine against hospitalization
(Mw et al., 2021) compared the odds ratio for the present-season influenza vaccination in the recorded
positive cases against the negative control groups. Influenza-related illnesses prevention can reduce
healthcare resource utilization, morbidity, and mortality among vulnerable populations. Populations with
developing or limited immune responses need preventive measures like immunization to reduce
hospitalizations. Even though the increased diversification of influenza viruses has resulted in many
variants in circulation, the Northern and Southern Hemisphere vaccine strains maintain a moderating
effect in preventing hospitalizations. Administering influenza vaccines can effectively reduce medically
attended respiratory illnesses in geographically diverse states (Mw et al., 2021). The researchers
interviewed participants to understand previous hospitalizations, vaccination status, influenza illness
symptoms, and immunosuppressive drug use, which were later used to evaluate the effectiveness of the
vaccine effectiveness. The study focused on the adult population and reported higher vaccine
effectiveness in preventing influenza-related hospitalizations between senior citizens and young adults.
The proposed study aims to provide additional evidence to report on influenza vaccine effectiveness in
children under five years. Besides, certified medical records rather than self-reporting will help the
investigator identify the vaccination status in children.

Influenza virus infection is a significant healthcare burden for the adult and pediatric populations.
Children under five years are more likely to contract and spread influenza-related respiratory illnesses in
households and communities. Globally, influenza surveillance networks implement case-control studies
to ascertain the effectiveness of influenza vaccines. The investigations involve enrolling patients with
influenza-like illnesses for clinical tests based on predetermined inclusion criteria. The patients who test
negative serve as controls, while the confirmed cases serve as intervention groups. Evidence-based
studies conducted between 2005 and 2019 estimated the overall influenza vaccine effectiveness against
pediatric hospitalizations to be between 54.78% (lowest) and 72.6% (highest). The study confirms the
relevance of full influenza vaccination as well as the significant benefits offered by the vaccine to
children younger than five years (Dawa et al., 2018).

With more than 70 percent of vaccinated persons receiving protection against influenza-related
respiratory illnesses in the 2019/20 seasons, vaccination remains a reliable clinical intervention. National
guidelines and recommendations by government agencies should be used to promote vaccination among
vulnerable populations. Given the reported variation in influenza viruses, national guidelines should
indicate the vaccine types and target populations and address potential biases to ensure desired outcomes
(Rose et al., 2020).

Although influenza vaccine effectiveness and efficacy assessment methods have changed over
time, there are challenges that limit their performance. The increasingly dynamic evolution of influenza
viruses creates the need for routine updates to vaccine strains. Besides, new vaccine products are being
developed to generate enhanced protection for vulnerable populations whose immune profiles continue to
change. The increased variation of influenza vaccine effectiveness across subtypes, population groups,
and regions creates an urgent need for scientific studies to generate evidence-based insights for public
health agencies and policymakers to understand vaccine protection (Emukule, n.d.).

2.3 Influenza-Related Disease Burden

Their recent study (Otieno et al., 2022) identified a significant healthcare burden of influenza-related
respiratory illnesses among young infants and pregnant women in Kenya. The study reported a lower
incidence of laboratory-confirmed influenza all through the post-partum period compared to a twofold
higher incidence during pregnancy. Additionally, the prevalence of influenza-related illnesses was more
among HIV-infected pregnant women compared to the uninfected ones. The findings focused on
maternal vaccination to alleviate the influenza-related illness burden on pregnant and postpartum women
and avert possible transmission to infants. Children under five years are eligible for influenza
vaccination, but infants younger than six months face a substantial burden of related respiratory illnesses
because they are not eligible for influenza vaccination.

The prevalence of influenza in infants younger than three months is considerably lower compared to the
incidence in children aged three-to-six months. These varying prevalence rates imply that antibody
transfer from expectant mothers protects infants in their early months. Also, pregnant women vaccination
using inactivated influenzas antidotes reduces the burden of influenzas-related illnesses among infants.
Findings from (Recommendations for Influenza Vaccine Composition, n.d.) illustrate the burden
associated with influenza-related respiratory illnesses in infants and pregnant women. However, the study
did not account for children aged six months to five years. On the other hand, the study did not consider
the direct impacts of influenza vaccination on infants but focused on maternal vaccination. The proposed
study will rely on findings from scholarly literature and empirical analysis to evaluate and report
influenza-related illnesses' burden on children younger than five years.

There is a possibility of 2.7—4.7 severe acute respiratory illness hospitalization per 1000 influenza cases
among children under five years. The hospitalization rate for children under five years is seven times
higher compared to individuals older than five years. The findings show the extent of the influenza
disease burden among younger children who are not eligible for vaccination and suggest the need for a
framework for recruiting expectant women for influenza vaccination. Although vaccine-related
complications predispose pregnant mothers to health-related risks, vaccinating them protects infants from
the burden of influenza-related respiratory illnesses. Influenza hospitalization rates are higher in
underserved populations with unique needs than in rural and urban populations with access to healthcare
services. Data from healthcare utilization surveys can provide insights into the disease burden associated
with medically attended influenza incidences in different regions. However, such surveys are limited to
healthcare settings and cannot capture the actual burden of influenza-related respiratory illnesses in
regions with healthcare disparities (Emukule, n.d.).

Influenza is a global and local healthcare threat. Even though Kenya has implemented an elaborate
epidemiological surveillance network for national influenza tracking and management, higher prevalence
rates still exist among children younger than five years. Influenza-related complications and illnesses
account for a considerable disease burden on mortality. Thus, vaccine effectiveness studies are vital for
prevention and ensuring influenza pandemic preparedness to safeguard the vulnerable population from
adverse health impacts (Dennis et al., 2013).

Clinical diagnoses for influenza patients with medically attended respiratory illnesses reveal the presence
of multiple viral agents with co-infection frequencies exceeding 20 percent. About 25% of medically
attended influenza cases in Kenya report underlying medical conditions. However, there is no mandatory
screening for chronic illnesses in Kenya, implying that many influenza patients with comorbidities are at
high risk. On the other hand, HIV patients are at risk for adverse influenza pandemics, which can
significantly reduce their quality of life. Vaccination is a crucial intervention for controlling the spread of
influenza and alleviating communities from related respiratory illnesses burden (Matheka et al., 2013).
The proposed study will report how influenza vaccination influences the number of related medically
attended respiratory illnesses among children under five years at high risk of developing severe influenza
illness.

