Predicting Tuberculosis Outcomes Using Routine Surveillance Data in Chiang Mai, Thailand: Retrospective Cohort Study.

BACKGROUND

Tuberculosis (TB) remains one of the leading causes of death from a single infectious disease worldwide. In Thailand, persistent gaps in early detection and access to TB care remain important public health challenges, particularly among populations in rural and remote areas.

OBJECTIVE

This study aimed to develop and evaluate predictive models for TB outcomes using routine surveillance data to support risk stratification, thereby informing public health decision-making in Chiang Mai Province, Thailand.

METHODS

A retrospective cohort study was conducted using data from 5557 TB cases registered in the National Tuberculosis Information Program in Chiang Mai Province from 2020 to 2024. Models were developed to predict treatment success, mortality, and time to treatment initiation. We evaluated model performance using the area under the receiver operating characteristic curve, the Harrell Concordance Index, and error-based metrics for time-to-treatment prediction. Scenario analyses were conducted under predefined assumptions to assess projected changes in detection-related indicators, treatment coverage, and mortality relative to provincial baseline indicators.

RESULTS

The models demonstrated good predictive performance. Mortality was associated with HIV co-infection (hazard ratio 5.80) and was highest among older adults with HIV co-infection (hazard ratio 12.30). Treatment delays were longer among individuals living in rural or remote areas and among those without health insurance, with mean delays ranging from 8 to 18 days. These findings suggest distinct patterns in TB outcomes, with mortality more closely related to clinical vulnerability and treatment delay more closely related to health care access factors. Under modeled scenarios, the models projected that detection-related indicators would increase by 25%, treatment coverage would increase by 15%, and mortality would decrease by 20% relative to provincial baseline conditions.

CONCLUSIONS

Predictive modeling using routine TB surveillance data may support risk stratification and provide insights into treatment outcomes and delays. The findings highlight the combined roles of clinical vulnerability and health care access factors in TB outcomes and support further evaluation of data-driven approaches to inform targeted TB interventions in resource-limited settings.