A new phase space analysis algorithm for the early detection of syncope during head-up tilt tests

In this study, we evaluate the ability to automatically predict syncope and differentiate between patients with positive response to head up tilt test (HUTT) and other with negative response, using only 12-min RR-interval time series for the supine position, preceding HUTT. An original method, based on the analysis of heart rate variability dynamics and a combination of phase space (PS) analysis and kernel support vector machine (KSVM), is proposed. The dynamic behavior of the RR-interval time series was analyzed using reconstructed phase space (RSP). Parameters computed from the phase space area such as the phase space density and indices derived from the recurrence quantification analysis were computed. Only parameters, displaying a statistical difference, are used for further classification using KSVM, to identify negative and positive patients. By applying a cross validation procedure repeated 10 times using 1/3 of the population in the training step, we determined the capability of correctly classifying positive patients. An optimal configuration maximizing the sensitivity for the early detection of positive response was found leading to 95% of sensitivity and 47% of specificity. RPS combined with KSVM demonstrate the interest to take into account the dynamics of the RR series and their capability to predict tilt test's outcome using only pre-HUTT data.