The case-control studies are commonly used in public health research to assess the influence of factors
associated with outcomes of interventions or diseases. In a case-control study, individuals with the target
outcome of interest are considered as the first group of cases. The researcher constructed the second
group, identified as the controls. The control group is similar to the first group, but they need more
outcomes of interest. After identifying the case and control groups, the researcher ascertains the presence
of historical factors and reports any everyday exposure in the cases compared to the controls. The
researcher hypothesizes a potential link between the exposure and the outcome of interest if it is more
dominant in the cases (Tenny et al., 2022).

Case-control studies have become the standard design for evaluating the annual estimates of influenza
vaccine effectiveness. This researcher investigated the effectiveness of inactivated influenza vaccine in
children aged six months to 12 years during the 2015-2016 influenza season. The findings suggested that
the vaccine's effectiveness was statistically significant in preventing influenza-related illnesses and
hospitalization among the research subjects. Additionally, they reported a 47% influenza vaccine
effectiveness against type A influenza and a 34% vaccine effectiveness against influenza type B viruses.
The study used a larger sample and demonstrated the relevance of vaccine effectiveness in illness
prevention. Besides, the case-control study linked vaccine effectiveness with fewer medically attended
respiratory illnesses, according to a 2019 study (Shinjoh et al., 2019).

This case-control study was conducted to evaluate the effectiveness of influenza vaccination in
preventing influenza-related hospitalizations among children aged 6 to 59 months. The study used data
from the US Influenza Vaccine Effectiveness Network from the 2010-2011 through 2015-2016 influenza
seasons. The study found that influenza vaccination was associated with a 48% reduction in the risk of
influenza-associated hospitalization among children aged 6 to 59 months during the study period. This
finding was consistent across multiple influenza seasons and influenza vaccine types. The authors
concluded that influenza vaccination is an effective strategy for preventing influenza-related
hospitalizations in young children. The study supports current annual influenza vaccination
recommendations for all children six months and older (Thompson et al., 2019, pp. 2010–2016).

This case-control study evaluated the effectiveness of inactivated influenza vaccine (IIV) in preventing
influenza-related hospitalizations among children under five years old in Kenya during the 2014-2015
influenza season. The study used data from the Kenya Medical Research Institute-Wellcome Trust
Research Programme. The study found that IIV was associated with a 65% reduction in the risk of
influenza-related hospitalization among children under five during the 2014-2015 influenza season. This
finding was consistent across all age groups under five years old. The authors concluded that IIV is an
effective strategy for preventing influenza-related hospitalizations in young children in Kenya. The study
provides valuable information for public health policymakers and practitioners in Kenya and other low-
resource settings (Katz et al., 2016).

This case-control study was conducted to evaluate the effectiveness of influenza vaccination in
preventing laboratory-confirmed influenza among young children in the United States during the 2010-
2011 influenza season. The study used data from the US Influenza Vaccine Effectiveness Network. The
study found that influenza vaccination was associated with a 45% reduction in the risk of laboratory-
confirmed influenza among children aged 6 to 59 months during the 2010-2011 influenza season. This
finding was consistent across different subtypes of the influenza virus. The authors concluded that
influenza vaccination effectively prevents laboratory-confirmed influenza among young children in the
United States. The study supports current recommendations for annual influenza vaccination for all
children aged six months and older (Ferdinands et al., 2014, pp. 2010–2012).

2.4 Interrupted Time Series Analysis

An interrupted time series analysis uses pre-test and post-test observations to examine how the
outcome varies with time following an intervention. Time is the exogenous assignment variable. The
impact of the intervention can be evaluated by examining trend-level changes at different time points or
after the intervention. Also, this can be analyzed by examining the change in slope after the intervention.
Data variability, regression to the mean, and secular changes can be controlled in an interrupted time
series by including underlying trends. Interrupted time series models examine overall effects and can be
applied in cases with unaffected control groups (Lopez Bernal et al., 2019).

A 2007 study used an interrupted time series analysis to evaluate the decline in pneumonia admissions
after routine childhood immunization. The researchers used data from an inpatient database, where the
admission rates were specified as the outcome variable. The findings revealed a 39 percent decline in
admission rates at the beginning of the intervention and a 65 percent decline at the end of the eight-year
study. The research contributes to the growing body of evidence revealing the strength of interrupted time
series analysis in showing the impact of vaccination in lowering the number of medically attended
vaccine-preventable respiratory illnesses in children younger than two years (Grijalva et al., 2007).

A 2016 study used interrupted time series analysis to evaluate the impact of the universal childhood
influenza vaccination program in Israel on influenza-related hospitalizations and deaths. The study found
that the vaccination program significantly reduced influenza-related hospitalizations and deaths in
children under five years old during the study period. This interrupted time series study evaluated the
impact of Israel's universal childhood influenza vaccination program on influenza-related hospitalizations
and deaths in children under five. The program was introduced in the 2010-2011 influenza season and
offered free influenza vaccination to all children aged six months to 8 years old. The study found that the
vaccination program significantly reduced influenza-related hospitalizations and deaths in children under
five years old during the study period. Specifically, the program was associated with a 73% reduction in
influenza-related hospitalizations and a 65% reduction in influenza-related deaths in this age group. The
authors concluded that the universal childhood influenza vaccination program in Israel has effectively
reduced influenza-related hospitalizations and deaths in children under five years old. The study provides
valuable information for policymakers and public health practitioners seeking to implement or improve
influenza vaccination programs for children. Overall, this study provides strong evidence for the
effectiveness of universal childhood influenza vaccination programs in reducing influenza-related
hospitalizations and deaths in children. Interrupted time series analysis helps strengthen the study's
conclusions by accounting for potential confounding factors and examining trends over time(Segaloff et
al., 2019, pp. 2017–2018).

2.5 Machine Learning Algorithms

Alan Turing hypothesized the first machine that could learn and develop artificial intelligence in the
1950s, which is when machine learning first saw practical implementation (Turing 1950 Article - A. M.
Turing (1950) Computing Machinery and Intelligence. Mind 49, n.d.). This review aims to highlight the
benefits of ML algorithms for predicting the outcome of intervention in healthcare settings. Machine
learning has been employed in a variety of ways since its inception, including face detection for security
services (Wati & Abadianto, 2017), enhancing productivity, and lowering risk in public transportation
(Ellis et al., 2014),(Omrani, 2015), and more recently in several areas of healthcare and biotechnology
(Siddiqui et al., 2020),(Woldaregay et al., 2019),(Esteva et al., 2017),(Rajpurkar et al., 2017),(Robot‐
assisted Surgery for Benign Distal Ureteral Strictures: Step ‐by ‐step Technique Using the SP® Surgical
System - Kaouk - 2019 - BJU International - Wiley Online Library, n.d.),(Tang et al., 2017). As
healthcare expands into the modern realm of technology, numerous breakthroughs appear. The
advancement of the profession, including faster, more accurate, and more straightforward diagnosis,
depends on artificial intelligence- and machine learning-based methodologies and applications (Habehh
& Gohel, 2021).

