Patient selection for proton therapy in lung cancer: Nationwide external validation and model update of a prognostic model for 2-year mortality including mean heart dose

Purpose: To externally validate a prognostic model for the prediction of 2-year mortality in locally advanced non-small cell lung cancer (NSCLC) patients treated with (chemo)radiotherapy in multiple Dutch centers, to enable selection of patients for proton therapy. Materials and methods: A published prognostic model for 2-year all-cause mortality after (chemo)radiotherapy for irresectable NSCLC, including mean heart dose, gross tumor volume, and smoking, was selected. External validation and model update was performed on cohorts from 5 centers. Model performance was assessed in terms of discrimination and calibration. In case the model performance differed substantially across centers, this was followed by an internal-external cross-validation (IECV) procedure. Ridge regression was used to fit the final set of predictors to the combined dataset and bootstrapping was applied to perform further internal validation. Results: A total of 554/1094 (51 %) patients died within 2 years. Performance of the prognostic model differed substantially across centers. The IECV showed very low levels of heterogeneity in all three performance measures, indicating performance was consistent across centers. The final model, including the mean heart dose and gross tumor volume, showed moderate discrimination (AUC 0.64; 95 %CI 0.61-0.68) and good calibration (calibration-in-the-large 0.0;- 0.12-0.12, calibration slope 1.04 [0.78-1.30]). Conclusion: This nation-wide external validation study of a prognostic model for 2-year mortality resulted in a model that performs consistently in patient cohorts across the country. The model has been adopted by the national health authority for model-based selection and reimbursement of proton therapy for lung cancer.