External validation of the preoperative risk evaluation for partial nephrectomy (PREP) score

Abstract

Introduction

This study aims to externally validate the preoperative risk evaluation for partial nephrectomy (PREP) score, recently introduced to predict the risk of major complications after partial nephrectomy based on patient characteristics, in a large multi-institutional cohort of robot-assisted partial nephrectomy (RAPN) patients.

Materials and Methods

A retrospective review was performed on the IRB-approved, multi-institutional United States Kidney Cancer Data Network (US-KIDNET) to identify 10 154 patients who underwent RAPN from 2018 to 2025. Major complications were defined as Clavien–Dindo grade III–V within 30 days. Patients were assigned to risk categories for major complications based on weighted comorbidities, as defined by the PREP score, and the alignment of predicted complication rates with actual complication rates was evaluated. Discrimination was evaluated using the area under the receiver operating characteristic curve (AUC). Calibration was examined using the Hosmer–Lemeshow (HL) goodness-of-fit test. Subgroup analysis was performed by robotic platform and surgical approach. Exploratory univariate and multivariate logistic regression were performed to evaluate associations between individual PREP components and major complication rates. All computations were performed in Python (v3.8).

Results

A total of 8615 patients who underwent RAPN were included in the final validation cohort. The PREP score did not perform well in this robotic surgery cohort, with limited overall discrimination (AUC 0.547; 95% CI 0.518–0.580). On calibration, the model's predicted probabilities were not well-aligned with observed outcomes (HL χ² = 150.2; p < 0.001). Performance did not differ significantly across robotic platform or surgical approach subgroups. In the adjusted analysis, coronary artery disease (CAD, aOR 2.22), chronic obstructive pulmonary disease (COPD, aOR 2.75), chronic kidney disease (CKD, aOR 1.48), and obesity (aOR 1.73) were independently associated with major complications, but not congestive heart failure (CHF; aOR 0.50, p = 0.36).

Conclusion

Our real-world external validation for the PREP score demonstrates that this score does not perform well in a large, multicentre, robotic cohort. Our findings suggest that coefficient re-estimation, rather than simple recalibration, may be required.