A 2020 study conducted to predict hypotensive events in the ICU settings found that out of the 181,000
minutes of monitoring of 400 study participants, the algorithm used exhibited 94% accuracy, 96%
specificity, and 85% sensitivity in the prediction of hypertension well within 30 minutes of the incidents.
The study demonstrated the high potential of the ML model to predict hypotensive events based on a
short-term hypotensive physiological history of a patient (Moghadam et al., 2020). Another 2019 study
researched developing ML algorithms for predicting mortality in spinal epidural abscesses in hospitals
and 90-day post-discharge mortality. The study deployed five different ML algorithms; elastic-net
penalized logistic regression, random forest, neural network, support vector machine, and stochastic
gradient boosting, which were then assessed on the following metrics: decision curve, calibration, and
overall performance. The stochastic gradient boosting performed best across discrimination, calibration,
decision curve analysis, and c-statistic=0.89. The study concluded that the ML predictions showed some
potential for internal prediction of 90-day mortality, and recommendations were made for future studies
to validate the algorithms externally using independent populations (Karhade et al., 2019).

A retrospective study done in 2020 to assess the use of a real-time time phased ML model for the
prediction of sepsis in critical care found that the predictive sepsis model provided an intuitive way to
illustrate the impact of the input features, which made it more interpretable in clinical settings (Li et al.,
2020). A 2021 study leveraged the use of electronic medical records (EMR) to predict neutropenic events
in chemotherapy patients using ML algorithms. The final proposed model validated known risk factors
and offered insights on significant indicators and treatment profiles that elevated risk (Wiberg et al.,
2021).

A prediction of an illness's likely course and or outcome is referred to as a prognosis. It is a crucial job in
the clinical patient management process. The primary prognostic tasks for cancer forecasts are described
in this study (Cruz & Wishart, 2007). The general definition of the prognosis task includes the following
variations based on these predictive tasks: (Collen, 1986) disease susceptibility prediction (or predicting
the likelihood of developing any disease before the actual occurrence of the disease); (Hasman et al.,
2014), disease recurrence prediction (or predicting the likelihood of redeveloping the disease after it has
been resolved); and (Thakkar & Davis, 2006) prediction of survivability (or predicting an outcome after
the diagnosis of the disease in terms of life expectancy, survivability, disease progression, etc.). This has
led the researcher to explore ML algorithms to predict the changes in pupation based on the introduction
of an intervention.

RNN algorithms have been recently used in prediction; a 2019 study done to determine crop yield, the
study deployed a prediction model using convolutional neural networks (CNNs) and recurrent neural
networks (RNNs) for crop yield prediction based on environmental data and management practices. Three
distinguishing characteristics of the CNN-RNN make it a potentially helpful approach for additional
agricultural yield prediction investigations. Without knowing the genotype of the seeds, the CNN-RNN
model was created to reflect how environmental factors change over time and how seeds develop
genetically (PDF), TensorFlow: A System for (Large-Scale Machine Learning, n.d.). The model showed
that the yield prediction might be generalized to unproven contexts without a significant loss in forecast
accuracy. (Andrade et al., 1999). When used with the backpropagation technique, the model could show
how much the weather, the precision of weather predictions, soil conditions, and management techniques
may account for the difference in crop yields (Archontoulis et al., 2016).

Recurrent Neural Network (RNN), a time-series analysis model, has been widely employed for time-
series forecasting tasks in several disciplines since the emergence of deep learning techniques (Elman,
1991). Due to their adaptability in capturing nonlinear relations, RNNs have drawn some interest.
Traditional RNN, however, needs help capturing long-term dependencies and is plagued by the issue of
vanishing gradients (Bengio et al., 1994). The recurrent gated unit (GRU) (Cho et al., 2014) and its
version, the long-short-term memory (LSTM) (Gers et al., 2000), have recently been used for sequence
modeling, which bypasses the restriction and has had significant success in a variety of applications.

2.6 Summary

This section provided a review of scholarly literature aligning with the dissertation’s purpose of
evaluating the impacts of influenza vaccine administration on the number of medically attended
respiratory illnesses in children older than six months and younger than five years. Vaccine effectiveness
and respiratory disease burden emerged as the major themes in the literature. While the studies
considered for the review provided evidence-based findings relevant to this dissertation project, a study
has yet to be conducted at the four demonstration sites to assess the impact of influenza vaccine
administration on the number of related respiratory illnesses. Also, it emerged that computational
techniques could be applied alongside public health study designs like case-control evaluations to
ascertain the effectiveness of vaccines. Researchers like Grijalva et al. (2007) employed an interrupted
time series model to evaluate the effects of vaccination on the number of vaccine-preventable respiratory
illnesses. The use of ITS will enable the researcher to achieve the first two objectives in realizing
vaccination's impact in the sites described above.

A 2021 study conducted to predict pre-term birth using regression and ML demonstrated the
effectiveness of the algorithm in establishing new predictors that were not achievable by traditional
regression analysis. In their data set, machine learning found more variables related to pre-term birth than
logistic regression did. Machine learning revealed diabetes as the most significant predictor of pre-term
birth during the first trimester, whereas logistic regression indicated past abortions (which include
miscarriages) as the strongest predictor. Because the prevalence of prior abortions was significantly
greater than that of diabetes, a history of prior abortions (including miscarriages) may be a more
significant predictor of pre-term birth (Arabi Belaghi et al., 2021).

The studies described in this chapter under the use of healthcare settings theme demonstrate the potential
for machine learning algorithms to help healthcare systems target interventions more effectively by
identifying patients at high risk of adverse outcomes and tailoring interventions to their specific needs.
Using large and complex datasets, machine learning algorithms can help identify patterns and
relationships that may not be apparent using traditional statistical methods. This can lead to more
accurate predictions and more effective interventions in healthcare settings. This evidence of the
successful applications of ML algorithms has led the researchers to decide to use ML algorithms to
achieve the third objective. Although we plan to use RNN, it should be noted that overall. In contrast, the
use of RNNs for evaluating the impact of influenza vaccine administration may be an exciting avenue for
this research, it appears that more traditional statistical methods, such as interrupted time series analysis,
have been more commonly used in this area to date. The next chapter will present the proposed
methodology and design for the study. It will include a detailed discussion about the data collection
procedure, analysis techniques, reliability, and validity concerns.

