Workshop on Epidemiology of Infectious Diseases: Emerging Challenges - IMS

Workshop on Epidemiology of Infectious Diseases: Emerging Challenges
(4 - 8 Jan 2010)

~ Abstracts ~

Estimating and projecting HIV prevalence for countries with generalized HIV/AIDS epidemics
Leontine Alkema, National University of Singapore

The Joint United Nations Programme on HIV/AIDS, UNAIDS, uses a deterministic susceptible-infected model for estimating HIV prevalence, and for making short term projections. I will discuss the UNAIDS model and the Bayesian melding approach which is used to assess the uncertainty in the UNAIDS estimates and projections. In this approach, prevalence time series data from antenatal clinics, as well as information on the input parameters of the UNAIDS model are used to derive probabilistic HIV prevalence estimates and projections. A calibration method is used to take into account population prevalence estimates.

This presentation is based on a collaborative research project with UNAIDS in 2007 and is joint work with Adrian E. Raftery and Samuel J. Clark (University of Washington, US).

Challenges for national-scale disease spread and control simulators: parameterization and economic considerations
Roman Carrasco, National University of Singapore

New advances in computer capacity have allowed for the development of complex national-scale simulation models for pandemic and bioterrorism preparedness and epidemics control. These models are capable of representing persons and their contact networks as individual agents (agent-based models) bringing valuable insight on the effectiveness of school closures, use of anti-virals and vaccines on the containment of a pandemic. These complex national-scale simulators present two main caveats: they are rarely parameterized or validated from observed past epidemics and they overlook economic considerations at the micro (agent) and macro level.

In this presentation, I will discuss several promising methods and research approaches that would help overcome those caveats: (i) the application of statistical inference methods that can improve the inference and analysis of complex models such as approximate Bayesian computation, particle filtering, maximum likelihood methods and regression meta-modeling; (ii) experimental psychology learning and imitation algorithms that would allow to evaluate the potential behaviors of the agents during the course of a pandemic and how they can undermine or boost management efforts; and (iii) the integration of the spread of an epidemic with partial equilibrium economic models will de illustrated.

The application of Bayesian computational techniques to agent-based models and the inclusion of behavioral models of the agents is a new and exciting venue of research that will enhance our predictive capability regarding the management of potential future pandemics, bioterrorist attacks and disease epidemics in general.

Incorporating partnership and gap lengths in STI modeling - why it matters and how to do it
Mark Chen, Tan Tock Seng Hospital

In classical models of STI transmission, risk behavior is often quantified through some variant of mean partner change rates. However, partnership behaviour can also be framed in terms of partnership and gap lengths, the latter being a measure of the time spent between successive partnerships. We have previously shown using the pair modelling framework that the ?gap? between partnerships is a key determinant of the transmission potential of pairing behaviour, but that partnerships of medium length can potentially be more efficient than shorter partnerships even for infections like gonorrhoea which have short infectious periods. In this presentation, we explore in greater detail some of these concepts. Firstly we illustrate the importance of intra-pair re-infections in Susceptible-Infectious-Susceptible (SIS) type STIs, and how this phenomenon is inadequately accounted for in classical models of STI transmission. Secondly, we draw on concepts proposed by others to show how partnership data can be adjusted for to account for both the number of partners reported and the types of partnership behaviours reported. Finally, we use empirical data on sexual behaviour to illustrate that risky gap and partnership behaviour are correlated with higher partner mean partner change rates, and that both may act synergistically to provide better explanations for the higher STI risk in some population subgroups.

Predicting the H1N1 outbreak in Singapore, on-line and in real-time: can we do the same for dengue?
Alex Cook, National University of Singapore

As the novel strain of influenza A (H1N1) arrived in Singapore, we set up a sentinel network of family-doctor volunteers, who sent daily reports of influenza-like and other acute respiratory illnesses that they had seen. These we used to form daily predictions of the future trajectory of the pandemic in Singapore, by fitting a compartmental model to the data using a technique called particle filtering, which forecast the timing of the peak to within a week. These predictions were made publicly available on the internet, updated daily as new reports came in, and featured in the national press. They were used in planning essential service continuity, implementing other studies at appropriate times, and we hope in communicating the current status of the outbreak to both the medical community and the general public. In this talk I describe how we developed the network and created the forecasts, and present some of the difficulties in attempting to reproduce the system for an endemic, vector-borne disease such as dengue.

