Computational Epidemiology From Disease Transmission Modeling To Vaccination Decision Making 1st Ed
Computational Epidemiology From Disease Transmission Modeling To Vaccination Decision Making 1st Ed
Computational Epidemiology From Disease Transmission Modeling To Vaccination Decision Making 1st Ed
Jiming Liu
Shang Xia
Computational
Epidemiology
From Disease Transmission Modeling
to Vaccination Decision Making
Health Information Science
Series editor
Yanchun Zhang, Victoria University, Melbourne, Victoria, Australia
Editorial Board
Riccardo Bellazzi, University of Pavia, Italy
Leonard Goldschmidt, Stanford University Medical School, USA
Frank Hsu, Fordham University, USA
Guangyan Huang, Victoria University, Australia
Frank Klawonn, Helmholtz Centre for Infection Research, Germany
Jiming Liu , Hong Kong Baptist University, Hong Kong
Zhijun Liu, Hebei University of Engineering, China
Gang Luo, University of Utah, USA
Jianhua Ma, Hosei University, Japan
Vincent Tseng, National Cheng Kung University, Taiwan
Dana Zhang, Google, USA
Fengfeng Zhou, Shenzhen Institutes of Advanced Technology, Chinese Academy
of Sciences, China
With the development of database systems and networking technologies, Hospital
Information Management Systems (HIMS) and web-based clinical or medical
systems (such as the Medical Director, a generic GP clinical system) are widely
used in health and clinical practices. Healthcare and medical service are more data-
intensive and evidence-based since electronic health records are now used to track
individuals’ and communities’ health information. These highlights substantially
motivate and advance the emergence and the progress of health informatics research
and practice. Health Informatics continues to gain interest from both academia and
health industries. The significant initiatives of using information, knowledge and
communication technologies in health industries ensures patient safety, improve
population health and facilitate the delivery of government healthcare services.
Books in the series will reflect technology’s cross-disciplinary research in IT and
health/medical science to assist in disease diagnoses, treatment, prediction and mon-
itoring through the modeling, design, development, visualization, integration and
management of health related information. These technologies include information
systems, web technologies, data mining, image processing, user interaction and
interfaces, sensors and wireless networking, and are applicable to a wide range of
health-related information such as medical data, biomedical data, bioinformatics
data, and public health data.
Series Editor: Yanchun Zhang, Victoria University, Australia;
Editorial Board: Riccardo Bellazzi, University of Pavia, Italy; Leonard Gold-
schmidt, Stanford University Medical School, USA; Frank Hsu, Fordham Uni-
versity, USA; Guangyan Huang, Victoria University, Australia; Frank Klawonn,
Helmholtz Centre for Infection Research, Germany; Jiming Liu, Hong Kong Baptist
University, Hong Kong, China; Zhijun Liu, Hebei University of Engineering, China;
Gang Luo, University of Utah, USA; Jianhua Ma, Hosei University, Japan; Vincent
Tseng, National Cheng Kung University, Taiwan; Dana Zhang, Google, USA;
Fengfeng Zhou, Shenzhen Institutes of Advanced Technology, Chinese Academy
of Sciences, China.
Computational Epidemiology
From Disease Transmission Modeling
to Vaccination Decision Making
Jiming Liu Shang Xia
Department of Computer Science Department of Computer Science
Hong Kong Baptist University Hong Kong Baptist University
Kowloon, Hong Kong Kowloon, Hong Kong
This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
This book is dedicated to all the people
around the world, who came together to fight
against the novel coronavirus (COVID-19)
pandemic.
Jiming Liu and Shang Xia
Preface
The title of this preface may look a bit unusual for a research monograph.
Nevertheless, this was indeed the kind of questions that came to our mind when
we were writing it.
This book was born in a special time. As it was being written and published, the
world was enduring one of its greatest challenges in decades, if not in centuries. The
novel coronavirus, known as COVID-19, had rapidly spread to around 200 countries
and territories in 6 continents (with only Antarctica untouched) within a few months,
resulting in more than 5 million people infected and over 300,000 deaths (as of
May 2020). All the people, no matter where they were and who they were, found
themselves caught right amid this most unprecedented global crisis, with devastating
casualties, country lockdowns, service/business shutdowns, and possible economic
meltdown.
The world is truly in a state of emergency, a time of great uncertainty and anxiety.
Yet, as in the history of human civilization, we all should be hopeful that
humankind will be able to learn and prevail in the end. There will be no exception
this time. One of the important lessons that we can probably learn from the
Mother Nature in this global fight against COVID-19 is that only by being
united as one, as humanity, working together to remove the barriers of races,
vii
viii Preface
who may study the book in the regular order of Chaps. 1–7 and then revisit Chaps. 2–
6 to extend some of the topics and problems.
xi
xii Acknowledgements
of the university, to make it the most conducive place for scholarship. He would
like to thank Hong Kong Research Grants Council (RGC) for the funding support
over the years; a number of grants have been awarded to specifically support his
team’s research on understanding and solving epidemiological problems through the
exciting routes of computer science, machine learning, and artificial intelligence.
Last but the foremost, he would like to express his deepest thanks to his wife
M.L. and his daughters I.Y.Y. and B.Y.X. for their long-lasting love and the most
wonderful time.
Shang Xia would like to express his sincere gratitude to Prof. Jiming Liu for his
enlightening, patience, motivation, enthusiasm, and profound knowledge. Without
his encouragement and persistence, this book could not be accomplished. He would
like to express his sincere gratefulness to Computer Science Department at Hong
Kong Baptist University (HKBU), where he acquired his PhD degree, benefited a lot
from the most inspirational guidance, and enjoyed a fulfilling campus life. For this
rewarding journey in Hong Kong, he would sincerely express his heartfelt gratitude
to Dr. Benyun Shi, Dr. Li Tao, and Dr. Yang Liu, from whom he benefited their
collaboration and support. The sincere thanks also go to Prof. Xiao-Nong Zhou and
the National Institute of Parasitic Diseases at Chinese Center for Disease Control
and Prevention for the great support for his academic career and research. Last but
not least, he would like to thank his family: his wife Yao Q.Q. and his daughters
Yoyo and Xiuxiu for their caring, love, and support in this wonderful life.
Both authors wish to express their special thanks to Dr. Yang Liu for his great
efforts in proofreading the manuscript and offering excellent editorial suggestions
and help.
1 Paradigms in Epidemiology. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Methodological Paradigms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Recent Developments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 Infectious Diseases and Vaccination . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.4 Objectives and Tasks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.4.1 Modeling Infectious Disease Dynamics . . . . . . . . . . . . . . . . . . . . . . . 8
1.4.2 Modeling Vaccine Allocation Strategies. . . . . . . . . . . . . . . . . . . . . . . 8
1.4.3 Modeling Vaccination Decision-Making . . . . . . . . . . . . . . . . . . . . . . 9
1.4.4 Modeling Subjective Perception . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2 Computational Modeling in a Nutshell . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.1 Modeling Infectious Disease Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.1.1 Infectious Disease Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.1.2 Age-Specific Disease Transmissions . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.2 Modeling Contact Relationships . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.2.1 Empirical Methods. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.2.2 Computational Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.3 Case Study. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.3.1 Hong Kong H1N1 Influenza Epidemic . . . . . . . . . . . . . . . . . . . . . . . . 25
2.3.2 Age-Specific Contact Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.3.3 Validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.4 Further Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3 Strategizing Vaccine Allocation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.1 Vaccination . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.1.1 Herd Immunity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.1.2 Vaccine Allocation Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.2 Vaccination Priorities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.3 Age-Specific Intervention Priorities. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.3.1 Modeling Prioritized Interventions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
xiii
xiv Contents
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
Abbreviations
xv
Notation
xvii
xviii Notation
Epidemiology deals with “the study of the occurrence and distribution of health-
related states or events in specified populations, including the study of the deter-
minants influencing such states, and the application of this knowledge to control
the health problems” [1]. As defined by MacMahon et al. [2], epidemiology is
interdisciplinary by nature, concerning the sciences of etiology, genetics, biology,
pharmacy, geography, ecology, as well sociology and human behavior. Epidemio-
logical studies motivated by combating infectious diseases mainly focus on four
aspects of challenges, as follows: (1) pattern analysis, by investigating the spatio-
temporal distributions of the observed disease occurrences; (2) causal inference, by
identifying and evaluating associated impact factors; (3) forecasting and prediction,
by evaluating the dynamics of infectious diseases with reference to different sce-
narios; and (4) policy analytics, by exploring and conducting effective intervention
measures.
Toward these ends, the pioneers in epidemiology have provided much useful
knowledge to guide efforts in infectious disease control. As pointed out by Merrill
[3], epidemiology has evolved from supernatural practices to research based on
scientific foundations, from ad hoc reports to systematic investigations of public
health events and problems, from ignorance of the causes of diseases to a scientific
understanding of their hidden factors, determinants, and outcomes, and from lacking
feasible means for solving public health problems to having effective approaches to
disease intervention.
Developmental milestones in infectious disease epidemiology can be dated
back to the work of Hippocrates (460–377 BC), who examined the influence of
environments and attempted to explain how diseases transmit and cause infection
in a group of host individuals [3]. Other early studies include the work of John
Graunt (1620–1674), who described disease mortality rates by applying statistical
and census methods [4], and Thomas Sydenham (1624–1689), who studied disease
distribution patterns, moving from an observational to an analytical perspective [5].
In the nineteenth century, John Snow (1813–1858) traced the sources of disease
outbreaks (e.g., cholera in Soho, London, in 1854) and thereafter pointed out the
associations of disease outbreaks with social and natural environments [6]. To
more formally describe the dynamics of disease transmission, Ross (in 1911) and
MacDonald (in 1957) developed a set of mathematical equations and proposed
a threshold indicator, named the basic reproduction number, to quantitatively
characterize the extent of disease transmission [7].
• Empirical Methods
The paradigm of empirical observation and investigation is well suited to the
early stage of epidemiological studies. As mentioned by Rothman et al. [9], it
typically involves (1) collecting observational data about disease transmission,
i.e., when, where, and who, and associated impact factors, e.g., the characteristics
of disease pathogens and host individuals at the microscopic scale, and of etiolog-
ical and meteorological environments at the macroscopic scale; (2) qualitatively
describing or quantitatively analyzing observational data to establish associative
1.1 Methodological Paradigms 3
or causal relationships between impact factors and disease transmission; and (3)
conducting further experiments or field investigations to test epidemiological
hypotheses, usually relating the proposed causes to the observed effects, the
findings of which may serve as the foundation for planning and implementing
disease intervention.
• Theoretical Methods
The theoretical paradigm in epidemiological studies involves the use of mathe-
matical tools, and is focused on generalizing and characterizing the processes of
disease transmission and their interrelationships with various impact factors [10].
Mathematical equations or models are typically constructed to quantitatively
describe the dynamics of disease transmission and estimate possible outcomes.
By evaluating different conditions under which the models reach convergent,
stable, or equilibrium states, public health authorities can potentially make long-
term projections and informed decisions on disease intervention. Theoretical
epidemiology sometimes draws on certain assumptions and simplifications about
the real processes of disease transmission. Meanwhile, it may also require
mathematical operations to derive model constructs of the behaviors of various
diseases, and use these to infer the disease dynamics and the corresponding
intervention measures.
• Computational Methods
With the developments in artificial intelligence, machine learning, data analytics,
data mining, and geographic science and information systems, the computational
paradigm has rapidly emerged in epidemiological studies. Computational meth-
ods are aimed to better characterize and understand the real processes of disease
4 1 Paradigms in Epidemiology
The above-mentioned methods have been in vogue for several decades and have
been used to make great contributions to our understanding and ability to combat
infectious diseases. However, there remain a number of challenges. As schemati-
cally illustrated in Fig. 1.2, these challenges come from emerging and re-emerging
infectious diseases, which are significantly correlated with multiple impact factors
and their interacting effects, such as genetic mutation of disease pathogens/parasites
[12], human socio-economic and behavioral changes [13], and environmental and
ecological conditions [14, 15].
Fig. 1.2 Some interacting components (in circles) and their associated impact factors that can
affect the transmission of infectious diseases
1.2 Recent Developments 5
Now we examine influenza as an example. It has been shown that a wide range
of factors are involved in the dynamic processes of these outbreaks [16], which may
include the following: (1) pathogenic factors, such as viral genetic recombination
and the expression of pathogens; (2) host factors, such as the immunity of people
at different ages; (3) social and behavioral factors, such as people’s movement
or travel activities; and (4) policy factors, such as disease intervention measures.
Furthermore, these factors closely interact with each other. For instance, disease
pathogens are carried by humans as they travel, which accelerates the recombination
of different types of viruses. However, the implementation of disease intervention,
e.g., school closures, changes people’s contact behaviors, effectively cutting off the
route of disease transmission.
This highlights how various factors can interrelate and interact at various scales.
Crucially, the coupling and interactive relationships among those impact factors
can determine the intrinsic (yet possibly hidden) spatial, temporal, and social
mechanisms of disease transmission. These mechanisms can involve systemic
characteristics, such as feedback, saturation, bifurcation, and chaos, thus posing new
challenges for comprehensive epidemiological investigations [17].
Effective intervention measures rely on biomedical understanding of disease
pathogens/parasites, descriptive studies of spatio-temporal patterns of disease occur-
rences, and causal analysis of impact factors. In addition, predictive explorations
of the trends of disease transmission, i.e., the mechanistic interactions among
the components of the transmission process, are also key to understanding and
combating infectious diseases. For example, an early warning system for an
emerging infectious disease, like COVID-19, requires knowledge about the possible
geographic routes of disease transmission, such as human air-travel networks
[18]. The prevention of zoonotic and vector-borne diseases, like COVID-19 and
malaria, requires both environmental and ecological changes of animal/vector
species to be addressed [19], as well as human migrant and mobile behaviors
[20]. Furthermore, the effectiveness of disease intervention measures depends on
the efficacy of resource allocation, compliance of targeted host populations, and
responsive feedback to environmental modifications.
In addition to the above-mentioned challenges, epidemiological studies also
face new opportunities in the present and future data-centric era, enabled by the
confluence of data from various sources and the development of modeling and
analytical tools in data science [21]. For example, a global disease surveillance
system connects the health agencies of its member countries and partners at
different levels, via local, regional, national, and international organizations [22].
This surveillance system can be used for managing and sharing historical records
and reports on when and where specific people have been infected by certain kinds
of diseases.
Other data sources are also helpful for analyzing and modeling potential disease
transmission. For example, remote sensing data from satellites can readily be
utilized for mapping the meteorological and ecological conditions of local or global
environments [23, 24].
6 1 Paradigms in Epidemiology
Faced with the threat of infectious diseases, implementing timely and effective
disease intervention measures is critical for preventing mortality and debilitating
morbidity, and reducing the socio-economic losses. Various types of intervention
measures have been widely studied and adopted for these purposes. For example,
immediate isolation/quarantine can prevent transmission during an influenza-like
epidemic [27, 28]. The mass prophylactic use of antiviral drugs can reduce the
vulnerability of susceptible individuals exposed to infectious diseases [29]. Inter-
ventions by social distancing (e.g., school closures and workplace shutdowns) can
lower the frequency of contacts among the host population and, hence, reduce the
probability of transmitting diseases between susceptible and infectious individuals
[30, 31].
Besides the above-mentioned intervention measures, vaccination has been
regarded as one of the most effective methods for preventing infectious diseases,
due to the effect of vaccine-induced herd immunity (i.e., immunizing a certain
portion of the host population provides indirect protection for the unimmunized
individuals [32]). That is to say, to prevent a potential outbreak, the vaccination
coverage in a host population needs to be above a critical level for inducing the
effect of herd immunity, known as herd immunity threshold. In practice, it remains
a continual challenge for public health authorities to achieve such a threshold of
vaccination coverage for preventing disease outbreaks.
