Additive and logistic EuroSCORE performance in high risk patients

We compared the performances of the additive and logistic EuroSCORE in predicting mortality in high-risk cardiac surgical patients, at a single institution. Both models were applied to 6535 patients, operated on at the Western Infirmary, Glasgow from March 1994 to August 2004. Calibration and discrimination were assessed using the Hosmer-Lemeshow [HL] Chi-square test and areas under the ROC curve. Overall mortality was 2.95%. Predicted mortalities were 4.1% [additive] and 5.2% [logistic]. Actual mortality was 0.6% in the low risk (additive EuroSCORE 1-2), 2.1% in the medium risk (EuroSCORE 3-5) and 7% in the high-risk groups (EuroSCORE 6 plus). Actual mortality increased beyond a predicted risk of 8-10%. At the low risks both systems slightly over-estimated mortality, with the logistic EuroSCORE being more accurate. At EuroSCOREs between 10-20, the additive EuroSCORE under-estimated risk, while the logistic EuroSCORE over-estimated mortality. Both systems were inaccurate at high risk. The HL statistics were 11.15 [P<0.64] for the additive and 37.78 [P<0.47] for the logistic models. ROC curve areas were 0.749+/-0.04 [additive] and 0.746+/-0.03 [logistic]. The additive EuroSCORE model remains a simple system for cardiac risk assessment. The logistic EuroSCORE was not more accurate even in high-risk patients.