Clinical-radiomic model development for local recurrence prediction in stereotactic body radiation therapy-treated lung cancer: comparative analysis with radiobiological models.

Accurate prediction of local recurrence (LR) after stereotactic body radiation therapy (SBRT) for lung cancer remains challenging, and conventional radiobiological models have shown limited predictive performance. This study aimed to develop a clinical-radiomic model for LR prediction and compare its performance with existing radiobiological models. A total of 153 SBRT-treated lung cancer patients were retrospectively analyzed. Radiobiological models were used to estimate 2-year and 3-year local control (LC) probabilities, which were subsequently converted to LR probabilities using the relationship P(LR) = 1 - P(LC) for comparison with other models. Radiomic feature selection and model construction were performed using LASSO regression with tenfold cross-validation. A clinical-radiomic model was built by integrating clinical factors with the radiomic score (Radscore), followed by nomogram development. Model performance was evaluated using ROC analysis and calibration plots, with internal validation performed through bootstrap resampling. Clinical utility was evaluated using bootstrapped ROC, calibration, and decision curve analysis (DCA). The clinical-radiomic model showed higher AUC values than the radiobiological and radiomic models alone for both 2-year and 3-year LR predictions (all P < 0.05). Nomogram-derived risk scores differed significantly between the recurrence and non-recurrence groups. Bootstrap resampling suggested stable discriminatory performance, acceptable calibration, and potential decision-analytic performance. The clinical-radiomic model may show potential for individualized LR risk stratification in lung cancer patients receiving SBRT. However, further external validation in prospective multicenter studies is required before routine clinical application.