JMIR Publications

ABSTRACT

Background:

Wearable devices have the potential to the aid of the diagnosis of chronic obstructive pulmonary disease (COPD) by utilizing daytime monitored vital signs, such as heart rate variability (HRV) and respiration (R). However, conventional daytime monitoring in a single day may be influenced by factors such as motion artifacts and emotions. And continuous monitoring of night-sleep HRV and R to assist the diagnosis of COPD has not been reported.

Objective:

To explore and compare the effects of the continuously monitored night-sleep HRV, heart rate (HR), and R on the diagnosis of COPD.

Methods:

We recruited patients with different severity of COPD and healthy controls between July 2020 and November 2022, constituting the case and control groups, respectively. The vital signs including HRV, HR, and R were recorded using non-contact bed sensors from 22:00 pm to 8:00 am of the following day, and the recordings of each subject lasted for at least 30 days. Instead of monitoring these parameters in a single day, we obtained statistical means of HRV, HR, and R as features over time-scales of 7, 14, and 30 days based on the continuous monitoring. Binary logistic regression was employed to discriminate between the case group and the control group, and the performance was assessed in terms of receiver operating characteristic (ROC) curve. In addition, the effects that the statistical means of HRV, HR, and R over different time of recordings have on COPD diagnosis was evaluated.

Results:

A total of 146 individuals were enrolled in this study, including 37 COPD patients in the case group and 109 participants in the control group. The averaged number of continuous night-sleep monitoring was 56.5(31.8-114.8) (days) {P50(P25-P75)}, corresponding to a cumulative person-time of 13,260, including 4,608 cases and 8,652 controls. Using the features with regarding to the statistical means of HRV, HR, and R over 1, 7, 14, and 30 days, binary logistic regression classification to COPD yielded the accuracy, Youden’s index and the area under ROC curve (AUC) of 0.958, 0.904 and 0.989, respectively. Numerical also demonstrated that the performance of classification was directionally proportional to the numbers of night-sleep monitoring. Regarding the importance of features for diagnosis, the statistical means of R, HRV, and HR followed the order of R > HRV > HR. Specifically, the statistical means of the duration of respiration rate faster than 21 times/min (RRF), the power of high frequency band 0.15~0.40 Hz in HRV (HF), and the respiration rate (RR) were identified the top 3 most significant features for classification, corresponding to the cut-off values of 0.1 min, 1316.3 (nU), and 16.3 times per min, respectively.

Conclusions:

Continuous night-sleep monitoring has significant potential for the diagnosis of COPD. As the number of nights-sleep monitoring increases from 1 to 30 days, the statistical means of HRV, HR, and R show a better reflection of an individual's health condition, compared to monitoring vital signs in a single day/night. This improvement leads to higher performance in aiding COPD diagnosis. Furthermore, numerical results highlight that the statistical means of RRF, HF, and RR are crucial features for diagnosing COPD, demonstrating the importance of monitoring both HRV and R during night-sleep.