Predicting Orthostatic Tolerance Using Tilt TestAgnostic ECG Data

This feasibility study aims to explore the potential of using Machine Learning to predict orthostatic tolerance time (OTT) using features extracted from 24-hours Holter ECG data, independent of tilt test results, collected the day before the exam. Data relevant to 179 Head-Up Tilt (HUT) tests from male participants in various Head-Down Tilt ed rest campaigns, with different level of cardiovascular deconditioning, were used for this purpose. Promising results were obtained using K-nearest Neighbour Regressor following the Random Forest classifier, with a mean absolute error of $7.57(6.97; 8.69) \mathrm{min}$, resulting in 80% accuracy, 85% recall and 86% precision in classifying subjects with OTT $\gt15 \mathrm{~min}$. Once confirmed, these results could help in avoiding HUT test in subjects whose OTT is considered normal.