Unveiling TB Patient Mobility: Trajectory Reconstruction and Geographic Hubs of Transmission in Kampala

Keywords: Tuberculosis, Deep Learning, Human Mobility, Spatio-temporal Analysis, Public Health
Abstract Type: Paper Abstract

Abstract

Tuberculosis (TB) remains a major public health challenge due to its high global burden, drug-resistant strains, and complexities in diagnosis and treatment, particularly in resource-limited settings. Since TB transmission often occurs not only through close social relationships but also through casual contact in daily life, understanding the mobility patterns of TB patients is crucial for informing prevention strategies. While many studies have examined human mobility, few have focused specifically on TB patients. In this study, we collected Cell Phone Record (CDR) data from TB patients in Kampala, Uganda. To address the sparsity of CDR data, we developed a transformer-based model, BERT4Traj, designed for reconstructing patient trajectories using BERT-like masking and prediction mechanisms. This model captures spatiotemporal patterns and user background information to infer detailed trajectory paths, allowing us to reconstruct patient movements with finer temporal granularity. We further examine how the mobility patterns of TB patients differ from one another and how these patterns evolve over time, particularly in relation to health care visits. Additionally, we explore geographic hubs where TB patients may have contact, identifying areas at higher risk of transmission. These insights into the spatial-temporal mobility patterns of TB patients can help optimize public health interventions by focusing on critical locations and ensuring patients receive continuous care.