Human behavior can have a significant impact on infective disease dynamics. In turn, a complex interplay of disease spread, awareness of the disease, and population beliefs affect human behavior \cite{Funk2010}. The mobility of a person, whether that person is infected or not, is a particularly important factor of disease spread \cite{Rizzo2014}. Awareness-induced changes in movement patterns, such as a decision to avoid unsafe infected areas, often have a detrimental effect and might lead to even higher disease spreading, since they result in bringing the infection into previously isolated communities \cite{Wang2012,Meloni2011}. At the same time, international travel restrictions have been shown to have a limited impact on disease spreading, due to the high heterogeneity of human mobility patterns \cite{bajardi2011human}. In fact, it is this heterogeneity, both in terms of population behavior and a-priori infections, that drives disease development. In her discussion of HIV and other STDs transmission Aral argues that bridge groups, such as truckers, the police and the military personnel, transmit infections from highly infected groups, e.g., sex workers, to previously uninfected populations \cite{Aral2000}. Our work is founded on the above observation, and we propose a model that explicitly takes the transmission of risk into account. While previous models consider artificial simulations \cite{Buscarino2014} and long-distance \cite{Merler2010} or multiscale \cite{balcan2009multiscale} mobility networks in order to quantify possible outcomes of different metapopulations movement patterns on disease spread, we build our model upon individual mobility and interactions, as recorded by fine-grain cellular network traces.
Our work relies on mobile phone call records for estimating risk transfer in a population. The suitability of CDRs for tracking population movements and identification of spatial events in populations has been shown by Bengtsson et al. \cite{Bengtsson2011} and Candia et al. \cite{Candia2008}. Furthermore, when it comes to infectivity modeling, in \cite{Eames2009} Eames et al. show that simple interaction potential measures, such as the total number of a user’s connections (total degree), perform almost as well as more complex measures of interaction, such as individually weighted links. In further work the total node degree might be used to approximate a user’s potential for contact. Finally, in this work we do not modify the interaction network over time. Such modifications, orthogonal to our approach, are discussed in \cite{Kamp2010}, and can be accounted for by having a time-dependent contact network.