Model-based estimates of chikungunya epidemiological parameters and outbreak risk from varied data types.

Assessing the factors responsible for differences in outbreak severity for the same pathogen is a challenging task, since outbreak data are often incomplete and may vary in type across outbreaks (e.g., daily case counts, serology, cases per household). We propose that outbreaks described with varied data types can be directly compared by using those data to estimate a common set of epidemiological parameters. To demonstrate this for chikungunya virus (CHIKV), we developed a realistic model of CHIKV transmission, along with a Bayesian inference method that accommodates any type of outbreak data that can be simulated. The inference method makes use of the fact that all data types arise from the same transmission process, which is simulated by the model. We applied these tools to data from three real-world outbreaks of CHIKV in Italy, Cambodia, and Bangladesh to estimate nine model parameters. We found that these populations differed in several parameters, including pre-existing immunity and house-to-house differences in mosquito activity. These differences resulted in posterior predictions of local CHIKV transmission risk that varied nearly fourfold: 16% in Italy, 28% in Cambodia, and 62% in Bangladesh. Our inference method and model can be applied to improve understanding of the epidemiology of CHIKV and other pathogens for which outbreaks are described with varied data types.