The task is challenging due to a series of reasons. For one thing, although
significant progress has been made over the years in vaccine development, the
capacity for providing adequate and timely vaccine doses remains a concern,
especially when encountering emerging infectious diseases, e.g., 2009 influenza A
(H1N1) [33]. Supply restrictions can arise due to many factors, including the time
needed for finalizing vaccine compositions, to respond to the constantly evolution
of new disease strains [34], the limited capacity for vaccine manufacturing and
logistics [35], and the difficulties in access to and uptake of vaccines due to
poor delivery infrastructures and economic constraints, especially in developing
countries [36]. In such situations, public health authorities in charge of vaccination
1.4 Objectives and Tasks 7
programs face the question of how to allocate a finite number of available vaccine
doses to most effectively prevent disease transmission. For example, the World
Health Organization (WHO) has strongly suggested that each country should
respond to a possible shortage of vaccine supplies by deciding in advance which
groups should have access [37].
Furthermore, the public acceptance of a vaccination program will crucially affect
the actual level of vaccine uptake: any loss of confidence in vaccine safety and
efficacy will lead to huge gaps between the level of public vaccination willingness
and the level needed to contain disease transmission. Historically, societies have
experienced several events of vaccine refusal, e.g., the pertussis vaccine scare in
the 1970s [38], the decline of measles-mumps-rubella (MMR) vaccine uptake in
the 1990s [39, 40], and the rise and popularity of anti-vaccination movements [41,
42]. The rejection of vaccination and the subsequent decline of vaccine uptake have
brought about outbreaks of certain vaccine-preventable diseases that were thought
to no longer be threats to humankind [43, 44].
In view of this, an in-depth understanding of individuals’ voluntary vaccination
compliance is urgently required. It has been found that public acceptance of
vaccination, which amounts to individuals’ decisions on whether or not to take
vaccines, are affected by a mixture of cultural, behavioral, and socio-economic
factors. For example, the public may have doubts about vaccine safety and efficacy
due to scare stories around the adverse effects of vaccination [45, 46]. Or, behaving
in their own self-interest, individuals may be inclined not to get vaccinated if
enough other people have been vaccinated [47, 48]. The affordability and convenient
accessibility of new vaccines are also of importance for individuals considering
vaccination, especially in developing countries [49, 50].
Furthermore, the rapid emergence of online social media, e.g., Facebook and
Twitter, allows opinions, whether for or against vaccination, to spread broadly
and immediately among the population [51]. Therefore, social influences play an
increasingly important role in individuals’ vaccination decisions. In this regard, an
individual’s decision on whether or not to vaccinate himself/herself is no longer a
personal affair, but will affect the decisions of others, and collectively determine the
final coverage of a vaccination program.
Clearly, there exists an urgent need for more systematic studies of vaccination
at both population and individual levels, and thereby improve the efficacy of
vaccination programs for preventing the outbreak of infectious diseases.
The heterogeneity of the host population means that the disease-preventing effects
of vaccination in individuals of different ages can vary markedly. An immediately
related practical question is how to allocate a finite number of vaccine doses to
1.4 Objectives and Tasks 9
It has long been observed that the spread of awareness about an epidemic via social
media can affect individuals’ opinions and behaviors concerning an epidemic. In the
case of an emerging infectious disease, it can be difficult for individuals to become
informed about the disease and/or a newly developed vaccine prior to their decision-
making. In such a case, the spread of awareness about disease severity and vaccine
safety could affect individuals’ subjective perception about vaccination and, hence,
substantially affect their decisions [54].
To gain a better understanding of individuals’ voluntary vaccination, we develop
a belief-based decision model to evaluate the effect of the spread of awareness
on individuals’ decision-making and on the effectiveness of disease intervention.
Compared with the existing studies on modeling individual-level vaccination
decision-making, this belief-based model has the following unique properties:
• Unlike existing decision models that represent decision-making as a binary
problem, we consider the role of uncertainty in individuals’ vaccination decision-
making. Specifically, the situation in which an individual has made no firm
decision can be considered as a state of “yet to decide”, due to uncertainty. In this
regard, we introduce three belief variables to characterize the possible decision
response from an individual, namely that he/she will accept or reject the vaccine,
or has not yet decided.
• We further consider the fact that individuals’ decisions depend on their subjective
perception about whether or not vaccination is acceptable. Moreover, awareness
of disease severity and vaccine safety can spread from person to person—akin
to a disease itself—and will substantially affect their subjective perception of
vaccination.
• To model the spread of awareness, we utilize various real-world online social
networks to characterize the structure of individuals’ social relationships. There-
after, we further extend Dempster-Shafer Theory (DST) to computationally
model the propagation and evolution of individuals’ beliefs, as well as their
decision-making, having incorporated the awareness obtained from their socially
interconnected neighbors.
We investigate the effect of the spread of awareness on individuals’ vaccination
decision-making with respect to three considered impact factors, based on a series of
simulations of the 2009 Hong Kong H1N1 influenza epidemic. First, the reporting
rates of severe infection and adverse effects of vaccination are used to represent the
frequencies of these topics, which tend to draw public attention. Next, we consider
the coefficient of awareness fading, a parameter used to quantify the information
flows among individuals. Finally, we examine the effect of disease reproduction
number, which corresponds to the severity of an epidemic.
The simulation results show that the reporting rates will determine the number of
vaccinated individuals and the time at which they receive vaccination. A higher
fading coefficient will significantly reduce individuals’ vaccination willingness.
12 1 Paradigms in Epidemiology
1.5 Summary
There are many parallels between epidemiological studies and systems studies.
That is, we can view the problems of epidemiology from a systems perspective. In
systems studies, the goal of modeling is to develop representations or frameworks,
in mathematical or computational languages, that are abstracted from and yet
allow for characterization of certain real-world observations. In epidemiological
studies, the essence of modeling further entails two important aspects: (1) Problem-
driven conceptual modeling, which translates certain real-world problems in an
epidemiological domain into conceptual models in a theoretical or computational
domain. As an example, a metapopulation-based compartmental model may be
utilized to describe disease dynamics in an age-structured host population. (2) Data-
oriented real-world grounding, which requires us to discover ways to embody
the developed conceptual models, i.e., model parameterization, by obtaining and
utilizing real-world data and statistical analysis of the real-world observations. For
instance, the effect of age-specific population contact patterns may be inferred from
data by utilizing a computational method. Furthermore, we may parameterize and
validate the contact-based model with a real-world scenario, such as the 2009 Hong
Kong H1N1 influenza epidemic.
In this chapter, we provide an overview of the most essential models for
characterizing infectious disease transmission, namely, epidemic models and con-
tact relationship models. By doing so, we show how some of the questions in
infectious disease epidemiology can be approached by modeling and characterizing
the dynamics of infectious disease transmission [55].
(a) (b)
Metapopulation level
Susceptible
Infectious
Recovered
(c)
Fig. 2.1 Epidemic models: (a) population-level compartmental models, (b) individual-level
network-based models, and (c) metapopulation-level models
the number of contact relationships one has will determine his/her corresponding
chance of getting infected and, hence, the route of disease transmission through the
network.
Network-based models provide a way of incorporating individual-level het-
erogeneity. However, they also have limitations. For example, to construct a
contact network of disease transmissions, one must know in advance the detailed
characteristics of each individual and the disease-associated contact relationships.
In practice, it is difficult, if not impossible, to have such prior knowledge, sometimes
even for a small group of host individuals.
Thus far, existing studies have addressed the problem of representing individuals’
contact relationships primarily through statistical means, by collecting and analyz-
ing empirical data on individuals’ contact activities in certain places/regions for a
fixed time period [69]. These studies may provide static and empirical descriptions
of individuals’ contact relationships (who meets whom, when, where, and how
often). However, this approach also has limitations, as follows: (1) it is hard to define
the type of contacts responsible for disease transmission, because this depends on
the type of disease; (2) it is not always feasible to have such an empirical description
of contact relationships for a certain host population; and (3) individuals’ actual
contacts may change during the spread of an infectious disease, invalidating the
empirical knowledge basis on which disease transmissions are predicted.
In this regard, the challenge lies in the lack of accurate and reliable descrip-
tions of individuals’ contact relationships, without which the accuracy of model
predictions cannot be improved. As discussed in the following subsection, one
promising alternative is to develop a computational method to infer individuals’
contact relationships from the socio-demographic data of the population. Such
inferred contact relationships can serve as a good foundation for modeling age-
specific disease dynamics and the corresponding intervention strategies.
Metapopulation-Level Models
As aforementioned, population-level compartmental models based on differential
equations provide a deterministic description of the dynamics of disease transmis-
sion in a host population, and have the advantage of being tractable. However, the
assumption that individuals are homogeneously mixed neglects the variations in
the underlying disease transmission. In contrast, individual-level models based on
contact networks do address the host population’s heterogeneity, but the disease
dynamics resulting from a stochastic process are often intractable and sensitive to
the specific settings for simulations. In addition, it remains challenging to construct
realistic contact networks.
As a solution to bridge the gap between population-level compartmental mod-
els and individual-level network models, metapopulation-level models have been
developed, which modify the conventional compartmental models by further taking
into account the structure of a host population (Fig. 2.1c) [70, 71]. These models
address the heterogeneity of the host population by subdividing it into several
subpopulations according to certain characteristics of the individuals, such as their
age, occupation, and geographic location [72]. Once this is done, the infection
20 2 Computational Modeling in a Nutshell
dS
= −λS
dt
dI
= λS − γ I (2.1)
dt
dR
= γI
dt
where S, I , and R denote the numbers of individuals in the susceptible, infectious,
and recovered/immunized compartments, respectively. The parameter λ is the
infection rate, which denotes the proportion of the susceptible population that will
become infected during the present time step. The parameter γ is the recovery rate,
which describes the rate at which the infectious individuals recover and, therefore,
become immunized from secondary infection. In detail, λ can be viewed as a
composite of three factors:
I
λ=β· ·α (2.2)
N
where α describes the infectivity of infected individuals and β is the susceptibility
of uninfected individuals. N is the total size of the host population. In this case, the
basic reproduction number R0 can be computed as follows:
α·β
R0 = (2.3)
γ
2.2 Modeling Contact Relationships 21
dSi
= −λi Si
dt
dIi
= λi Si − γi Ii (2.4)
dt
dRi
= γi Ii
dt
where γi represents the rate of recovery, corresponding to the duration of disease
infection. λi is the infection rate, which denotes the probability of being infected for
susceptible individuals in age group i. For each time step, λi can be calculated as
follows:
n
Ij
λi = μ · βi · cij · αj · (2.5)
Nj
j =1
where cij describes the contact frequency between individuals in age groups i and
j , αj measures the infectivity for individuals in group j , which is the probability
of disease transmission when an infectious individual has contacts with other
susceptible individuals, βi denotes the susceptibility for individuals in group i,
which represents the probability of being infected when a susceptible individual is
exposed to infectious contacts, and μ is a constant disease transmission rate for all
age groups and can be estimated from R0 in the initial stage of disease transmission.
(STDs) [79, 80], as well as air-borne diseases [81, 82]. However, although both
infection tracing and contact tracing can provide a contact network to represent
the potential routes of disease transmission, such a network covers only a subset
of the host population, as it is focused on the contact relationships immediately
surrounding the infected individuals.
• Survey-Based Studies
Survey-based studies record individuals’ contact activities in detail for the
entire host population. For example, Mossong et al. collected diaries from
individuals in eight European countries that recorded the characteristics of those
individuals’ daily contacts with reference to their contactees’ age and sex, and
the location, duration, frequency, and occurrence of physical contacts [83]. Read
et al. launched a detailed diary-based survey of contact in terms of a group of
adults’ casual and close contact encounters [84]. Hens et al. carried out a 2-day
population survey in Belgium to mine social contact patterns for epidemic models
[85]. Such survey-based studies have provided comprehensive descriptions of
individuals’ contact patterns, which can be used to improve the accuracy of
epidemic models for infectious disease control.
The above-mentioned empirical studies can provide quantitative descriptions
of real networks of individuals’ contact relationships. However, there are several
limitations to epidemic modeling and disease control. First, it is difficult to define
the connections of a contact network in a way that generally represents the routes of
disease transmission, as the types of contact responsible for infection are disease-
dependent; e.g., influenza spreads through the air, while HIV is transmitted via
sexual contact. Next, it is time- and labor-intensive (and thus impractical) to
obtain detailed information from the entire population with respect to their contact
behaviors. Therefore, it is not always feasible to acquire such an empirical character-
ization of contact patterns within a specific host population. Further, survey-based
studies can only provide static and empirical descriptions of individuals’ contact
relationships. Finally, individuals’ potential behavioral changes, in response to
both disease transmission and adopted disease interventions, are not taken into
account by surveys, although these can significantly affect the dynamics of disease
transmission.
• Contact Networks
Contact networks are essentially graphs, where a graph is a collection of nodes,
which are joined by a set of connections, called links. Each link denotes a
relationship or an interaction between the nodes it joins. Contact networks
can be further categorized as undirected or directed, and the directedness of a
network is highly epidemiologically relevant, as it indicates the possible paths of
propagation of a contagion.
• Contact Matrix
In a contact network of size N (i.e., with N nodes), a compact way to specify
all contact relationships is to utilize an adjacency matrix C, in which elements
cij = 1 if a link connects nodes i and j , and zero otherwise. Such a
matrix C is also known as a contact matrix, and can be used to describe the
individuals’ contact relationships, such as the occurrence of pairwise contacts or
the frequencies of their contacts [52]. Both contact networks and contact matrices
are computational characterizations of individuals’ contact relationships. Note
that contact networks that are undirected or directed correspond to contact
matrices that are symmetric or asymmetric, respectively.
Computational methods to infer contact networks from the demographic and
social characteristics of a host population are in practice very useful for infec-
tious disease modeling. For example, in studying the contact network of STDs,
simulation-generated networks have been used to characterize the patterns of
observed contact relationships, e.g., the role of hub individuals with a large number
of partners [86, 87]. Potterat et al. described a risk network for individuals with
HIV infection in Colorado Springs, U.S.A., by analyzing the community-wide
HIV/AIDS contact-tracing records during the 1980s and 1990s, i.e., sexual and
drug-injection partners [88].
In modeling outbreaks of air-borne diseases, simulation-generated contact net-
works have been used to capture the demographic and social characteristics of
the considered host population [89]. For example, Halloran et al. simulated the
stochastic spread of smallpox in a community of 2000 people, in which each
individual was generated from population’s age distribution and household size
that can be derived from census data [90]. Meyers et al. investigated the dynamics
of SARS by generating a contact network based on the social setting in the city
of Vancouver [91]. Eubank et al. explored dynamic bipartite graphs to model
the physical contact patterns corresponding to individuals’ movements between
different locations [92, 93].
In this section, we examine a case study of the 2009 Hong Kong H1N1 influenza
epidemic. We discuss how to characterize the heterogeneous contact relationships
within an age-structured host population. Note that in this case, an empirical
2.3 Case Study 25
In Hong Kong, the first H1N1 influenza (also known as human swine influenza,
HSI for short) infection case was an imported case confirmed on May 1, 2009
[94]. The first reported local case (i.e., the first indigenous HSI infection without
an epidemiological link to imported patients) was laboratory-confirmed on June 10,
2009. As of September 2010, there were over 36,000 laboratory-confirmed cases
of HSI in Hong Kong [95]. Figure 2.2a shows the daily number of newly reported
H1N1 infection cases over the period of 200 days since the disease onset in early
May 2009, as reported by the Centre for Health Protection (CHP) of Hong Kong
[96].
Figure 2.2b further gives the proportions of disease infections in different
age groups. In what follows, we present a data-driven computational model
for investigating the dynamics of disease transmission among those age-specific
subpopulations.
