Influenza Forecasting With Mathematical Models

The populations of cities are not laboratory mice, so actions to protect their health often cannot be rigorously tested. But people who are responsible for protecting population health must nonetheless make decisions.

At best, those decisions are based on evidence from previous similar events in similar populations, but often this evidence is scanty or not applicable to the situation at hand. That has led to the popularity of mathematical models that simulate populations and allow policymakers to ask "what if" questions about the interventions that they are considering.

Khazeni and colleagues have developed a sophisticated model to examine the health impacts and cost-effectiveness of various influenza mitigation strategies in a city similar to New York City for influenza A H5N1 and novel H1N1. The model quantifies the impact of virus epidemiologic features and mitigation strategies on illness and costs-something of interest to us as the New York City Department of Health and Mental Hygiene prepares for fall influenza season. They conclude that expanded use of an adjuvanted vaccine would be the most effective and cost-effective strategy to limit death due to H5N1 influenza. For H1N1, they conclude that vaccinating one third of the population would shorten a pandemic and save more lives if implemented earlier in the season.

Often the greatest utility of predictive mathematical models is not any single conclusion drawn from them but rather their ability to clarify assumptions about the dynamics of disease in a population and to determine which inputs have the greatest impact on the outcome of interest. Policymakers can then focus their attention on the most pressing objective (for example, vaccination or antiviral distribution), and epidemiologists can determine which indicators should most intensively be monitored and studied. For example, Khazeni and colleagues find that the population vaccination coverage required to mitigate spread of H5N1 increases by 10% as vaccine efficacy decreases by 10%. Knowing the dynamics of this relationship is important for policymakers, who can use the information to make decisions on the intensity and resources poured into vaccination campaigns.

Unfortunately, sometimes the inputs that most affect outcomes in models are also subject to the most uncertainty. In Khazeni and colleagues' model, 2 important inputs with a high degree of uncertainty are the impact of nonpharmaceutical interventions (for example, school closure) and the efficacy of the vaccine. Varying vaccine efficacy dramatically changes the population coverage required to shorten the pandemic; therefore, numerical predictions of the model are only as good as the strength of the assumed value of this variable.