K S Berg1, R Stenseth, H Pleym, A Wahba, V Videm. 1. Department of Laboratory Medicine, Children's and Women's Health, Norwegian University of Science and Technology, Department of Cardiothoracic Anaesthesia and Intensive Care, St. Olav University Hospital, Trondheim, Norway.
Abstract
BACKGROUND: Several models for prediction of early mortality after open-heart surgery have been developed. Our objectives were to develop a local mortality risk prediction model, compare it with the European System for Cardiac Operative Risk Evaluation (EuroSCORE), and investigate whether the addition of intra-operative variables could enhance the accuracy of risk prediction. METHODS: All 5029 patients undergoing open-heart surgery in 2000-2007 were included in the study. Logistic regression with bootstrap methods was used to develop a pre-operative risk prediction model for in-hospital mortality. Next, several intra-operative variables were added to the pre-operative model. Calibration and discrimination were assessed, and the model was internally validated for prediction in future datasets. We thereafter compared the pre-operative model with the additive and logistic EuroSCOREs. RESULTS: Our pre-operative model included eight risk factors that are routinely registered in our department: age, gender, degree of urgency, operation type, previous cardiac surgery, and renal, cardiac, and pulmonary dysfunction. The model estimated mortality accurately throughout the dataset except in the 1% of patients at extremely high risk, in which mortality was somewhat overestimated. The estimated shrinkage factor was 0.930. The areas under the receiver operating characteristic curve for our pre-operative model and the logistic EuroSCORE were 0.857(0.823-0.891) and 0.821(0.785-0.857) (P=0.02). There was no significant difference in performance between the pre-operative and the intra-operative model (P>0.10). CONCLUSION: Our pre-operative model was simple and easy to use, and showed good predictive ability in our population. Internal validation indicated that it would accurately predict mortality in a future dataset.
BACKGROUND: Several models for prediction of early mortality after open-heart surgery have been developed. Our objectives were to develop a local mortality risk prediction model, compare it with the European System for Cardiac Operative Risk Evaluation (EuroSCORE), and investigate whether the addition of intra-operative variables could enhance the accuracy of risk prediction. METHODS: All 5029 patients undergoing open-heart surgery in 2000-2007 were included in the study. Logistic regression with bootstrap methods was used to develop a pre-operative risk prediction model for in-hospital mortality. Next, several intra-operative variables were added to the pre-operative model. Calibration and discrimination were assessed, and the model was internally validated for prediction in future datasets. We thereafter compared the pre-operative model with the additive and logistic EuroSCOREs. RESULTS: Our pre-operative model included eight risk factors that are routinely registered in our department: age, gender, degree of urgency, operation type, previous cardiac surgery, and renal, cardiac, and pulmonary dysfunction. The model estimated mortality accurately throughout the dataset except in the 1% of patients at extremely high risk, in which mortality was somewhat overestimated. The estimated shrinkage factor was 0.930. The areas under the receiver operating characteristic curve for our pre-operative model and the logistic EuroSCORE were 0.857(0.823-0.891) and 0.821(0.785-0.857) (P=0.02). There was no significant difference in performance between the pre-operative and the intra-operative model (P>0.10). CONCLUSION: Our pre-operative model was simple and easy to use, and showed good predictive ability in our population. Internal validation indicated that it would accurately predict mortality in a future dataset.