Severity classification of obstructive sleep apnea using only heart rate variability measures with an ensemble classifier

Abstract

Identifying non-polysomnographic predictors of sleep apnea severity is an on-going challenge in the sleep research. In this study, we detect the severity of obstructive sleep apnea (OSA) condition using features derived from a single lead electrocardiogram (ECG) signal. We have explored 17 time domain, frequency domain, and nonlinear Heart Rate Variability (HRV) features to distinguish subjects in the severe and non-severe OSA groups. An optimal set of nine features including Poincare plot based metrics was found to provide significant discriminative information (p-value < 0.05) for classification. On the test dataset of 16 ECG records, a classification accuracy, sensitivity, and specificity of 87.5%, 100%, and 83.33% respectively, have been achieved using Subspace Discriminant Ensemble classifier. Results demonstrate that severe OSA detection using a single lead ECG signal can be more practical yet simpler compared to existing techniques.