Development and validation of a primary care–accessible risk prediction model for severe pulmonary TB

SUMMARY

Background

China has a high TB burden and limited primary care. We therefore developed a risk prediction model using routine clinical and lab indicators to flag severe TB cases early.

Methods

In a case–control study of 2,012 TB patients (499 severe, 1,513 non-severe) from four hospitals (2012–2022), patients were split into training (n = 1,408) and validation (n = 604) sets. Univariable and multivariable logistic regression identified predictors to build a nomogram (i.e., a tool that integrates patient-specific data to predict the probability of a specific clinical event), with model performance assessed using receiver operating characteristic (ROC) curves, calibration, and decision curves.

Results

Multivariable analysis identified seven independent predictors: advanced age, history of TB, lymphocytopenia, monocytosis, neutrophilia, hyponatremia, and hypoalbuminemia. The area under the ROC curve (AUC) for the training set was 0.834 (95% confidence interval [CI]: 0.796–0.872), with a sensitivity of 89.0% and specificity of 63.4%; the validation set AUC was 0.807 (95% CI: 0.746–0.867), with sensitivity 82.0% and specificity 71.7%. The calibration curve showed high consistency between predicted probabilities and actual observations. Decision curve analysis demonstrated clinical net benefit of the model within the threshold probability range of 5%–95%.

Conclusion

This model, based on routine indicators, effectively identifies high-risk severe TB cases, offering a practical and cost-efficient tool for early patient stratification in primary care.