Johan Nilsson, Mattias Ohlsson, Lars Thulin, Peter Höglund, Samer A.M Nashef and Johan Brandt Risk factor identification and mortality prediction in cardiac surgery using artificial neural networks The Journal of Thoracic and Cardiovascular Surgery 132, 12-19 (2006) Abstract: Objective: Artificial neural networks (ANNs) is a non-linear technology useful for complex pattern recognition problems. This study aimed to develop a method to select risk variables and predict mortality after cardiac surgery by using ANNs. Methods: Prospectively collected data from 18362 patients undergoing heart surgery at 128 European institutions in 1995 (the EuroSCORE database) was used. Models to predict the operative mortality were constructed using ANNs. For calibration a 6-fold cross-validation technique was used, and for testing a 4-fold cross-testing was performed. Risk variables were ranked and minimized in number by calibrated ANNs. Mortality prediction with 95% confidence limits for each patient was obtained by the bootstrap technique. The area under the receiver operating characteristics (ROC) curve was used as a quantitative measure of the ability to distinguish between survivors and non-survivors. Subgroup analysis of surgical operation categories was performed. The results were compared with those from logistic EuroSCORE analysis. Results: The operative mortality was 4.9%. ANNs selected 34 of the total 72 risk variables as relevant for mortality prediction. The ROC area for ANNs (0.81) was larger than the logistic EuroSCORE model (0.79; p=0.0001). For different surgical operation categories there were no differences in the discriminatory power for the ANNs (p=0.15) but significant differences were found for the logistic EuroSCORE (p=0.0072). Conclusions: Risk factors in a ranked order contributing to the mortality prediction were identified. A minimal set of risk variables achieving a superior mortality prediction was defined. The ANN model is applicable independent of the cardiac surgical procedure. LU TP 05-50