Health Services Research & Development

Veterans Crisis Line Badge
Go to the ORD website
Go to the QUERI website

2006 HSR&D National Meeting Abstract


3081 — A Comparison of Four AMI Mortality Risk Adjustment Models

Author List:
Greiner GT (Northwest Center for Outcomes Research in Older Adults)
Lowy E (Northwest Center for Outcomes Research in Older Adults)
Maynard C (Northwest Center for Outcomes Research in Older Adults)
Sales AE (Northwest Center for Outcomes Research in Older Adults)
Fihn SD (Northwest Center for Outcomes Research in Older Adults)

Objectives:
Previous research has shown that patients who are admitted to VA facilities with acute myocardial infarction (AMI) are dissimilar to patients admitted to non-VA hospitals. Several models for risk adjustment have been developed for use when comparing the outcomes of patients in different settings. The purpose of this study was to compare the model developed by the Ischemic Heart Disease Quality Enhancement Research Initiative (IHD QUERI) in collaboration with the Office of Quality and Performance with those published using NRMI (National Registry for Myocardial Infarction), or CRUSADE (Can Rapid Risk Stratification of Unstable Angina Patients Suppress Adverse Outcomes with Early Implementation of the ACC/AHA Guidelines Quality Improvement Initiative) data, and the Joint Commission on Accreditation of Healthcare Organizations (JCAHO).

Methods:
Data were obtained from the FY2004 External Peer Review Program retrospective review of all AMI patients discharged from VA medical centers, supplemented with administrative data from the Austin Automation Center. We excluded veterans with AMI onset after hospital admission, and those transferred from non-VA facilities. For each of the four models estimated using logistic regression, we computed the area under the Receiver Operating Characteristic (ROC) curve to derive a c-statistic for model comparison. Each model was applied to three different cohorts: 1) All acute coronary syndrome patients (including ST-elevation MI, Non- ST-elevation MI, and unstable angina), 2) AMI patients only, and 3) AMI patients aged 65 and older.

Results:
Across all three cohorts, the NRMI-based model consistently had the best model fit (c-statistic of 0.79 for all ACS, 0.79 for AMI and 0.73 for AMI aged 65+) followed respectively and consistently by IHD QUERI (0.78 for all ACS, 0.78 for AMI and 0.72 for AMI aged 65+), CRUSADE (0.76 all ACS, 0.76 for AMI and 0.70 for AMI aged 65+), and JCAHO (0.74 all ACS, 0.74 for AMI and 0.0.67 for AMI aged 65+).

Implications:
All models performed within a similar range with none performing above 0.80 and only one below 0.70. While similar, there are clear differences in overall performance as the NRMI-based model consistently had the best fit across the three cohorts.

Impacts:
When quality of care for AMI is compared within VHA and with systems outside of VHA, it is essential to perform risk-adjustment. These results provide guidance to investigators and administrators evaluating the quality of care for AMI.