Accounting for total variation and robustness in profiling health care providers
share
Monitoring outcomes of health care providers, such as patient deaths, hospitalizations and readmissions to hospital, is important for assessing the quality of health care. Analyses based on random effects and fixed effects are commonly used to assess outcomes and identify providers with unusually poor (or good) outcomes. However, these approaches have shortcomings that can lead to unfair assessments of providers. One primary issue is that they do not consider variation between providers that is outside the providers' control, for example, due to unobserved patient characteristics that vary between providers. In this article, we review the existing fixed- and random-effects methods, and propose a smoothed empirical null approach that accounts for the total variation and adapts to different provider sizes. The linear model with random provider effects provides an illustration of the proposed empirical null method. This extends easily to other distributions to describe, for example, survival or binary outcomes. We also generalize the empirical null method to allow for some variation being due to quality of care. Methods are illustrated with numerical simulations and an example of a mortality measure in the Dialysis Facility Compare program.
READ FULL TEXT