Model-based evaluation and cost-effectiveness analysis of Methicillin-resistant Staphylococcus aureus intervention policies
Ben Cooper, Mahidol University, Thailand

Aims

Hospitals are faced with a wide choice of methicillin-resistant Staphylococcus aureus (MRSA) control methods, with little available evidence on cost-effectiveness to aid in decision making. We aimed to use transmission dynamic models to assess the cost-effectiveness of MRSA infection control programmes in intensive care units (ICUs) from the perspective of a healthcare decision maker who manages resources at a national level.

Methods

MRSA infections increase morbidity and mortality and can increase duration of hospital stay. Interventions that prevent MRSA infections will change both costs and health outcomes: costs due to intervention-related expenditures as well as savings from infections avoided; and health benefits due to avoided morbidity and reduced risk of mortality. We developed a dynamic, stochastic, individual-based model of MRSA transmission in ICUs to enable comparison of control policies. We synthesized evidence from multiple sources, and propagated
parameter uncertainty through the model. Incremental costs and benefits of the different policy options were evaluated and Cost-Effectiveness Acceptability Frontiers and the expected value of perfect information were derived.

Results

Screening all patients on admission with an expensive, rapid molecular test was the most cost-effective screening option. Policies including patient isolation were cost-effective, whilst decolonization of positive patients was cost saving. The optimal decision depended on the decision maker?s willingness to pay for health benefits. Using the United Kingdom's National Health Service willingness to pay threshold of £20,000-£30,000 per quality adjusted life year, screening all patients using a rapid molecular test, followed by decolonisation of identified positives had the greatest probability of being cost-effective in the absence of resistance to the decolonizing agents. The expected value of perfect information on the parameters was high, reflecting large uncertainties in transmission parameters and decolonisation effectiveness.

Conclusions

These models allow comparison of the many available interventions and can provide a useful guide to policy makers as well as allowing us to quantify the expected value of reducing key uncertainties, thus providing a rational basis for setting future research priorities.

Studies of pandemic and seasonal influenza in Hong Kong
Ben Cowling, The University of Hong Kong, Hong Kong

2009 pandemic influenza A (H1N1) emerged in early 2009 and rapidly spread around the world. In Hong Kong, community transmission was first identified in mid-June, and the government immediately proactively closed schools, kindergartens and childcare centers for 2 weeks initially. The closures were subsequently extended to the summer vacation. Analysing data on confirmed cases and hospitalizations associated with novel H1N1, we show that pandemic influenza transmissibility was relatively low throughout the closure period and summer vacations, while the age mix of cases varied before and after school closures and vacations. Serogic surveys through the course of the epidemic suggest that around 15% of the general population were infected with H1N1 during the first wave. Household studies provide detailed information on the transmission dynamics of respiratory viruses. We conducted a secondary transmission study in 100 households with index cases laboratory-confirmed to have seasonal or pandemic influenza A virus infection in July and August 2009, with detailed information on viral shedding and humoral antibody responses. We also report preliminary results of a vaccine trial in 110 households, where seasonal influenza vaccination may have interesting consequences for susceptibility to pandemic influenza.

Robust analyses of epidemic type data, with application to earthquake and infectious disease outbreaks
Jason Fine, University of North Carolina at Chapel Hill, USA

We consider regression modelling for a single realization of a modulated renewal process, as commonly encountered in "outbreak" data in earthquake and infectious disease applications. Previous work has focused almost exclusively on fitting parametric models using point process likelihood. We propose a general estimation framework using pseudo martingale estimating equations naively treating gap times as independent, which is generally applicable to semiparametric and parametric models. It facilitates a semiparametric extension of a popular parametric earthquake model, the ETAS model. Interestingly, a careful asymptotic analysis shows that ignoring dependence amongst gap times generally yields estimators which are asymptotically equivalent to those based on i.i.d. data. Simulations and empirical analyses of Taiwanese earthquake sequences illustrate the methodolog's practical utility. To relax the intensity assumpion for point processes, a rate model is proposed, where the covariate process based on previous event history may not capture the dependencies in the sequence. We consider partial likelihood based inferences under a semiparametric multiplicative rate model developed in the pseudo martingale set-up. Under an intensity model, gap times may naively be used in the partial likelihood with variance estimation employing the observed information matrix. Under a rate model, the gap times cannot be treated as independent and studying the partial likelihood is much more challenging, involving ergodic assumptions and limit theory for alpha mixing sequences. Novel variance estimators are proposed which extend cluster variance estimators and block bootstrapping to partial likelihood. Simulation studies demonstrate the improved performance of our inferences relative to intensity based analyses. An application to the spread of smallpox in 1967 in the town of Abakaliki, Nigeria, illustrates our approach.

