Development and Validation of a Random Survival Forest-Based Model for Predicting All-Cause Mortality in Peritoneal Dialysis Patients: A Retrospective Cohort Study

Background Patients undergoing peritoneal dialysis (PD) face elevated risks of all-cause mortality. We aimed to develop and validate a prediction model for all-cause mortality in PD patients using random survival forest (RSF) methodology. Methods In this retrospective cohort study, patients receiving maintenance PD between 1 January 2017 and 31 December 2019 were enrolled as the training cohort (n = 221), with those treated from 1 January 2020 to 31 December 2022 serving as the temporal validation cohort (n = 256). Eligible participants aged 18–80 years had received PD for ≥ 3 months with complete baseline data. We collected demographic characteristics, laboratory parameters (hemoglobin, albumin, alkaline phosphatase, urea nitrogen, creatinine, electrolytes, etc.), and PD-specific metrics (Kt/V, peritonitis rate). RSF analysis was employed to identify mortality-associated variables. Results The prediction model incorporated multiple clinical and biochemical variables, demonstrating robust discriminative ability in the training set (C-index 0.931, 95% CI 0.894–0.968). Temporal validation maintained satisfactory performance (C-index 0.733, 95% CI 0.635–0.831), with age, hypoalbuminemia, and elevated high-sensitivity C-reactive protein (hsCRP) emerging as key predictors. A clinically applicable nomogram was developed to estimate 3-year survival probabilities. Conclusions This RSF-based model reliably predicts all-cause mortality in PD patients, providing a valuable tool for risk stratification and personalized management. External validation is warranted to confirm generalizability.