Era-specific prognostic factors and an interpretable machine-learning survival model for renal cell carcinoma brain metastases

BACKGROUND: Brain metastases (BM) in renal cell carcinoma (RCC) are associated with poor prognosis and limited clinical guidance. We aimed to identify prognostic factors for overall survival (OS) in RCC BM and develop an interpretable machine-learning (ML) model for individualized risk prediction. PATIENTS AND METHODS: We retrospectively analyzed 929 patients with histologically confirmed clear cell RCC BM treated with stereotactic radiosurgery or surgical resection and stratified them by treatment era [interferons (IFN), tyrosine kinase inhibitors (TKI), and immune checkpoint inhibitors (IO)]. Univariate Cox regression analyses were carried out, and a prognostic classification tree was constructed for the IO era. ML analyses used a comprehensive survival modeling framework integrating multiple algorithms and ensemble strategies, with final model selection based on nested cross-validation performance and interpretation using SHapley Additive exPlanations. Model performance was evaluated using the concordance index (C-index), time-dependent area under the receiver operating characteristic curve (AUC), Brier score, and calibration metrics, and an interactive web-based calculator was developed. RESULTS: Median OS increased from 0.8 years in the IFN era to 1.1 years in the TKI era and 2.0 years in the IO era. The CoxNet (elastic net, α = 0.2) survival model achieved a C-index of 0.64, a 6-month AUC of 0.75, and a 95% confidence interval of 0.47-0.92 in the test cohort. Model interpretation identified extracranial disease status and age at BM intervention as the dominant predictors, with additional contributions from functional status and intracranial disease burden. Limitations include a lack of patient-level systemic therapy details, tumor volume measurements, and laboratory variables used in prior prognostic models. CONCLUSIONS: This study reports the largest cohort of patients with RCC BM and presents an interpretable ML model for individualized survival prediction to support clinical decision making.