Validation and quality measurements for EuroSCORE and EuroSCORE II in the Spanish cardiac surgical population: a prospective, multicentre study

Since its development in the late 1990s, the European System for Cardiac Operative Risk Evaluation (EuroSCORE) has been the predictive model of choice for estimating mortality after cardiac surgery. As outcomes from cardiac surgery improved, the EuroSCORE showed a loss of calibration, and a revised version of the model was developed, EuroSCORE II. The objectives of this study were to examine the validity of both scores in the Spanish population, and to depict the performance of both models on a funnel plot. A prospective multicentre study was performed, with requests to participate sent to all centres in Spain. Participating centres reported the EuroSCORE, EuroSCORE II and the actual mortality of each patient. Incomplete data were requested to get a zero incidence of lost data. Calibration of models was evaluated with the Hosmer-Lemeshow goodness-of-fit test, and discrimination with the areas under the receiver operating characteristic (ROC) curve. A funnel plot was constructed using mortality data from the 2010 European Registry, to represent risk-adjusted mortality. Twenty Spanish centres participated in the study; 4034 patients undergoing cardiac surgery between 1 October 2012 and 31 March 2013 were collected. Prevalence of risk factors was analysed. The observed mortality rate was 6.5%. The mean additive EuroSCORE was 6.5. The mean expected mortality rate was 9.8% for the logistic EuroSCORE, and 5.7% for EuroSCORE II. Areas under the ROC curves were EuroSCORE: 0.77 [95% confidence interval (CI): 0.75-0.80], EuroSCORE II: 0.79 (95% CI: 0.76-0.82). Results for the goodness-of-fit test were EuroSCORE: 33.02 (P < 0.001), EuroSCORE II: 38.98 (P < 0.001). Risk-adjusted mortality is far beyond the lower bound of the CI if EuroSCORE is used as the reference model, and is between the confidence limits, but near to the upper bound when EuroSCORE II is used. Spanish cardiac surgical patients have a high-risk profile. Areas under the ROC curve show good discrimination for both models. Predicted mortality using EuroSCORE II more closely matches actual mortality than that predicted by the original EuroSCORE. Both models show statistically significant differences from the actual mortality rate, with EuroSCORE overpredicting and EuroSCORE II underpredicting mortality. The funnel plot illustrates risk-adjusted mortality clearly out of boundaries when EuroSCORE is used, and near underprediction when the reference is EuroSCORE II.