Machine learning model using heart rate variability for the prediction of vasovagal syncope

Abstract Background and Aims: Vasovagal syncope (VVS) is a reversible condition that can cause serious events, such as cerebral hemorrhage due to head trauma. Its occurrence is difficult to predict; therefore, this study aimed to develop machine learning (ML) models using heart rate variability (HRV) to predict the occurrence of VVS. Methods and Results We investigated 80 patients with syncope who underwent Holter monitoring during the head-up tilt test. Twenty-six ML models for the prediction of VVS using HRV parameters were examined. We randomly split the subjects into training and testing sets at a 7:3 ratio (training set: 46 positive and 9 negative subjects; testing set: 21 positive and 4 negative subjects). The XGB classifier achieved the highest performance compared with other algorithms (accuracy, 0.72; area under the curve, 0.801; sensitivity, 0.857; specificity, 0.621). When examining the time taken to predict syncope occurrence using the developed prediction model, it was found that the model could forecast approximately 3 min (equivalent to 193.2 beats) prior to the onset of symptoms. Conclusion The ML model utilizing HRV parameters was proficient in preemptively predicting VVS incidents. In practical applications, this approach can effectively enhance VVS prediction by leveraging the HRV, thereby aiding in the prevention of unforeseen accidents.