Chapter 3: Methodology

3.1.0 Introduction

This chapter explains the various methods used to achieve the desired goals of the study. It mainly
addresses how the data will be obtained and the methodology used in data analysis.

3.2.0 Study design

This will be a retrospective case-control study designed to ascertain the impact of the influenza vaccine
on the rates of medically attended respiratory illness. The procedures for carrying out this assessment are
given in further detail below.

3.2.1 Study sites

Projects sites: Mombasa and Nakuru counties will be the primary study sites. The assessment will be
conducted in each of the two counties; two sub-counties will be selected to participate in the project:
Njoro and Nakuru North sub-counties in Nakuru, Jomvu, and Likoni sub-counties in Mombasa.
Mombasa and Nakuru. The initial project sites were selected as outlined below.

The assessment will be conducted in 10 immunizing facilities in each sub-counties participating in the
influenza vaccine demonstration project. Immunizing facilities were selected based on their
representativeness of the population served, facility size, and type. In each of the study sub-counties, the
following five categories of health facilities (HFs) were explicitly targeted: leading sub-county hospitals,
other public hospitals, public health centers, public dispensaries, and private hospitals. From each of
these categories, the selected HFs included one sub-county hospital, 1 other public hospital, 3 public
health centers, 3 public dispensaries, and 2 private HFs. In cases where there were more than the required
number of HFs for a specific category, a random sample was selected.
Control sites: Control facilities will be selected from the Molo sub-county in Nakuru and Changamwe
sub-county in Mombasa. These two sub-counties did not participate in the vaccine demonstration project.
The process for selection of the participating HFs in each of these sub-counties will be similar to the one
described above for the sub-counties that participated in the vaccine demonstration project.

Selection of target population: The target population is children aged <=5 years with reported medically
attended respiratory illness during an inpatient or outpatient visit in the selected HFs.

3.3.0 Data Collection and Sources

The data sources for quantifying the rates of medically attend respiratory illness among children under 5
years old in the selected HFs will be obtained from DHIS or logs and tally sheets for the periods outlined
as follows; the baseline (pre-intervention) period will be between October 2018 to October 2019, while
the 2-year intervention period will be between November 2019 to November 2021 and lastly the 1-year
post-intervention period will be marked between December 2021 to December 2022. The variables of
interest will be the number of cases of ; (1) upper respiratory illness, (2) lower respiratory illness, (3)
chronic respiratory illness, (4) during an inpatient or (5) outpatient visit, (6) in respective counties for the
target population.

Exclusion criteria for respiratory illness history records/log review.


Facility record review will NOT be conducted if:
I. Facilities are not included in the two counties or the demonstration project.
II. Hospital administrators do not grant permission to access and review registration logs and patient
hospital records.

3.4.0 Data analysis

3.4.1 Implementation of ITS analysis

Simple pre-to-post comparisons may be used in some research; however, this approach ignores temporal
trends that could have an impact on the outcomes or how they should be interpreted. They specifically
fail to account for trends that might have already started before the intervention. Simple pre-to-post
comparisons would suggest a higher average outcome for the pre-interruption phase in Figure 1.
However, a look at the entire series indicates that the outcome was already slowly falling before the
interruption. Additionally, keep in mind that given the aggregate of the results, ITS approaches are often
ecological (e.g., rates, counts).
Figure 1. Example situation in which a simple pre- to post-comparison would be misleading when not
controlling for pre-interruption trend

In an ITS study design, a segmented regression technique is used to assess the time series of interest as
the outcome measure, with dummy variables inserted as model predictors to represent the series' pre- and
post-interruption or intervention phases. The regression is followed by the single-ITS analysis (SITSA)
model (Penfold & Zhang, 2013):

(Eq 1) y = α + β1T + β2X + β3XT + ε

Where y = outcome variable, α = intercept, β = coefficients, T = time (1, 2, 3, …, N), X = study phase (0
during pre-interruption and 1 during post-interruption), XT = time after interruption (0 during pre-
interruption and 1, 2, 3, …, N during post-interruption), and ε = error or residual. The coefficient (β) for
T indicates the pre-interruption slope. The primary coefficients of interest are β2 and β3 (for X and XT),
which respectively indicate the change in level from pre- to post-interruption and the change in slope
from pre- to post-interruption (Penfold & Zhang, 2013).

The post-interruption slope may be calculated by adding coefficients 1 and 3 (Linden, 2015). Statistical
significance is attained using post-estimation techniques. Covariance can also be used to determine the
standard error for the total. Conclusions can be reached with SITSA based on statistical evidence rather
than a visual evaluation of time series graphs. After considering any secular trend in the pre-interruption
period, it should be possible to discern a change in level and slope between the pre-and post-phases if an
interruption or intervention has an impact on the outcome. Figure 2 displays instances of basic post-
interruption effects.
Figure 2. Examples of simple effects of interruption/intervention include level change, slope change,
and both.

The ITS design can be reinforced by integrating a comparable control series in a comparative design
utilizing multi-interrupted time series analysis, even though the majority of researchers only use SITSA
(MITSA). MITSA is built on a regression model with additional dummy variables, similar to the single
series case (Linden, 2015):

(Eq 2) y = α + β1T + β2Xt + β3XT + β4Z + β5ZT + β6ZX + β7ZTX + ε

where Z = binary indicator for treatment (1) or control (0) series, ZT = 1, 2, 3, …, N for the treatment
series and is always 0 for the control series, ZX = 0 for pre-interruption observations and 1 for post-
interruption observations in the treatment series and is always 0 for the control series, and ZXT = 0 for
pre-interruption and 1, 2, 3, …, N for post-interruption of treatment series and is always 0 for control. All
other variables are as indicated in Eq 1, and coefficients 1 to 3 represent characteristics of the control
series as described in Eq 1. The pre-interruption phase level and slope variations between the treatment
and control series are indicated by coefficients 4 and 5, respectively. These coefficients will not be
significant (e.g., p > 0.05) in situations when the series were comparable before the interruption. The
remaining coefficients are used to determine differences between the treatment and control series; 6
denotes the level difference between the two in the post-interruption phase, and 7 denotes the change in
slope difference from pre- to post-interruption a "difference-in-differences of slopes"; (Linden, 2015).
Data preparation: The first step in an ITS analysis is to prepare the data for analysis. This may involve
cleaning the data to remove any missing or erroneous data points, checking for outliers or extreme values,
and aggregating the data to the appropriate period (e.g., monthly or weekly).

