Geospatial risk mapping and time-series forecasting of malaria transmission among under-fives in Bauchi and Oyo States, Nigeria

Abstract

Malaria remains a major public health challenge in Nigeria, with marked geographic and seasonal variability influencing transmission dynamics. Understanding spatial and temporal heterogeneity is essential for optimising malaria control strategies and improving the targeting of interventions. This study investigated under-five malaria transmission patterns in two epidemiologically distinct Nigerian states using integrated time-series forecasting and Bayesian geospatial risk mapping. A two-stage analytical framework was applied to monthly confirmed under-five malaria cases obtained from the District Health Information System 2 (DHIS2) between January 2014 and December 2024. At the state level, autoregressive integrated moving average models with exogenous variables (ARIMAX) were developed using climatic variables obtained from the NASA Prediction of Worldwide Energy Resources (POWER) database. Candidate lag structures were identified using cross-correlation analysis, and competing models were compared using the Akaike Information Criterion (AIC), Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE). Final models were validated using out-of-sample data before generating forecasts for January-December 2025. At the Local Government Area (LGA) level, Bayesian Besag-York-Mollié (BYM) models were used to estimate annual relative malaria risk. Bauchi exhibited a substantially higher malaria burden and stronger seasonal transmission than Oyo throughout the study period. One-month lagged rainfall improved forecasting performance in both states, whereas temperature did not improve predictive accuracy and was excluded from the final models. Forecasts for 2025 projected continued seasonal transmission, with substantially higher malaria incidence in Bauchi than in Oyo. Bayesian spatial analysis identified persistent high-risk LGAs in Bauchi, particularly Warji, despite declining risk over time, while Ido remained the principal persistent hotspot in Oyo. Integrating ARIMAX forecasting with Bayesian geospatial risk mapping provides a comprehensive framework for characterising malaria transmission dynamics and identifying persistent transmission hotspots. The findings highlight the value of combining routine surveillance data with climate information to strengthen malaria early warning systems and support geographically targeted malaria control strategies in Nigeria. Not applicable.