M. Bubak et al. (Eds.): ICCS 2008, Part I, LNCS 5101, pp. 590–599, 2008. © Springer-Verlag Berlin Heidelberg 2008 An Individual-Based Model of Influenza in Nosocomial Environments Boon Som Ong 1 , Mark Chen 2 , Vernon Lee 2 , and Joc Cing Tay 1,* 1 ROSS Scientific Pte Ltd Innovation Centre, Units 211-212, 16 Nanyang Drive Singapore 637722 * joccing@ross-scientific.com 2 Department of Clinical Epidemiology, Tan Tock Seng Hospital Moulmein Road, Singapore 30843 Abstract. Traditional approaches in epidemiological modeling assume a fully mixed population with uniform contact rates. These assumptions are inaccurate in a real epidemic. We propose an agent-based and spatially explicit epidemiol- ogical model to simulate the spread of influenza for nosocomial environments with high heterogeneity in interactions and susceptibilities. A field survey was conducted to obtain the activity patterns of individuals in a ward of Tan Tock Seng Hospital in Singapore. The data collected supports modeling of social be- haviors constrained by roles and physical locations so as to achieve a highly precise simulation of the ward’s activity. Our results validate the long-standing belief that within the ward, influenza is typically transmitted through staff and less directly between patients, thereby emphasizing the importance of staff- oriented prophylaxis. The model predicts that outbreak size (and attack rate) will increase exponentially with increasing disease infectiousness beyond a cer- tain threshold but eventually tapers due to a target-limited finite population. The latter constraint also gives rise to a peak in epidemic duration (at the threshold level of infectiousness) that decreases to a steady value for increasing infec- tiousness. Finally, the results show that the rate of increase in distinct cumu- lated contacts will be highest within the first 24 hours and gives the highest yield for contact tracing among patients that had prolonged periods of non- isolation. We conclude that agent-based models are a necessary and viable tool for validating epidemiological beliefs and for prediction of disease dynamics when local environmental and host factors are sufficiently heterogeneous. Keywords: Agent-based modeling, Spatially-explicit model, Epidemiology, In- fluenza, Contact patterns. 1 Introduction During the SARS crisis, hospitals were found to be especially vulnerable to outbreaks [1-3]. Hospitals are also susceptible to nosocomial influenza, and rapid cross- infection between healthcare workers and patients can occur [4-6]. In spite of this, there has been little work in simulating the potential spread of infections in the hospital setting, with a granularity that allows policy makers and infection control