(a)
(b)
Fig. 2.2 The 2009 Hong Kong H1N1 influenza epidemic. The confirmed cases of the H1N1
infection reported by the Centre for Health Protection (CHP) of Hong Kong for the first 200 days
since the disease onset. (a) The dynamics of disease transmission in terms of the daily number of
newly infected cases. (b) The proportions of reported infections in different age subpopulations
represent children mixing with adults, mainly in households, and a wider contact
“plateau” among adults, which accounts for contacts occurring in workplaces.
Individuals’ cross-age contacts exhibit specific patterns that correspond to the
likelihoods of individuals’ mixing together within certain social settings (school,
household, workplace, and general community), which in turn depend on the
socio-demographic structure of the population (age distribution, school attendance,
2.3 Case Study 27
household size, and working population). These patterns suggest that disease
transmissions through social contacts mainly occur in certain typical social contact
settings. In addition, individuals’ social contacts may change due to either their self-
initiated behaviors (e.g., avoidance of public places) or governmental compulsory
policies (e.g., school closures and workplace shutdowns).
Based on the above empirical findings, a reasonable strategy is to devise a
computational means for directly inferring individuals’ setting-specific contact
patterns from the demographic structure of the considered population. Then, these
setting-specific patterns can be used to estimate the overall social contacts that
account for disease transmissions, by combining the patterns via their respective
coefficients, to reveal the proportions of individuals’ contacts within different social
settings.
Specifically, we can define and compute the contact frequency between a pair of
individuals in age groups i and j , i.e., cij , as the total number of contacts between
two age groups, Cij = Cj i , divided by the product of their population sizes, Ni and
Nj :
Cij Cj i
cij = = = cj i (2.6)
Ni Nj Nj Ni
Note that matrix C with the elements of cij is the overall contact matrix,
which describes individuals’ cross-age contact frequencies. Based on this definition,
matrix C is symmetric for cij = cj i .
Next, we calculate the probability for individuals of different ages mixing
within certain social settings, i.e., individuals sharing the same places, namely
households, schools, workplaces, and general communities. Then, we generate four
matrices accounting for the specific patterns of individuals’ contacts within each
social setting, which are represented by CH for contacts within households, CS
for schools, CW for workplaces, and CG for general communities. Thus, we can
estimate the overall matrix of individuals’ cross-age contact frequencies as a linear
combination of the four setting-specific matrices:
C = r H CH + r S CS + r W CW + r G CG (2.7)
rH + rS + rW + rG = 1 (2.8)
Household School
(a) (b)
Workplace
(c) (d)
(e) (f)
Fig. 2.3 Contact patterns inferred from the census data of Hong Kong. We consider disease
transmissions among individuals between 0 and 85+ years old and divide these into 18 age groups.
The contact matrices are generated corresponding to the likelihoods of individuals’ mixing within
respective social settings: household (CH ), school (CS ), workplace (CW ), and general community
(CG ). The overall contact matrix is calculated as the linear combination of the four setting-specific
contact matrices. The combination coefficient of each matrix denotes the proportion of effective
contacts occurring in that social setting
2.3 Case Study 29
Specifically, Fig. 2.3a describes the contacts in households, in which the main
diagonal and two secondary diagonals correspond to the contacts within families and
between parents and children. Figure 2.3b shows the pattern of contacts in schools,
in which the strong diagonal elements among individuals below 20 years old
indicate that students are more inclined to have contact with same-age individuals.
Figure 2.3c presents the pattern of contacts in workplaces, in which the contacts are
more frequent among individuals aged between 20 and 65. Figure 2.3d gives the
pattern of individuals’ random contacts with each other in general communities.
We normalize the elements of the four generated contact pattern matrices, such
that their total numbers of contacts are equal. For the overall contact matrix (see
Fig. 2.3e), the coefficients used for combining the setting-specific matrices, rΦ and
Φ ∈ {H, S, W, G}, can be approximately estimated as the fraction of disease
infections occurring in the respective social settings.
Based on the population size of each age group (Fig. 2.3f), it has been shown that
31% of infections during the 2009 Hong Kong H1N1 influenza epidemic occurred
in households [98]. In addition, we assume that the other three contact matrix
coefficients are identical to those empirical estimations for the respective social
settings [99, 100], as follows: 0.24 in schools, 0.16 in workplaces, and 0.29 in
general communities.
2.3.3 Validation
(a)
(b)
Fig. 2.4 The baseline dynamics of disease transmission. We calibrate our epidemic model based
on the real-world scenario of the 2009 Hong Kong H1N1 influenza epidemic, as reported by the
Centre for Health Protection (CHP) of Hong Kong for the first 200 days since the disease onset.
(a) The dynamics of disease transmission in terms of the daily number of newly infected cases
reported as a proportion of the total number of infections. (b) A comparison of the observation and
model predictions in terms of disease infection rates by age
2.4 Further Remarks 31
and school-age students between 0 and 19 years old constituted a large proportion of
the infection cases, while that of adults was relatively small, as shown in Fig. 2.4b.
2.5 Summary
In the control and prevention of infectious diseases, one of the most effective
measures is vaccination. A key question for public health authorities is the optimal
allocation of a finite number of available vaccine doses to most effectively reduce
disease incidence. This is closely related to the question of which age groups or
subpopulations will be most vulnerable and should be vaccinated first. Thus, if
we can answer this question, we will be in a strong position to combat infectious
diseases.
In this chapter we focus on the above questions. Building on the age-structured
compartmental model of infectious diseases introduced in the preceding chapter,
we discuss a computational means for identifying and prioritizing the target
subpopulations for effective vaccine allocation [55]. Through a series of simulation-
based experiments, we examine the performance of such a vaccine allocation
strategy by considering different epidemic scenarios as well as other intervention
strategies.
3.1 Vaccination
Vaccination has long been recognized as one of the most effective methods for the
control and prevention of infectious diseases. In this section, we review the basic
notions and existing methods of vaccination. Specifically, we discuss the principal
idea behind vaccination, in terms of achieving the herd immunity effect. We also
look at some of the existing vaccination strategies that have been designed for
disease control and prevention.
Generally, the effect of vaccination is judged in two respects: (1) how effectively
it can directly immunize vaccinated individuals; and (2) how effectively it can
indirectly protect unvaccinated individuals from getting infected by reducing disease
transmission. To discuss effectiveness, we need to further introduce the concept of
herd immunity, which is often mentioned in the literature although with several
slightly different meanings. Here, we adopt the definition as given by John et
al. [111], which is that herd immunity refers to “the proportion of subjects with
immunity in a given population,” which may be due to natural recovery from
infection, vaccine-induced immunization, or a combination of both. In essence, herd
immunity indicates the proportion of immunized individuals in a host population
who will be able to resist the spread of a disease. Due to herd immunity, it will not
be necessary for everyone in a host population to be vaccinated to prevent outbreaks,
because the size of the susceptible population has been reduced, and so has the
probability of disease transmission from the infectious to the susceptible. This
indirect protection of unvaccinated individuals, as a result of vaccine-induced herd
immunity, is often referred to as the herd immunity effect, which is schematically
illustrated in Fig. 3.1.
Next, we introduce the notion of a herd immunity threshold. This indicates the
critical portion of a host population that must be vaccinated/immunized to prevent
disease transmission. Studies on a herd immunity threshold have focused primarily
on how to determine the minimum vaccination coverage needed to prevent a poten-
tial outbreak. It should be noted that in epidemiology, the prevention of an outbreak
entails that the average number of secondary infections per infectious individual
should be less than one, which is equivalent to saying the basic reproduction number
R0 < 1.0 (see also Sect. 2.1.1). In this regard, the concept of a herd immunity
threshold reflects the vaccination coverage that leads to R0 = 1.0. In a standard
SIR model, which assumes that individuals of a host population have uniform and
random contacts with each other, the herd immunity threshold, denoted by θ , for
random vaccination (assuming 100% vaccine effectiveness), can readily be written
as follows:
1
θ =1− (3.1)
R0
As can be seen, the usefulness of the notion of the herd immunity threshold is that
it provides a way to examine the effectiveness of vaccination programs for disease
control, that is, to evaluate the level of vaccination coverage in a host population.
Similar results for vaccination in different stages were also reported by Myliusa et
al. [123] and Bansal et al. [124].
Generally, the methods outlined in the preceding chapter adopt a fixed objective
function to be optimized and an optimal strategy to be computed for the given period
of time. To function, these methods make several assumptions. Specifically, they
assume we are able to make accurate predictions about disease dynamics, and that
we know in advance the number of available vaccine doses (needed to set up the
constraints) and the timing of vaccine release (needed to determine the optimization
time period). In real-life applications, such assumptions are evidently rather strong,
as we generally have limited knowledge about disease-associated epidemiological
features, e.g., infectivity and transmissibility, during the spread of an emerging
disease. In addition, we may not know the quantity and timing of vaccine supply.
As a result, the optimization-based methods may not be practically feasible.
Several recent studies have been performed to address the vaccine allocation
problem by focusing on disease transmission rather than on future predictions.
Importantly, the authors of these studies allowed vaccine allocation to be dynam-
ically/adaptively adjusted in relation to the dynamics of disease transmission,
e.g., based on the time-varying, age-specific incidences of infection revealed from
surveillance data [125], or the real-time monitoring of hospitalization and infection-
induced death [126]. The findings of these studies have partly solved the problem
of obtaining detailed knowledge of an emerging disease (i.e., the parameters
used in an epidemic model). For example, proxy indicators may be used for
determining vaccine allocation, e.g., the group-specific confirmations of infection,
or the hospitalization and mortality rates [127].
So far, in this section we have reviewed some of the key concepts and methods
used in planning and evaluating the effectiveness of vaccination programs in terms
of vaccination coverage in a host population. Another important issue is that
the actual coverage of a vaccination program could be affected by individuals’
willingness to be vaccinated (for more details, see Chaps. 4–6).
In a real-world situation, the vaccine supply is usually limited. This is partly because
the quantity of available vaccine doses may be insufficient to meet the actual
demand, and partly because the vaccine’s release after the disease onset may be
delayed. Thus, an important concern for public health authorities is to achieve the
goals of vaccination by making the best use of the finite number of available vaccine
doses. For example, in the United States, the National Vaccine Advisory Committee
(NVAC) and the Advisory Committee on Immunization Policy (ACIP) have set the
goals of vaccination as being to weaken health effects, including severe morbidity
and death, and minimize socio-economic effects [128]. Accordingly, the NVAC and
ACIP have recommended a priority vaccination for vaccine workers, health-care
providers, and the ill elderly, followed by healthy people aged 2–64 [129].
3.3 Age-Specific Intervention Priorities 37
In this book, we consider vaccine allocation for the purpose of most effectively
minimizing disease transmission in the whole population. In this case, the key
problem is how to determine the relative vaccine allocation priority for each of the
subpopulations corresponding to their respective roles in disease transmission. That
is, the challenge for effectively allocating a finite number of vaccine doses is herein
translated into the question of how to adaptively adjust vaccine allocation to various
subpopulations with reference to the immediate situation of disease transmission,
such as the disease incidence and prevalence rates in different subpopulations,
the possible changes of individuals’ contact relationships, and the total number of
available vaccine doses and the time of vaccine release.
During the 2009 Hong Kong H1N1 influenza epidemic, the government announced
the immediate closure of all primary schools and kindergartens when the first non-
imported case was confirmed, and at the same time, targeted children between
the ages of 6 months and 6 years as the priority groups for vaccination [130].
Relevant to this, a pertinent question is how we can systematically determine the
relative priorities of subpopulations for disease interventions during the time course
of disease transmission, where such intervention measures may include vaccine
allocation, contact reduction, or a combination of both.
The aim of our work is to provide a method to enable optimal prioritization
of subpopulations for disease interventions, using a combination of age-specific
vaccine allocation with setting-specific contact reduction. We evaluate the effects
of disease interventions for containing disease transmission by measuring the
reduction in the reproduction number. By doing so, we show which subpopulations
should be prioritized for vaccination, so as to generate the greatest reduction in the
number of disease transmissions. We do so by considering the marginal effects of
reducing the reproduction number for different cases of vaccine allocation by age,
and reducing contacts by social setting.
For demonstration, we use a compartmental model to describe the dynamics
of an influenza-like disease transmission in an age-structured host population. We
parameterize the epidemic model with the epidemiological data from the 2009 Hong
Kong H1N1 influenza epidemic, and further implement our method to identify
the relative priorities of subpopulations for disease interventions in Hong Kong.
Additionally, we carry out a series of simulation-based experiments with different
settings of disease intervention.
Compared with existing optimization-based approaches, our method has the
following features:
• The strategy utilizes prior knowledge about individuals’ age-specific susceptibil-
ity and infectivity, real-time disease prevalence in each of the subpopulations, and
the basic patterns of individuals’ cross-age contact frequency within each social
38 3 Strategizing Vaccine Allocation
We now introduce the standard epidemic model and our detailed computational
method for identifying the relative priorities of age-specific subpopulations for
intervention measures.
First, we use vector I(k) = [I1 (k), . . . , IN (k)]T to denote the number of
infectious individuals in each of N age groups at the kth generation of disease
infection. Then, we characterize the dynamics of disease infection from generation
k to generation k + 1 as follows:
where K is the reproduction matrix, also known as next generation matrix (NGM)
[131].
For the earlier-mentioned SIR model, the reproduction matrix K can be written
as follows:
where matrix S describes the size of the susceptible population in each age group;
it has elements S1 , S2 , . . . , SN in the diagonal and zeros elsewhere. Matrix B
summarizes individuals’ age-specific susceptibility in the diagonal elements of
β1 , β2 , . . . , βN , and zeros elsewhere. Matrix A gives the age-specific infectivity
of infected individuals with the diagonal elements of α1 , α2 , . . . , αN , and zeros
elsewhere. Matrix C, known as the contact matrix (see Eq. (2.7) in Sect. 2.3.2),
describes the frequency of contacts between two age groups. Moreover, during the
3.3 Age-Specific Intervention Priorities 39
course of disease transmission, the susceptible populations will decrease in size over
time, and therefore matrices S and K will change dynamically.
In epidemiology, the effective reproduction number Rt refers to the number of
new infection cases caused by a typical infectious individual in a completely suscep-
tible population. By constructing the NGM, the effective reproduction number Rt in
the context of an age-structured host population can be approximately estimated as
the dominant eigenvalue of reproduction matrix K:
Rt = ρ (K) (3.4)
Given the condition that matrices S, B, C, and A are all symmetric, Rt can be
approximately calculated as:
Rt = x T1 Ky 1 (3.5)
where x 1 and y 1 are the top left and right eigenvectors of reproduction matrix K (the
corresponding top eigenvalue is Rt ). Specifically, we choose a normalized format of
each eigenvector in which the elements are positive and sum to one. As proposed
by Wallinga et al. [106], y 1 and x 1 approximately correlate to the number of new
infections in each age group, I:
y1 ∝ I
(3.6)
x 1 ∝ S−1 B−1 AI
Due to the lack of knowledge about the reporting rate of disease infections, i.e.,
the ratio of confirmed cases to the overall infections, we approximate the susceptible
population size Si using the population size Ni , based on the assumption that the
number of infections is relatively small in the host population. Therefore, we can
prioritize each age group as follows:
2
dRt αi Ii
∝ (3.11)
dSi βi Ni
Therefore, with the above equation, we can determine the relative priorities
of age groups for vaccine allocation with respect to their age-specific infectivity,
susceptibility, population sizes, and current disease prevalence.
dRt
∝ x T1 SBCψ Ay 1 (3.13)
dr ψ
3.3 Age-Specific Intervention Priorities 41
Then, based on the inferred setting-specific contact matrices, the relative priority
for contact reduction that targets social setting ψ can be computed as follows:
dRt ψ
∝ αi Ii cij αj Ij (3.14)
dr ψ
Thus, with the above equation, we will be able to further estimate the relative
priorities of social settings for contact reduction, having taken into consideration
individuals’ age-specific infectivity, current disease prevalence, and setting-specific
contact patterns.
