New indices for sleep apnea detection from long-time ECG recordings

We used our computer program enabling detection of sleep apnea using long-time one-channel ECG signal recordings. It allows the calculations of commonly accepted six heart rate variability (HRV) parameters in time domain. We also introduced additional 34 indices which were created as a combination of selected or all basic six indices of HRV. For testing we used 70 sample recordings from the Physionet database containing single ECG signals 7 to 10 hours long. The analysis was performed on samples lasting 10000 seconds. The efficiency of the software was evaluated using the Receiver Operating Characteristic (ROC) method. For basic 6 HRV indices we found that the highest accuracy of discrimination was achieved for standard deviation of successive differences (88.5%). The area under the ROC curve was 0.89. The sensitivity and specifity were 96% and 70%, respectively. For one of the newly proposed indices which was average sum of square of all six base indices the accuracy was at the level of 90%. The area under the ROC curve was 0.85. The sensitivity and specifity were 98% and 70%, respectively.