In Feb. 2012, SRTR and OPTN jointly hosted a consensus conference on transplant program quality and surveillance. One recommendation was to study Bayesian hierarchical modeling strategies. Coincident with the conference, CMS commissioned a report from the Committee of Presidents of Statistical Societies entitled, "Statistical Issues on Assessing Hospital Performance," which also recommended a Bayesian approach. SRTR is thus exploring Bayesian modeling strategies. The current methodology yields an observed/expected ratio (O/E) with a P value from a hypothesis test of whether the program is performing as expected. The Bayesian approach produces a distribution of the probability that a center is underperforming; the figure shows 3 examples.

The posterior mean hazard ratio is indicated on each histogram along with a 95% credible interval. A decision rule can then be constructed to identify programs for further scrutiny. SRTR explored a decision rule to review programs with a >75% probability that they are performing >25% worse than expected or a >10% probability that they are performing >150% worse than expected. We compared this approach with the current flagging methodology against final determinations from past OPTN reviews from 2007 (Table 1).

Simulations suggest that the Bayesian flag has a low false-positive flag rate for all center volumes and better detects underperformance for mid-volume programs. SRTR will continue exploring this approach.

« Back to 2013 American Transplant Congress