Rhythm or morphology-which information in the ECG is more useful with respect to detection of sleep apnea?

This study investigates and compares the utility of different electrocardiographic features quantifying variability of either heart rhythm or of ECG morphology for detection of sleep apnea. For heart rhythm, the sequence of RR-intervals was analyzed. To characterize ECG morphology, signal vectors were extracted from the QRS- and the T-wave region. Their projection on the first principal component yields series of scalar coefficients which were used as parameters of morphology. From the derived series as well as from a demodulated version of these sequences, we calculated a measure of similarity as well as a spectral parameter from temporal segments of 5 minutes duration. The diagnostic accuracy of these features was evaluated by means of ROC-analysis and a second order polynomial classifier. Best results were obtained from the local similarity index of the projection coefficients from the T-wave region (sensitivity 87.2%, specificity 87.5%). A slight performance increase was achieved by considering additional features from the QRS-region and the RR-intervals (88.8%/93.1%). It is concluded that although relevant information is embedded in the heart rhythm, the variability of ECG morphology reflect apnea related patterns slightly better.