Ideally, the data used for robust spatial prediction of disease distribution should be both high-resolution and spatially expansive. However, such in-depth and geographically broad data are rarely available in practice. Instead, researchers usually acquire either detailed epidemiological data with high resolution at a small number of active sampling sites, or more broad-ranging but less precise data from passive case surveillance. We propose a novel inferential framework, capable of simultaneously drawing insights from both passive and active data types. We developed a Bayesian latent point process approach, combining active data collection in a limited set of points, where in-depth covariates are measured, with passive case detection, where error-prone, large-scale disease data are accompanied only by coarse or remotely-sensed covariate layers. Using the example of malaria, we tested our method's efficiency under several hypothetical scenarios of reported incidence in different combinations of imperfect detection and spatial complexity of the environmental variables. We provide a simple solution to a widespread problem in spatial epidemiology, combining latent process modelling and spatially autoregressive modelling. By using active sampling and passive case detection in a complementary way, we achieved the best-of-both-worlds, in effect, a formal calibration of spatially extensive, error-prone data by localised, high-quality data.
Keywords: Bayesian modelling; N-mixture models; disease mapping; imperfect detection; latent point process; spatial epidemiology.