Neurally mediated syncope prediction based on changes of cardiovascular performance surrogates: Algorithms comparison

Two methodologies for neurally mediated syncope (NMS) prediction, based on the joint analysis of the electrocardiogram (ECG) and photoplethysmogram (PPG), are compared. Several features that characterize the variations in the inotropic, chronotropic, vascular tone and blood pressure surrogates were extracted and fed into two prediction models. The first method is based on the combination of the Minkowski distance metric with a threshold-based approach to evaluate the changes in the extracted features regarding the patient orthostatic stable state. The second method implements a SVM classification model to identify pre-syncope events. The output of the classification model is regularized using a “Firing power” (FP) measure and a threshold-based approach is used to generate alarms. Both methodologies were validated in 43 subjects using a three-way data split approach. The results achieved by the presented methodologies show that the first methodology is able to predict syncope episodes with better accuracy (sensitivity (SE) of 100% and specificity (SP) of 92%) while maintaining a low rate of false alarms (FPRh: 0.146h^-1) and good prediction time (aPTime: 217.58s).