Victorio Carosella, Hugo Grancelli, Walter Rodríguez, Miguel Sellanes, Miguel Cáceres, Hernán Cohen Arazi, César Cárdenas, Carlos Nojek
Background During the last decades, several risk assessment models have been applied to predict the risk of mortality after cardiac surgery; however, none of them have been developed in Latin American populations. These models have inferior performance when applied to patient groups other than the ones on whom they were developed. Objective To perform external and temporal validation of a local risk score for cardiac surgery [Argentinean System for Cardiac Operative Risk Evaluation (ArgenSCORE)] and compare it to the EuroSCORE. Material and Methods A total of 5268 consecutive adult patients undergoing cardiac surgery were included from June 1994 to December 2009. The risk model was developed through logistic regression on the data of 2903 patients who underwent cardiac surgery between June 1994 and December 1999 at a center. Prospective internal validation was performed on 708 patients between January 2000 and June 2001. External and temporal validation of the recalibrated model were performed between February 2000 and December 2009, evaluating model discrimination and calibration in patients operated on at four centers different from the one where the score had been originally developed. The method was also compared to the EuroSCORE. Results The external validation was performed on 1657 patients, mean age was 62.8±13.3 years and global mortality was 4.58%. The ArgenSCORE showed both good discriminatory power with an area under the ROC curve of 0.80 and predictive capacity for risk assessment in all patients (observed mortality 4.58% vs. expected mortality 4.54%; p=0.842). The EuroSCORE showed good discriminatory power (area under the ROC curve of 0.79) but overestimated the risk (observed mortality 4.58% vs. expected mortality 5.23%; p <0.0001). Conclusions The ArgenSCORE showed an adequate capacity to predict in-hospital mortality in cardiac surgery 10 years after being developed. The score can be applied to populations with similar geographic characteristics, showing a better performance compared to an established international risk stratification model.