Simpson's paradox and the impact of donor-recipient race-matching on outcomes post living or deceased donor kidney transplantation in the United States

BACKGROUND: Race is a prognostic indicator in kidney transplant (KT). However, the effect of donor-recipient race-matching on survival after KT remains unclear. METHODS: Using the United Network for Organ Sharing (UNOS) database, a retrospective study was conducted on 244,037 adults who received first-time, kidney-alone transplantation between 2000 and 2019. All patients were categorized into two groups according to donor-recipient race-matching, and the living and deceased donor KT (LDKT and DDKT) were analyzed in subgroups. RESULTS: Of the 244,037 patients, 149,600 (61%) were race-matched, including 107,351 (87%) Caucasian, 20,741 (31%) African Americans, 17,927 (47%) Hispanics, and 3,581 (25%) Asians. Compared with race-unmatching, race-matching showed a reduced risk of overall mortality and graft loss (unadjusted hazard ratio (HR) 0.86, 95% confidence interval (CI) 0.84–0.87; and unadjusted HR 0.79, 95% CI: 0.78–0.80, respectively). After propensity score-matching, donor-recipient race-matching was associated with a decreased risk of overall graft loss (P < 0.001) but not mortality. In subgroup analysis, race-matching was associated with higher crude mortality (HR 1.12, 95% CI: 1.06–1.20 in LDKT and HR 1.11, 95% CI: 1.09–1.14 in DDKT). However, race-matching was associated with a decreased risk of graft loss in DDKT (unadjusted HR 0.97, 95% CI: 0.96–0.99), but not in LDKT. After propensity score-matching, race-matching had better outcomes for LDKT (patient survival, P = 0.047; graft survival, P < 0.001; and death-censored graft survival, P < 0.001) and DDKT (death-censored graft survival, P = 0.018). Nonetheless, race-matching was associated with an increased adjusted mortality rate in the DDKT group (P < 0.001). CONCLUSION: Race-matching provided modest survival advantages after KT but was not enough to influence organ offers. Cofounding factors at baseline led to a contorted crude conclusion in subgroups, which was reversed again to normal trends in the combined analysis due to Simpson's paradox caused by the LDKT/DDKT ratio.