Visual inspection: The next step is to inspect the data to look for any patterns or trends over time,
including the level and trend of the outcome variable (e.g., number of respiratory illness cases) before and
after the intervention period. This can help identify any potential confounding factors or other changes in
the population or healthcare system that may affect the outcome variable.

Model specification: The ITS model will be specified to estimate the effect of the intervention on the
outcome variable. The model will include a pre-intervention trend (using data from the baseline period),
an intervention effect, and a post-intervention trend (using data from the post- intervention period).

Intervention effect: The intervention effect will be estimated by examining the change in the level and
slope of the outcome variable at the time of the intervention relative to the pre- intervention trend. This
can be done using a regression model or another statistical technique.

Model fit: The first step is to assess the overall fit of the ITS model to the data. This will be done by
visually inspecting the predicted values from the model against the actual values, and by calculating
measures of goodness of fit using the root mean squared error (RMSE).

Reporting: The results of the ITS analysis will be reported, including estimates of the intervention effect,
any statistical tests or confidence intervals, and a discussion of the implications of the findings. The
limitations of the analysis will also be discussed, including any potential biases or confounding factors that
may affect the results.

Overall, ITS analysis is a powerful technique for evaluating the impact of an intervention on a time-series
outcome variable, such as the number of respiratory illness cases in a population. By accounting for pre-
existing trends and other time-varying factors, ITS analysis can provide robust estimates of the effect of
the intervention and help guide public health policy and practice.
3.4.2 Implementation of Recurrent Neural Network (RNN) algorithm
The architecture of traditional Recurrent neural networks, also known as RNNs, is a class of neural
networks that allow previous outputs to be used as inputs while having hidden states. They are typical as
follows:

Fig. 3: A simple recurrent neural network (RNN) and its unfolded structure through time t. Each arrow
shows a complete connection of units between the layers. To keep the figure simple, biases are not shown.

A simple RNN has three layers: input, recurrent hidden, and output. The input layer has N input units.
The inputs to this layer are a sequence of vectors through time t such as
. . . , 𝒙𝒕−𝟏, 𝒙𝒕, 𝒙𝒕+𝟏, . . . , 𝐰𝐡𝐞𝐫𝐞 𝒙𝒕 = (𝒙𝟏, 𝒙𝟐, . . . , 𝒙𝑵). The input units in a fully connected RNN are
connected to the hidden units in the hidden layer, where the connections are defined with a weight matrix
WIH. The hidden layer has M hidden units 𝒉𝒕 = (𝒉𝟏, 𝒉𝟐, . . . , 𝒉𝑴), that are connected to each other
through time with recurrent connections, Figure 3. Small non-zero elements can be used to initialize
hidden units, enhancing overall performance and network stability (Sutskever et al., n.d.).The system's
"memory" or state space is described by the hidden layer as

(Eq 3) ℎ𝑡 = 𝑓𝐻(𝑜𝑡),
where
(Eq 4) 𝑜𝑡 = 𝑊𝐼𝐻𝑋𝑡 + 𝑊𝐻𝐻ℎ𝑡−1 + 𝑏ℎ,

fH(·) is the hidden layer activation function, and bh is the bias vector of the hidden units. The hidden
units are connected to the output layer with weighted connections WHO. The output layer has P units yt
= (y1, y2, …, yP ) that are computed as

(Eq 5) 𝑦𝑡 = 𝑓𝑂(𝑊𝐻𝑂ℎ𝑡 + 𝑏𝑜)

where 𝒇𝑶(·) is the activation function and bo is the bias vector in the output layer. Since the input-target
pairs are sequential through time, the above steps are repeated consequently over time t = (1, …, T ).
Equations (3) and (5) demonstrate that an RNN comprises a set of nonlinear state equations that can be
iterated over time. Based on the input vector, the hidden states offer a prediction at the output layer for
each timestep. The hidden state of an RNN is a set of values that, independent of the influence of any
external factors, compile all the specific knowledge required about the network's previous states over
numerous timesteps. At the output layer, this integrated information can define the network's future
behavior and enable precise predictions (Sutskever et al., 2011)

Data preparation: The first step is to prepare the data for analysis. This may involve cleaning the data to
remove any missing or erroneous data points, checking for outliers or extreme values, and aggregating
the data to the appropriate period (e.g., monthly or weekly). The data must also be split into training and
testing sets to train and evaluate the RNN model.

Feature engineering: The next step is to engineer the features or predictors that will be used to train the
RNN model. This may involve selecting relevant variables or covariates, such as age, sex, vaccination
status, and environmental factors, and transforming the data to a suitable format for the RNN algorithm,
such as a time series or panel data format.

Model specification: The RNN model will be specified based on the features and outcome variables of
interest. The model may include one or more layers of recurrent or convolutional neural networks, along
with activation functions, regularization techniques, and other hyperparameters. The model will be
trained using a suitable optimization algorithm, such as stochastic gradient descent, and will be evaluated
using Mean Absolute Error (MAE), or the accuracy, precision, recall, and F1 score,

Model training and evaluation: The RNN model will be trained on the training data using the specified
features and outcome variables. The performance of the model will be evaluated on the testing data, and the
model parameters may be fine-tuned or adjusted based on the evaluation results. This process may involve
using cross-validation or other techniques to minimize overfitting and ensure the generalizability of the
results.

Reporting: The results of the RNN analysis will be reported, including the model's performance on the
testing data, the estimated effect of the intervention on the outcome variable, and any statistical tests or
confidence intervals. The limitations of the analysis will also be discussed, including any potential biases
or confounding factors that may affect the results.

Overall, using an RNN machine learning algorithm to analyze time series data can provide a powerful
tool for evaluating the impact of a vaccination program on the number of respiratory illnesses. By
accounting for potential confounding factors and analyzing trends over time, this approach can provide
valuable insights into the effectiveness of the vaccination program and help guide public health policy
and practice.

3.5.0 Ethical considerations

This study will be approved by the ethics committee of the National Commission for Science,
Technology & Innovation (NACOSTI) and Scientific Ethics Review Unit (SERU); Kenya Medical
Research Institute (KEMRI)-Scientific and Ethics Review body. All data will be de-identified to protect
the privacy of study participants.

3.6.0 Limitations

The study may be subject to certain limitations, including potential confounding factors not accounted
for in the analysis, the possibility of incomplete or inaccurate data, and the potential for unmeasured
changes in healthcare delivery or other factors that may affect the outcome variable.

