Clinical risk stratification of drug-resistant tuberculosis in diabetes

Patients with pulmonary tuberculosis and diabetes mellitus (PTB-DM) are vulnerable to drug-resistant tuberculosis. This study developed and temporally validated a simple risk stratification tool for early identification of drug resistance in PTB-DM patients. This retrospective cohort included 1266 PTB-DM patients admitted between January 2023 and June 2025. Drug resistance was defined as resistance to at least one first-line anti-tuberculosis drug. Although multidrug-resistant or rifampicin-resistant tuberculosis (MDR/RR-TB) was of particular clinical interest, the primary model was developed for any drug resistance to ensure statistical stability given the limited number of MDR/RR-TB events; MDR/RR-TB was evaluated as a secondary outcome in sensitivity analyses. Patients were divided into derivation and temporal validation cohorts. Multivariable logistic regression was used to estimate individual risk, and predicted probabilities were used to classify patients into low-, intermediate-, and high-risk groups. Model performance was assessed by ROC analysis, calibration, Brier score, decision curve analysis, and cumulative detection curves. Drug resistance occurred in 229 patients (18.1%). Retreatment, previous treatment interruption, longer diabetes duration, higher HbA1c, and lower albumin were retained in the final model. The AUC was 0.769 in the derivation cohort and 0.748 in the validation cohort. In the validation cohort, drug-resistance rates increased from 9.4% to 23.4% and 50.9% across the three risk groups. Cumulative detection curves showed a higher probability of drug-resistance detection in the high-risk group. This simple risk stratification tool showed acceptable performance in temporal validation and may help identify PTB-DM patients who could benefit from prioritised DST and intensified management. However, external multi-centre validation is still required before routine use.