Next, we consider the case of a campaign against the spread of an infectious disease,
in which multiple intervention measures will be implemented at the same time. In
view of this, we are interested in evaluating the effect of vaccination and contact
reduction being implemented simultaneously. We estimate the marginal reduction of
the effective reproduction number, d 2 Rt , corresponding to simultaneous vaccination
of the susceptible population dS and the reduction of effective social contacts dC:
d 2 Rt ∝ x T1 (dS)B(dC)Ay 1 (3.15)
Specifically, when selecting age group i for vaccination and social setting ψ
for contact reduction, the effect of implementing the two intervention measures on
reducing the effective reproduction number Rt can be evaluated as follows:
d 2 Rt dS
∝ x T1 BCψ Ay 1 (3.16)
dSi dr ψ dSi
That is,
d 2 Rt αi ψ
∝ c α I
ij j j (3.17)
dSi dr ψ Si
In this section, we demonstrate our vaccine allocation strategy using the real-world
scenario of the 2009 Hong Kong H1N1 influenza epidemic, by determining the
relative priorities of the age-specific subpopulations for vaccination.
Figure 3.2 shows the H1N1 scenario in terms of the daily numbers of reported
infections in different age groups during the spread of the disease in Hong Kong.
Accordingly, Fig. 3.3 presents the relative priorities for vaccine allocation in those
age groups during the course of disease transmission.
Generally, individuals between 0 and 29 years old are the most important
subpopulation for containing disease transmission by vaccination. However, for
each specific age group, the identified priorities vary between different stages of
disease transmission. For the first month since the disease onset, up to day 25, we
can observe that individuals between 0 and 29 years of age are targeted as the
top priority subpopulation for vaccination. This describes the situation in which
outbreaks will appear among school-age students due to their high frequency of
contacts. Subsequently, on day 50, it can be observed that the relative priorities
of individuals aged between 0 and 9 and those between 20 and 29 are increased,
while the priorities of individuals between 10 and 19 are reduced. When disease
Fig. 3.2 The numbers of reported infections in different age groups during the 2009 Hong Kong
H1N1 influenza epidemic
3.4 Case Study 43
Fig. 3.3 Relative priorities of age-specific subpopulations for vaccine allocation during the course
of disease transmission
infection peaks near day 120, individuals between 0 and 19 are predicted to become
the dominant priority for vaccination, which agrees with the real-world observation
that children and school-age students accounted for a large proportion of the new
infections in this stage (Fig. 3.2). Finally, in the decay stage of the epidemic,
children between 0 and 9 will become the subpopulation with the highest priority
for vaccination. However, it should be pointed out that vaccination is more effective
in the initial stage of disease transmission than in the stage of decay.
Figure 3.4 shows the relative priorities of the social settings for disease interven-
tion by contact reduction among individuals. Overall, the reduction of individuals’
contacts within schools is identified as the key measure for containing disease trans-
mission throughout the whole period of disease transmission. Disease transmissions
within households and workplaces account for a relatively large proportion during
the initial and the decay stages of disease spread (between day 25 and day 100 and
between day 150 and day 200, respectively), however, when infection peaks around
day 120, interactions within these environments account for a relatively small
proportion of disease transmissions. As mentioned before, the estimated proportion
of infections occurring in schools was not the largest among the four considered
social settings (i.e., the empirical estimation for schools was approximately 24%,
and for households 31%). However, the disease infections that had already occurred
in schools played a significant role in the disease transmission among the host popu-
lation. Therefore, the reduction of individuals’ effective contacts in the social setting
of schools (through school closures as well as school sanitation and disinfection)
should be implemented immediately after the disease onset, and conducted for the
whole period of disease transmission.
44 3 Strategizing Vaccine Allocation
Fig. 3.4 Relative priorities of social settings for contact reduction during the course of disease
transmission
Figure 3.5 further presents the relative priorities of the age groups and social
settings when vaccine allocation and contact reduction are implemented simultane-
ously. Generally, vaccination of individuals between 0 and 19 and contact reduction
in schools are the most important measures for containing disease transmission.
For people of older ages, contact reduction in households, workplaces, and general
communities is more important than contact reduction in schools. Specifically, as
shown in Fig. 3.5a, on day 1 of disease transmission, contact reduction in schools
and vaccination of individuals aged between 5 and 19 should be the top priority
to contain disease transmission. On day 60, as indicated in Fig. 3.5b, individuals
between 15 and 19 are identified as the target subpopulation for vaccination,
followed by individuals aged 10–14 and 5–9. At this stage, contact reduction in
the social setting of schools remains the top priority. When disease infection peaks
at approximately day 120, as illustrated in Fig. 3.5c, the age groups 5–9 and 10–14
are predicted to become the most important subpopulations for vaccination. In the
final stage of disease transmission on day 180, as shown in Fig. 3.5d, vaccinating
children becomes more important.
By comparing the results of disease intervention using only vaccination (Fig. 3.3)
with those using simultaneous vaccination and contact reduction (Fig. 3.5), we
can observe that the age group 20–29 has a higher priority for vaccine allocation
in the case of vaccination only than in the case of adopting both intervention
measures. This is mainly due to the interplay of the two intervention measures, in
that the effects of reducing effective contacts in schools can prevent or delay disease
transmission to other age groups.
3.4 Case Study 45
(a) (b)
(c) (d)
Fig. 3.5 Relative priorities of subpopulations and social settings for simultaneous implementation
of age-specific vaccination and setting-specific contact reduction during the course of disease
transmission
The results in Fig. 3.6 show that vaccination and contact reduction can ameliorate
disease prevalence by reducing the incidence rate at the peak of an outbreak. In the
case of disease intervention by contact reduction only, we can clearly observe from
the time course of the infectious population sizes that reducing individuals’ contacts
in different social settings can lead to distinctly different results. Contact reduction
in schools (the blue solid curve) outperforms those in the other three social settings,
in terms of preventing the occurrence of an infection outbreak and lowering the
incidence rate at the peak of disease prevalence. Contact reduction in households
(the red solid curve) and workplaces (the yellow solid curve) each has a similar
effect on disease control and a better performance than that in general communities
(the green solid curve). These results agree well with our previous prioritization
of social settings for contact reduction, in which schools are identified as the top
priority, followed by households and workplaces.
For the implementation of vaccination and contact reduction simultaneously,
the results are shown as the dashed curves in Fig. 3.6. With contact reduction in
schools, disease transmission would be almost eliminated (the blue dashed curve).
Vaccination when combined with contact reduction in households and workplaces
performs better than contact reduction in general communities in reducing disease
prevalence, which, in turn, is still better than vaccination only. In addition, the
Workplace only
0.008 Community only
Vaccine only
Vaccine & household
Vaccine & school
Vaccine & workplace
Vaccine & community
0.004
0
0 50 100 150 200
Days
Fig. 3.6 Disease dynamics under the disease interventions of age-specific vaccination and setting-
specific contact reduction. The baseline dynamics of disease transmission without any intervention
(black solid curve); contact reduction only in schools (blue solid curve), households (red solid
curve), workplaces (yellow solid curve), and general communities (green solid curve); vaccination
only (black dashed curve); vaccination and contact reduction in schools (blue dashed curve),
households (red dashed curve), workplaces (yellow dashed curve), and general communities (green
dashed curve) (Color figure online)
3.5 Further Remarks 47
implementation of contact reduction in schools only (the blue solid curve) leads
to a lower incidence rate at the peak of disease outbreak than the simultaneous
implementation of vaccination and contact reduction in general communities (the
green solid curve), and also delays the time of disease outbreak.
The two disease intervention measures studied here are age-specific vaccine allo-
cation and social setting-specific contact reduction. Unlike previous studies in
which statistical means are utilized to describe individuals’ contact relationships,
in this chapter we have discussed how to infer setting-specific contact matrices
by decomposing individuals’ actual disease transmission-related contacts into their
contacts within four specific social settings, i.e., school, household, workplace,
and the general community. Therefore, the contact-reduction method of disease
intervention can be interpreted as the reduction of the proportion of individuals’
contacts in a certain social setting. Furthermore, the changes of individuals’ overall
contacts can be interpreted as the changes of contact proportions within different
social settings. We have also examined the reproduction number to evaluate the
effects of implementing different intervention measures on containing disease
transmission. Therefore, we can identify the priority subpopulations (in terms of
age groups and social settings) based on the marginal reduction of the reproduction
number caused by changes of the susceptible population sizes by age (vaccination)
and of the contact proportions by social setting (contact reduction).
We have evaluated our method with respect to the real-world scenario of the
2009 Hong Kong H1N1 influenza epidemic and, thereafter, examined the relative
priorities of subpopulations for age-specific vaccination and setting-specific contact
reduction in Hong Kong. Our study has practical value for public health authorities
in preparing and assessing their intervention measures for controlling an infectious
disease. First, the age distribution of new infections will always be available from
an epidemic surveillance system, e.g., the CHP in Hong Kong. Next, the basic
patterns of individuals’ contacts in different social settings will depend mainly on
the socio-demographic characteristics of the population, which can be derived either
through statistical means or by computational methods from the census data for
the host population. Finally, disease control will be more effective when multiple
intervention measures are implemented simultaneously.
It should be pointed out that the results from our method partly depend on
the accuracy of the age distribution of new infections reported by the surveillance
system. In addition, other potential factors that also affect the results may include the
reporting rates of infection, which may vary for individuals in different age groups
due to their physical and biological conditions, and the time needed for the case
confirmations, which may lead to a delayed response to disease transmission.
48 3 Strategizing Vaccine Allocation
3.6 Summary
Although vaccination has long been regarded as one of the most effective methods
in controlling infectious diseases, public concerns about the safety and efficacy of
vaccines will significantly affect the effective coverage of a vaccination program.
In the real world, it remains highly challenging to achieve adequate and lasting
vaccination coverage in a host population, especially for a voluntary vaccination
program. An example is the decline of MMR vaccine uptake in Britain [132],
subsequent to a controversial study associating the vaccine with the development
of autism. Even though other studies contested this association, the public remained
doubtful about the safety of the MMR vaccine, leading to a decrease in vaccine
uptake, which was in turn followed by outbreaks of measles [132]. Similarly, public
losses of confidence in the safety of pertussis vaccine have led to declines in
its uptake in many countries, subsequently resulting in a series of large pertussis
outbreaks [133].
In the remaining three chapters, we aim to explain how individuals make their
decisions once a voluntary vaccination program becomes available, and what could
be the dominant factors in their decision-making [53, 54]. Specifically, we focus our
discussions on the following three key aspects:
• Risks and Benefits of Vaccination
Vaccination immunizes susceptible individuals to prevent infection. However,
vaccinated individuals also face the potential risk of adverse effects of the
vaccine, which may sometimes lead to severe complications. In this regard,
individuals making vaccination decisions typically focus on several determinants
associated with the risks and benefits of vaccination, such as the perceived risk
of disease infection, the perceived safety and efficacy of the vaccine, and the
socio-economic costs associated with vaccination and disease infection (e.g.,
the monetary cost of vaccination, medical expenses for treatment in the case of
infection, and possible absence from work[134]).
The perceived risks affecting individuals’ decision-making are twofold: the risk of
being infected for unvaccinated individuals, and the risk of a vaccine’s adverse
effects for vaccinated individuals. It has been found that individuals who worry
about being infected or believe themselves vulnerable to infection will be more
inclined to accept vaccination, and vice versa [140, 141]. Similar patterns have also
been found for individuals’ intention to be vaccinated when they felt that a pan-
demic would be severe and long-lasting [142]. Moreover, studies have shown that
individuals with positive attitudes to vaccination, and who believe that vaccination
can reduce their risk of infection, were motivated to be vaccinated [143, 144]. In
contrast, individuals who worried about the vaccine’s potential adverse effects and
doubted its efficacy were found to have a lower level of vaccine uptake [145, 146].
4.2 Game-Theoretic Modeling of Vaccination Decision-Making 51
Game theory provides a useful tool for characterizing the dilemma associated with
voluntary vaccination. That is, for each individual, it would be better not to take
the risk of vaccination, while benefiting from the herd immunity generated from
the rest of the population keeping the vaccination coverage high. Game-theoretic
analysis assumes that individuals possess perfect rationality to maximize their own
gains by adjusting their decision of whether to vaccinate. If applied to a vaccination
program, game theory implies that the host population’s individuals will adjust their
willingness to vaccinate by balancing the costs and benefits associated with the
status of a spreading epidemic and the safety and efficacy of vaccines.
There are two types of costs associated with an individual’s vaccination decision:
(1) the cost of vaccination (e.g., the potential risk of a vaccine’s adverse effects or
the expense of vaccine administration); and (2) the cost of infection if not vaccinated
(e.g., disease complications, expenses for treatment, or absence from work). We let
ξ and ζ denote the costs associated with vaccination and infection, respectively, and
use λ̂i to represent the perceived risk of being infected by a susceptible individual
i (see the definition of λi in Sect. 2.1.2). Then, we can introduce a cost function for
individual i with a decision σi , as follows:
where ξ denotes the cost associated with being vaccinated, and λ̂i · ζ denotes the
cost associated with rejecting vaccination.
52 4 Explaining Individuals’ Vaccination Decisions
Game-theoretic analysis
Risk of infection: ˆ
Decision equilibrium
Cost ratio: rc
Fig. 4.1 A decision process for characterizing individuals’ voluntary vaccination. A group of
socially interconnected individuals can make decisions by minimizing all of the associated costs
Next, without loss of generality, we let rc = ξ/ζ describe the ratio of ξ and ζ .
Thus, we can further transform the cost function Fi (·) into the following:
As illustrated in Fig. 4.1, we assume that individuals estimate the risk of disease
infection based on their perception of disease severity, as reflected in the perceived
disease transmission rate, β̂, as well as their neighbors’ vaccination decisions, as
represented by Nivac and Ninon for the numbers of neighbors who make the decision
to vaccinate or not, respectively. In addition, vaccinated individuals are assumed to
be successfully immunized from disease infection, while unvaccinated individuals
can be infected and thus transmit the disease. Therefore, the perceived infection risk,
λ̂i , can be computed corresponding to the proportion of unvaccinated neighbors as
follows:
Ninon
λ̂i = β̂ · (4.3)
Nivac + Ninon
vaccination (σi = 1) if rc < λ̂i , reject vaccination (σi = −1) if rc > λ̂i , and keep
his/her decision unchanged from the previous step if rc = λ̂i . We can express this
cost-minimized choice of individual i, σ̂i , in the following form:
⎧
⎨ +1, if rc < λ̂i
σ̂i = −1, if rc > λ̂i (4.4)
⎩
unchanged, if rc = λ̂i
If all individuals follow the same strategy of minimizing their cost functions, after
a certain number of iterations of decision-making a steady state will be reached, in
which all individuals will have no incentive to change their decision in the next step.
Fig. 4.2 The vaccination thresholds for preventing disease outbreaks, with respect to: (1) R0 =
1.2; (2) R0 = 1.6; and (3) R0 = 2.0
4.4 Further Remarks 55
8 .0
4 6
Reproduction number
Fig. 4.3 Vaccination coverage at the steady state of individuals’ decision-making. We investigate
the effect of individuals’ decision-making being motivated by cost minimization, with this effect
measured by the resulting vaccination coverage, through varying the values of cost ratio rc between
0 and 1.0. The disease severity (basic reproduction number R0 ) is set as: (1) R0 = 1.2; (2) R0 =
1.4; (3) R0 = 1.6; (4) R0 = 1.8; and (5) R0 = 2.0
Game-theoretic analysis has been used in various studies to investigate the collective
effect of individual-level vaccination choices on the population-level vaccination
coverage, that is, the coverage of vaccination achieved at the equilibrium state of
individuals’ strategy adjustment. It has been found that the vaccination coverage
resulting from individuals’ collective self-interested choices will always be below
the optimal level of vaccination for the whole population (i.e., the threshold
of vaccination for the eradication and elimination of disease infection). Specific
examples are the analysis of vaccination campaigns against smallpox [47], seasonal
influenza [48], and more recently the H1N1 epidemic [161]. In the real world,
individuals’ choices deviate greatly from those that would result from perfect
rationality.
