Application of oxygen saturation variability analysis for the detection of exacerbation in individuals with COPD: A proof-of-concept study

Abstract

Background: Individuals with chronic obstructive pulmonary disease (COPD) commonly experience exacerbations, which may require hospital admission. Early detection of exacerbations, and therefore early treatment, could be crucial in preventing admission and improving outcomes. Our previous research has demonstrated that the pattern analysis of peripheral oxygen saturation (S_p O₂ ) fluctuations provides novel insights into the engagement of the respiratory control system in response to physiological stress (hypoxia). Therefore, this pilot study tested the hypothesis that the pattern of S_p O₂ variations in overnight recordings of individuals with COPD would distinguish between stable and exacerbation phases of the disease.

Methods: Overnight pulse oximetry data from 11 individuals with COPD, who exhibited exacerbation after a period of stable disease, were examined. Stable phase recordings were conducted overnight and one night prior to exacerbation recordings were also analyzed. Pattern analysis of S_p O₂ variations was carried examined using sample entropy (for assessment of irregularity), the multiscale entropy (complexity), and detrended fluctuation analysis (self-similarity).

Results: S_p O₂ variations displayed a complex pattern in both stable and exacerbation phases of COPD. During an exacerbation, S_p O₂ entropy increased (p = 0.029) and long-term fractal-like exponent (α2) decreased (p = 0.002) while the mean and standard deviation of S_p O₂ time series remained unchanged. Through ROC analyses, S_p O₂ entropy and α2 were both able to classify the COPD phases into either stable or exacerbation phase. With the best positive predictor value (PPV) for sample entropy (PPV = 70%) and a cut-off value of 0.454. While the best negative predictor value (NPV) was α2 (NPV = 78%) with a cut-off value of 1.00.

Conclusion: Alterations in S_p O₂ entropy and the fractal-like exponent have the potential to detect exacerbations in COPD. Further research is warranted to examine if S_p O₂ variability analysis could be used as a novel objective method of detecting exacerbations.

Keywords: Pulse Oximetry; SpO2; entropy; physiological measurement; respiratory.

FIGURE 1

Representative 90‐minute S_pO₂ signals recorded from an individual with COPD at (a) stable phase and (b) a day prior to clinical diagnosis of exacerbation (exacerbation phase). X‐axis is the data points of the pulse oximeter signals recording (1 sample every 4 seconds), and Y‐axis is the S_pO₂ (%)

FIGURE 2

Multiscale entropy (MSE) graph describing the overall complexity of the individuals with COPD at stable phase and exacerbation. The error bars are calculated sample error of the mean values

FIGURE 3

ROC curve for classifying COPD phase (stable or exacerbation) based on S_pO₂ variability indices

FIGURE A1

Two examples of DFA graphs on S_pO₂ variability data showing the linear trend when plotting scale and detrended fluctuations on a log‐log scale. This graph represents the stable phase (red dots) and the exacerbation phase (blue dots) of a participant with COPD. α1 and α2 are short‐term and long‐term scaling exponent respectively

FIGURE B1

(a) Correlation between mean S_pO₂ and S_pO₂ Sample Entropy in individuals with COPD in Stable phase. (b) Correlation between mean S_pO₂ and S_pO₂ Sample Entropy in individuals with COPD in Exacerbation phase. There is no significant correlation between mean S_pO₂ and S_pO₂ entropy in individuals with COPD. This is unlike previous reports in healthy individuals where Entropy of S_pO₂ exhibits a significant inverse correlation with mean S_pO₂. For more information please see (Bhogal & Mani, ; Costello et al., 2020). Sample Entropy is calculated at scale 1 with m = 2 and r = 0.2

FIGURE C1

Bland‐Altman plot of sample entropy (SampEn) in 90 min vs. 60 min S_pO₂ signal duration