A Hierarchical Bayes Approach to a Study of Hospital Variation in Surgical Procedures

, , , and

BibTeX reference

There is increased interest in rating types of hospitals or geographical regions containing hospitals on the basis of their performance in the provision of certain types of medical services or procedures. Another important aspect of such a study could be the identification of exceptional hospitals or regions. The outcome variable of interest is often a binary one representing the type of procedure used or a successful outcome of a service. Because of the hierarchical nature of the data (for example, patients, doctors, hospitals, hospital type within hospitals, regions or small areas, etc.), a hierarchical model should be used. Such studies lend themselves very easily to a multiple logistic regression model with mixed effects; that is, patient information such as age, gender and hospital information such as caseload are considered as fixed effects while hospital and/or region are modeled as random effects. The hierarchical Bayes approach proposed here also allows for the standardization of the random effects which permits the use of normal probability plots for the detection of outliers and exceptional cases. For a large well-defined hospital population, Simons et al (1997) recently reported statistically significant differences in surgical choices in the treatment of rectal cancer seemingly due to hospital type and caseload within hospital type. Their data are used here to illustrate the utility of hierarchical Bayes techniques for parameter estimation and outlier detection in a logistic regression model with random effects in such a study.

, 20 pages


G-2002-22.pdf (200 KB)