56 4 Explaining Individuals’ Vaccination Decisions
4.5 Summary
Fig. 5.1 An integrated decision process for characterizing individuals’ voluntary vaccination. We
extend the existing game-theoretic analysis by incorporating the effect of social influences. By
doing so, we can investigate the steady state of individuals’ decision-making and examine the
effect of social influences on voluntary vaccination and, hence, the effectiveness of disease control
5.1 Social Influences on Vaccination Decision-Making 59
= Ninon / ·
12 2
ιnon
i wij (5.2)
j ∈Ninon
We use σ˜i to denote the formalized social opinion resulting from the social influ-
ences of individual i’s neighbors. As a modification of the standard SIT definition
(where σ˜i corresponds to the opinion backed by stronger social influences), σ˜i , being
either acceptance or rejection of vaccination, will be determined by comparing the
influences of the two opposite opinions. We let Διi denote the discrepancy between
ιvac
i and ιnon
i . Then, we normalize Διi as follows:
ιvac − ιnon
Διi = i i
. (5.3)
ιi + ιnon
vac
i
60 5 Characterizing Socially Influenced Vaccination Decisions
Fig. 5.2 A decision process for characterizing individuals’ voluntary vaccination. We extend the
existing decision analysis by incorporating the effect of social influences. A group of socially
interconnected individuals can make decisions by evaluating the decisions of others. We utilize
SIT to characterize the effect of social influences on individuals’ decision-making with reference
to their interaction relationships
5.1 Social Influences on Vaccination Decision-Making 61
Δι
0 0.5
Δ
Fig. 5.3 Social opinion based on the influence of two opposite opinions. For individual i, σ̃i
is formalized as acceptance of vaccination with the probability P (Διi ) or otherwise with the
probability 1 − P (Διi ). In the Fermi function, ν denotes individuals’ responsiveness to the
discrepancy Διi
where P (Διi ) denotes the probability that social opinion Διi is to accept vaccina-
tion, and 1 − P (Διi ) to reject vaccination. Here P (Διi ) is computed from the Fermi
function as follows:
1
P (Διi ) = (5.5)
1 + exp (−ν · Διi )
The Fermi function is a sigmoid function that has been widely used for describing
individuals’ behavioral changes as a response to the payoff discrepancy between
two different choices. Here, ν describes individuals’ responsiveness to the effect
discrepancy, i.e., the difference in effect of the two opposite opinions. As shown
in Fig. 5.3, a larger value of ν means that the choice backed by a stronger
social influence will be more likely to dominate social opinion even if the effect
discrepancy, Διi , is relatively small.
Next, we introduce a probability, rf , termed the individual’s conformity rate, that
indicates the degree of individuals’ tendency to adopt the social opinion of his/her
interconnected neighbors, which corresponds to how likely individual i is to convert
his/her cost-minimized choice (σ̂i ) to the social influence-formalized opinion (σ̃i ).
62 5 Characterizing Socially Influenced Vaccination Decisions
σ̃i , with rf
σi = (5.6)
σ̂i , with 1 − rf
(b)
(c)
the associated costs. In this simulation setup, when rf = 1.0, the vaccination
coverage at the steady state of decision-making converges to approximately 2% for
individuals’ vaccination willingness at the initial level of 30% (Fig. 5.4a), 50% at
the level of 45% (Fig. 5.4b), and 97% at the level of 60% (Fig. 5.4c).
In addition, the effect of varying conformity rate rf (individuals’ tendency to
be swayed by social opinions) can also be observed from the changing vaccination
decisions depending on vaccination-associated costs (cost ratio rc ). When individ-
uals become more susceptible to social influences (gradually increasing conformity
rate rf ), as shown in Fig. 5.4a, the effect of such influences tends to increase
the vaccination coverage when the cost of vaccination is low (0 < rc ≤ 0.5).
Conversely, when the cost of vaccination is relatively high (0.5 < rc ≤ 1.0), the
effect of social influences will reduce the resulting vaccination coverage at the
steady state of individuals’ decision-making. Furthermore, when the conformity
rate rf approaches 1.0, the vaccination coverage will decrease/increase sharply
and finally converge to a fixed level that depends on individuals’ initial level of
vaccination willingness.
Based on the earlier-mentioned SIR model (see also Sect. 2.1.2), we investigate
the effect of social influences on disease control by evaluating disease infection rates
(the percentage of individuals being infected as a result of disease transmission) with
respect to different vaccination coverage levels resulting from individuals’ decision-
making.
Figure 5.5 shows the disease infection rates with respect to the interplay of
individuals’ cost minimization and the effect of social influences on vaccination
decision-making (the values of cost ratio rc and conformity rate rf each ranging
from 0 to 1.0). Regarding our considered epidemic scenario (basic reproduction
number R0 = 1.6), the simulation results in Fig. 5.5a–c show that disease infection
can be eliminated given a relatively low cost of vaccination (cost ratio 0 < rc ≤ 0.8)
and a moderate effect of social influences (conformity rate 0 < rf ≤ 0.6).
Specifically, when individuals are less susceptible to social influences (confor-
mity rate rf < 0.6), the effectiveness of disease control is determined by the
relative cost of vaccination (cost ratio rc ) in that a lower vaccination cost leads
to a reduction in the disease infection rate, due to higher vaccination coverage.
However, as individuals’ tendency to be affected by social influences increases
(conformity rate 0.8 ≤ rf < 1.0), the effect of vaccination cost on disease control
is weakened accordingly, while individuals’ initial level of vaccination willingness
matters more. In the extreme case of rf = 1.0 (i.e., individuals are absolute
followers of social influences), the disease infection rate increases to 46% for the
initial level of vaccination willingness of 30%, as shown in Fig. 5.5a. If the initial
level of vaccination willingness is set as 45%, as shown in Fig. 5.5b, the infection
rate will be higher than in the situation with the initial level of 60%, as shown in
Fig. 5.5c, where cost ratio rc > 0.8 and conformity rate 0.2 ≤ rf < 0.8.
In addition, we examine the steady-state vaccination coverage and the resulting
infection rate corresponding to different initial levels of individuals’ vaccination
willingness prior to their decision-making, the results of which are shown in Fig. 5.6.
We note that individuals’ initial level of willingness affects the converged level
5.2 Case Study 65
(b)
(c)
Conformity rate rf =1
1
0.8
Vaccin tion coverage
0.6
0.4
0.2
0
0.2 0.4 0.6 0.8 1.0
Initial vaccination willingness
(a)
Conformity rate rf =1
1
0.8
Disease attack rate
0.6
0.4
0.2
0
0.2 0.4 0.6 0.8 1.0
Initial vaccination willingness
(b)
Fig. 5.6 The effect of individuals’ initial levels of vaccination willingness when individuals are
all social followers (conformity rate rf = 1.0). (a) Vaccination coverage at the steady state of
individuals’ decision-making. (b) The resulting effects on epidemic control in terms of disease
attack rate
5.3 Further Remarks 67
in Fig. 5.6a. That is, in the situation of individuals being absolute social followers,
there is a threshold value of individuals’ initial level of vaccination willingness,
which can be used to evaluate the effectiveness of a voluntary vaccination program
for eliminating the epidemic, as in Fig. 5.6b.
Figure 5.7 shows the vaccination coverage at the steady state of individuals’
decision-making with respect to different disease reproduction numbers. Here, we
observe a similar effect of social influences in all three considered situations: the
effect of social influences will increase the vaccination coverage when the relative
cost of vaccination rc is low (see Fig. 5.7a–c), reduce it when rc is relatively high
(see Fig. 5.7g–i), and bring the coverage to a certain level when individuals become
strong followers of social influences (conformity rate rf approaches 1.0). The
simulation results further show that when the effect of social influences is relatively
weak (conformity rate 0 < rf ≤ 0.6), the vaccination coverage will increase in the
case of relatively severe disease transmission, i.e., large reproduction numbers (e.g.,
R0 = 2.0). However, if the effect of social influences is strengthened (conformity
rate rf approaches 1.0), the vaccination coverage at the steady state of individuals’
decision-making is mainly determined by individuals’ initial level of vaccination
willingness, rather than the related costs and disease severity.
1 R0=1.2 1 1
R =1.6 0.8
0.8 0
0.8
R0=2.0
0.6 0.6
0.6
0.4 0.4 0.4
0.2 0.2 0.2
0 0 0
1.0 1.0 1.0
0.8 0.8 0.8
0.6 0.6 0.6
0.4 0.4 0.4
0.2 0.2 0.2
0 0 0
(c)
(a) (b)
1 1 1
0 1.0 0 0
0.8 1.0 1.0
0.8 0.8
0.6 0.6 0.6
0.4 0.4 0.4
0.2 0.2 0.2
0 0 0
1 1 1
0 0 0
1.0 1.0 1.0
0.8 0.8 0.8
0.6 0.6 0.6
0.4 0.4 0.4
0.2 0.2 0.2
0 0 0
Fig. 5.7 Sensitivity analysis of the vaccination coverage at the steady state of individuals’
decision-making, with respect to: (1) R0 = 1.2; (2) R0 = 1.6; and (3) R0 = 2.0
5.4 Summary
It has long been observed that the spread of awareness about an epidemic affects
individuals’ behaviors or opinions. In the context of an emerging infectious disease,
it would be unrealistic for individuals to have prior knowledge about a newly
developed vaccine when deciding whether to be vaccinated. In this case, the spread
of awareness on social media about both disease severity and vaccine safety will
affect individuals’ subjective perception of vaccination and, hence, substantially
influence their decisions.
In this chapter, we address the further development of decision models enabling
us to characterize individual-level voluntary vaccination; the models should incor-
porate the factor of subjective perception, which is partly shaped by social media
[54]. In addition, we examine how such individuals’ vaccination decisions affect
efforts to ensure adequate vaccination coverage for infectious disease control.
individuals, rippling through their interaction relationships, which will in turn affect
their vaccination decisions.
For such a situation, we construct a new individual-level model to characterize
vaccination decision-making. At the same time, we use an epidemic model to
describe the dynamics of disease transmission as a result of individuals’ voluntary
vaccination. Based on our model, we investigate how the spread of awareness
affects the changes of individuals’ vaccination decisions during an infectious disease
outbreak.
m : 2Θ → [0, 1]
m(φ) = 0 (6.1)
m(A) = 1
A⊆2Θ
m(φ) = 0
m(Y es) ∈ [0, 1]
(6.2)
m(No) ∈ [0, 1]
m(Θ) = 1 − m(Y es) − m(No)
where m(Y es) describes an individual’s belief that he/she should be vaccinated to
prevent infection, m(No) represents the belief that he/she should reject vaccination
having considered the potential adverse effects, and m(Θ) denotes the belief that
he/she is yet to decide whether or not to be vaccinated (due to uncertainty about the
disease and vaccine). Based on the above formulation, an individual will decide to
76 6 Understanding the Effect of Social Media
be vaccinated with the probability of m(Y es), reject the vaccine with the probability
of m(No), and have no firm decision with the probability of m(Θ). In addition, we
assume individuals who decide to be vaccinated will be vaccinated immediately
and, therefore, will either be successfully immunized or suffer from vaccine adverse
effects. Those with no decision will review their decision state in the next time step.
to continuously update his/her perception of the disease and vaccine and then make
his/her vaccination decision. Based on the obtained awareness, the individual will
update his/her belief values (denoted by mi ) by combining the present belief values
(denoted by mi ) with the newly received evidence mei . This can be expressed in the
following form (with ⊕ denoting the combination operation):
Specifically, based on the assumption that the multiple sources are independent,
the belief-value update corresponding to the extended Dempster rule of combination
is as follows:
where mi (SB ) · mei (SC ) represents the basic belief mass associated with
SB ∩SC =φ
the conflict between present beliefs and the newly received evidence. In the
Dempster combination rule, the denominator, 1 − mi (SB ) · mei (SC ), is a
SB ∩SC =φ
normalization factor, which attributes the conflict probability mass to the universal
set m(Θ).
By doing so, we utilize DST to develop a belief-based decision model for
characterizing individuals’ vaccination decisions in the presence of uncertainty.
Furthermore, we extend the classical DST framework by incorporating the spread of
awareness in a structured host population, in which the certainty regarding a piece
of evidence will decay as it is transmitted from person to person.
For illustration, Fig. 6.2 shows individuals’ belief-value updates calculated with
respect to the spread of awareness about two independently reported negative events
on a synthetic lattice network, in which the fading coefficients are set as: 0.1
(Fig. 6.2a) and 0.2 (Fig. 6.2b), respectively.
(a)
(b)
Fig. 6.3 The Human Swine Influenza Vaccination Programme (HSIVP) in Hong Kong. This
program was launched on December 1, 2009. As of March 13, 2010, more than 180,000 doses of
HSI vaccine were administered to persons of various groups. The rate of adverse events following
immunization (AEFI) was evaluated as 17.8 per 100,000 vaccinated individuals, i.e., the reporting
rate of AEFI was estimated as 0.0178%. (a) The daily number of vaccinated individuals since the
beginning of HSIVP. (b) The reported cases of infection and vaccine adverse effects
6.3 Case Study 81
Fig. 6.4 A partial snapshot of individuals’ social network. We utilize a network structure to
represent individuals’ interaction relationships, based on the data of a Facebook-like online
community. In this network, the nodes denote individuals and the links represent their interactions
in terms of sending and receiving messages
We run our model with the aforementioned parameterization under various settings
to reveal the interplay between the dynamics of disease transmission and individ-
uals’ vaccination. In doing so, we investigate the effect of the spread of awareness
of disease severity and vaccine safety in a host population by investigating various
settings of the reporting rates of negative events and κ, the coefficient of awareness
fading f , and the disease basic reproduction number R0 .
As shown in Fig. 6.6, we investigate the reporting rates for the negative events of
severe infections () and vaccine adverse effects (κ) with respect to two levels: 1%
and 0.1%. Here, we set = 0.01 and κ = 0.001 for the situation of “disease scare”,
and similarity, = 0.001 and κ = 0.01 for the situation of “vaccine scare”. A higher
reporting rate of severe disease infections will prompt individuals to be vaccinated
(Fig. 6.6a, dash curve), which will in turn reduce disease transmission (Fig. 6.6b,
dash curve). Moreover, vaccination in the early stage will be more effective than in
the later stages. We can observe that when = 0.01, the difference in the number of
vaccinated individuals between the situations of κ = 0.001 and κ = 0.01 (Fig. 6.6a,
dashed curve and solid curve, respectively) is relatively small in the early stage of
disease transmission (before day 10). After that, the vaccination dynamics when
82 6 Understanding the Effect of Social Media
Susceptible
Infectious
Recovered
Vaccinated
(a) (b)
(c) (d)
Fig. 6.5 Monte Carlo simulations of disease transmission and voluntary vaccination. (a) The
dynamics of disease transmission in terms of the sizes of susceptible, infectious, recovered, and
vaccinated populations. (b) The reported cases of severe disease infection and vaccine adverse
effects. (c) The average belief values about vaccination in a host population. (d) The dynamics of
voluntary vaccination in terms of the daily number of vaccinated individuals
= 0.01 and κ = 0.001 will peak on day 15 with more than 4% individuals
being vaccinated, while that of = 0.01 and κ = 0.01 will peak on day 11 at
2%. Accordingly, we can observe that the disease dynamics in the situations of
= 0.01, κ = 0.001 and = 0.01, κ = 0.01 (Fig. 6.6b, dashed curve and solid
curve, respectively) have a similar incidence rate at the peak of disease infection,
but the durations of the disease transmission period are different.
