External validation of radiobiological models for local control prediction in lung cancer patients treated with stereotactic body radiation therapy.

Background: The debate regarding the accuracy of radiobiological models for local control (LC) prediction in lung cancer patients undergoing stereotactic body radiation therapy (SBRT) remains unresolved. The study seeks to externally validate the predictive efficacy of radiobiological models using single-institutional SBRT database. Methods: The cohort comprised 153 patients diagnosed with primary or metastatic lung cancer who underwent SBRT. The study employed three radiobiological models to estimate the probability of 2-year LC, including the Liu model, Klement model, and Ohri model. Furthermore, the likelihood of 3-year LC was predicted using the Liu model, Klement model, Gucken model, and Santiago model. The performance of the prediction models was assessed through the AUC values of the receiver operating characteristic (ROC) curve and the calibration plots. Results: Local recurrence was observed in 38.6% of patients (59/153) within two years, and in 43.1% (66/153) within three years after the radiotherapy. The ROC curves indicated discriminative power for all the 2-year and 3-year models, with the exception of the Klement model. The Ohri model showed a significantly improved discriminative ability than the Klement model for 2-year prediction, while it was not statistically significant when compared to the Liu model. However, no significant differences were found among the four models in terms of 3-year LC prediction. The calibration plots, using the Hosmer-Lemeshow goodness-of-fit test, confirmed that the predicted probabilities of the models were in agreement with the actual observation with Conclusion: Considering the balance between prediction accuracy and model simplicity, it is recommended to utilize the Ohri model for 2-year LC prediction and either the Gucken model or Santiago model for 3-year LC prediction.