Bayesian Methods for Assessing Transplant Program Performance

Based on recommendations from a recent consensus conference and a report commissioned by the Centers for Medicare & Medicaid Services to the Committee of Presidents of Statistical Societies, the Scientific Registry of Transplant Recipients (SRTR) plans to adopt Bayesian methods for assessing transplant program performance. Current methods for calculating program‐specific reports (PSRs) often generate implausible point estimates of program performance, wide confidence intervals and underpowered conventional statistical tests. Although technically correct, these methods produce statistical summaries that are prone to misinterpretation. The Bayesian approach assumes that performance of most programs is about average and few programs perform much better or much worse than average; thus, strong evidence is required to conclude that performance is extremely good or poor. In Bayesian statistics, inference is performed via a posterior probability distribution, which reflects both the available data and prior beliefs about what model parameter values are most likely. In the PSRs, the posterior distribution of a program‐specific hazard ratio will show whether a program is likely to be performing better or worse than average. Bayesian‐derived PSRs will be available for preview by programs on the private SRTR website in mid‐2014 and will likely replace current methods for public reporting in early 2015. This viewpoint describes the rationale for the decision of the Scientific Registry of Transplant Recipients to switch to a Bayesian methodology to assess transplant program performance, provides examples in comparison with the current approach, and discusses the advantages and limitations of the Bayesian approach. Also see the authors' article on page 1310 and an editorial by Schold and Axelrod on page 1231.