A new model outperforming RPA and DS-GPA scores for individualized survival prediction of patients following whole brain irradiation for brain metastasis

Purpose: Survival after whole brain radiation therapy (WBRT) in patients with multiple brain metastases (BM) is currently predicted by group-based scoring systems with limited usability for decision. We aimed to develop a more relevant individualized predictive model than Radiation Therapy Oncology Group - Recursive Partitioning Analysis (RTOG-RPA) and Diagnosis - Specific Graded Prognostic Assessment (DS-GPA) for patients with limited life-expectancy.

Methods: Based on a Discovery cohort of patients undergoing WBRT, multivariable piecewise Cox regression models with time cut-offs at 1 and 3 months were developed to predict overall survival (OS). A final parsimonious model was defined, and an external validation cohort was used to assess its discrimination and calibration at one, six, and 12 months.

Results: In the 173-patient Discovery cohort, the majority of patients had primary lung cancer (56%), presence of extracranial disease (ECD) (75%), Eastern Cooperative Oncolgy Group - Performance Status (ECOG-PS) score 1 (41%) and no intracranial hypertension (ICH) (74%). Most patients were classified as the RPA class II (48%). The final piecewise Cox model was based on primary site, age, ECD, ECOG-PS and ICH. An external validation of the model was carried out using a cohort of 79 patients. Individualized survival estimates obtained with this model outperformed the RPA and DS-GPA scores for overall survival prediction at 1-month, 6-months and 12- months in both Discovery and Validation cohorts. A R/Shiny web application was developed to obtain individualized predictions for new patients, providing an easy-to-use tool for clinicians and researchers.

Conclusion: Our model provides individualized estimates of survival for poor prognosis patients undergoing WBRT, outperforming actual scoring systems.