Tuberculosis remains a major public health concern in Uganda, one among the thirty high TB burden countries globally. Despite national progress, gaps persist due to asymptomatic disease, diagnostic limitations, and uneven access to healthcare within the country. This study implemented the Epi-control platform, an AI-driven predictive modelling tool, to predict community-level hotspots and support data-driven active case-finding (ACF). Using retrospective chest X-ray screening data, we integrated demographic, environmental, and human development indicators from open-source databases to model TB risk at sub-parish level. A proprietary Bayesian modelling framework was deployed and validated by comparing TB yields between predicted hotspots and non-hotspot locations. Across Uganda, the model identified significantly higher TB yields in hotspot areas (risk ratio = 1.69, 95% CI 1.41-2.02; p < 0.001). The Central and Western regions showed the highest concentrations of hotspots, consistent with their population density and urbanization patterns. The results show that the model prioritized areas with higher observed ACF yield in this retrospective dataset, supporting its potential operational use for screening prioritization under similar implementation conditions. The results demonstrate that AI-based predictive modelling can enhance the efficiency of ACF by targeting high-risk areas for screening. Integrating such predictive tools within national TB programmes may support screening planning and resource prioritization; prospective evaluation and external validation are needed to assess generalisability and incremental impact.
Amanya et al. (Wed,) studied this question.