Early prediction of tilt test outcome, with support vector machine non linear classifier, using ECG, pressure and impedance signals

The tilt test is a valuable clinical tool for the diagnosis of Vasovagal Syncope. No practical system has been implemented to predict the tilt test outcome at the beginning in the procedure. Our objective was to evaluate and benchmark, over a sufficient database, the predictive performance of the proposed parameters in the literature. We analyzed a database of 727 consecutive cases of tilt test. Previously proposed features were measured from heart rate and systolic/diastolic pressure, in several representative signal segments. A support vector machine (SVM) was used to predict the test outcome with the available features. Also the inclusion of additional physiological signals (impedance) was intended to improve the performance. The predictive performance of the nonline-arly combined previously proposed features was limited (p<;0.03 and area under ROC curve 0.57±0.12), especially in the beginning of the test, which is the most clinically relevant period. The improvement with additional available physiological information and SVM was limited (area under ROC curve 0.59±0.22). We conclude that the existing methods for tilt test outcome prediction knowledge should be considered with caution.