DaRenCa risk score: A prognostic model for recurrence in clear cell renal cell carcinoma

Objective This study aimed to investigate whether a machine‐learning model improves the assessment of postsurgical recurrence‐free survival in patients with non‐metastatic clear cell renal cell carcinoma (ccRCC) compared with a Cox proportional hazards (CPH) approach. Patients and Methods Patients undergoing curative surgery for non‐metastatic ccRCC between 2010 and 2018 were identified from the DaRenCa Study‐3, a nationwide register‐based cohort study. Three recurrence prediction models were developed: an extreme gradient boosting (XGBoost) model, a feature‐matched CPH model and a pathology‐based CPH model. The data set was divided into training and test cohorts. Missing data were addressed using multiple imputation for the CPH models, whereas XGBoost handled missing values inherently. Model performance was evaluated using the concordance index (C‐index) with 1000 bootstrap resamples. The XGBoost model was also compared with the Leibovich nomogram. Results Among 2782 patients, with a median follow‐up of 7.3 years, 13.7% developed a recurrence. In the test cohort, the XGBoost model showed higher discrimination than both CPH models. Compared with the best performing pathology‐based CPH model, XGBoost demonstrated a paired bootstrap difference in Uno's C‐index of 0.022 (95% CI 0.005–0.038). The model also identified a subgroup of patients with a very low risk of recurrence (<3% after 10 years) and demonstrated improved clinical risk stratification, with clearer separation between risk groups, higher hazard ratios between groups and larger differences in 5‐year recurrence‐free survival compared with established models. This improved risk stratification could reduce follow‐up imaging by approximately 11% compared with current EAU guideline recommendations. Limitations include the retrospective design and lack of external validation. Conclusion The XGBoost model provided improved prediction of recurrence compared with CPH models and the Leibovich nomogram, supporting more precise risk stratification. With external validation, this approach may help reduce unnecessary imaging after surgery.