Identification of arterial oxygen intermittency in oximetry data

Figures

• 

  Figure 1. Illustrative arterial oxygen saturation time series from patients without intradialytic high-frequency and high-amplitude oximetry patterns (IHHOP) (a), and with IHHOP (b). The insets show 5 min segments of both time series. The dashed line marks the 90% oxygen saturation level, below which the patient is hypoxemic.

• 

  Figure 2. Distribution of values of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\epsilon }_{opt}$$\end{document}ϵopt calculated from 5 min segments of SaO2 with 1.0 Hz sampling frequency, with a rolling window step size of 1 min, the vertically defined regions mark the clustering of the segments in three categories: no IHHOP (light blue), maybe IHHOP (light green) and with IHHOP (light red). On the right side, there are illustrative oxygen saturation 5-min segments for different values of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\epsilon }_{opt}$$\end{document}ϵopt. The y-axis is omitted to emphasize relative differences between the segments, all have the same scale for oxygen saturation values.

• 

  Figure 3. (a) Arterial oxygen saturation time series. (b) Oxygen Desaturation Density (ODD), ODD was calculated by looking at 10 min windows. (c) Rolling window analysis for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\epsilon }_{opt}$$\end{document}ϵopt, with a moving rolling window of 10 min long. The blue line corresponds to the value of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\epsilon }_{opt}$$\end{document}ϵopt calculated from a 1 Hz SaO2, and the orange line was calculated from SaO2 with a 0.1 Hz sampling frequency. The shaded areas correspond to periods in which the patient was reported asleep, color-coded to the intensity of the observed IHHOP.

• 

  Figure 4. Boxplot with the values of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\epsilon }_{opt}$$\end{document}ϵopt for different SAS intensities calculated by their respective ODD value. ****p-value ≤ 0.0001 for a t-test independent sample.

• 

  Figure 5. ROC curve for different binary classification algorithms for identifying SAS events based on \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\epsilon }_{opt}$$\end{document}ϵopt values. The stars in the figure represent the points chosen for individual analysis of sensitivity, specificity, PPR, and NPR, for the best performing classifier. For each scenario, there is the AUC for the best performing classifier.