MIDAS Consultation on Social Behavior Abstracts Meeting Report

Brookings Institution
June 20-21, 2006
Behavioral Epidemiology Conference

Temporal waves in mortality from influenza in 1918

Derek Cummings
The incidence of death due to influenza in the pandemic of 1918 exhibited multiple temporal peaks in many locations across the globe. The cause of these waves is not known. Several mechanisms have been hypothesized to be involved including seasonal forcing, genetic changes of the virus over time and behavioral changes in response to the epidemic. Here we examine the spatial temporal distribrution of these temporal waves across the United States. The pattern of these temporal waves varied across the United States and within states. We present an analysis of these patterns and two simple models of the impact of behavioral changes and seasonality on the epidemic to determine which mechanisms might create spatial temporal patterns that are consistent with the data.


Effects of Behavioral Changes in Smallpox and Influenza Models

Sara Del Valle
Communicable diseases are highly sensitive to how rapidly people reduce their contact activity patterns and to the precautions that the population takes to reduce the transmission of the disease. Recent experiences with the SARS epidemics show that an outbreak of a deadly disease would generate dramatic behavioral changes.

However, models for infectious diseases have focused on analyzing the impact of traditional intervention strategies such as isolation of infectives, contact tracing, quarantine of contacts, ring vaccination, and mass vaccination. In this talk I will present a model in which some individuals lower their daily contact activity rates once an epidemic has been identify in their community. I will demonstrate that even gradual and mild behavioral changes can have a dramatic impact in slowing the epidemic and reducing the total number of cases. I conclude that for simulations of infectious diseases to be useful, they must consider the impact of behavioral changes. This is especially true if the model predictions are being used to guide public health policy.


Coupled Contagion Dynamics of Fear and Disease: Mathematical and Computational Explorations

Joshua M. Epstein, Jon Parker
We model two interacting contagion processes: one of disease and one of fear about the disease. Individuals can "contract" fear through contact with individuals who are infected with the disease (the sick), infected with fear only (the scared), and infected with both fear and disease (the sick and scared). Scared individuals--whether sick or not--may remove themselves from circulation with some probability, which affects the course of the disease epidemic proper. If we allow individuals to recover from fear and return to circulation, the coupled dynamics become quite rich, and include multiple waves of infection, such as occurred in the 1918 flu.


Epidemiological modeling and human contact patterns

Stephen Eubank
Before we can incorporate crisis behavior into epidemiological models, we need to understand behavior in normal circumstances. Fortunately, it is possible to estimate human contact patterns under normal conditions in large urban areas using activity surveys. I will discuss how we can use this seemingly ego-centric data to construct global social networks. There are some perhaps surprising features of the resulting networks that I will trace to individual activity patterns. I will illustrate the effects that variations in network estimates have on the spread of infectious disease, and why it is so important to understand what has happened historically during crises. I will conclude with a few comments on getting access to and working with the data.


Behaviorally Realistic Analysis

Baruch Fischhoff
Analyses should seek to achieve behavioral realism by recognizing: (a) the role of expert judgment in how knowledge is elicited and represented, (b) the scientific basis for predicting the behaviors shaping the risks (e.g., whether people observe quarantines, report suspected cases, or abandon potentially exposed pets), and (c) the informational needs of those depending on their results. The talk will consider the intellectual challenges facing the individuals attempting cope with avian flu in their personal lives and in their professional lives, whether as analysis or policy makers.


Modeling social conflicts of public health strategies

Alison Galvani
Public health strategies that are optimal for the community are not necessarily optimal for the individual. For example, if most of the community is vaccinated, it can be best for an individual to refuse vaccination since the individual receives the benefits of reduced disease prevalence while avoiding anticipated adverse effects of the vaccine. This type of conflict can undermine public health programs.

Epidemiological models that ignore economic constraints or incentives are likely to perform poorly in predicting the evolution of epidemics. Similarly, economic and behavioral models that neglect the temporal dynamics of disease are equally ineffective. I will discuss the integration of epidemiology, mathematical modeling and economics to provide an improved predictive framework.


Under What Conditions Should We Model Human Behavior?

