A radiomics-clinical nomogram for predicting individualised treatment duration in newly diagnosed pulmonary tuberculosis.

AIM: The treatment duration for pulmonary tuberculosis (PTB) varies considerably based on disease severity, pathogen characteristics, and host immune status. This study aimed to develop predictive models for individualised treatment duration in newly diagnosed PTB patients, thereby supporting personalised therapeutic strategies.

MATERIALS AND

METHODS

A retrospective cohort of 242 newly diagnosed PTB patients was analysed and randomly divided into training and testing cohorts. Radiomic features were selected via Least Absolute Shrinkage and Selection Operator (LASSO)-Cox regression, while clinical indicators were identified through univariate Cox regression. Three models-a radiomics model, clinical model, and radiomics-clinical combined model-were constructed using Cox proportional hazards regression. Model performance was assessed using the C-index, with the optimal model visualised through a nomogram and forest plot. The time-dependent area under the curve (AUC) was used to evaluate predictive performance over time of the optimal model. Kaplan-Meier (K-M) survival analysis was performed based on the Rad score to stratify patients into high- and low-risk groups.

RESULTS

Two radiomic features and three clinical variables were incorporated into the final models. The combined model outperformed the radiomics-only and clinical-only models, achieving C-indices of 0.81 (training cohort) and 0.79 (testing cohort), with time-dependent AUCs consistently above 0.75. Calibration curves demonstrated good agreement between predicted and observed outcomes, and K-M analysis confirmed that the Rad score effectively stratified patients by treatment duration (P<.05).

CONCLUSION

The radiomics-clinical model demonstrated superior predictive performance, robustness, and clinical applicability compared to single-feature models. This approach provides a practical tool for personalising treatment duration and supports more precise management of PTB patients through effective risk stratification.