Extract Prediction Factors of Vasovagal Syncope from Non-Medicine Tilt-Table Testing: An Artificial Neural Network Prediction Model

The occurrence of syncope could easily lead to accidents and falls, which endangering the safety of patients enormously. Early diagnosis helps to reduce the harm caused by the disease. However, the causes of syncope are very complex, and its diagnosis is difficult. Tilting-table test is the appropriate way to assess the recurrent unexplained syncope. Some of the Tilting-table test using vasodilator agent to increase the positive rate and shorten the time, but pharmacological challenge test will decrease the specificity and increase opportunities for false positive. In this study, hemodynamic and body content data collected from cases of non-pharmacological Tilting-table, were analyzed to extract factors which are related to vasovagal syncope. These factors were applied in the neural networks to build up a model which can predict vasovagal syncope. Among total collection of 60 cases, 30 of vasovagal syncope confirmed cases and another 30 to Tilting-table test were the negative cases. The study use of non- pharmacological Tilting-table test collect the hemodynamic data from cases when they were supine, 3 minutes after tilt up test and prior to end of tilt up test. The results showed that patients with vasovagal syncope have lower body weight, body mass index and the mean blood pressure during supine position. During tilt test, there were higher heart rate variability low frequency / high frequency ratio (HRV LF/HF), lower baroreflex sensitivity (BRS) and total peripheral resistance index (TPRI) than negative group during presyncopal period. The body mass index, supine heart rate, Cardiac Index(CI) while tilt up 3 minutes , Left Ventricular Work Index(LVWI) while tilt up 3 minutes, and the mean blood pressure (MBP) while tilt up 3 minutes were selected from logistic regression analysis, which are related factors of vasovagal syncope. The model using those factors build by ANN show that the training group's sensitivity was 92%; specificity was 95% and accuracy was 93.75%.; the validation group have 100% accuracy, ROC threshold is 0.517, ROC area under the curve is 0.979; the overall sensitivity of 93%; specificity of 96%, accuracy was 95.5%. The model has good performance to predict vasovagal syncope.