Street-level ambient fine particulate matter and tuberculosis: A machine learning and Bayesian spatiotemporal analysis.

Tuberculosis (TB) remains a leading cause of death from infectious diseases globally, yet evidence linking short-term ambient fine particulate matter (PM2.5) exposure to TB incidence at fine spatial scales is limited. We investigated the street-level associations between short-term PM2.5 exposure and TB onset in Jinan, China, involving a total of 8291 newly diagnosed active TB cases from 2015 to 2019. Daily street-level PM2.5 concentrations were estimated using an extreme gradient boosting (XGBoost) model integrated with geographic information systems (GIS). A Bayesian hierarchical spatiotemporal model with a population offset was applied to quantify exposure-response relationships, reporting Incidence Rate Ratios (IRRs). Stratified analyses were conducted by age, sex, season, and residential location. Results indicated that each 10 μg/m3 increase in PM2.5 at lag 3 days was associated with a higher TB incidence (IRR = 1.02, 95% CI: 1.01-1.03), with effects persisting to lag 6 (IRR = 1.01, 95% CI: 1.00-1.02). Subgroup analyses revealed significantly stronger effects among the elderly (aged ≥65 years), females, rural residents, and during cold seasons (November-April). Sensitivity analyses and SHAP analysis confirmed the robustness of the findings and the interpretability of the exposure model. These findings underscore the critical need for targeted public health interventions that prioritize protection of high-risk subgroups (the elderly, females, and rural residents), and implement heightened pollution control during cold seasons to mitigate the TB burden.