Predictive modeling and spatiotemporal analysis of TB in Argentina: Advancing control efforts through machine learning.

OBJECTIVES

To improve prediction and understanding of TB dynamics in Argentina, identifying key risk factors and high-incidence areas to inform surveillance and public health control strategies.

STUDY

DESIGN

Retrospective observational study.

METHODS

We applied (i) machine learning models (Histogram-Based Gradient Boosting, XGBoost, Random Forest, and Logistic Regression) to predict treatment outcomes, (ii) time series models (ARIMA, SARIMAX, and LSTM) to forecast weekly TB case counts, and (iii) spatial analysis tools (LISA, Moran's I) to identify high-incidence clusters.

RESULTS

Weekly TB notifications increased after the onset of the COVID-19 pandemic (t = 4.75, p = 2.10 × 10-6), with LISA revealing two significant clusters (p < 0.05). HGB predicted treatment outcomes (ROC-AUC = 0.86; top predictors: HIV, treatment duration, age). LSTM outperformed SARIMAX (RMSE 3.50 → 2.88 in Salta-Jujuy; 3.82 → 0.89 in Autonomous City of Buenos Aires (CABA)-Buenos Aires; -18 %/-77 %). A sex-difference test indicated higher infection among men (p < 0.05).

CONCLUSIONS

Combining ML and spatial tools enhances TB monitoring by supporting early identification of high-risk areas, improving epidemiological surveillance, and enabling targeted, data-driven public health interventions in Argentina.