External validation of a prognostic model in routine practice for short- and long-term survival in peritoneal dialysis

BackgroundThere are several indices to predict survival at dialysis start but tools to predict mortality for prevalent patients are lacking. This study provides evidence for external validity of the Cohen model to assess 6-, 12-, and 18-months survival of prevalent peritoneal dialysis (PD) patients.MethodsProspective cohort study of 464 PD patients in a university-based program between 2015 and 2019. Survival probabilities were compared to observed survival. Discrimination and calibration were assessed through predicted risk-stratified observed survival, cumulative area under the curve, Somer's Dxy, and a calibration slope estimate.ResultsDiscrimination performance was moderate with c-statistic of 0.73 to 0.74 for all 3 time points. The model over predicted mortality risk with the best predictive accuracy for 6-month survival. The difference between observed and mean predicted survival at 6, 12, and 18 months was 3.1%, 5.5%, and 11.0%. Kaplan-Meier curves showed good discrimination between low- and high-risk patients with hazard ratios [95% confidence interval (CI)]: C4 vs C1 32.0 [4.3-236.5]. Miscalibration of the model was the greatest for the highest risk patient group in whom 12 and 18 months predicted survival was 15% and 28% lower than observed survival.ConclusionsThe Cohen prognostic model can identify PD patients at high risk for death over 6, 12, and 18 months. Given it overestimates mortality risk for the highest risk patients, care must be taken to not use predictions to withhold treatment but rather to risk stratify and identify those who may benefit from enhanced kidney supportive care. This miscalibration provides an imperative to refine the tool for PD patients.