In addition, we investigate the effect of awareness fading corresponding to
different fading coefficients f , the results of which are shown in Fig. 6.7. We
note that awareness fading can affect the dynamics of individuals’ vaccination in
terms of the number of vaccinated individuals and the time of individuals’ being
vaccinated. In our simulation, when the fading coefficient f = 0.1, the daily
6.3 Case Study 83
(a)
(b)
Fig. 6.6 The effect of reporting rates of negative events (severe disease infections and vaccine
adverse effects κ). (a) The dynamics of voluntary vaccination (daily number of vaccinated
individuals). (b) The dynamics of disease transmission (daily number of infectious individuals)
84 6 Understanding the Effect of Social Media
Vaccination dynamics
0.1
f =0.1
f =0.4
Population size (100%)
0.08
f =0.7
0.06
0.04
0.02
0
0 10 20 30 40 50
Days
Disease dynamics
0.1
f =0.1
f =0.4
Population size (100%)
0.08
f =0.7
0.06
0.04
0.02
0
0 10 20 30 40 50
Days
Fig. 6.7 The effect of fading coefficient (f ). (a) The dynamics of voluntary vaccination (daily
number of vaccinated individuals). (b) The dynamics of disease transmissions (daily number of
infectious individuals)
6.4 Further Remarks 85
It is well known that the spread of awareness about an epidemic will affect
individuals’ health-related behavior. For example, individuals who are aware of
the risk of infection may take measures to reduce their susceptibility or distance
their social contacts to protect themselves [175, 176]. In the context of vaccination,
the spread of awareness about severe infections and vaccine adverse effects will
affect individuals’ perception of disease severity and vaccine safety and, hence,
substantially change their vaccination decisions.
As a further step from the studies in the preceding chapters, here we have con-
sidered a belief-based characterization of individuals’ vaccination decisions. In our
model, we have correlated individuals’ subjective assessment of disease severity and
vaccine safety with the dynamics of disease transmission and voluntary vaccination
by exploring the awareness of reported negative events. Unlike existing belief-based
studies, e.g., that of Coelho et al. [177], we have characterized individuals’ belief-
value updates as a result of the spread of awareness in a structured host population
(a Facebook-like online community). In this case, we can represent the situation
in which individuals collect health-related information from online social media,
and make vaccination decisions according to their obtained awareness from socially
interconnected neighbors. Additionally, instead of a binary polarization of decisions,
86 6 Understanding the Effect of Social Media
(a)
(b)
Fig. 6.8 The effect of disease reproduction number R0 . (a) The dynamics of voluntary vaccination
(daily number of vaccinated individuals). (b) The dynamics of disease transmissions (daily number
of infectious individuals)
6.5 Summary 87
i.e., either to vaccinate or not, we have introduced a third decision response, “yet to
decide”, and associated it with the belief value of uncertainty. By doing so, we have
utilized an extended DST to characterize individuals’ belief-value updates in the
presence of uncertainty.
By computationally characterizing the effect of the spread of awareness, this
work has some practical implications for understanding individuals’ vaccination
decision-making and for improving the effectiveness of a vaccination program. A
growing number of individuals use Internet-based communication services to obtain
and share health-related information [178]. This represents the growing power of
data on users’ online communication to help track events in real time during an
epidemic, e.g., detecting a pandemic of influenza by monitoring related queries to
online search engines [179]. Salathe et al. collected individuals’ Twitter updates to
assess public sentiments toward a novel vaccine [180]. Henrich et al. used online
comments as obtained from the websites of CBC, Vancouver Sun, and Global
and Mail to capture public attitudes about the H1N1 vaccine [181]. Thus, online
social media provide effective tools to access real-time data to evaluate the public
perception of diseases and vaccines. These can be used to estimate public acceptance
of a prospective vaccination program, which will enable public health authorities to
make plans in advance to improve the effectiveness of vaccination strategies.
So far, our study has provided a modeling framework that incorporates the spread
of awareness with the belief-based characterization of decision-making. It should
be pointed out that the simulation results may depend on the social network used
in our example (a Facebook-like online community). In our model, the spread of
awareness only mediates individuals’ localized interactions (i.e., between socially
interconnected neighbors), while the global effect of public media is not taken into
account. For future work, it will be interesting to extend the current model by adding
a globalized spread of awareness, wherein each individual will become aware of
a reported negative event with a certain probability. Additionally, by utilizing a
DST-based characterization of vaccination decisions, the combination rule used to
describe individuals’ belief-value updates will determine the resulting decisions.
The above-mentioned issues are worthy of further investigation.
6.5 Summary
The preceding chapters have, in many aspects, alluded to a new way of viewing
and pursuing epidemiological studies in the modern context. Starting from cer-
tain empirical observations of a real-world infectious disease situation, beset by
unanswered questions, we then represent it using an abstract modeling language,
parameterize our models with the real-world data, and thereafter analytically and
quantitatively examine the underlying associated risk factors and the implications
for evidence-based policies and practice.
In this chapter, we continue to build on the above-mentioned discussions, and
further generalize them through the prism of systems epidemiology, a term coined
here to encompass the latest methodological development in epidemiology [182].
We must begin by asking precisely what systems epidemiology is, and how the
notion of systems is relevant to epidemiology. These key questions will be answered
in the following discussions on systems thinking.
Systems thinking is a philosophical and methodological perspective that draws
on the fundamental notions of systems theory, which views a system as an
integration of components with the interacting relationships among them and the
environments in which they reside [183, 184]. As mentioned by Maani et al.
[185], systems thinking emphasizes two fundamental concepts, i.e., complexity
and entirety. Systems complexity is generated from the structure of integrated
components, i.e., how the constituent components are organized and interact with
each other and with the environment. Systems entirety is derived from the dynamic
behaviors of a system as a whole, which is to say, how a complex system of
interacting components behaves and exhibits emergent properties at the system
level, rather than being a simple behavioral aggregation of its basic components.
Fig. 7.1 The four essential steps (in ovals) for performing the complex systems approach in
combating infectious diseases. The directional arrows show their functional interrelationships
7.3 Systems Modeling in Practice 93
1. Last JM, Harris SS, Thuriaux MC, et al. A Dictionary of Epidemiology. New York: Oxford
University Press; 2014.
2. MacMahon B, Trichopoulos D. Epidemiology: Principles and Methods. Philadelphia:
Lippincott Williams and Wilkins; 1997.
3. Merrill RM. Introduction to Epidemiology. Burlington: Jones and Bartlett Learning; 2015.
4. Rothman KJ. Lessons from John Graunt. Lancet. 1996;347(8993):37–9. https://doi.org/10.
1016/S0140-6736(96)91562-7.
5. Koutouvidis N, Marketos S, Beveridge A. The contribution of Thomas Sydenham (1624-
1689) to the evolution of psychiatry. Hist Psychiatry. 1995;6(24):513–20. https://doi.org/10.
1177/0957154X9500602408.
6. Cameron D, Jones IG. John Snow, the broad street pump and modern epidemiology. Int J
Epidemiol. 1983;12(4):393–96. https://doi.org/10.1093/ije/12.4.393.
7. Smith DL, Battle KE, Hay SI, et al. Ross, Macdonald, and a theory for the dynamics and
control of mosquito-transmitted pathogens. PLoS Pathog. 2012;8(4):e1002588. https://doi.
org/10.1371/journal.ppat.1002588.
8. Zadoks J. Methodology of epidemiological research. Ann Rev Phytopathol. 1972;10(1):253–
76. https://doi.org/10.1146/annurev.py.10.090172.001345.
9. Rothman KJ, Greenland S, Lash TL. Modern Epidemiology. Philadelphia: Lippincott
Williams & Wilkins; 2008.
10. Miettinen OS. Theoretical Epidemiology: Principles of Occurrence Research in Medicine.
Wiley: New York; 1985
11. Marathe M, Vullikanti AKS. Computational epidemiology. Commun ACM. 2013;56(7):88-
96. https://doi.org/10.1145/2483852.2483871.
12. Morse SS, Mazet JA, Woolhouse M, et al. Prediction and prevention of the next pandemic
zoonosis. Lancet. 2012;380(9857):1956–65. https://doi.org/10.1016/s0140-6736(12)61684-
5.
13. Morens DM, Folkers GK, Fauci AS. The challenge of emerging and re-emerging infectious
diseases. Nature. 2004;430(6996):242. https://doi.org/10.1038/nature02759.
14. Fauci AS, Morens DM. The perpetual challenge of infectious diseases. N Engl J Med.
2012;366(5):454–61. https://doi.org/10.1056/NEJMc1204960SA2.
15. Protopopoff N, Van Bortel W, Speybroeck N, et al. Ranking malaria risk factors to guide
malaria control efforts in African highlands. PLoS One. 2009;4(11):e8022. https://doi.org/
10.1371/journal.pone.0008022.
16. Flahault A, Zylberman P. Influenza pandemics: Past, present and future challenges. Public
Health Rev. 2010;32(1):319–40. https://doi.org/10.1007/BF03391605.
37. World Health Organization. WHO guidelines on the use of vaccines and antivi-
rals during influenza pandemics; 2004. Accessed 10 Aug 2004. Website,
Available from: http://www.who.int/csr/resources/publications/influenza/WHO_CDS_CSR_
RMD_2004_8/en/index.html.
38. Gangarosa EJ, Galazka A, Wolfe C, et al. Impact of anti-vaccine movements on pertussis
control: The untold story. Lancet. 1998;351(9099):356–61. https://doi.org/10.1109/4.772418.
39. Roberts RJ, Sandifer QD, Evans MR, et al. Reasons for non-uptake of measles, mumps, and
rubella catch up immunisation in a measles epidemic and side effects of the vaccine. Br Med
J. 1995;310(6995):1629–39. https://doi.org/10.1136/BMJ.310.6995.1629.
40. Anderson P. Another media scare about MMR vaccine hits Britain. Br Med J.
1999;318(7198):1578. https://doi.org/10.1136/BMJ.318.7198.1578.
41. Poland GA, Jacobson RM. Understanding those who do not understand: A brief review of the
anti-vaccine movement. Vaccine. 2001;19(17-19):2440–5. https://doi.org/10.1016/S0264-
410X(00)00469-2.
42. Hobson-West P. ‘Trusting blindly can be the biggest risk of all’: Organised resistance to
childhood vaccination in the UK. Sociol Health Illn. 2007;29(2):198–215. https://doi.org/10.
1111/j.1467-9566.2007.00544x.
43. Jansen VA, Stollenwerk N, Jensen HJ, et al. Measles outbreaks in a population with declining
vaccine uptake. Science. 2003;301(5634):804. https://doi.org/10.1126/science.1086726.
44. Poland GA, Jacobson RM. The age-old struggle against the antivaccinationists. N Engl J
Med. 2011;364(2):97–9. https://doi.org/10.1056/NEJMp1010594.
45. Streefland PH. Public doubts about vaccination safety and resistance against vaccination.
Health Policy. 2001;55(3):159–72. https://doi.org/10.1016/S0168-8510(00)00132-9.
46. Freed GL, Clark SJ, Hibbs BF, et al. Parental vaccine safety concerns: The experiences of
pediatricians and family physicians. Am J Prev Med. 2004;26(1):11–4. https://doi.org/10.
1016/j.amepre.2003.09.004.
47. Bauch CT, Galvani AP, Earn DJD. Group interest versus self-interest in smallpox vacci-
nation policy. Proc Natl Acad Sci. 2003;100(18):10564–7. https://doi.org/10.1073/pnas.
1731324100.
48. Galvani AP, Reluga TC, Chapman GB. Long-standing influenza vaccination policy is in
accord with individual self-interest but not with the utilitarian optimum. Proc Natl Acad Sci.
2007;104(13):5692–7. https://doi.org/10.1073/pnas.0606774104.
49. Jadhav SS, Gautam M, Gairola S. Emerging markets and emerging needs: Developing coun-
tries vaccine manufacturers’ perspective and its current status. Biologicals. 2009;37(3):165–8.
https://doi.org/10.1016/j.biologicals.2009.02.009.
50. Friede M, Palkonyay L, Alfonso C, et al. WHO initiative to increase global and equitable
access to influenza vaccine in the event of a pandemic: Supporting developing country
production capacity through technology transfer. Vaccine. 2011;29(S1):A2–7. https://doi.
org/10.1016/j.vaccine.2011.02.079.
51. Keelan J, Pavri V, Balakrishnan R, et al. An analysis of the human papilloma virus vaccine
debate on MySpace blogs. Vaccine. 2010;28(6):1535–40. https://doi.org/10.1016/j.vaccine.
2009.11.060.
52. Liu Y, Gu Z, Xia S, et al. What are the underlying transmission patterns of COVID-19
Outbreak? An age-specific social contact characterization. EClinicalMedicine. 2020;100354.
https://doi.org/10.1016/j.eclinm.2020.100354.
53. Xia S, Liu J. A computational approach to characterizing the impact of social influence on
individuals’ vaccination decision making. PLoS One. 2013;8(4):e60373. https://doi.org/10.
1371/journal.pone.0060373.
54. Xia S, Liu J. A belief-based model for characterizing the spread of awareness and its impacts
on individuals’ vaccination decisions. J R Soc Interface. 2014;11(94):20140013. https://doi.
org/10.1098/rsif.2014.0013.
55. Xia S, Liu J, Cheung W. Identifying the relative priorities of subpopulations for containing
infectious disease spread. PLoS One. 2013;8(6):e65271. https://doi.org/10.1371/journal.
pone.0065271.
100 References
56. Hethcote HW. The mathematics of infectious diseases. SIAM Rev. 2000;42(4):599–653.
https://doi.org/10.1137/S0036144500371907.
57. Keeling MJ, Rohani P. Modeling Infectious Diseases in Humans and Animals. Princeton:
Princeton University Press; 2011.
58. Edmunds WJ, O’callaghan C, Nokes D. Who mixes with whom? A method to determine the
contact patterns of adults that may lead to the spread of airborne infections. Proc R Soc B
Biol Sci. 1997;264(1384):949–57. https://doi.org/10.1098/rspb.1997.0131.
59. Diekmann O, Heesterbeek JAP, Metz JA. On the definition and the computation of the basic
reproduction ratio R0 in models for infectious diseases in heterogeneous populations. J Math
Biol. 1990;28(4):365–82. https://doi.org/10.1007/BF00178324.
60. Heesterbeek JAP. A brief history of R0 and a recipe for its calculation. Math Philos Found
Biol Biomed Science. 2002;50(3):189–204. https://doi.org/10.1023/A:1016599411804.
61. Bailey NT. The Mathematical Theory of Infectious Diseases and Its Applications. London:
Griffin; 1975.
62. Anderson RM, May RM. Infectious Diseases of Humans: Dynamics and Control. New York:
Oxford University Press; 1992.
63. Coburn BJ, Wagner BG, Blower S. Modeling influenza epidemics and pandemics: Insights
into the future of swine flu (H1N1). BMC Med. 2009;7(1):30. https://doi.org/10.1186/1741-
7015-7-30.
64. Keeling MJ, Grenfell BT. Understanding the persistence of measles: Reconciling theory,
simulation and observation. Proc R Soc B Biol Sci. 2002;269(1489):335–43. https://doi.org/
10.1098/rspb.2001.1898.
65. Keeling MJ, Eames KT. Networks and epidemic models. J R Soc Interface. 2005;2(4):295–
307. https://doi.org/10.1098/rsif.2005.0051.
