Applying time, frequency and nonlinear features from nocturnal oximetry to OSA diagnosis.

This study is aimed to improve the diagnostic ability of blood oxygen saturation (SaO(2)) in obstructive sleep apnea (OSA) detection. We studied 74 patients suspected of suffering from OSA. Ten characteristics were derived from each SaO2 recording: arithmetic mean, variance, skewness and kurtosis from both time and frequency domains, central tendency measure and Lempel-Ziv complexity. The diagnostic ability of each feature was assessed by means of a receiver operating characteristics (ROC) analysis. Additionally, forward stepwise logistic regression (LR) was applied. The kurtosis in the time domain and the nonlinear measure of complexity were automatically selected. This methodology reached 93.2% sensitivity, 80.0% specificity and 87.8% accuracy, improving the results from each feature individually. Our study showed that common statistics in the time and frequency domains and nonlinear features could provide additional and complementary information to help in OSA diagnosis.