University of Washington - Department of Statistics
Advisor: Mark Handcock
Influenza pandemics pose a serious global health concern. The recent A (H1N1) influenza pandemic caused 18,500 lab-confirmed deaths, and mutation of the A (H5N1) "avian" influenza virus could also cause a pandemic with an estimated 60% case mortality rate in humans, requiring fast analysis of intervention and containment strategies. When a new influenza virus emerges with pandemic potential, stochastic simulation models are used to assess the effectiveness of different strategies. One such model is operated by University of Washington researchers in the Center for Statistics and Quantitative Infectious Disease (CSQUID), and two other research groups operate similar large-scale simulators. These detailed simulation models are based on Census and transportation data and assume that people make random contacts within their households, classes and schools, and workgroups and workplaces. The model assumptions regarding social contact behavior are not informed by surveys of social contact behavior. We develop a realistic, dynamic model of social contact behavior within a high school based on friendship network data and a survey of contact behavior in high schools. In our model, students are more likely to contact their friends and make longer durations of contacts to friends, and the probability of transmission during a contact is a function of contact duration. We compare different versions of the contact network model to assess the impact of various network structures on the disease transmission process, and we select a final model based on these comparisons. We perform disease simulations over this contact network and compare epidemic outcomes to simulations over a random mixing model. We then simulate a targeted antiviral prophylaxis intervention strategy and a grade closure intervention strategy and compare estimated intervention impact under the two models. We find our network-based model to estimate a lower probability of epidemic, lower final sizes, and a later peak date than a random mixing model. We show that the estimated impact of the two interventions is different under the network-based model than a random mixing scenario and the direction and magnitude of the difference varies with transmission probability. Our findings have implications for policy recommendations based on models assuming random mixing. We recommend further exploration of network structure on disease dynamics, consideration of the potential bias created by the random mixing assumption, and further work to integrate network structure into current simulators.