Fuzzy Logic-Based Model to Stratify Cardiac Surgery Risk

Background: Medical practice is usually performed in a context of uncertainty, where expert knowledge has shown to be efficient in the decision-making process. Objective: The aim of this study was to develop and validate a fuzzy logic-based model to predict cardiac surgery mortality risk. Methods: Four hundred and fifty patients undergoing cardiac surgery were prospectively included in the study and mortality risk was predicted based on five scores: 1) “clinical expert” opinion, 2) fuzzy logic-based system according to expert knowledge, 3) Parsonnet, 4) Ontario and 5) EuroSCORE. The fuzzy logic model was developed in the following stages: expert selection of different mortality predictive variables, tables of influence among variables, construction of a fuzzy cognitive map (FCM) and its implementation in an artificial neuronal network, expert-determined patient risk score, test set risk calculation based on fuzzy predictors, validation set risk using calibrated FCM, and comparison with the other scores according to the level of agreement and precision with ROC curves. Results: The calibrated model was used to predict the outcome of the validation set (360 patients), based on the FCM score and risk predicted by Parsonnet, Ontario and EuroSCORE. The ROC areas showed that FCM had at least the same performance as other scores to predict mortality (ROC=0.793 vs. 0.775, 0.767, 0.741 and 0.701 for EuroSCORE, “expert”, Ontario and Parsonnet, respectively). Conclusions: A fuzzy logic-based system employing expert knowledge and the implementation of an expert system is postulated to predict cardiac surgery mortality risk. The model not only mimicked the outcomes obtained by the “expert”, but had the same performance as others risk scores.