Roz D. Lasker
The objective of modelers of deadly infectious disease outbreaks is to help planners and policy makers identify the most effective containment strategies. The objective of people experiencing such an outbreak is to take actions that have the best chance of protecting themselves and their families. As I will illustrate with findings from the Redefining Readiness study, these objectives can conflict - with potentially serious consequences - if the actions that modelers and planners decide that people should take are not feasible for them, are risky or harmful for them, go against their natural protective inclination, or come from officials they don't trust. Should we invest our energy trying to predict and model human behavior under such suboptimal conditions, where a substantial number of people face daunting barriers or trade-offs trying to protect themselves? Or would it be more worthwhile to try to develop better policies by identifying feasible conditions that would make it possible for the greatest number of people to protect themselves and modeling the impact of that behavior on disease containment? In this session, I will share specific information currently being generated by the Redefining Readiness local demonstration sites that can support this kind of modeling, paying particular attention to issues related to "social distancing."


Bayesian Analysis of the 1918 Influenza Pandemic in Baltimore, MD and Newark NJ, USA

Yue Yin 1, Donald Burke 1;2, Derek Cummings 1, Thomas A. Louis 1
1 Johns Hopkins Bloomberg School of Public Health; 2 University of Pittsburgh

Though the influenza pandemic of 1918 has been thoroughly studied, several features remain unexplained. With an emphasis on formal statistical estimates of model parameters for a stochastic S-I-R model, we have used a conducted Bayesian analyses of the population level, daily incidence and influenza related deaths in Baltimore, MD and Newark, NJ for the fall of 1918. We find that a stochastic S-I-R model fits the Baltimore incidence poorly, the duration in the asymptomatic, infective state is longer than commonly assumed and results are quite sensitive to prior distributions. The model fits the Newark incidence well, asymptomatic duration is compatible with biologic understanding and results are sturdy. In each city, the best fit to data is produced by a time-varying rt, fmixing rate g£fdisease transmissibility g, and Rt, the reproductive number, that have their peak in mid-epidemic. In Baltimore these functions are highly peaked; in Newark they are relatively flat. We speculate on what produces the poor fit in Baltimore and the need for time-varying rt and Rt, but have no direct information to support our speculations. Candidate causes include inhomogeneous mixing (spatial spread or aggregation of homogeneous, sub-epidemics), time-varying mixing rate (due to changes in social dynamics) and time-varying infectivity (unlikely in this short time frame). An understanding of the mechanism driving this temporal pattern and the substantial differences between Baltimore and Newark can aid in decision making during a new pandemic. As important, formal statistical analyses document assumptions, process complex data to produce estimates, associated uncertainties and structure sensitivity analyses. Formal analyses extract estimates from "live" data and from the output of agent based models, serving to update information on inputs to micro-simulations.


Escaping the Flu: A Historical Assessment of Nonpharmaceutical Disease Containment Strategies Employed by Selected Communities During the Second Wave of the 1918-1920 Influenza Pandemic

Howard Markel, Alexandra Stern, J. Alexander Navarro, Joseph R. Michalsen
In the absence of adequate stocks of an effective vaccine and/or antiviral drugs, the United States may have to rely on nonpharmaceutical interventions (NPI) to contain the spread of an infectious disease outbreak until pharmacological means become available. Because many of these NPI are costly and socially disruptive, their effectiveness and practicality need to be understood before their implementation or incorporation into a response plan.

We undertook a historical evaluation of these NPI as employed by American communities during the second wave (September-December 1918) of the 1918-1920 influenza pandemic. A team of medical historians from the University of Michigan Medical School's Center for the History of Medicine visited these communities to access and collect available primary source material from libraries, archives, and other private and public holdings. We selected 7 communities that reported relatively few if any cases of influenza, and no more than one influenza-related death while NPI were enforced during the second wave of the 1918-1920 influenza pandemic: San Francisco Naval Training Station, Yerba Buena Island, California; Gunnison, Colorado; Princeton University, Princeton, New Jersey; Western Pennsylvania Institution for the Blind, Pittsburgh, Pennsylvania; Trudeau Tuberculosis Sanatorium, Saranac Lake, New York; Bryn Mawr College, Bryn Mawr, Pennsylvania; and Fletcher, Vermont. Because of the apparently reduced morbidity and low mortality these communities experienced during the second wave of the pandemic, we have labeled them "provisional influenza escape communities." "Provisional" means that we cannot definitively determine on the basis of the historical evidence available to us if these communities sustained their low morbidity and mortality rates because of policy decisions made by their community leaders and public health officials, because the virus skipped some communities altogether and varied in its behavior in other communities (viral normalization patterns), or because of other factors such as population density, geography, and good fortune.

