Abstract
This article explores four key questions about statistical models developed to describe the recent past and future of vector-borne diseases, with special emphasis on dengue: (1) How many variables should be used to make predictions about the future of vector-borne diseases? (2) Is the spatial resolution of a climate dataset an important determinant of model accuracy? (3) Does inclusion of the future distributions of vectors affect predictions of the futures of the diseases they transmit? (4) Which are the key predictor variables involved in determining the distributions of vector-borne diseases in the present and future? Examples are given of dengue models using one, five or 10 meteorological variables and at spatial resolutions of from one-sixth to two degrees. Model accuracy is improved with a greater number of descriptor variables, but is surprisingly unaffected by the spatial resolution of the data. Dengue models with a reduced set of climate variables derived from the HadCM3 global circulation model predictions for the 1980s are improved when risk maps for dengue's two main vectors (Aedes aegypti and Aedes albopictus) are also included as predictor variables; disease and vector models are projected into the future using the global circulation model predictions for the 2020s, 2040s and 2080s. The Garthwaite-Koch corr-max transformation is presented as a novel way of showing the relative contribution of each of the input predictor variables to the map predictions.