PREDICTING MORTALITY AFTER CORONARY-ARTERY BYPASS-SURGERY - WHAT DO ARTIFICIAL NEURAL NETWORKS LEARN

Objective. To compare the abilities of artificial neural network and l ogistic regression models to predict the risk of in-hospital mortality after coronary artery bypass graft (CABG) surgery. Methods. Neural ne twork and logistic regression models were developed using a training s et of 4,782 patients undergoing CABG surgery in Ontario, Canada, in 19 91, and they were validated in two test sets of 5,309 and 5,517 patien ts having CABG surgery in 1992 and 1993, respectively. Results. The pr obabilities predicted from a fully trained neural network were similar to those of a ''saturated'' regression model, with both models detect ing all possible interactions in the training set and validating poorl y in the two test sets. A second neural network was developed by cross -validating a network against a new set of data and terminating networ k training early to create a more generalizable model. A simple ''main effects'' regression model without any interaction terms was also dev eloped. Both of these models validated well, with areas under the rece iver operating characteristic curves of 0.78 and 0.77 (p > 0.10) in th e 1993 test set. The predictions from the two models were very highly correlated (r = 0.95). Conclusions. Artificial neural networks and log istic regression models learn similar relationships between patient ch aracteristics and mortality after CABG surgery.