ABSTRACT
Many studies, focused on identifying new biomarkers for coronary artery disease (CAD) risk computation and monitoring, suggested a potential diagnostic role for fatty acids (FA). In the present study, we explored the potential diagnostic role of FA by using a data mining approach based on fourth generation artificial neural networks (ANN). Forty-one male subjects were enrolled. According to coronary angiography, 31 displayed CAD and 10 did not (non-CAD, control group). FA analysis was performed on plasma samples using a gas chromatography-mass spectrometry system and analyses were performed by an ANN method. The variables most closely related to CAD were low levels of alpha-linolenic acid, eicosapentaenoic acid, eicosatetraenoic and docosahexaenoic acids. High levels of 1,1-dimethoxyhexadecane, total dimethyl acetals and docosatetraenoic acid were related to non-CAD condition. This subset of variables, which were most closely correlated to the target diagnosis, achieved a consistent predictive rate. The average accuracy obtained was 76.5%, with 93% of sensitivity and 60% of specificity. The area under the ROC curve was equal to 0.79. In conclusion, our study highlighted the association between different plasma FA species, CAD and non-CAD conditions. The specific subset of variables could be of interest as a new diagnostic tool for CAD management.
Link to Full Text