The heterogeneous survival benefit of whole brain radiotherapy (WBRT) in brain metastatic non-small cell lung cancer (NSCLC) was prospectively evidenced in the Quality of Life after Treatment for Brain Metastases (QUARTZ) trial, resulting in inconsistent guideline recommendations and diverse clinical practices for giving WBRT. The objective of this study was to develop and externally validate an individual prediction model to demonstrate the added survival benefit of WBRT to assist decision making when giving WBRT is undetermined.
For model development, we collected 479 brain metastatic NSCLC patients unfit for surgery or stereotactic radiotherapy techniques at Siriraj Hospital. Potential predictors were age, sex, performance status, histology, genetic mutation, neurological symptoms, extracranial disease, previous systemic treatment, measurable lesions, further systemic treatment, and WBRT. Cox proportional hazard regression was used for survival analysis. We used multiple imputations to handle missing data and a backward selection method for predictor selection. Bootstrapping was used for internal validation, while model performance was assessed with discrimination (c-index) and calibration prediction accuracy. The final model was transformed into a nomogram and a web-based calculator. An independent cohort from Sawanpracharak Hospital was used for external validation.
In total, 452 patients in the development cohort died. The median survival time was 4.4 (95% CI, 3.8–4.9) months, with 5.1 months for patients who received WBRT and 2.3 months for those treated with optimal supportive care (OSC). The final model contained favorable predictors: female sex, KPS > 70, receiving additional systemic treatment, and WBRT. Having active extracranial disease, experiencing neurological symptoms, and receiving previous systemic treatment were adverse predictors. After optimism correction, the apparent c-index dropped from 0.71 (95% CI, 0.69–0.74) to 0.70 (95% CI, 0.69–0.73). The predicted and observed values agreed well in all risk groups. Our model performed well in the external validation cohort, with a c-index of 0.66 (95% CI, 0.59–0.73) and an acceptable calibration.
This model (