Abstract

Abstract Background and Aims Survival analysis is a cornerstone in medical research. For this purpose Kaplan-Meier is the most widely used statistical test, but the presence of competing risks violates the fundamental assumption that the censoring mechanism is independent of survival time. This leads to overestimation of the cumulative probability of cause-specific failure. Cumulative incidence estimate and competing risks analysis are preferred. The purpose of this study was to compare different survival analysis methods: Kaplan-Meier and cumulative incidence function estimates in a cohort of Peritoneal Dialysis (PD) patients. Method The survival of 115 incident patients on PD in a university hospital was evaluated after establishing 2 cohorts: patients starting renal replacement therapy with PD (PD first; n=85) and patients switching to PD on the first 6 months of dialysis (PD transfer; n=30). Kaplan-Meier, cumulative incidence function, cause-specific and subdistribution hazards were performed. The event of interest was death and the competing risk events were transfer to hemodialysis and renal transplantation. Results Besides higher residual renal function (RRF) and kt/V in the PD first group, there were no other significant differences between groups. There were 22 deaths. PD first group had a better survival with both Kaplan-Meier (log-rank test, p=0.013) and cumulative incidence function (p=0.021) approaches. The Cox regression model showed, as protecting variables, higher albumin (HR=0.174; CI95% 0.054-0.562), higher RRF (HR=0.785; CI95% 0.666-0.925) and PD first (HR=0.350; CI95% 0.132-0.927). Higher Charlson Index predicted worse outcome (HR=1.459; CI95% 1.159-1.835). PD as first dialysis therapy was associated with 65.0 % lower risk of death comparing with PD transfer. The subdistribution multivariable model found higher Charlson Index (HR=1.389; CI95% 1.118-1.725) and lower RRF (HR=0.798; CI95% 0.680-0.936) were statistically associated with death, but not PD transfer or albumin. This result differs from the obtained using the cause-specific hazard model. Analyzing the competing events, patients submitted to renal transplantation had a lower Charlson Index. Conclusion The probability of death was overestimated by the Kaplan-Meier method. The bias of Kaplan-Meier is especially great when the hazard of the competing risks is large. This study consisted on a statistical critical analysis of a real medical example, broader clinical conclusions related with “PD first initiative” should be cautious in this context. It is primordial to recognize the presence of competing risks in studies with multiple outcomes, as in Peritoneal Dialysis studies, to estimate cumulative incidence and yield more accurate results. This study shows how different conclusions are attained with different statistical methodology and its relevance in clinical context.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call