Social contact network modeling for the spread of infectious diseases in Singapore
Xiuju Fu, Institute of High Performance Computing

Heterogeneities in societies affect epidemics in real life. Social contacts show one of these heterogeneities due to people exposing themselves with different ways and time duration to the variation of infectious agents at concentrating locations. Simultaneously, people also transport the agents between different locations when moving and contacting. Hence, social contact network study provides an opportunity to investigate the effect of such heterogeneity. In this talk, the modeling of social contact network of Singapore for investigating infectious disease spread is introduced. The social contacts are analyzed and evaluated with the appropriate data collected as the inputs for contact network construction and later disease transmission analyses. The results from the contact networks, disease simulations and modeling will then be presented. The aim of the work is to develop preparedness plans and epidemic intervention in different scenarios.

On applications of Richards model to epidemic modeling
Ying-Hen Hsieh, China Medical University, Taiwan

Management of infectious diseases, such as the 2009 pH1N1 influenza pandemic, often imposes great challenge for mathematical modeling due to limited information on outbreak/pathogen, stochastic variations of disease epidemiology, and changing case definitions and surveillance practices. The Richards model and its variants are useful to fit the cumulative epidemic curve in order to obtain estimates for the turning points of the outbreak, the basic reproductive number R₀, as well as the final outbreak size for a wave of infections in the absence of artificial interventions. Examples of its applications to SARS [1-2], dengue [3-4], and 2009 pH1N1 outbreaks will be presented.

References
1. Hsieh YH, Lee JY, Chang HL. SARS epidemiology. Emerging Infectious Diseases 2004;10(6):1165-1167.
2. Hsieh YH, Cheng YS. Real-time forecast of multi-wave epidemic outbreaks. Emerging Infectious Diseases 2006;12(1):122-127.
3. Hsieh YH and Ma S. Intervention Measures, Turning Point, and Reproduction Number for Dengue, Singapore, 2005. Am J Trop Med Hyg. 2009;80:66-71.
4. Hsieh YH, Chen CWS. Turning Points, reproduction number, and Impact of Climatological events on Multi-Wave Dengue Outbreaks. Trop. Med. Internat. Health. 2009;16(4):1-11.

Optimal design of influenza transmission studies in households
Brendan Klick, University of Hong Kong, Hong Kong

The recent outbreak of H1N1 swine flu and the continued threat posed by the possible emergence of H5N1 avian flu coupled with the already large mortality and morbidity associated with seasonal influenza have led to a rise of clinical trials planned which examine the efficacy of antivirals and nonpharmaceutical interventions to prevent secondary transmission of influenza in the past decade. These studies frequently use RT-PCR methods to permit laboratory-confirmation of influenza virus infection from nasal and throat swabs collected on home-visits. Using a Monte Carlo simulation approach calibrated on empirical data from field studies conducted in Hong Kong, we examine the optimal number of home-visits in terms of cost benefit to the trial. In general, more home-visits lead to more accurate and robust estimates of secondary influenza infection rates at the cost of a smaller overall sample size, thus increasing standard error of parameter estimates. In terms of accurately estimating secondary attack rates, we find that as the money available for a clinical trial increases the benefits of more home visits becomes greater. However, in terms of power to detect a difference in secondary attack rate between intervention and control groups using either the relative risk or odds ratio for a study at any cost level, we found that a design with only one home visit optimally timed after the primary infection will be most powerful.

