Epidemic and intervention modelling – a scientific rationale for policy decisions? Lessons from the 2009 influenza pandemic

Abstract

Problem

Outbreak analysis and mathematical modelling are crucial for planning public health responses to infectious disease outbreaks, epidemics and pandemics. This paper describes the data analysis and mathematical modelling undertaken during and following the 2009 influenza pandemic, especially to inform public health planning and decision-making.

Approach

Soon after A(H1N1)pdm09 emerged in North America in 2009, the World Health Organization convened an informal mathematical modelling network of public health and academic experts and modelling groups. This network and other modelling groups worked with policy-makers to characterize the dynamics and impact of the pandemic and assess the effectiveness of interventions in different settings.

Setting

The 2009 A(H1N1) influenza pandemic.

Relevant changes

Modellers provided a quantitative framework for analysing surveillance data and for understanding the dynamics of the epidemic and the impact of interventions. However, what most often informed policy decisions on a day-to-day basis was arguably not sophisticated simulation modelling, but rather, real-time statistical analyses based on mechanistic transmission models relying on available epidemiologic and virologic data.

Lessons learnt

A key lesson was that modelling cannot substitute for data; it can only make use of available data and highlight what additional data might best inform policy. Data gaps in 2009, especially from low-resource countries, made it difficult to evaluate severity, the effects of seasonal variation on transmission and the effectiveness of non-pharmaceutical interventions. Better communication between modellers and public health practitioners is needed to manage expectations, facilitate data sharing and interpretation and reduce inconsistency in results.

Background

Outbreak analysis and mathematical modelling have played an important role in the planning of the public health response to infectious disease outbreaks, epidemics and pandemics. These tools can help quantify the risk to human health posed by a new infectious organism, rapidly analyse and interpret limited data in the early stages of an epidemic, and use such analysis to predict future developments. All of these actions are necessary to evaluate the potential benefits of specific control measures. Statistical and mathematical models integrate and synthesize epidemiological, clinical, virologic, genetic and sociodemographic data to gain quantitative insights into patterns of disease transmission.¹

Soon after the emergence of A(H1N1)pdm09 in North America in 2009, the World Health Organization (WHO) convened an informal mathematical modelling network of public health experts and mathematical modelling groups in academic institutions. This network worked collaboratively to characterize the dynamics and impact of the pandemic and demonstrate the potential outcome of various interventions in different settings. This work was published in formats suitable for various audiences, including technical experts, policy-makers and the general public. Emphasis was on adapting and interpreting experiences from developed countries for application to low-resource settings.²

In this paper we provide an overview of the analysis and mathematical modelling undertaken during and following the 2009 pandemic, with an emphasis on research of relevance to public health planning and decision-making.

Pre-pandemic planning

Mathematical models have been used by ministries of health and governments to inform influenza pandemic planning in many developed countries. Planning assumptions – in which disease severity (e.g. the case-fatality ratio) and the transmission characteristics (e.g. the basic reproductive number, R₀) of the influenza virus are based on past pandemics (e.g. 1918, 1957, 1968) or potential pandemic viral strains (e.g. highly pathogenic avian influenza subtype H5N1) – are modelled to estimate the potential incidence trajectory of infected and fatal cases and the likely impact of control measures. Such information makes it possible to determine the medical and non-medical interventions required, the feasibility of containment and the optimal size of the medication stockpile and best use of pharmaceuticals once a pandemic begins.^3,4

Modelling during the 2009 pandemic

During the 2009 A(H1N1) pandemic, members of the influenza modelling community worked closely with public health agencies and ministries of health. Efforts focused on rapidly quantifying transmission to provide evidence for WHO pandemic phase changes;⁵ assessing severity⁶ and seasonality;^7,8 interpreting epidemiologic trends over time; measuring antigenic changes in the virus⁹ and assessing the potential impact of interventions.^10,11 Modellers in public health agencies also provided input into study design and helped to identify key data to address public health challenges.^12,13

Although mathematical modelling was used for planning purposes and to explore mitigation options in many countries of the Americas (e.g. Canada, Mexico and the United States of America), Europe (e.g. France, Germany, the Netherlands and the United Kingdom of Great Britain and Northern Ireland), Asia (e.g. China and Japan) and the Pacific (Australia and New Zealand), it was not sophisticated simulation modelling, but rather, real-time statistical analyses based on mechanistic transmission models and the interpretation of emerging epidemiologic and virologic data that most often informed policy decisions on a day-to-day basis. These results were widely disseminated in peer-reviewed publications, yet much of the advice and guidance derived from the modelling was never formally published but was presented instead during face-to-face meetings with national policy-makers, with occasional documentation in meeting minutes or reports.