3.7.0 Conclusion

The use of interrupted time series analysis to evaluate the impact of a vaccine administration intervention
can provide valuable information on the effectiveness of the intervention and help guide public health
policy and practice. By accounting for potential confounding factors and examining trends over time, this
study aims to provide robust evidence for the impact of the influenza vaccination program on the number
of medically attended respiratory illnesses in children aged 0 to 5 years.
Using RNN analysis to evaluate the impact of a vaccine administration intervention can provide valuable
information on the effectiveness of the intervention and help guide public health policy and practice. Using
an RNN algorithm to analyze time-series data, this study aims to provide robust evidence for the impact of
the influenza vaccination program on the number of medically attended respiratory illnesses in children
aged 0 to 5 years.
Chapter 4: References

Arabi Belaghi, R., Beyene, J., & McDonald, S. D. (2021). Prediction of pre-term birth in nulliparous women

using logistic regression and machine learning. PloS One, 16(6), e0252025.

https://doi.org/10.1371/journal.pone.0252025

Belongia, E. A., Simpson, M. D., King, J. P., Sundaram, M. E., Kelley, N. S., Osterholm, M. T., & McLean, H.

Q. (2016). Variable influenza vaccine effectiveness by subtype: A systematic review and meta-

analysis of test-negative design studies. The Lancet. Infectious Diseases, 16(8), 942–951.

https://doi.org/10.1016/S1473-3099(16)00129-8

Bengio, Y., Simard, P., & Frasconi, P. (1994). Learning long-term dependencies with gradient descent is

difficult. IEEE Transactions on Neural Networks, 5(2), 157–166.

https://doi.org/10.1109/72.279181

Bernal, J. L., Cummins, S., & Gasparrini, A. (2017). Interrupted time series regression for the evaluation

of public health interventions: A tutorial. International Journal of Epidemiology, 46(1), 348–355.

https://doi.org/10.1093/ije/dyw098

Brousseau, N., Green, H., Andrews, N., Pryse, R., Baguelin, M., Postema, A., Ellis, J., & Pebody, R. (2015).

Impact of influenza vaccination on respiratory illness rates in children attending private boarding

schools in England, 2013–2014: A cohort study. Epidemiology and Infection, 143, 1–11.

https://doi.org/10.1017/S0950268815000667

Campbell, A. P., Ogokeh, C., Weinberg, G. A., Boom, J. A., Englund, J. A., Williams, J. V., Halasa, N. B.,

Selvarangan, R., Staat, M. A., Klein, E. J., McNeal, M., Michaels, M. G., Sahni, L. C., Stewart, L. S.,

Szilagyi, P. G., Harrison, C. J., Lively, J. Y., Rha, B., Patel, M., & New Vaccine Surveillance Network

(NVSN). (2021). Effect of Vaccination on Preventing Influenza-Associated Hospitalizations Among

Children During a Severe Season Associated With B/Victoria Viruses, 2019–2020. Clinical

Infectious Diseases, 73(4), e947–e954. https://doi.org/10.1093/cid/ciab060


CDC. (2020, October 6). Benefits of Flu Vaccination During 2018-2019 Flu Season. Centers for Disease

Control and Prevention. https://www.cdc.gov/flu/about/burden-averted/2019-2020.htm

CDC. (2022, October 3). Flu Symptoms & Complications. Centers for Disease Control and Prevention.

https://www.cdc.gov/flu/symptoms/symptoms.htm

Cho, K., van Merrienboer, B., Gulcehre, C., Bahdanau, D., Bougares, F., Schwenk, H., & Bengio, Y. (2014).

Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

(arXiv:1406.1078). arXiv. https://doi.org/10.48550/arXiv.1406.1078

Clemens, J., Brenner, R., Rao, M., Tafari, N., & Lowe, C. (1996). Evaluating New Vaccines for Developing

Countries: Efficacy or Effectiveness? JAMA, 275(5), 390–397.

https://doi.org/10.1001/jama.1996.03530290060038

Collen, M. F. (1986). Origins of medical informatics. The Western Journal of Medicine, 145(6), 778–785.

Cruz, J. A., & Wishart, D. S. (2007). Applications of machine learning in cancer prediction and prognosis.

Cancer Informatics, 2, 59–77.

Dawa, J. A., Chaves, S. S., Nyawanda, B., Njuguna, H. N., Makokha, C., Otieno, N. A., Anzala, O.,

Widdowson, M.-A., & Emukule, G. O. (2018). National burden of hospitalized and non-

hospitalized influenza-associated severe acute respiratory illness in Kenya, 2012-2014. Influenza

and Other Respiratory Viruses, 12(1), 30–37. https://doi.org/10.1111/irv.12488

de Hoog, M. L. A., Venekamp, R. P., Damoiseaux, R. A. M. J., Schilder, A. G. M., Sanders, E. A. M., Smit, H.

A., & Bruijning-Verhagen, P. C. J. L. (2019). Impact of Repeated Influenza Immunization on

Respiratory Illness in Children With Pre-existing Medical Conditions. Annals of Family Medicine,

17(1), 7–13. https://doi.org/10.1370/afm.2340

Dennis, J., Ramsay, T., Turgeon, A. F., & Zarychanski, R. (2013). Helmet legislation and admissions to

hospital for cycling related head injuries in Canadian provinces and territories: Interrupted time

series analysis. BMJ (Clinical Research Ed.), 346, f2674. https://doi.org/10.1136/bmj.f2674


Ellis, K., Godbole, S., Marshall, S., Lanckriet, G., Staudenmayer, J., & Kerr, J. (2014). Identifying Active

Travel Behaviors in Challenging Environments Using GPS, Accelerometers, and Machine Learning

Algorithms. Frontiers in Public Health, 2.

https://www.frontiersin.org/articles/10.3389/fpubh.2014.00036

Elman, J. L. (1991). Distributed representations, simple recurrent networks, and grammatical structure.

Machine Learning, 7(2), 195–225. https://doi.org/10.1007/BF00114844

Emukule, G. O. (n.d.). The epidemiological and economic burden of influenza in Kenya: Implications for

public health action. 182.

