Empirical Validation of an Adaptive Spatial Hierarchical Bayesian SEIR Model for HIV Transmission Dynamics in Kenya

This study presents the empirical validation of an adaptive spatial hierarchical Bayesian SEIR model developed for modeling HIV transmission dynamics in Kenya. Using comprehensive HIV surveillance data from 15 strategically selected Kenyan counties spanning 2015-2024 (1,620 county-month observations), the model was validated through comparative analysis with benchmark approaches, sensitivity testing, and policy scenario evaluation. Data sources included the Kenya AIDS Indicator Survey, National AIDS Control Council, Kenya Health Information System, and Kenya National Bureau of Statistics. The validation framework employed multiple performance metrics including Root Mean Square Error (RMSE), Mean Absolute Error (MAE), correlation coefficients, and Area Under the Curve (AUC) for hotspot detection. Results demonstrated that the proposed model achieved superior predictive accuracy with 48% improvement in prediction errors (RMSE = 0.035) compared to non-spatial approaches and 22% improvement over static spatial models. Diebold-Mariano tests confirmed statistically significant improvements (p < 0.001) across all benchmark comparisons. Sensitivity analysis revealed transmission rate as the most influential parameter (first-order Sobol index = 0.34), while the model maintained robust performance with up to 20% missing data. The validated model identified substantial spatial heterogeneity in basic reproduction numbers ranging from 0.3 in Wajir to 3.2 in Homabay counties. These findings validate the model for evidence-based resource allocation, early warning systems for emerging transmission hotspots, and adaptive HIV program management in Kenya and similar resource-limited settings.