Predicting survival outcomes in renal cell carcinoma spinal metastases: a multicenter evaluation of existing prognostic systems.

Abstract

Survival prediction models for patients with spinal metastases are crucial for guiding clinical decision-making and optimizing treatment strategies. Renal cell carcinoma spinal metastases (RCC-SM) present unique challenges due to their distinct biological behavior and variable response to systemic therapies. To externally validate existing prognostic scoring systems for predicting survival in patients with RCC-SM using multicenter data from China. Retrospective external validation study. 103 patients with RCC-SM who underwent surgical treatment at three specialized spine oncology centers in China between 2015 and 2023. Survival at 90 days, 180 days, and 1 year postsurgery, assessed using area under the curve (AUC), calibration intercept and slope, and Brier scores. Six prognostic scoring systems were evaluated, including Tomita, revised Tokuhashi, revised Katagiri, New England Spinal Metastasis Score, Skeletal Oncology Research Group (SORG) nomogram, and SORG machine learning (ML) model. Discrimination and calibration were assessed using ROC curves, calibration plots, and Brier scores. Cox regression identified independent prognostic factors. The study was funded by Henan Province Key Science and Technology Project (252102311081). A total amount of RMB 20,000 ($2,740) was received. SORG ML demonstrated the highest discriminative ability for 90-day survival (AUC: 0.765), while revised Tokuhashi performed best for 180-day survival (AUC: 0.754), and revised Katagiri for 1-year survival (AUC: 0.806). However, nearly all models exhibited underestimation of survival probabilities, particularly in high-risk subgroups. Independent prognostic factors included American Spinal Injury Association grade, visceral metastases, preoperative systemic therapy, preoperative radiotherapy, and neutrophil-to-lymphocyte ratio. Existing prognostic models for RCC-SM show varying predictive accuracy, with SORG ML and revised Katagiri performing best for short- and long-term survival, respectively. However, recalibration is needed to address underestimation, particularly in East Asian populations. Future models should incorporate dynamic treatment responses and molecular biomarkers to improve predictive accuracy and clinical utility.