Early outbreak investigations provided data that proved critical for characterizing the epidemiology of infection with A(H1N1)pdm09 in communities, schools and households. They made it possible to estimate R₀, serial intervals and age-specific clinical attack rates and to track the temporal distribution of secondary infections.⁵ These parameters were essential in assessing the burden of infection with A(H1N1)pdm09 and disease severity. Early rapid analyses with limited data performed to inform policy decisions were then followed by more detailed studies that made use of more reliable and complete data. For example, retrospective analyses of publicly available epidemiologic and virologic data from several countries provided a unique opportunity to compare the spread of the same virus in different countries and to determine if differences in latitude, temperature, humidity, population age structure or mixing patterns affected transmission dynamics.¹⁴

Policy decisions about the optimal use, effectiveness and cost-effectiveness of pharmaceutical (e.g. antivirals or vaccines) and non-pharmaceutical interventions (e.g. school closures, social distancing measures, masks) were heavily influenced by the results of mathematical modelling.^11,15 Since antivirals and vaccines were in short supply or unavailable in many countries at the start of the 2009 pandemic (and, in some countries, throughout the pandemic), modelling provided guidance for the optimal use of such interventions to reduce transmission by targeting school-aged children and other high transmitters, or to reduce morbidity and mortality by targeting high-risk individuals, such as those with chronic underlying conditions or pregnant women.

School closure was a policy option considered in some countries. Although A(H1N1)pdm09 caused milder disease than initially expected, some countries, such as Argentina and Japan, closed all schools early in their epidemic by extension of or overlap with school holidays, while others closed only certain schools. Modelling proved useful in weighing the potential health benefits of school closures against their social and economic costs. During the pandemic, modelling groups in several countries, including Australia, China (Hong Kong Special Administrative Region), France, Japan, the Netherlands, the United Kingdom and the United States, confidentially shared unpublished results with WHO and other WHO Members States via the WHO mathematical modelling network to inform decision-making.²

Lessons and challenges

It is difficult to reliably assess the extent to which modelling informed decision-making during the 2009 pandemic. This is because modellers and biostatisticians in most countries provided advice as part of highly interactive multidisciplinary advisory groups, whose contributions often consisted of presenting formal modelling results and a mechanistic dynamic perspective on the unfolding epidemic. Furthermore, policy-makers needed to weigh not only the potential health benefits of different interventions, but also the economic, social, political and ethical costs associated with particular policy options. What is certain, however, is that the insights gained from statistical modelling informed policy in many countries.

Despite good achievements, several challenges remain. To set realistic expectations, improved communication between policy-makers and the public about what modelling can and cannot deliver is essential. It is also important to effectively communicate how prediction differs from scenario modelling. Scenarios are useful in planning for assessing the effectiveness of interventions and various policy options, but they are not predictions. The failure to communicate uncertainty was problematic and led to misunderstanding of modelling results during the 2009 pandemic.

Political pressures during the 2009 pandemic were intense. The data available often failed to match the information needs of policy-makers. Key decisions, such as how much vaccine to purchase, had to be made despite great uncertainty surrounding the likely overall health impact of the pandemic. Analyses conducted in “real time” using limited data are always subject to substantial uncertainty, and central estimates and worst-case assessments are invariably subject to change as more data become available.

As expected, fundamental data gaps early in the pandemic, especially on population infection rates over time, made it very difficult to accurately assess its impact and disease severity. Many countries had reliable and timely data on the demand for primary health care due to influenza-like illness but very limited data on the proportion of individuals who were becoming infected and seeking health care. As a result, the numbers of symptomatic cases who were seeking medical care could not be used to estimate the overall incidence of influenza infection in the community. Real-time serosurveillance data could have filled this gap, but such data were not available in any country before the first peak of pandemic influenza activity.¹³ Other data gaps also made it difficult to evaluate the likely impact of seasonal variation on transmission^7,8 or the effectiveness of many non-pharmaceutical interventions, particularly in low-resource settings.

Several important lessons were learnt from the 2009 pandemic (Box 1). Chief among them is that modelling is not a substitute for data. Rather, modelling provides a means for making optimal use of the data available and for determining the type of additional information needed to address policy-relevant questions. We must not, however, take too negative a view of achievements in 2009. Modellers provided a quantitative framework for analysing surveillance data and for understanding both the dynamics of the pandemic and the impact of the interventions. Arguably, it was such timely yet straightforward data analysis and interpretation that most informed the policy decisions made during the first months of the pandemic, rather than sophisticated pandemic simulation modelling of the type used for pre-pandemic planning.

Box 1. Summary of main lessons learnt.

• Better serosurveillance and monitoring of community illness attack rates could have filled data gaps (e.g. not knowing the underlying infection attack rate over time) that made it difficult to estimate disease severity and to predict peak pandemic activity.

• Sharing and analysis of detailed epidemiologic data during the pandemic was crucial for informing decisions, but data from low-resource countries was limited.

• Communication between modelling groups and policy-makers was good in several countries but could be improved further.

In future, better coordination will be needed not only among modellers and modelling groups, but also with clinicians, epidemiologists, virologists and public health decision-makers. It will also be important to reduce inconsistencies and build consensus across modelling groups. These goals will be facilitated by the establishment of national and international modelling networks such as those that were created in 2009.