Development and validation of a diagnostic clinical prediction model for suspected pulmonary tuberculosis in adults: A retrospective analysis

Objective: Early and accurate diagnosis of pulmonary tuberculosis (PTB) is critical for disease control. This study aimed to develop and validate a clinical prediction model to differentiate adult inpatients with suspected PTB. Methods: A retrospective cohort of 602 adults admitted to Changsha Central Hospital (January-October 2019) with suspected PTB was randomly divided into a modeling group (n = 421) and an internal validation group (n = 181). Univariate and multivariate logistic regression analyses in the modeling cohort identified independent clinical and radiological predictors. Model performance was assessed by the area under the receiver operating characteristic curve (AUC), calibration (Hosmer-Lemeshow test), and decision curve analysis. External validation was performed using 213 similar patients from Hunan Chest Hospital (January-October 2019). Results: Five independent predictors were retained: age (years), erythrocyte sedimentation rate (mm/h), upper-lobe or dorsal-segment infiltration on chest imaging, presence of multifocal polymorphic pulmonary lesions, and pleural thickening (all p < 0.05). In the modeling cohort, AUC = 0.977 (95% confidence interval (CI): 0.964-0.989); at the optimal cutoff probability of 0.611, sensitivity was 95.9% and specificity 89.4%. Calibration was satisfactory in both internal (Hosmer-Lemeshow p = 0.815) and external (p = 0.973) validation cohorts. Decision curve analysis demonstrated net clinical benefit across relevant threshold probabilities. Conclusions: The proposed five-variable model exhibited excellent discrimination, calibration, and clinical utility in both internal and external cohorts. Its simplicity and reliance on routinely available data make it a practical adjunct for clinicians in diagnosing PTB among adults with suspected disease.