Risk prediction in coronary artery surgery: a comparison of four risk scores

OBJECTIVE: To determine which of four proposed risk scores best predicts immediate outcome of cardiac surgery. DESIGN: Observational cohort study. SETTING: Sir Charles Gairdner Hospital (a university teaching hospital), Perth, Western Australia, 18 March 1993 to 5 March 1996. SUBJECTS: 927 consecutive patients undergoing surgery for coronary artery disease. OUTCOME MEASURES: Patient risk scores (by methods of Parsonnet et al., Higgins et al., Tremblay et al. and Tu et al.); in-hospital mortality; postoperative hospital stay > 14 days; receiver operating characteristic (ROC) curves comparing sensitivity and specificity in predicting adverse outcomes for each risk score. RESULTS: In-hospital mortality rate was 3.5% and mean postoperative hospital stay was 10.7 days. The four scores had similar predictive abilities, with mean areas under the ROC curves (95% confidence intervals) for mortality and postoperative stay > 14 days, respectively: 0.70 (0.62-0.78) and 0.70 (0.65-0.75) for the Parsonnet score; 0.68 (0.59-0.77) and 0.70 (0.64-0.75) for the Higgins score; 0.68 (0.59-0.77) and 0.67 (0.62-0.73) for the Tremblay score; and 0.68 (0.60-0.76) and 0.69 (0.64-0.75) for the Tu score. CONCLUSION: Any of the scores may be used to estimate perioperative risk and to compare outcome between coronary surgery units, but none has sufficient specificity and sensitivity to identify specific individuals who will experience an adverse outcome. Further development of risk assessment is needed before adverse outcome can be accurately predicted in cardiac surgical patients.