The Impact of the Unstructured Contacts Component in Influenza Pandemic Modeling

Individual based models have become a valuable tool for modeling the spatiotemporal dynamics of epidemics, e. g. influenza pandemic, and for evaluating the effectiveness of intervention strategies. While specific contacts among individuals into diverse environments (family, school/workplace) can be modeled in a standard way by employing available socio-demographic data, all the other (unstructured) contacts can be dealt with by adopting very different approaches. This can be achieved for instance by employing distance-based models or by choosing unstructured contacts in the local communities or by employing commuting data. Methods/Results. Here we show how diverse choices can lead to different model outputs and thus to a different evaluation of the effectiveness of the containment/mitigation strategies. Sensitivity analysis has been conducted for different values of the first generation index G(0), which is the average number of secondary infections generated by the first infectious individual in a completely susceptible population and by varying the seeding municipality. Among the different considered models, attack rate ranges from 19.1% to 25.7% for G(0)= 1.1, from 47.8% to 50.7% for G(0)= 1.4 and from 62.4% to 67.8% for G(0)= 1.7. Differences of about 15 to 20 days in the peak day have been observed. As regards spatial diffusion, a difference of about 100 days to cover 200 km for different values of G(0) has been observed. Conclusion. To reduce uncertainty in the models it is thus important to employ data, which start being available, on contacts on neglected but important activities (leisure time, sport mall, restaurants, etc.) and time-use data for improving the characterization of the unstructured contacts. Moreover, all the possible effects of different assumptions should be considered for taking public health decisions: not only sensitivity analysis to various model parameters should be performed, but intervention options should be based on the analysis and comparison of different modeling choices.