Comparison of three methods of presurgical risk stratification in cardiac surgery

Abstract

Background: During the past decade several models were derived to predict the outcome of heart surgery on the basis of clinical and angiographic preoperative data in adult patients with acquired heart disease. The purpose of our study was to verify and compare the predictive accuracy of hospital mortality according to the following 3 models: Parsonnet (Newark Beth Israel Medical Center, NBI score), Higgins (Cleveland Clinic Foundation, CCF score), and Roques (French Multicenter study, French Score), in a setting independent from the one where the models were derived. Methods: For each of the 516 patients undergoing heart surgery in our institution between January 1992 and December 1993, we calculated the presurgical risk according to the 3 models. Then we divided the patients into 3 tertiles of the same dimension, after having ordered the patients on the basis of the mean score obtained by the 3 models. We then compared the predicted with the observed mortality by means of the accuracy index and the overestimation histogram. Results: Overall observed mortality (1.5%) was similar to the mortality predicted by the NBI score (1.5±2.1%, p=n.s.), by the CCF score (1.7±2.0%, p=n.s.), and by the French score (1.9±2.5%, p=n.s.). All 3 models identify a significantly increasing risk from the first to the third tertile. The accuracy of the prediction of global mortality is close to 1 and equal with the 3 models. The accuracy index decreases from the first to the third tertile, showing the loss of the predicting power in the patients at high risk. The accuracy index in the 8 patients who died is similarly low (NBI score: 0.06±0.06; CCF score: 0.13±0.07; French score: 0.10±0.07; p=n.s.). Conclusion: The global predicted mortality obtained by the 3 models is not significantly different from the observed mortality, and therefore, the global accuracy is similar and high. The predictive accuracy is reliable for patients who will survive but is poor for those who will die. The models for presurgical risk stratification are useful to compare the results among different institutions or different surgeons, or to monitor the results over time in the same institution, but they cannot be used to predict accurately the individual risk of hospital mortality.