Constructing Probabilistic Human Mobility Patterns for Estimating Total Exposures and Pathogen Transmission Using Aggregate Mobile Phone Data

While phone location data is widely collected by providers and shared under special agreements with data aggregators, it is not available for research or routine surveillance purposes. Moreover, raw phone data may expose personal travel patterns which would not be ethically or lawful to use. One large data aggregator, Advan (https://advanresearch.com/) provides aggregate traffic patterns at weekly and monthly temporal resolution for many points of interest. This aggregate data is mostly used for commercial applications, but we argue that it can be also used to develop a model of probabilistic human mobility patterns at weekly or monthly resolution. Here we provide a stochastic algorithm that using the aggregate data provides realistic human mobility patterns at the intended temporal resolution. In the proposed model we associate a "person" with both a unique census block as the residence and another census block as the workplace. These human mobility patterns are dynamic and can be used in exposure models and in epidemiological models of epidemic spreads. Because these models do not assume any exposure or pathogen transmission mechanisms, they can be coupled to any external exposure fields or transmission model of infectious disease. Here we show how they can used to build a probabilistic model of exposure for an arbitrary exposure filed E(X, t) and a SIR contagion model. This probabilistic model can be developed at monthly or weekly time resolutions depending on the data source available and the specific programmatic needs.