Binomial geostatistical model.

Malaria remains a public health concern. Monitoring the fine-scale heterogeneity of the malaria burden enables more targeted control efforts. Although malaria indicator surveys (MIS) have been crucial in evaluating the progress of malaria control interventions, they are only designed to provide a cross-sectional national and regional malaria disease burden. Recent advances in geostatistical methods allow us to interpolate national survey data to describe subnational disease burden that is crucial in informing targeted control. A binomial geostatistical model employing Markov chain Monte Carlo (MCMC) parameter estimation methods is used to understand the spatial drivers of malaria risk in Kenya and to predict malaria risk at a fine-scale resolution, including identifying hotspots. A total of 11,549 children aged six months to 14 years from 207 clusters were sampled in this survey and used in the present analysis. The national malaria prevalence based on the data was 8.4%, with the highest in the lake endemic zone (18.1%) and the lowest in the low-risk zone (<1%). The analysis shows that elevation, proportion of insectcide treated net (ITN) distributed, rainfall, temperature and urbanization covariates are all significant predictors of malaria transmission. The 5x5 Km resolution maps show that malaria is heterogeneous in Kenya, with hotspot areas in the lake endemic area, the coastal areas, and some parts of the shores of Lake Turkana and Kajiado. The high-resolution malaria prevalence maps produced as part of the analysis have shown that Kenya has additional malaria hotspots, especially in areas least expected. These findings call for a rethinking of malaria burden classification in some regions for effective planning, implementation, resource mobilization, monitoring, and evaluation of malaria interventions in the country.