Historical research is fraught with all the problems and limitations of retrospective studies. The researcher may be helped or hindered by numerous investigators, recorders, and collectors of information who preceded him or her and generally performed their work without a common reference framework or even sets of uniform definitions and concepts.

The historian must also rely upon archivists who may or may not have preserved this material and cataloged it in a way that aids retrieval.

These issues are some, but hardly the only, limitations of any historical study, including this one. Nevertheless, history represents an essential arrow in the quiver of human inquiry.

One would like to think that the 7 communities we identified fared better than others because of the NPI they enacted. We cannot prove that for any of them, although the case is, perhaps, strongest for the Naval Training Station at Yerba Buena Island and, possibly, Gunnison, Colorado. Further complicating our task, in addition to the quality and quantity of information available for study, is the fact that some of these communities were sparsely populated and geographically isolated, and all of them were subject to the vagaries of how the influenza virus normalized in affected populations.

Limited by these factors, we have reached two major conclusions:

(1) Protective sequestration (the shielding of a defined and still healthy group of people from the risk of infection from outsiders), if enacted early enough in the pandemic, crafted so as to encourage the compliance of the population involved without draconian enforcement measures, and continued for the lengthy period of time at which the area is at risk, stands the best chance of protection against infection.

When implemented successfully, protective sequestration also involves quarantine of any outsider who seeks entry, self-sufficiency in the supplies necessary for daily living, enforcement of regulations when necessary (including fining and jailing), and the ability of those sequestered to entertain themselves and maintain some semblance of a normal life.

(2) Available data from the second wave of the 1918-1920 influenza pandemic fail to show that any other NPI (apart from protectivesequestration) was, or was not, effective in helping to contain the spread of the virus. American communities engaged in virtually the same menu of measures, including: 1) the isolation of ill persons; 2) the quarantine of those suspected of having direct contact with the ill; 3) social distancing measures, such as the cancellation of schools and mass gatherings; 4) reducing an individual's risk for infection, (e.g., face masks, hand washing, respiratory etiquette); and 5) public health information campaigns and risk communications to the public. Despite these measures, most communities sustained significant illness and death; whether these NPI lessened what might have been even higher rates had these measures not been in place is impossible to say on the basis of available historical data. Moreover, we could not locate any consistent, reliable data that would support the conclusion that face masks, as available and as worn during the 1918-1920 influenza pandemic, conferred any protection to the populations that wore them.

However inconclusive are the data from 1918, the collective experiences of American communities from the pandemic are truly noteworthy, especially in light of the fact that faced with a pandemic today we would likely rely on many of these same NPI to attempt to contain the spread of the infection until pharmacological supplies of vaccine and antivirals were available.


Examples of Prevalence Elastic Behavior

Mead Over, World Bank
Prevalence elastic behavior refers to systems that generate cycles of change in response to disease frequence, with one example being the response to HIV/AIDS. As prevalence increases, the government implements active control and prevention programs. As they become effective, people may change behavior that generates a rise in cases. Examples of HIV/AIDS, condom use and price, and pandemic influenza, suggest that unless we know the impact of effective free treatment on risk behavior, we have no real idea of the net effects of the policy on morbidity or mortality. Successful interventions against a pandemic may lead to people's unwillingness to implement intervention strategies.

Vaccines: Integrated Economic and Epidemiological Models

Michael Kremer, Harvard University; Christopher Snyder, Dartmouth College; Heidi Williams, Harvard University
Vaccines interfere with the transmission of a disease, and this positive externality leads vaccine developers to capture a private benefit which is smaller than the social benefit of their innovation. In this paper, we build an integrated economic and epidemiological model which formalizes this externality and yields several results. First, the externalities from vaccination are greatest for rare diseases. As a conseqence, the ratio of the social value of the vaccine relative to the private incentives for vaccine research and development (R&D) is largest in the case of rare diseases, and indeed can be arbitrarily large in percentage terms for sufficiently rare diseases. Hence, holding constant the total burden of disease, firms will find developing vaccines for the more common but less serious diseases like the flu more profitable than for rarer but more deadly diseases.