Competing Risk Analysis in Renal Allograft Survival: A New Perspective to an Old Problem. Issue 3 (22nd February 2021)

• Record Type: Journal Article
• Authors: El Ters, Mireille; Smith, Byron H.; Cosio, Fernando G.; Kremers, Walter K.
• Is Part Of: Transplantation. Volume 105:Issue 3(2021)
• Page Start: 668
• Page End: 676
• Publication Date: 2021-02-22
• Subjects: Transplantation of organs, tissues, etc -- Periodicals; Transplantation immunology -- Periodicals; 617.95
• DOI: 10.1097/TP.0000000000003285
• Languages: English
• ISSNs: 0041-1337

Abstract

Abstract : Background: Graft survival after kidney transplant (KTX) is often estimated by the Kaplan-Meier (KM) method censoring for competing endpoints, primarily death. This method overestimates the incidence of graft loss. Methods: In 3157 adult KTX recipients followed for a mean of 79.2 months, we compared kidney and patient survival probabilities by KM versus competing risk analysis (CRA). These methods are extended to comparing different regression methods. Results: Compared with CRA, the probabilities of death and graft loss (censored for the other outcome) were substantially higher by KM. These differences increased with increasing follow-up time. Importantly, differences in graft losses were magnified in subgroups with greater probabilities of death. Among recipients with diabetes, the probabilities of graft loss at 20 years were 57% by KM and 32% by CRA, while for non-diabetes mellitus corresponding values were 44% and 35%. Similar results are noted when comparing older versus younger recipients. Finally, we find that the Fine-Gray method assumptions are violated when using age and gender as covariates and that the alternative method of Aalen-Johansen may be more appropriate. Conclusions: CRA provides more accurate estimates of long-term graft survival and death, particularly in subgroups of recipients with higher rates of the competing event. Overestimation of risk by KM leads to both quantitative and qualitative misinterpretations of long-term KTX outcomes. When using regression analyses, care should be taken to check assumptions to guide the choice of appropriate method.