Emukule, G. O., Paget, J., van der Velden, K., & Mott, J. A. (2015). Influenza-Associated Disease Burden in

Kenya: A Systematic Review of Literature. PLoS ONE, 10(9), e0138708.

https://doi.org/10.1371/journal.pone.0138708

Esteva, A., Kuprel, B., Novoa, R. A., Ko, J., Swetter, S. M., Blau, H. M., & Thrun, S. (2017). Dermatologist-

level classification of skin cancer with deep neural networks. Nature, 542(7639), 115–118.

https://doi.org/10.1038/nature21056

Ferdinands, J. M., Olsho, L. E. W., Agan, A. A., Bhat, N., Sullivan, R. M., Hall, M., Mourani, P. M.,

Thompson, M., Randolph, A. G., & Pediatric Acute Lung Injury and Sepsis Investigators (PALISI)

Network. (2014). Effectiveness of influenza vaccine against life-threatening RT-PCR-confirmed

influenza illness in US children, 2010-2012. The Journal of Infectious Diseases, 210(5), 674–683.

https://doi.org/10.1093/infdis/jiu185

Fretheim, A., & Tomic, O. (2015). Statistical process control and interrupted time series: A golden

opportunity for impact evaluation in quality improvement. BMJ Quality & Safety, 24(12), 748–

752. https://doi.org/10.1136/bmjqs-2014-003756

Gers, F. A., Schmidhuber, J., & Cummins, F. (2000). Learning to Forget: Continual Prediction with LSTM.

Neural Computation, 12(10), 2451–2471. https://doi.org/10.1162/089976600300015015


Grijalva, C. G., Nuorti, J. P., Arbogast, P. G., Martin, S. W., Edwards, K. M., & Griffin, M. R. (2007). Decline

in pneumonia admissions after routine childhood immunisation with pneumococcal conjugate

vaccine in the USA: A time-series analysis. Lancet (London, England), 369(9568), 1179–1186.

https://doi.org/10.1016/S0140-6736(07)60564-9

Habehh, H., & Gohel, S. (2021). Machine Learning in Healthcare. Current Genomics, 22(4), 291–300.

https://doi.org/10.2174/1389202922666210705124359

Hasman, A., Mantas, J., & Zarubina, T. (2014). An abridged history of medical informatics education in

europe. Acta Informatica Medica: AIM: Journal of the Society for Medical Informatics of

Bosnia & Herzegovina: Casopis Drustva Za Medicinsku Informatiku BiH, 22(1), 25–36.

https://doi.org/10.5455/aim.2014.22.25-36

Influenza (Seasonal). (n.d.). Retrieved November 29, 2022, from https://www.who.int/news-room/fact-

sheets/detail/influenza-(seasonal)

Katz, M. A., Lebo, E., Emukule, G. O., Otieno, N., Caselton, D. L., Bigogo, G., Njuguna, H., Muthoka, P. M.,

Waiboci, L. W., Widdowson, M.-A., Xu, X., Njenga, M. K., Mott, J. A., & Breiman, R. F. (2016).

Uptake and Effectiveness of a Trivalent Inactivated Influenza Vaccine in Children in Urban and

Rural Kenya, 2010 to 2012. The Pediatric Infectious Disease Journal, 35(3), 322–329.

https://doi.org/10.1097/INF.0000000000001035

Li, L., Jiang, Y., & Huang, B. (2021). Long-term prediction for temporal propagation of seasonal influenza

using Transformer-based model. Journal of Biomedical Informatics, 122, 103894.

https://doi.org/10.1016/j.jbi.2021.103894

Linden, A. (2015). Conducting Interrupted Time-series Analysis for Single- and Multiple-group

Comparisons. The Stata Journal, 15(2), 480–500. https://doi.org/10.1177/1536867X1501500208

Lopez Bernal, J. A., Andrews, N., & Amirthalingam, G. (2019). The Use of Quasi-experimental Designs for

Vaccine Evaluation. Clinical Infectious Diseases, 68(10), 1769–1776.


https://doi.org/10.1093/cid/ciy906

Mameli, C., Cocchi, I., Fumagalli, M., & Zuccotti, G. (2019). Influenza Vaccination: Effectiveness,

Indications, and Limits in the Pediatric Population. Frontiers in Pediatrics, 7, 317.

https://doi.org/10.3389/fped.2019.00317

Matheka, D. M., Mokaya, J., & Maritim, M. (2013). Overview of influenza virus infections in Kenya: Past,

present and future. The Pan African Medical Journal, 14, 138.

https://doi.org/10.11604/pamj.2013.14.138.2612

Mw, T., Hk, T., Ch, T., M, G., Tm, M., As, M., Et, M., Rk, Z., Fp, S., Db, M., Sm, O., Rj, G. K., Jr, B., Jm, F.,

Mm, P., & undefined. (2021). Influenza Vaccine Effectiveness Against Hospitalization in the United

States, 2019-2020. The Journal of Infectious Diseases, 224(5), 813–820.

https://doi.org/10.1093/infdis/jiaa800

Nunes, M. C., Cutland, C. L., Jones, S., Downs, S., Weinberg, A., Ortiz, J. R., Neuzil, K. M., Simões, E. A. F.,

Klugman, K. P., & Madhi, S. A. (2017). Efficacy of Maternal Influenza Vaccination Against All-

Cause Lower Respiratory Tract Infection Hospitalizations in Young Infants: Results From a

Randomized Controlled Trial. Clinical Infectious Diseases, 65(7), 1066–1071.

https://doi.org/10.1093/cid/cix497

Omrani, H. (2015). Predicting Travel Mode of Individuals by Machine Learning. Transportation Research

Procedia, 10, 840–849. https://doi.org/10.1016/j.trpro.2015.09.037

Otieno, N. A., Nyawanda, B. O., McMorrow, M., Oneko, M., Omollo, D., Lidechi, S., Widdowson, M.,

Flannery, B., Chaves, S. S., Azziz‐Baumgartner, E., & Emukule, G. O. (2022). The burden of

influenza among Kenyan pregnant and postpartum women and their infants, 2015–2020.

Influenza and Other Respiratory Viruses, 16(3), 452–461. https://doi.org/10.1111/irv.12950

Penfold, R. B., & Zhang, F. (2013). Use of interrupted time series analysis in evaluating health care

quality improvements. Academic Pediatrics, 13(6 Suppl), S38-44.


https://doi.org/10.1016/j.acap.2013.08.002

Perin, J., Mulick, A., Yeung, D., Villavicencio, F., Lopez, G., Strong, K. L., Prieto-Merino, D., Cousens, S.,

Black, R. E., & Liu, L. (2022). Global, regional, and national causes of under-5 mortality in 2000–

19: An updated systematic analysis with implications for the Sustainable Development Goals.

