Mobility resolution needed to inform predictive epidemic models for spatial transmission from mobile phone data

Mobility flows extracted from mobile phone data have been extensively used in recent years to inform spatial epidemic models for the study of various infectious disease epidemics, including Malaria, Cholera, and Ebola. Most recently, the COVID-19 pandemic marked a historic shift, as it led to the sharing of unprecedented fine-scale mobility data. This abundancy of data illuminated the geographical variability in transmission patterns and underscored the importance of the use of mobility data for public health questions. Little attention has been devoted however to (i) the definition of the mobility process that is relevant to the epidemic spread, and (ii) the mobility data resolution that is needed to describe the invasion dynamics. We take advantage of a real-world dataset, gathered from mobile phone users in Senegal to define three epidemiological couplings between locations, based on different characterizations of the mobility process, and at varying resolution levels. They are based respectively on: (i) the total number of displacements between any two municipalities on two consecutive calls (Displacement-based D ); (ii) the number of calls made by residents in each location (Location-based L ); (iii) the most visited location of residents during daytime (Most visited location-based C ). To assess the impact of the different coupling definitions on the epidemic diffusion, we use them to inform mobility in a spatial epidemic model. We found that preserving any displacement on the observed trajectories from mobile phone data does not capture the epidemiological link between different locations, for infections where daily mobility is important (e.g. airborne or direct contact diseases). Most importantly, we found that at the country scale, places in which individuals spend most of their time including workplaces, schools or particular point of interests like restaurants or theater and are the dominant driver of disease diffusion. In fact, tracking in detail individual activities beyond home and all visited locations during the day does not add epidemiological important information. Novel paradigms for the release of mobile phone data to researchers can therefore be envisioned that strengthen privacy and confidentiality, while at the same time providing enough details - specifically aggregated home-visited locations coupling - to inform predictive epidemic models.