Comparing and integrating human mobility data sources for measles transmission modeling in Zambia.

Quantifying population mobility is crucial in developing accurate models of infectious disease dynamics. Increasingly, multiple data sources are available to describe individual and population mobility in a single location; however, there are no methods to systematically integrate these data. Combining information from these data sets may be valuable and help mitigate inherent biases in each data set due to sampling, censoring, and recall. We examined four commonly used data sources (mobile phone records, travel survey, Demographic and Health Survey, and Facebook location information) to quantify subnational travel patterns in Zambia. First, we explored summary metrics of mobility from each data set. Estimates of the probability of a trip and location of travel varied across data sets, with some data quantifying twice the frequency of travel as others. Then, we developed a departure-diffusion model that is able to produce a single estimate of travel by combining these data sets. When multi-data set models included mobile phone records, this data source dominated estimates given the broader spatial coverage. We then used a metapopulation model to simulate a measles outbreak to identify how these different data sets and models would impact estimates of the spatial spread of a highly infectious disease. We found that using travel survey data to parameterize mobility resulted in the introduction of cases in 98% of districts compared to less than 50% when mobile phone data or Facebook data were used. This study highlights the need for methods that facilitate integrating multiple data sets to improve validity of mobility estimates and resultant infectious disease transmission dynamics.