Adopting mathematical modelling for public health decision making
Vernon Lee, National University of Singapore

Mathematical modelling has been widely used in recent years to show the computational effectiveness of various strategies against diseases. The most powerful use of such models is their direct application to decision making to influence public health measures. This presentation describes several unique scenarios where different models may be useful to public health decision making. Firstly, the use of health economic models have been increasing, and we present 2 models for influenza and human papillomavirus that will be useful in determining the potential burden of disease and the importance and quantity of preparedness measures to be adopted. Another use of models is for to provide real time information to assist with public health policies, and we show how a simple real time predictive mathematical model can have substantial impact on public health planning. And finally, models can be used to show the effectiveness of clinical and public health interventions, and we show how a model used during pandemic influenza validated the preventive measures used in the field.

HIV and STDs: What can we do using mathematics?
Jie Lou, Shanghai University, China

It includes two models:

1) Since cancer remains a significant burden for HIV-infected individuals, gaining insight into the epidemiology and mechanisms that underlie AIDS-related cancers can provide us with a better understanding of cancer immunity and viral oncogenesis. We studied an HIV-1 dynamical model incorporating the AIDS-related cancer cells. We discuss the existence, the stability properties and the biological meanings of its steady states, in particular for the positive one: cancer-HIV-healthy cells steady state. We find conditions for Hopf bifurcation of the positive steady state, leading to periodic solutions, sequences of period doubling bifurcations and appearance of chaos. Further, chaos and periodic behavior alternate. Our results are consistent with some clinical and experimental observations.

2) We examine epidemic threshold and dynamics for sexually transmitted diseases (STDs) spread using a multiple susceptible-infected--removed-susceptible ODE model on scale-free network. We derive the threshold for the epidemic to be zero in infinite scale-free network. For a hard cut off scale-free network, we also prove the stability of disease-free equilibrium and the persistence of STDs infection. The effects of two immunization schemes, including proportional scheme and targeted vaccination, are studied and compared. We find that targeted strategy compare favorably to a proportional scheme in terms of effectiveness. Theory and simulations both prove that an appropriate condom using has prominent effect to control STDs spread on scale-free networks.

Estimates of pandemic influenza H1N1-2009 in Singapore
Stefan Ma, Ministry of Health

The first case of pandemic influenza A(H1N1) [H1N1-2009] was detected in Singapore on 26 May 2009, one month after the first cases of novel influenza A(H1N1) was reported in the US. Knowing the size of the outbreak would provide very useful indicator for health policy makers to plan and allocate healthcare resources. The procedures of estimating number of cases of H1N1-2009 which made use of the number of polyclinic attendances for acute respiratory infection and influenza-like illness and the weekly prevalence of H1N1-2009 will be discussed.

On the estimation of R₀ from the initial phase of an outbreak of a vector-borne infection
Eduardo Massad, University of São Paulo, Brazil

The magnitude of the basic reproduction ratio R₀ of an epidemic can be estimated in several ways, namely, from the final size of the epidemic, from the average age at first infection, or from the initial growth phase of the outbreak. In this paper, we discuss this last method for estimating R₀ for vector-borne infections. Implicit in these models is the assumption that there is an exponential phase of the outbreaks, which implies that in all cases R₀ > 1. We demonstrate that an outbreak is possible, even in cases where R0 is less than one, provided that the vector-to-human component of R₀ is greater than one and that a certain number of infected vectors are introduced into the affected population. This theory is applied to two real epidemiological dengue situations in the Southeastern part of Brazil, one in which R₀ is less than one, and one in which R₀ is greater than one. In both cases, the model mirrors the real situations with reasonable accuracy.

Quantifying effect of responses used in influenza H1N1 2009 swine flu outbreak in Australia using an individual-based model
George Milne, The University of Western Australia, Australia

The role of simulation models in determining optimal responses to emergent diseases; with application to swine-origin H1N1.

We modify a highly detailed simulation model of a actual community of 30,000 individuals previously developed to quantify the mitigating impact of alternative interventions for a H5N1 pandemic.

These modifications reflect the known biology of H1N1 2009.

Specifically we have examined interventions used in Australia to mitigate the illness attack rate.

These include social distancing measures (school closures and home isolation) and the use of antivirals for treatment and prophylaxis.

Results indicate optimal intervention strategies to be used until a suitable vaccine becomes available.

