OBJECTIVE: To revalidate a local model for prediction of in-hospital mortality after coronary surgery several years after its introduction and the EuroSCORE in a specific area within its original scope. To assess the specific advantages of one type of instrument over the other in a definite context. STUDY DESIGN AND SETTING: Data from consecutive patients undergoing a first isolated coronary artery bypass between November 2001 and November 2003 in five hospitals in Catalonia were prospectively collected. RESULTS: The study included 1,605 patients. Areas under the receiver-operating characteristics curves were around 0.75 for both models. Calibration was low for both models and the local model significantly overestimated risk. The ordering of operating centers by performance was identical with each strategy but the centers labeled as outliers differed. CONCLUSION: (1) Evaluation of performance of individual hospitals was consistent using both systems and almost identical when they were internally recalibrated, (2) The impact of the benchmark population characteristics on model performance may be greater than that of risk factors considered for score calculation, (3) Promoting the use of a widely used instrument as the EuroSCORE might be sufficient for most evaluations. Local scales can be useful to highlight locally relevant features and temporal trends.
OBJECTIVE: To revalidate a local model for prediction of in-hospital mortality after coronary surgery several years after its introduction and the EuroSCORE in a specific area within its original scope. To assess the specific advantages of one type of instrument over the other in a definite context. STUDY DESIGN AND SETTING: Data from consecutive patients undergoing a first isolated coronary artery bypass between November 2001 and November 2003 in five hospitals in Catalonia were prospectively collected. RESULTS: The study included 1,605 patients. Areas under the receiver-operating characteristics curves were around 0.75 for both models. Calibration was low for both models and the local model significantly overestimated risk. The ordering of operating centers by performance was identical with each strategy but the centers labeled as outliers differed. CONCLUSION: (1) Evaluation of performance of individual hospitals was consistent using both systems and almost identical when they were internally recalibrated, (2) The impact of the benchmark population characteristics on model performance may be greater than that of risk factors considered for score calculation, (3) Promoting the use of a widely used instrument as the EuroSCORE might be sufficient for most evaluations. Local scales can be useful to highlight locally relevant features and temporal trends.