Identification and Quantification of Apneas by Computer-based Analysis of Oxygen Saturation

Abstract Manual detection and quantitation of apneas from an all-night polysomnogram is very time-consuming. Because SaO2 changes with virtually every apnea event, we reasoned that by identifying cyclical SaO2 changes, we could calculate (1) an apnea-hypopnea index that would correlate very well with the manually derived apnea-hypopnea index, and (2) the duration of apnea-hypopnea events. We developed a computer algorithm to scan and detect dips in SaO2 data digitally stored as a time series by computer throughout overnight studies. Desaturations detected by computer were compared with the events detected manually in 9 all-night polysomnograms from 6 patients with typical obstructive sleep apnea. Events detected by one method but not the other were subsequently verified to determine the overall number of apnea-hypopnea events present and to determine false positive and false negative rates for the 2 methods of detection. The total number of apneas was 4,008. Both methods agreed on 3,639 of them. Of 77 manually recorded apneas not detected by computer, 24 were subsequently discounted (manual false positives, 24 of 4,007 = 0.6%) and 53 confirmed (computer false negatives, 1.32%). Of 358 events not detected manually, 316 were confirmed (manual false negatives, 7.9%) and 42 discounted (computer false positives, 1.1%). Using the final manual scoring as the reference, the computer program detected apneas with a sensitivity of 97.9%, and the predictive value of a computer-detected event was 90.8%. for event duration, a regression was performed on 3,623 matched apneas-hypopnea events, giving a coefficient of r = 0.9431, p 10−6. We conclude that computer-based analysis of continuous SaO2 will detect apneas and hypopneas with a very high sensitivity and a very low false positive and false negative rate, and thus can be used to calculate event number and duration.