Mortality prediction in adult cardiac surgery patients: comparison of two risk stratification models

Abstract

OBJECTIVE. To assess and compare the two commonly applied models 'EuroSCORE and Parsonnet' in our local adult cardiac surgery patients, according to risk factor quantification related to mortality using a risk stratification protocol to assess the quality of cardiac surgical care.

DESIGN. Prospective study.

SETTING. Cardiac surgery centre in a regional hospital in Hong Kong.

PATIENTS. All adult patients undergoing coronary artery bypass graft and heart valve surgery at the Grantham Hospital were evaluated prospectively from November 1999 to July 2005.

MAIN OUTCOME MEASURES. In-hospital mortality was the defined end-point. Statistical analyses consisted of observed against expected mortality, Hosmer-Lemeshow goodness-of-fit test for calibration accuracy, and receiver operating characteristic curve for discrimination performance.

RESULTS. During the study period, 1247 patients underwent coronary artery bypass graft surgery and 1406 underwent heart valve surgery. Observed mortality rates in these two patient groups were 2.9% and 4.8% respectively. The expected mortality rates as predicted by the EuroSCORE were (mean±standard deviation) 4.0±3.3% and 5.2±3.0% respectively, and by the Parsonnet model were 5.9±4.2% and 7.3±4.4% respectively. EuroSCORE performed better than the Parsonnet model at predicting in-hospital mortality assessed by the Hosmer-Lemeshow goodness-of-fit test. The areas under the receiver operating characteristic curves in coronary artery bypass graft surgery were 0.76 for EuroSCORE and 0.74 for Parsonnet. The receiver operating characteristic curve areas in valve surgery were 0.77 for EuroSCORE and 0.79 for Parsonnet.

CONCLUSION. Despite significant geographic and demographic differences between European and Asian patients, in our local adult cardiac surgery patients, the EuroSCORE performed well with good calibration and discrimination in predicting mortality. There was a tendency for both models to over predict. However, the EuroSCORE can serve as a baseline for the development of a local risk model.