A Machine Learning Framework to Detect Syncope using the Active Stand

Background Vasovagal syncope (VVS) is the most common form of syncope, accounting for 50-60% of unexplained syncope. Currently diagnosis is achieved via clinical assessment combined with the Head-Up Tilt Test (HUT). Aim To examine the utility of the active stand test (AS) to identify those with a positive HUT or diagnosis of VVS. Design Retrospective study of hemodynamic responses to AS. Methods Continuous blood pressure responses to AS from 101 patients attending a Falls and Blackouts Unit were acquired, including: 37 controls (CON), 64 with a clinical diagnosis of VVS (VVS+) (34 tilt-positive (HUT+) and 30 tilt-negative (HUT-)) with a mean age of 25 ± 9 years. A total of 33 hemodynamic features were extracted with a subset of these entered into linear discriminant classifier. Classification accuracy was assessed using N-fold cross-validation. Results Results indicated that it was possible to classify the outcome of the HUT with sensitivity of 58.8%, specificity of 63.3% and an accuracy of 60.9%. Using a multivariate classifier it was possible to identify those with a positive diagnosis of VVS with a sensitivity of 84.3%, specificity of 72.9% and an accuracy of 80.2%. Conclusion This study highlights the existence of a unique AS hemodynamic response characterised by autonomic hypersensitivity exhibited by young patients prone to VVS which is detectable using a multi-parameter machine learning framework. With further verification, this approach may have applications in syncope and falls diagnosis, population studies and the tracking of treatment efficacy.