|
Activity Number:
|
104
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, August 4, 2008 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Health Policy Statistics
|
| Abstract - #301350 |
|
Title:
|
Model Comparison for Association Between Hospital Performance and Hospital Characteristics: Proportional or Hierarchical Binary Outcomes
|
|
Author(s):
|
Jacob J.H. Cheng*+ and Carlos F. Alzola and Nikolas Matthes
|
|
Companies:
|
Maryland Hospital Association and Data Insights and Maryland Hospital Association
|
|
Address:
|
6816 Deerpath Rd, Quality Indicator Project, Elkridge, MD, 21075,
|
|
Keywords:
|
Proportional Outcome ; Hierarchical Binary Outcome ; Linear Models ; Generalized Linear Mixed Models (GLMM) ; Generalized Estimating Equations (GEE) ; Hospital Performance and Characteristics
|
|
Abstract:
|
Hospital performance is usually measured by the proportion of patients who received appropriate care. Linear models are intuitive tools to study the association between performance and hospital characteristics but may violate normality and homoscedasticity assumptions. GEE and GLMM work on the binary patient-level outcomes. 2005--2006 data for heart failure care from 442 QIP hospitals were used. Models were evaluated by R2/RMSE with cross-validation and Type III test. Arcsine transformation improved the above problem in linear models resulting in closer inference to GEE and GLMM. GEE had the best power for marginal predictions while GLMM provided additional hospital-specific predictions. The difference between linear models and GEE/GLMM was minimal with enough sample size, leaving linear models attractive based on their flexibility for handling model diagnosis and nonlinearity detection.
|
