*Purpose: The C-statistic of risk-adjustment models is regularly used to assess accuracy of program evaluations or determine presence of unmeasured confounding. However, the C-statistic depends on variability in risk for individual transplants and may not provide information on accuracy of program evaluations or presence of unmeasured confounders.

*Methods: We conducted a simulation study to evaluate the relationship between the C-statistic and the accuracy of program evaluations with kidney-alone deceased donor transplants for adult recipients included in the January 2018 PSRs. The Spearman, or rank, correlation between the assigned tier (in the 5-tier rating system) and the true program-specific hazard ratio were estimated. Since programs with higher true hazard ratios should be placed in lower tiers, the Spearman correlation measured the accuracy of program evaluations. The correlation was stratified by the number of expected events at the program, i.e., program size, and the value of the C-statistic was changed by changing the level of variability in risk of individual transplants.

*Results: As illustrated in Table 1, C-statistics were similar with and without unmeasured confounding for the same level of variability. Further, the accuracy of the 5-tier system had no association with the C-statistic (Figure 1) but was strongly associated with the number of expected events. The associations were unaffected by presence of unadjusted risk factors, although the accuracy of program evaluations was worse, especially for larger programs.

*Conclusions: Thus, the C-statistic provides no information on accuracy of program evaluations or presence of unmeasured confounding.

Table 1. A comparison of the C-statistic, expected events, and observed 1-year survival for scenarios with and without unadjusted risks.

« Back to 2019 American Transplant Congress