Disaggregation Regression and Multi-Model Evaluation for Predicting Dengue Risk in Africa

Abstract

Dengue risk mapping is essential for estimating disease burden, and informing targeted surveillance and control efforts. Current approaches to risk mapping vary widely in their methodology, data sources, output metrics and applications. Many existing approaches focus on predicting ecological suitability and produce high-resolution risk maps based on environmental conditions, yet high-resolution incidence maps remain scarce, leaving a critical gap in guiding precise, location-specific interventions. The prediction of disease incidence or transmission intensity remains relatively uncommon in disease ecology, largely due to data limitations, reporting biases, and the inherent complexity that arises from transmission dynamics. In this study, we applied disaggregation regression modelling to downscale aggregated dengue case data from 14 countries in Central and South America, generating fine-resolution incidence estimates that we subsequently projected onto the African continent. We then compared the resulting predictions from the incidence-based risk map with three widely used approaches: vector suitability index, dengue environmental suitability index, and mechanistic transmission potential (Index P). The disaggregation model achieved relatively strong predictive accuracy within the training region (mean correlation = 0.72) and showed partial alignment with reported burden across Africa (Spearman {rho} = 0.33). Other risk maps exhibited similar or weaker correlations with reported cases in Africa, including {rho} = 0.33 for dengue environmental suitability, {rho} = 0.32 for transmission potential and {rho} = 0.23 for Aedes aegypti suitability. Disaggregation regression offers a valuable tool for translating reported case data into spatially explicit estimates of burden, bridging the gap between ecological risk and epidemiological relevance. While spatial agreement was high in parts of coastal West Africa across the different risk map approaches, notable divergences highlight the distinct assumptions underlying each framework. Authors SummaryDengue is a mosquito-borne viral disease with expanding global impact. Accurately mapping dengue risk is essential for identifying areas of high transmission and targeting interventions effectively. Most current approaches to mapping dengue risk focus on environmental suitability for the virus or its mosquito vector, rather than estimating actual disease burden. In this study, we used an incidence-based approach; disaggregation regression, to estimate dengue cases at high spatial resolution using national and regional case data from Latin America. We then applied the model to Africa, where surveillance data are limited, and compared its predictions to three other common types of dengue risk maps. Our results showed that while all approaches provided some insight into geographic risk patterns, they often highlighted different areas as priorities. Our incidence-based model captured both where dengue might occur and how intense transmission may be, helping bridge the gap between environmental and ecological suitability of transmission and real-world disease burden. This approach can support more informed decision-making in areas with limited surveillance and guide targeted control efforts.