Single centre, single domain validation of the EuroSCORE on a consecutive sample of primary and repeat CABG

Intra- and interdepartmental benchmarking require scoring systems with excellent performance on several properties: discrimination (resolution), reliability (calibration) and stability over the complete spectrum of peri-procedural risk. This single centre, single domain study validates the European system for cardiac operative risk evaluation (EuroSCORE) on an independent sample of primary and repeat coronary artery bypass grafting (CABG) patients and will evaluate these different properties.

The study population is a consecutive series of 2051 isolated primary and repeat CABG patients, inclusive of patients in cardiogenic shock or resuscitation, operated on in a single institution from January 1997 to July 2000. The age of the patients was 66+/-9 years, 77% were males and 7% were repeat procedures. The EuroSCORE was 5.0+/-3%, with a range from 0 to 22. The studied event was in-hospital death, defined as mortality during hospital stay, which was unlimited in time and included a stay in a secondary hospital without discharge home.

The EuroSCORE predicted 102 deaths versus 81 deaths observed (P=0.14, Fisher exact test). The EuroSCORE described only 20% of the variance of in-hospital mortality. The EuroSCORE created an area under the receiver operating characteristic curve of 0.83+/-0.03. The highest discriminative accuracy was obtained with 8% EuroSCORE risk (only 64% sensitivity and 87% specificity). Further exploration identified an over score in the EuroSCORE range 0-8 (57%, P<0.0001). There was an equal score (-2%, P=1) in the range 9-11, but an under score in the range 12-22 (-133%, P=0.003).

On the condition that these single centre results could be extended to any European cardiac surgery centre, it can be concluded that the overall acceptable performance of the EuroSCORE is the result of an over score in the lower risk and insufficient correction in the higher risk spectrum. The EuroSCORE is probably refined enough for improved informed consent versus aggregated results but should only be used for inter-institutional benchmarking with great caution, preferably below the 12% risk pivot.