66. Keeling M. The implications of network structure for epidemic dynamics. Theor Popul Biol.
2005;67(1):1–8. https://doi.org/10.1016/j.tpb.2004.08.002.
67. Tan Q, Liu Y, Liu J. Motif-aware diffusion network inference. Int J Data Sci Anal. 2018;1–13.
https://doi.org/10.1007/978-3-319-93040-4_50.
68. Dangerfield CE, Ross JV, Keeling MJ. Integrating stochasticity and network structure into an
epidemic model. J R Soc Interface. 2008;6(38):761–774. https://doi.org/10.1098/rsif.2008.
0410.
69. Salathe M, Kazandjieva M, Lee JW, et al. A high-resolution human contact network for
infectious disease transmission. Proc Natl Acad Sci. 2010;107(51):22020–5. https://doi.org/
10.1073/pnas.1009094108.
70. Grenfell B, Harwood J. (Meta) population dynamics of infectious diseases. Trends Ecol Evol.
1997;12(10):395–9. https://doi.org/10.1088/0034-4885/77/2/026602.
71. Yang X, Liu J, Cheung WKW, et al. Inferring metapopulation based disease transmission
networks. In: 18th Pacific-Asia Conference on Knowledge Discovery and Data Mining.
Berlin: Springer; 2014. p. 385–99.
72. Lloyd-Smith JO, Schreiber SJ, Kopp PE, et al. Superspreading and the effect of individual
variation on disease emergence. Nature. 2005;438(7066):355–9. https://doi.org/10.1038/
nature0415.
73. Dushoff J, Plotkin JB, Viboud C, et al. Vaccinating to protect a vulnerable subpopulation.
PLoS Med. 2007;4(5):e174. https://doi.org/10.1371/journal.pmed.0040174.
74. Wallinga J, Edmunds WJ, Kretzschmar M. Perspective: Human contact patterns and the
spread of airborne infectious diseases. Trends Microbiol. 1999;7(9):372–7. https://doi.org/
10.1016/S0966-842X(99)01546-2.
75. Stehle J, Voirin N, Barrat A, et al. High-resolution measurements of face-to-face contact
patterns in a primary school. PLoS One. 2011;6(8):e23176. https://doi.org/10.1371/journal.
pone.0023176.
76. Haydon DT, Chase-Topping M, Shaw D, et al. The construction and analysis of epidemic
trees with reference to the 2001 UK foot-and-mouth outbreak. Proc R Soc B Biol Sci.
2003;270(1511):121–7. https://doi.org/10.1098/rspb.2002.2191.
References 101
77. Riley S, Fraser C, Donnelly CA, et al. Transmission dynamics of the etiological agent of
SARS in Hong Kong: Impact of public health interventions. Science. 2003;300(5627):1961–
6. https://doi.org/10.1126/science.1086478.
78. Eames KT, Keeling MJ. Contact tracing and disease control. Proc R Soc B Biol Sci.
2003;270(1533):2565–71. https://doi.org/10.1098/rspb.2003.2554.
79. Bearman PS, Moody J, Stovel K. Chains of affection: The structure of adolescent romantic
and sexual networks. Sci Eng Ethics. 2004;110(1):44–91.
80. De P, Singh AE, Wong T, et al. Sexual network analysis of a gonorrhoea outbreak. Sex
Transm Infect. 2004;80(4):280–5. https://doi.org/10.1136/sti.2003.007187.
81. Lalvani A, Pathan AA, Durkan H, et al. Enhanced contact tracing and spatial tracking of
Mycobacterium tuberculosis infection by enumeration of antigen-specific T cells. Lancet.
2001;357(9273):2017–21. https://doi.org/10.1016/S0140-6736(00)05115-1.
82. Eichner M. Case isolation and contact tracing can prevent the spread of smallpox. Am J
Epidemiol. 2003;158(2):118–128. https://doi.org/10.1093/aje/kwg104.
83. Mossong J, Hens N, Jit M, et al. Social contacts and mixing patterns relevant to the spread
of infectious diseases. PLoS Med. 2008;5(3):e74. https://doi.org/10.1371/journal.pmed.
0050074.
84. Read JM, Eames KT, Edmunds WJ. Dynamic social networks and the implications for the
spread of infectious disease. J R Soc Interface. 2008;5(26):1001–7. https://doi.org/10.1098/
rsif.2008.0013.
85. Hens N, Goeyvaerts N, Aerts M, et al. Mining social mixing patterns for infectious disease
models based on a two-day population survey in Belgium. BMC Infect Dis. 2009;9(1):5.
https://doi.org/10.1186/1471-2334-9-5.
86. Doherty IA, Padian NS, Marlow C, et al. Determinants and consequences of sexual networks
as they affect the spread of sexually transmitted infections. J Infect Dis. 2005;191 Suppl
1:S42–54. https://doi.org/10.1086/425277.
87. Liljeros F, Edling CR, Amaral LAN. Sexual networks: Implications for the transmission of
sexually transmitted infections. Microbes Infect. 2003;5(2):189–96. https://doi.org/10.1016/
s1286-4579(02)00058-8.
88. Potterat J, Phillips-Plummer L, Muth S, et al. Risk network structure in the early epidemic
phase of HIV transmission in Colorado Springs. Sex Transm Dis. 2002;78 Suppl 1:i159–63.
https://doi.org/10.1136/sti.78.suppl1.i159.
89. Yu Z, Liu J, Zhu X. Inferring a district-based hierarchical structure of social contacts
from census data. PLoS One. 2015;10(2):e0118085. https://doi.org/10.1371/journal.pone.
0118085.
90. Halloran ME, Longini IM, Nizam A, et al. Containing bioterrorist smallpox. Science.
2002;298(5597):1428–32. https://doi.org/10.1126/science.1074674.
91. Meyers LA, Pourbohloul B, Newman ME, et al. Network theory and SARS: Predicting
outbreak diversity. J Theor Biol. 2005;232(1):71–81. https://doi.org/10.1016/j.jtbi.2004.07.
026.
92. Eubank S, Guclu H, Kumar VA, et al. Modelling disease outbreaks in realistic urban social
networks. Nature. 2004;429(6988):180–4. https://doi.org/doi:10.1038/nature02541.
93. Eubank S, Kumar VA, Marathe MV, et al. Structure of social contact networks and their
impact on epidemics. DIMACS Ser Discrete Math Theor Comput Sci. 2006;70:181–214.
https://doi.org/10.1090/dimacs/070/09.
94. The Hong Kong Government. 1st H1N1 case confirmed in HK; 2009. Accessed 1
May 2009. Website, Available from: https://www.news.gov.hk/isd/ebulletin/en/category/
healthandcommunity/090501/html/090501en05004.htm.
95. Center for Health Protection. Summary report on the surveillance of adverse events following
HSI immunisation and expert group’s comment on the safety of HSI vaccine in Hong Kong;
2010. Accessed 1 April 2011. Website, Available from: http://www.chp.gov.hk/files/pdf/hsi_
vaccine_aefi_report_en.pdf.
96. Center for Health Protection. Swine and seasonal flu monitor; 2010. Accessed 1 Oct 2010.
Website, Available from: https://www.chp.gov.hk/files/pdf/ssfm_26_11_09.pdf.
102 References
97. Census and Statistics Department of the Hong Kong Government. 2011 Hong Kong popula-
tion census; 2011. Accessed 1 April 2020. Website, Available from: http://www.census2011.
gov.hk.
98. Kwok KO, Leung GM, Riley S. Modelling the proportion of influenza infections within
households during pandemic and non-pandemic years. PLoS One. 2011;6(7):e22089. https://
doi.org/10.1371/journal.pone.0022089.
99. Fumanelli L, Ajelli M, Manfredi P, et al. Inferring the structure of social contacts from
demographic data in the analysis of infectious diseases spread. PLoS Comput Biol.
2012;8(9):e1002673. https://doi.org/10.1371/journal.pcbi.1002673.
100. Merler S, Ajelli M, Pugliese A, et al. Determinants of the spatiotemporal dynamics of the
2009 H1N1 pandemic in Europe: Implications for real-time modelling. PLoS Comput Biol.
2011;7(9):e1002205. https://doi.org/10.1371/journal.pcbi.1002205.
101. Wu JT, Ma ES, Lee CK, et al. The infection attack rate and severity of 2009 pandemic H1N1
influenza in Hong Kong. Clin Infect Dis. 2010;51(10):1184–91. https://doi.org/10.1086/
656740.
102. Wu JT, Ho A, Ma ESK, et al. Estimating infection attack rates and severity in real time during
an influenza pandemic: Analysis of serial cross-sectional serologic surveillance data. PLoS
Med. 2011;8(10):e1001103. https://doi.org/10.1371/journal.pmed.1001103.
103. Cowling BJ, Lau MSY, Ho LM, et al. The effective reproduction number of pandemic
influenza: Prospective estimation. Epidemiology. 2010;21(6):842–6. https://doi.org/10.1097/
EDE.0b013e3181f20977.
104. Cowling BJ, Fang VJ, Riley S, et al. Estimation of the serial interval of influenza.
Epidemiology. 2009;20(3):344–7. https://doi.org/10.1097/EDE.0b013e31819d1092.
105. Cauchemez S, Ferguson NM, Wachtel C, et al. Closure of schools during an influenza pan-
demic. Lancet Infect Dis. 2009;9(8):473–81. https://doi.org/10.1016/S1473-3099(09)70176-
8.
106. Wallinga J, van Boven M, Lipsitch M. Optimizing infectious disease interventions during an
emerging epidemic. Proc Natl Acad Sci. 2010;107(2):923–8. https://doi.org/10.1073/pnas.
0908491107.
107. Liu J, Xia S. Toward effective vaccine deployment: A systematic study. J Med Syst.
2011;35(5):1153–64. https://doi.org/10.1007/s10916-011-9734-x.
108. Keeling MJ, White PJ. Targeting vaccination against novel infections: Risk, age and spatial
structure for pandemic influenza in Great Britain. J R Soc Interface. 2011;8(58):661–70.
https://doi.org/10.1098/rsif.2010.0474.
109. Hashemian M, Qian W, Stanley KG, et al. Temporal aggregation impacts on epidemiological
simulations employing microcontact data. BMC Med Inform Decis Mak. 2012;12(1):132.
https://doi.org/10.1186/1472-6947-12-132.
110. Ferguson N. Capturing human behaviour. Nature. 2007;446(7137):733. https://doi.org/10.
1038/446733a.
111. John TJ, Samuel R. Herd immunity and herd effect: New insights and definitions. Eur J
Epidemiol. 2000;16(7):601–6. https://doi.org/10.1023/A:1007626510002.
112. Mounier-Jack S, Coker RJ. How prepared is Europe for pandemic influenza? Analy-
sis of national plans. Lancet. 2006;367(9520):1405–11. https://doi.org/10.1016/S0140-
6736(06)68511-5.
113. Straetemans M, Buchholz U, Reiter S, et al. Prioritization strategies for pandemic influenza
vaccine in 27 countries of the European Union and the Global Health Security Action Group:
A review. BMC Public Health. 2007;7(1):236. https://doi.org/10.1186/1471-2458-7-236.
114. Ng S, Wu P, Nishiura H, et al. An analysis of national target groups for monovalent 2009
pandemic influenza vaccine and trivalent seasonal influenza vaccines in 2009-10 and 2010-
11. BMC Infect Dis. 2011;11(1):230. https://doi.org/10.1186/1471-2334-11-230.
115. Coker R, Mounier-Jack S. Pandemic influenza preparedness in the Asia Pacific region.
Lancet. 2006;368(9538):886–9. https://doi.org/10.1016/S0140-6736(06)69209-X.
116. Germann TC, Kadau K, Longini IM, et al. Mitigation strategies for pandemic influenza in
the United States. Proc Natl Acad Sci. 2006;103(15):5935–40. https://doi.org/10.1073/pnas.
0601266103.
References 103
137. Lau JTF, Yeung NCY, Choia KC, et al. Factors in association with acceptability of
A/H1N1 vaccination during the influenza A/H1N1 pandemic phase in the Hong Kong general
population. Vaccine. 2010;28(29):4632–7. https://doi.org/10.1016/j.vaccine.2010.04.076.
138. Zijtregtop EAM, Wilschut J, Koelma N, et al. Which factors are important in adults’ uptake
of a (pre)pandemic influenza vaccine? Vaccine. 2009;28(1):207–27. https://doi.org/10.1016/
j.vaccine.2009.09.099.
139. Barriere J, Vanjak D, Kriegel I, et al. Acceptance of the 2009 A (H1N1) influenza vaccine
among hospital workers in two French cancer centers. Vaccine. 2010;28(43):7030–4. https://
doi.org/10.1016/j.vaccine.2010.08.021.
140. Rubin G, Potts H, Michie S. The impact of communications about swine flu (influenza A
H1N1v) on public responses to the outbreak: Results from 36 national telephone surveys in
the UK. Health Technol Assess. 2010;14(34):183–266. https://doi.org/10.3310/hta14340-03.
141. Maltezou HC, Dedoukou X, Patrinos S, et al. Determinants of intention to get vaccinated
against novel (pandemic) influenza A H1N1 among health-care workers in a nationwide
survey. J Infect. 2010;61(3):252–8. https://doi.org/10.1016/j.jinf.2010.06.004.
142. Hidiroglu S, Ay P, Topuzoglu A, et al. Resistance to vaccination: The attitudes and practices of
primary healthcare workers confronting the H1N1 pandemic. Vaccine. 2010;28(51):8120–4.
https://doi.org/10.1016/j.vaccine.2010.09.104.
143. Seale H, Heywood AE, McLaws ML, et al. Why do I need it? I am not at risk! Public
perceptions towards the pandemic (H1N1) 2009 vaccine. BMC Infect Dis. 2010;10(1):99.
https://doi.org/10.1186/1471-2334-10-99.
144. Wong LP, Sam IC. Factors influencing the uptake of 2009 H1N1 influenza vaccine in a
multiethnic Asian population. Vaccine. 2010;28(28):4499–505. https://doi.org/10.1016/j.
vaccine.2010.04.043.
145. Rubin GJ, Potts HW, Michie S. Likely uptake of swine and seasonal flu vaccines among
healthcare workers. A cross-sectional analysis of UK telephone survey data. Vaccine.
2011;29(13):2421–8. https://doi.org/10.1016/j.vaccine.2011.01.035.
146. Torun SD, Torun F. Vaccination against pandemic influenza A/H1N1 among healthcare
workers and reasons for refusing vaccination in Istanbul in last pandemic alert phase. Vaccine.
2010;28(35):5703–10. https://doi.org/10.1016/j.vaccine.2010.06.049.
147. Bauch CT, Earn DJD. Vaccination and the theory of games. Proc Natl Acad Sci.
2004;101(36):13391–4. https://doi.org/10.1073/pnas.0403823101.
148. Bauch CT. Imitation dynamics predict vaccinating behaviour. Proc R Soc B Biol Sci.
2005;272(1573):1669–75. https://doi.org/10.1098/rspb.2005.3153.
149. Reluga TC, Bauch CT, Galvani AP. Evolving public perceptions and stability in vaccine
uptake. Math Biosci. 2006;204(2):185–98. https://doi.org/10.1016/j.mbs.2006.08.015.
150. Reluga TC, Galvani AP. A general approach for population games with application to
vaccination. Math Biosci. 2011;230(2):67–78. https://doi.org/10.1016/j.mbs.2011.01.003.
151. Cojocaru MG. Dynamic equilibria of group vaccination strategies in a heterogeneous
population. J Glob Optim. 2008;40(1-3):51–63. https://doi.org/10.1007/s10898-007-9204-7.
152. Perisic A, Bauch CT. Social contact networks and disease eradicability under voluntary
vaccination. PLoS Comput Biol. 2009;5(2):e1000280. https://doi.org/10.1371/journal.pcbi.
