Predicting IPTp3 Coverage at Health-Facility Level Using XGBoost Machine Learning and Spatial Diagnostics in Kebbi State, Nigeria.

Malaria in pregnancy (MiP) remains a major public health challenge in Nigeria, contributing substantially to maternal anemia, low birth weight, and adverse neonatal outcomes. Effective delivery of intermittent preventive treatment in pregnancy (IPTp), particularly completion of at least three doses (IPTp3), is a core indicator of malaria prevention performance within antenatal care (ANC) services. This study examined the spatial distribution and facility-level determinants of IPTp3 coverage in Kebbi State, northwestern Nigeria, using routine health service and population data. Monthly facility-level MiP service delivery data (January 2022–November 2025) were obtained from the national DHIS2 platform and integrated with high-resolution 2025 population surfaces from WorldPop. Thiessen (Voronoi) polygons were generated around health facilities to delineate catchment areas, from which population characteristics were extracted. IPTp3 coverage was calculated as the ratio of cumulative IPTp3 doses to first ANC attendance (ANC1). Spatial aggregation and hotspot analysis were conducted at ward and local government area (LGA) levels. Facility-level predictive modelling employed a regularized XGBoost regression framework using aggregated service delivery, population, and facility structural variables. Marked spatial heterogeneity in IPTp3 coverage was observed across Kebbi State, with ward-level coverage ranging from near zero to values approaching one. Facility-level modelling demonstrated strong predictive performance (R² = 0.85; RMSE = 0.084; MAE = 0.061). Antenatal care attendance volume was the most influential predictor (≈31% of model gain), followed by LLIN distribution, malaria testing activity, reproductive-age female population size, and facility catchment area. Residual analysis revealed that while most LGAs performed within expected ranges, localized clusters of underperforming facilities persisted, particularly in Koko-Besse, Kalgo, and Suru, whereas Wasagu-Danko significantly outperformed model expectations. These findings highlight the combined influence of service delivery capacity and population context on IPTp3 uptake and provide a spatially explicit, policy-relevant framework for targeting malaria-in-pregnancy interventions at facility levels.