Statistical Modeling of a Recurrent but Rare Event

The statistical evaluation of provider performance is an increasingly important component of the US healthcare system as policy-makers seek to shift reimbursement from rewarding quantity to rewarding quality and discouraging overuse. Statistically valid and reliable measures of provider performance are needed to guard against provider reactions that result in unintended consequences for patients.  Both recent change to the Medicare dialysis reimbursement “bundle” payment for erythropoiesis-stimulating agent (ESA) drugs to manage anemia, and the identification of safety concerns associated with aggressive ESA use may result in increased frequency of red blood cell transfusion in the US chronic dialysis population. Thus, it is important to monitor dialysis facility-specific transfusion rates, relative to a national standard in order to identify specific facility treatment patterns that result in unnecessary  blood transfusions. A risk adjusted transfusion measure using Medicare claims data was developed as the  ratio of observed to expected number of transfusions; the number of expected transfusions is based on a proportional rates model and patient characteristics within each facility. Empirical null methods are used to classify more than 5000 facilities as having transfusion rates that are better (0.52%), no different (92.73%) or worse (6.75%) than the national average for 2012. The validity of this measure is tested by evaluating its association with other known quality measures, which include both dialysis facility outcomes and practices. This presentation will review the statistical methods applied, and describe our analysis and discuss ways to assess reliability and validity of the measure.