1000280.
153. Perisic A, Bauch CT. A simulation analysis to characterize the dynamics of vaccinating
behaviour on contact networks. BMC Infect Dis. 2009;9(1):77. https://doi.org/10.1186/1471-
2334-9-77.
154. Bauch CT, Bhattacharyya S. Evolutionary game theory and social learning can determine
how vaccine scares unfold. PLoS Comput Biol. 2012;8(4):e1002452. https://doi.org/10.1371/
journal.pcbi.1002452.
155. Fu F, Rosenbloom DI, Wang L, et al. Imitation dynamics of vaccination behaviour on social
networks. Proc R Soc B Biol Sci. 2011;278(1702):42–9. https://doi.org/10.1098/rspb.2010.
1107.
156. d’Onofrio A, Manfredi P, Poletti P. The impact of vaccine side effects on the natural history of
immunization programmes: An imitation-game approach. J Theor Biol. 2011;273(1):63–71.
https://doi.org/10.1016/j.jtbi.2010.12.029.
References 105
157. Mbah MLN, Liu J, Bauch CT, et al. The impact of imitation on vaccination behavior in social
contact networks. PLoS Comput Biol. 2012;8(4):e1002469. https://doi.org/10.1371/journal.
pcbi.1002469.
158. Zhang H, Zhang J, Li P, et al. Risk estimation of infectious diseases determines the
effectiveness of the control strategy. Physica D. 2011;240(11):943–8. https://doi.org/10.
1016/j.physd.2011.02.001.
159. Breban R, Vardavas R, Blower S. Mean-field analysis of an inductive reasoning game:
Application to influenza vaccination. Phys Rev E. 2007;76(3):031127. https://doi.org/10.
1103/PhysRevE.76.031127.
160. d’Onofrio A, Manfredi P, Salinelli E. Vaccinating behaviour, information, and the dynamics
of SIR vaccine preventable diseases. Theor Popul Biol. 2007;71(3):301–17. https://doi.org/
10.1016/j.tpb.2007.01.001.
161. Shim E, Meyers LA, Galvani AP. Optimal H1N1 vaccination strategies based on self-interest
versus group interest. BMC Public Health. 2011;11(S1):S4. https://doi.org/10.1186/1471-
2458-11-S1-S4.
162. Watts DJ, Dodds PS. Influentials, networks, and public opinion formation. J Consum Res.
2007;34(4):441–58. https://doi.org/10.1086/518527.
163. Nowak A, Szamrej J, Latané B. From private attitude to public opinion: A dynamic theory of
social impact. Psychol Rev. 1990;97(3):362. https://doi.org/10.1037/0033-295X.97.3.362.
164. Salathé M, Bonhoeffer S. The effect of opinion clustering on disease outbreaks. J R Soc
Interface. 2008;5(29):1505–8. https://doi.org/10.1098/rsif.2008.0271.
165. Bish A, Yardley L, Nicoll A, et al. Factors associated with uptake of vaccination against
pandemic influenza: A systematic review. Vaccine. 2011;29(38):6472–84. https://doi.org/10.
1016/j.vaccine.2011.06.107.
166. Young ME, Norman GR, Humphreys KR. Medicine in the popular press: The influence of
the media on perceptions of disease. PLoS One. 2008;3(10):e3552. https://doi.org/10.1371/
journal.pone.0003552.
167. Vance K, Howe W, Dellavalle RP. Social internet sites as a source of public health
information. Dermatol Clin. 2009;27(2):133–6. https://doi.org/10.1016/j.det.2008.11.010.
168. Signorini A, Segre AM, Polgreen PM. The use of Twitter to track levels of disease
activity and public concern in the US during the influenza A H1N1 pandemic. PLoS One.
2011;6(5):e19467. https://doi.org/10.1371/journal.pone.0019467.
169. Keelan J, Pavri-Garcia V, Tomlinson G, et al. YouTube as a source of information on
immunization: a content analysis. J Am Med Assoc. 2007;298(21):2482–4. https://doi.org/
10.1001/jama.298.21.2482.
170. Pandey A, Patni N, Singh M, et al. YouTube as a source of information on the H1N1 influenza
pandemic. Am J Prev Med. 2010;38(3):e1–3. https://doi.org/10.1016/j.amepre.2009.11.007.
171. Shafer G. A Mathematical Theory of Evidence. vol. 76. London: Princeton University Press;
1976.
172. Dempster AP. Upper and lower probabilities induced by a multivalued mapping. In: Classic
Works of the Dempster-Shafer Theory of Belief Functions. Berlin: Springer; 2008. p. 57–72.
173. Information Services Department of the Hong Kong Government. Human Swine Influenza
Vaccination Programme; 2010. Accessed 1 April 2011. Website, Available from: http://
www.info.gov.hk/gia/general/201003/17/P201003170168.htm.
174. Panzarasa P, Opsahl T, Carley KM. Patterns and dynamics of users’ behavior and interaction:
network analysis of an online community. J Am Soc Inf Sci Technol. 2009;60(5):911–32.
https://doi.org/10.1002/asi.v60:5.
175. Funk S, Gilada E, Watkinsb C, et al. The spread of awareness and its impact on epidemic
outbreaks. Proc Natl Acad Sci. 2009;106(16):6872–6877. https://doi.org/10.1073/pnas.
0810762106.
176. Fenichel EP, Castillo-Chavez C, Ceddia MG, et al. Adaptive human behavior in epidemi-
ological models. Proc Natl Acad Sci. 2011;108(15):6306–11. https://doi.org/10.1073/pnas.
1011250108.
106 References
177. Coelho FC, Codeco CT. Dynamic modeling of vaccinating behavior as a function of
individual beliefs. PLoS Comput Biol. 2009;5(7):e1000425. https://doi.org/10.1371/journal.
pcbi.1000425.
178. Kaplan AM, Haenlein M. Users of the world, unite! The challenges and opportunities of
social media. Bus Horiz. 2010;53(1):59–68. https://doi.org/10.1016/j.bushor.2009.09.003.
179. Ginsberg J, Mohebbi MH, Patel RS, et al. Detecting influenza epidemics using search engine
query data. Nature. 2009;457(7232):1012–4. https://doi.org/10.1038/nature07634.
180. Salathe M, Khandelwal S. Assessing vaccination sentiments with online social
media: Implications for infectious disease dynamics and control. PLoS Comput Biol.
2011;7(10):e1002199. https://doi.org/10.1371/journal.pcbi.1002199.
181. Henrich N, Holmes B. What the public was saying about the H1N1 vaccine: Perceptions
and issues discussed in on-line comments during the 2009 H1N1 pandemic. PLoS One.
2011;6(4):e18479. https://doi.org/10.1371/journal.pone.0018479.
182. Xia S, Zhou XN, Liu J. Systems thinking in combating infectious diseases. Infect Dis Poverty.
2017;6(1):144. https://doi.org/10.1186/s40249-017-0339-6.
183. Checkland P. Systems thinking, systems practice: Includes a 30-year retrospective. J Oper
Res Soc. 2000;51(5):647. https://doi.org/10.2307/254200.
184. Leischow SJ, Milstein B. Systems thinking and modeling for public health practice. Am
Public Health Assoc. 2006;96(3):403–5. https://doi.org/10.2105/ajph.2005.082842.
185. Maani K, Cavana RY. Systems Thinking, System Dynamics: Managing Change and Com-
plexity. New Zealand: Pearson Education New Zealand; 2007.
186. May RM. Simple mathematical models with very complicated dynamics. Nature.
1976;261(5560):459–67. https://doi.org/10.1038/261459a0.
187. Liu J, Jin X, Tsui KC. Autonomy Oriented Computing: From Problem Solving to Complex
Systems Modeling. Boston: Springer; 2006.
188. Liu J. Autonomous Agents and Multi-agent Systems: Explorations in Learning, Self-
organization and Adaptive Computation. Singapore: World Scientific; 2001.
189. Newman ME. Spread of epidemic disease on networks. Phys Rev E. 2002;66(1):016128.
https://doi.org/10.1103/PhysRevE.66.016128.
190. Chen H, Yang B, Liu J, et al. Mining spatiotemporal diffusion network: A new framework
of active surveillance planning. IEEE Access. 2019;7:108458–73. https://doi.org/10.1109/
ACCESS.2019.2927878.
191. Yang X, Liu J, Zhou XN, et al. Inferring disease transmission networks at a metapopulation
level. Malar J. 2014;2(1):8. https://doi.org/10.1186/2047-2501-2-8.
192. Shi B, Zhong J, Bao Q, et al. EpiRep: Learning node representations through epidemic
dynamics on networks. In: 2019 IEEE/WIC/ACM International Conference on Web
Intelligence. Piscataway: IEEE; 2019. p. 486–92.
193. Yu Z, Liu J, Wang X, et al. Efficient vaccine distribution based on a hybrid compartmental
model. PLoS One. 2016;11(5):e0155416. https://doi.org/10.1371/journal.pone.0155416.
194. Liu J, Yang B, Cheung WK, et al. Malaria transmission modelling: A network perspective.
Infect Dis Poverty. 2012;1(1):11. https://doi.org/10.1186/2049-9957-1-11.
195. Ming RX, Liu J, Cheung WK, et al. Stochastic modelling of infectious diseases for
heterogeneous populations. Infect Dis Poverty. 2016;5(1):107. https://doi.org/10.1186/
s40249-016-0199-5.
196. Yang B, Pei H, Chen H, et al. Characterizing and discovering spatiotemporal social contact
patterns for healthcare. IEEE Trans Pattern Anal Mach Intell. 2016;39(8):1532–46. https://
doi.org/10.1109/TPAMI.2016.2605095.
197. Chen H, Yang B, Pei H, et al. Next generation technology for epidemic prevention and
control: Data-driven contact tracking. IEEE Access. 2018;7:2633–42. https://doi.org/10.
1109/ACCESS.2018.2882915.
198. Shi B, Tan Q, Zhou XN, et al. Mining geographic variations of Plasmodium vivax for active
surveillance: A case study in China. Malar J. 2015;14(1):216. https://doi.org/10.1186/s12936-
015-0719-y.
References 107
199. Shi B, Liu J, Zhou XN, et al. Inferring Plasmodium vivax transmission networks from tempo-
spatial surveillance data. PLoS Negl Trop Dis. 2014;8(2):e2682. https://doi.org/10.1371/
journal.pntd.0002682.
200. Shi B, Xia S, Yang GJ, et al. Inferring the potential risks of H7N9 infection by spatiotem-
porally characterizing bird migration and poultry distribution in eastern China. Infect Dis
Poverty. 2013;2(1):8. https://doi.org/10.1186/2049-9957-2-8.
201. Zhu G, Liu J, Tan Q, Shi B. Inferring the spatio-temporal patterns of dengue transmission
from surveillance data in Guangzhou, China. PLoS Negl Trop Dis. 2016;10(4):e0004633.
https://doi.org/10.1371/journal.pntd.0004633.
202. Shi B, Zhan XM, Zheng JX, et al. Identifying key bird species and geographical hotspots of
avian influenza A (H7N9) virus in China. Infect Dis Poverty. 2018;7(1):97. https://doi.org/
10.1186/s40249-018-0480-x.
203. Zhang Y, Cheung WK, Liu J. A unified framework for epidemic prediction based on poisson
regression. IEEE Trans Knowl Data Eng. 2015;27(11):2878–92. https://doi.org/10.1109/
TKDE.2015.2436918.
204. Pei H, Yang B, Liu J, et al. Group sparse Bayesian learning for active surveillance on epidemic
dynamics. In: 32nd AAAI Conference on Artificial Intelligence. Menlo Park: AAAI; 2018.
p. 800–7.
205. Chen H, Yang B, Liu J. Partially observable reinforcement learning for sustainable
active surveillance. In: International Conference on Knowledge Science, Engineering and
Management. Berlin: Springer; 2018. p. 425–37.
206. Yang B, Guo H, Yang Y, et al. Modeling and mining spatiotemporal patterns of infection
risk from heterogeneous data for active surveillance planning. In: 28th AAAI Conference on
Artificial Intelligence. Menlo Park: AAAI; 2014. p. 493–9.
207. Shi B, Zheng J, Qiu H, et al. Risk assessment of malaria transmission at the border area of
China and Myanmar. Infect Dis Poverty. 2017;6(1):108. https://doi.org/10.1186/s40249-017-
0322-2.
208. Tan Q, Liu J, Shi B, et al. Public health surveillance with incomplete data-spatio-temporal
imputation for inferring infectious disease dynamics. In: 6th IEEE International Conference
on Healthcare Informatics. Piscataway: IEEE; 2018. p. 255–64.
209. Shi B, Xia S, Liu J. A complex systems approach to infectious disease surveillance and
response. In: International Conference on Brain and Health Informatics. Berlin: Springer;
2013. p. 524–35.
Index
G M
Game theory Machine learning
game-theoretic analysis, 12, 51, 55, 56, 58, group sparse Bayesian learning, 94
70 Poisson regression, 94
game-theoretic model, 12, 51–53 reinforcement learning, 94
minority game, 51 Macroscopic scale, 2
Geographic hotspots, 94 Mathematical modeling, 2, 15, 16, 31, 51, 67
Google Flu Trends, 6 Measles, 18, 49, 71
Measles–mumps–rubella (MMR), 7, 49
Metapopulation-based models
H subpopulations, 20 (see also Age-structured
Herd immunity host population)
herd immunity threshold, 6, 34, 35 Methodological paradigm
population-level herd immunity, 85 computational paradigm, 3
vaccine-induced herd immunity, 6 philosophical and methodological
Heterogeneity perspective, 89
heterogeneity of a host population, 8, 9, 18, theoretical paradigm, 3
32 (see also Age-structured host top-down reductionist approaches, 90
population) Microscopic scale, 2
heterogeneity of pathogens, 18 Mobility
HIV, 23, 24 human air-travel networks, 5
Human population mobile behaviors, 5
heterogeneous host population, 18 Morbidity, 6, 35, 36
human host population, 93 Mortality, 1, 6, 35, 36
Multidisciplinary, 93
Multiple scales, 91
I
Impact factors, 1–6, 11, 74, 93, 94
Individual-level model, 18–19, 75 N
Infection dynamics, 1, 2, 15, 93 Network models
Infection tracing, 22, 23 network-based disease model, 91
See also Contact tracing network inference, 94
Infectious disease control, 1, 20, 23, 31, 59, 71, network structure, 31, 59, 79, 81
92, 94 Next generation matrix (NGM), 38, 39
Infectivity, 8, 9, 20, 21, 29, 31, 32, 36, 38, 40,
41
Influenza
P
H1N1
Peak-incidence rate, 82
2009 Hong Kong H1N1 influenza
Predictive methods, 2
epidemic, 8, 10, 11, 25, 26, 29,
Prescriptive methods, 2
30, 32, 37, 42, 45, 47, 62, 67, 77
Problem solving, 9, 90, 91, 94
H7N9, 94
human swine influenza, 25, 42–45, 80
influenza-like disease, 32, 37, 56, 59, 70,
74, 87 R
seasonal influenza, 18 Real-world epidemic scenario, 12, 25
Integrated decision-making process, 10, 57, 62 Real-world problems, 15, 93, 94
Interdisciplinary methods, 90 Remote sensing data, 5
Internet-based media, 6 Reproduction matrix, 38, 39
Intervention measures, 3, 5, 6, 9, 16, 37, 38, See also Next generation matrix (NGM)
41, 44, 45, 47, 90 Reproduction number, 9, 12, 16, 22, 29, 37,
Intervention strategies 39–41, 47, 54, 55, 64, 74, 85, 86,
optimal prioritization, 37 88
prioritized interventions, 38–39, 45–47 See also Basic reproduction number (R0 )
112 Index