Adjusting for patient differences in predicting hospital mortality for percutaneous coronary interventions in the Clinical Outcomes Assessment Program.
Academic Article
Overview
abstract
BACKGROUND: The Clinical Outcomes Assessment Program (COAP) is a coordinated quality improvement program for percutaneous coronary interventions (PCIs) performed in Washington State hospitals. This study describes the development and testing of models for predicting hospital mortality in patients undergoing PCI. METHODS: The COAP PCI database contains extensive demographic, medical history, and procedural information. This study included 19,358 consecutive PCIs performed in 27 Washington hospitals in 1999 and 2000. The study population was randomly assigned to development (n = 11,591) and test (n = 7614) sets. Logistic regression mortality models were run in the development set and evaluated in the test set. RESULTS: The test and development sets were similar in demographic, medical history, and procedural characteristics. The overall hospital mortality rate was 1.6% and was similar in the test and development sets. By means of stepwise logistic regression analysis, cardiogenic shock, age, nonelective priority, elevated creatinine level, ejection fraction, number of diseased vessels, myocardial infarction <24 hours from admission, history of chronic obstructive pulmonary disease, male sex, history of peripheral vascular disease, history of PCI, and history of congestive heart failure were identified as predictors of hospital mortality. When applied to the test set, this model had excellent discrimination (c statistic = 0.87, 95% CI = 0.84-0.90). The model was also evaluated in the Northern New England PCI Registry, with very good results (c statistic = 0.85). CONCLUSION: Developing risk-adjusted models of mortality and other outcomes is an essential part of the quality improvement process for cardiac revascularization procedures. Because of the rapidly changing nature of PCI, modification of these models in the years to come will be required.