The Lancet Child & Adolescent Health, 6(2), 106–115. https://doi.org/10.1016/S2352-

4642(21)00311-4

Rajpurkar, P., Irvin, J., Zhu, K., Yang, B., Mehta, H., Duan, T., Ding, D., Bagul, A., Langlotz, C., Shpanskaya,

K., Lungren, M. P., & Ng, A. Y. (2017). CheXNet: Radiologist-Level Pneumonia Detection on Chest

X-Rays with Deep Learning. In ArXiv e-prints.

https://ui.adsabs.harvard.edu/abs/2017arXiv171105225R

Recommendations for influenza vaccine composition. (n.d.). Retrieved November 1, 2022, from

https://www.who.int/teams/global-influenza-programme/vaccines/who-recommendations

Robot‐assisted surgery for benign distal ureteral strictures: Step‐by‐step technique using the SP® surgical

system—Kaouk—2019—BJU International—Wiley Online Library. (n.d.). Retrieved December 7,

2022, from https://bjui-journals.onlinelibrary.wiley.com/doi/abs/10.1111/bju.14635

Rose, A., Kissling, E., Emborg, H.-D., Larrauri, A., McMenamin, J., Pozo, F., Trebbien, R., Mazagatos, C.,

Whitaker, H., & Valenciano, M. (2020). Interim 2019/20 influenza vaccine effectiveness: Six

European studies, September 2019 to January 2020. Eurosurveillance, 25(10), 2000153.

https://doi.org/10.2807/1560-7917.ES.2020.25.10.2000153

Sambala, E. Z., Cooper, S., Schmidt, B.-M., Walaza, S., & Wiysonge, C. S. (2021). Role of vaccines in

preventing influenza in healthy children. SAMJ: South African Medical Journal, 111(3), 206–207.

https://doi.org/10.7196/samj.2021.v111i3.14478

Segaloff, H. E., Leventer-Roberts, M., Riesel, D., Malosh, R. E., Feldman, B. S., Shemer-Avni, Y., Key, C.,

Monto, A. S., Martin, E. T., & Katz, M. A. (2019). Influenza Vaccine Effectiveness Against
Hospitalization in Fully and Partially Vaccinated Children in Israel: 2015-2016, 2016-2017, and

2017-2018. Clinical Infectious Diseases: An Official Publication of the Infectious Diseases Society

of America, 69(12), 2153–2161. https://doi.org/10.1093/cid/ciz125

Shinjoh, M., Sugaya, N., Furuichi, M., Araki, E., Maeda, N., Isshiki, K., Ohnishi, T., Nakamura, S., Yamada,

G., Narabayashi, A., Nishida, M., Taguchi, N., Nakata, Y., Yoshida, M., Tsunematsu, K., Shibata, M.,

Munenaga, T., Hirano, Y., Ookawara, I., … Keio Pediatric Influenza Research Group. (2019).

Effectiveness of inactivated influenza vaccine in children by vaccine dose, 2013-18. Vaccine,

37(30), 4047–4054. https://doi.org/10.1016/j.vaccine.2019.05.090

Siddiqui, M. K., Morales-Menendez, R., Huang, X., & Hussain, N. (2020). A review of epileptic seizure

detection using machine learning classifiers. Brain Informatics, 7(1), 5.

https://doi.org/10.1186/s40708-020-00105-1

Sutskever, I., Martens, J., & Hinton, G. (2011). Generating text with recurrent neural networks.

Proceedings of the 28th International Conference on International Conference on Machine

Learning, 1017–1024.

Tang, J., Liu, R., Zhang, Y.-L., Liu, M.-Z., Hu, Y.-F., Shao, M.-J., Zhu, L.-J., Xin, H.-W., Feng, G.-W., Shang,

W.-J., Meng, X.-G., Zhang, L.-R., Ming, Y.-Z., & Zhang, W. (2017). Application of Machine-

Learning Models to Predict Tacrolimus Stable Dose in Renal Transplant Recipients. Scientific

Reports, 7(1), Article 1. https://doi.org/10.1038/srep42192

Tenny, S., Kerndt, C. C., & Hoffman, M. R. (2022). Case Control Studies. In StatPearls. StatPearls

Publishing. http://www.ncbi.nlm.nih.gov/books/NBK448143/

Thakkar, M., & Davis, D. C. (2006). Risks, barriers, and benefits of EHR systems: A comparative study

based on size of hospital. Perspectives in Health Information Management, 3, 5.

Thompson, M. G., Kwong, J. C., Regan, A. K., Katz, M. A., Drews, S. J., Azziz-Baumgartner, E., Klein, N. P.,
Chung, H., Effler, P. V., Feldman, B. S., Simmonds, K., Wyant, B. E., Dawood, F. S., Jackson, M. L.,

Fell, D. B., Levy, A., Barda, N., Svenson, L. W., Fink, R. V., … PREVENT Workgroup. (2019).

Influenza Vaccine Effectiveness in Preventing Influenza-associated Hospitalizations During

Pregnancy: A Multi-country Retrospective Test Negative Design Study, 2010-2016. Clinical

Infectious Diseases: An Official Publication of the Infectious Diseases Society of America, 68(9),

1444–1453. https://doi.org/10.1093/cid/ciy737

Turing 1950 Article - A. M. Turing (1950) Computing Machinery and Intelligence. Mind 49: 433-460.

- Studocu. (n.d.). Retrieved December 7, 2022, from https://www.studocu.com/en-

us/document/emory-university/cognition/turing-1950-article/13121870

Vaccine Effectiveness: How Well Do Flu Vaccines Work? | CDC. (2022, August 25).

https://www.cdc.gov/flu/vaccines-work/vaccineeffect.htm

Verani, J. R., Baqui, A. H., Broome, C. V., Cherian, T., Cohen, C., Farrar, J. L., Feikin, D. R., Groome, M. J.,

Hajjeh, R. A., Johnson, H. L., Madhi, S. A., Mulholland, K., O’Brien, K. L., Parashar, U. D., Patel, M.

M., Rodrigues, L. C., Santosham, M., Scott, J. A., Smith, P. G., … Zell, E. R. (2017). Case-control

vaccine effectiveness studies: Data collection, analysis and reporting results. Vaccine, 35(25),

3303–3308. https://doi.org/10.1016/j.vaccine.2017.04.035

Wati, D. A. R., & Abadianto, D. (2017). Design of face detection and recognition system for smart home

security application. 2017 2nd International Conferences on Information Technology,

Information Systems and Electrical Engineering (ICITISEE), 342–347.

https://doi.org/10.1109/ICITISEE.2017.8285524

What Is Machine Learning (ML)? (2020, June 26). UCB-UMT.

https://ischoolonline.berkeley.edu/blog/what-is-machine-learning/

Woldaregay, A. Z., Årsand, E., Botsis, T., Albers, D., Mamykina, L., & Hartvigsen, G. (2019). Data-Driven

Blood Glucose Pattern Classification and Anomalies Detection: Machine-Learning Applications in


Type 1 Diabetes. Journal of Medical Internet Research, 21(5), e11030.

https://doi.org/10.2196/11030

You might also like