Post-treatment Lung Tuberculosis Sequelae: an Inexpensive Clinical-Laboratory Nomogram to Predict Tissue Destruction

Post-treatment lung destruction (LD) impairs quality of life in pulmonary tuberculosis (TB) survivors, yet early risk-stratification tools are lacking. We aimed to develop and internally validate a clinical-laboratory nomogram to predict LD at completion of standard anti-TB therapy.

In this retrospective cohort, we enrolled 205 treatment-naïve adults with pulmonary TB from April 2021 to April 2025. LD was defined on follow-up chest CT as extensive fibrosis, bronchiectasis with volume loss, or parenchymal destruction. Twenty-two baseline demographic, clinical, laboratory, and imaging variables were screened. Least absolute shrinkage and selection operator (LASSO; 10-fold cross-validation) was used for variable selection, followed by Akaike information criterion (AIC)-guided stepwise multivariable logistic regression. Model performance was compared with random forest (RF) and support vector machine (SVM) classifiers. Discrimination (area under the receiver-operating characteristic curve, AUC), calibration (bootstrap-corrected curve; Brier score), and clinical utility (decision-curve analysis, DCA) were assessed; internal validation used 1,000-sample bootstrap resampling.

LD occurred in 61/205 patients (29.8%). Nine predictors-silicosis, drug resistance, symptom-to-treatment delay, lymphocyte count, C-reactive protein, aspartate aminotransferase, γ-glutamyl transferase, albumin, and baseline atelectasis/cavity-composed the final model. The nomogram showed excellent discrimination (AUC = 0.93, 95% CI 0.897-0.971; optimism-corrected AUC = 0.93) and good calibration (Brier = 0.13). Across 10-40% risk thresholds, DCA indicated a higher net benefit than treat-all or treat-none strategies. Logistic regression slightly outperformed RF (AUC = 0.91) and SVM (AUC = 0.92) while retaining interpretability.

An inexpensive, easily applicable nomogram integrating routine clinical and laboratory indices accurately predicts post-treatment LD in TB patients. The tool can support personalized follow-up and timely interventions, warranting external validation in multicenter prospective cohorts.