Background: Prolonged ventilation is a serious complication after cardiac surgery, but few risk prediction models exist. Our objectives were to develop a specific risk prediction model based on pre-operative variables, to identify whether selected intraoperative variables could improve prediction, and to compare our model with the EuroSCORE.
Methods: Data from 5027 patients undergoing open-heart surgery in 2000-2007 were used for logistic regression model development. Internal validation was performed by bootstrapping. Discrimination and calibration were assessed with areas under the receiver operating characteristic curve (AUC) and the Hosmer-Lemeshow test. Our pre-operative model was compared with predictions based on the additive and logistic EuroSCORE.
Results: Age, previous cardiac surgery, peripheral arterial disease, left ventricular hypertrophy, chronic pulmonary disease, renal insufficiency, pre-operative hemoglobin concentration, urgent or emergency operation, and operation other than isolated coronary artery bypass grafting were identified as pre-operative predictors for prolonged ventilation (model I). Discrimination and accuracy were excellent (AUC: 0.848 and shrinkage factor: 94%). Calibration was good (Hosmer-Lemeshow test: P = 0.43). Inclusion of a few intraoperative variables somewhat improved the model, increasing shrinkage factors (96%) and discrimination ability (AUC model II = 0.870 and model III = 0.875 for two alternative such models). Our pre-operative model showed better performance than the logistic or additive EuroSCORE.
Conclusions: The pre-operative risk prediction model for prolonged ventilation with easily obtainable variables in routine clinical work performed well and was only slightly improved by inclusion of intraoperative variables. Performance was better than with the EuroSCORE.
© 2011 The Authors Acta Anaesthesiologica Scandinavica © 2011 The Acta Anaesthesiologica Scandinavica Foundation.