|
Activity Number:
|
83
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, August 9, 2004 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Health Policy Statistics
|
| Abstract - #301179 |
|
Title:
|
Optimal Designs for Bayesian Hospital Report Cards
|
|
Author(s):
|
Peter C. Austin*+
|
|
Companies:
|
Institute for Clinical Evaluative Sciences
|
|
Address:
|
G1 06, 2075 Bayview Ave., Toronto, ON, M4N 3M5, Canada
|
|
Keywords:
|
hospital report cards ; Bayesian models ; provider profiling ; decision theory ; quality of care
|
|
Abstract:
|
There is an increasing interest in measuring the quality of medical care that patients receive. Report cards publishing hospital-specific mortality rates for specific conditions or procedures are increasingly common. There is growing interest in the use of Bayesian methods in report cards. These allow the determination of the probability that the mortality rate at a specific hospital exceeds a specific threshold. The choice of probability level required to classify a hospital as having higher than acceptable mortality has not been justified. In profiling, misclassification can occur: hospitals that truly have acceptable mortality rates can be classified as having higher than acceptable mortality rates, while hospitals that truly have unacceptably high mortality can be classified as having acceptable mortality. We place the design of hospital report cards in a Bayesian decision-theoretic framework, by introducing loss functions to explicitly describe costs incurred by patients and providers when hospitals are misclassified. We then use Monte Carlo simulations to explicitly determine optimal probability levels that minimize the mean posterior cost of misclassification.
|