The relationship between observable and unobservable quantities in epidemic modelling
Hiroshi Nishiura, University of Utrecht, The Netherlands

When we analyze the empirical data set of an infectious disease epidemic, it is essential to understand what the observed data represent in an epidemiological sense. Since many epidemic models employ mechanistic assumptions for unobservable infection process, the unobservable mechanisms have to be inferred by modelling observable events. In this talk, two issues are discussed. First, we consider the dynamics of directly transmitted diseases with two types of infected population, i.e. asymptomatic and symptomatic individuals. We propose a revised form of the renewal process which permits us to estimate the transmission potential and explicitly clarify the difference between generation time and serial interval. Second, we consider the ideal length of reporting interval in epidemic data. We introduce simple algorithms to estimate the reproduction number as a function of time, adjusting the reporting interval to generation time of a disease and demonstrating a clear relationship among the generation time distribution, reporting interval and growth rate of an epidemic.

Agent based simulations for modelling STD transmission dynamic
Adrian Roellin, National University of Singapore

When modelling the transmission dynamics of sexually transmitted diseases (STD), agent based simulations are an attractive alternative to the classic deterministic compartmental models. The reason for this is that, if various aspects such as the complex interaction between formation and dissolving of sexual partnerships (in particular concurrent partnerships), gender, culturally motivated behaviour, core/non-core groups, etc. are to be included, compartmental models become mathematically increasingly intractable. Agent based simulations offer an elegant alternative as they allow to build in these aspects in a very natural way.

However, developing agent based simulations can be challenging and can consume a great part of the actual research project's time. We discuss some challenges by means of an R package prototype that allows to simulate models ranging from simple SIS models up to complex age and gender structured populations, allowing for screening and partnership notification programs. This software is currently being used to investigate the effect of such programs on the prevalence of /Chlamydia trachomatis/ and similar STDs.

[collaboration with N. Low, C. Althaus, J. Heijne (University of Bern), P. Barton (University of Birmingham)]

Infectious diseases epidemiology - a clinicians perspectives on some unanswered questions
Paul Ananth Tambyah, National University of Singapore

Infectious diseases are no longer thought of as historical curiosities. The current H1N1 2009 Influenza A pandemic together with recent epidemics of SARS, Avian Influenza and the Nipah virus have reminded us that Infectious Diseases continue to pose a threat to human health. These pose many challenges to clinicians especially in terms of recognizing the first case of a novel emerging pathogen. Many sentinel surveillance systems have been built along disease syndromes or unusual occurrences but few have been able to predict the occurrence of novel outbreaks. Once a novel pathogen emerges, there are always challenges of calibrating the response to the pathogen - this involves a fine balance between strict measures to protect healthcare workers and the general public and maintaining the normal functioning of hospitals, schools and institutions such as the military and police. In nosocomial infections, outbreaks are frequent events - many approaches have been taken from classic "shoe leather" epidemiology to modern molecular epidemiology to try to uncover the spread of these outbreaks. At the same time, interventions are often introduced reactively to outbreak situations and it is difficult to evaluate the efficacy of these interventions. These are some of the challenges that clinicians face and would appreciate help from our colleagues in mathematics and modeling.

Infectious diseases epidemiology - a clinicians perspectives on some unanswered questions
Julian Tang, National University of Singapore

Infectious diseases are no longer thought of as historical curiosities. The current H1N1 2009 Influenza A pandemic together with recent epidemics of SARS, Avian Influenza and the Nipah virus have reminded us that Infectious Diseases continue to pose a threat to human health. These pose many challenges to clinicians especially in terms of recognizing the first case of a novel emerging pathogen. Many sentinel surveillance systems have been built along disease syndromes or unusual occurrences but few have been able to predict the occurrence of novel outbreaks. Once a novel pathogen emerges, there are always challenges of calibrating the response to the pathogen - this involves a fine balance between strict measures to protect healthcare workers and the general public and maintaining the normal functioning of hospitals, schools and institutions such as the military and police. In nosocomial infections, outbreaks are frequent events - many approaches have been taken from classic "shoe leather" epidemiology to modern molecular epidemiology to try to uncover the spread of these outbreaks. At the same time, interventions are often introduced reactively to outbreak situations and it is difficult to evaluate the efficacy of these interventions. These are some of the challenges that clinicians face and would appreciate help from our colleagues in mathematics and modeling.

Best viewed with IE 7 and above