|
Activity Number:
|
410
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 6, 2003 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Section on Quality & Productivity
|
| Abstract - #300979 |
|
Title:
|
Cluster Analysis: A Tool for Provider Profiling with Multiple Outcomes
|
|
Author(s):
|
Kenneth C. Pietz*+
|
|
Companies:
|
Veterans Affairs Medical Center
|
|
Address:
|
2002 Holcombe Ave., Houston, TX, 77030-4211,
|
|
Keywords:
|
hierarchical models ; provider profiling ; cluster analysis
|
|
Abstract:
|
Provider profiling has become commonplace in the medical industry. Providers are ranked based on their predicted random effects using a hierarchical model. It is difficult to visualize overall patterns of care when several outcomes are modeled. The predicted provider effects can be regarded as realizations of random variables. Using two-dimensional cluster analysis, we compared the predicted effects for a population of primary care providers using medical encounters and outpatient pharmacy costs as outcomes. The population consisted of 227 providers caring for 101,630 patients in a network of VA hospitals. Two-level mixed models were used with Adjusted Clinical Groups as case-mix adjustment (fixed effect) and provider as random effect. Seven disjoint clusters in two dimensions were generated of the predicted provider effects. There were over- and under-users of both outcomes in three small clusters. The remaining four clusters corresponded roughly to the four quadrants in two-space. There was no pattern of association of high revisit rate with high pharmacy cost. Cluster analysis was found to be a useful tool for visualizing multiple aspects of provider performance.
|
