Continuous characterisation of exacerbation pathophysiology using wearable technologies in free-living outpatients with COPD: a prospective observational cohort study

Summary

Background

The most recent exacerbation of COPD (ECOPD) classification criteria relies in part on changes in respiratory rate (RR), heart rate (HR) and oxygen saturation (SpO₂). Despite this paradigm shift, a thorough understanding of exacerbation patterns is still lacking, as is the identification of physiological exacerbation biomarkers.

Methods

Using a convenience sampling approach, this prospective observational cohort study was conducted between February 2023 and January 2024. Continuous measurements of daytime/overnight respiratory (primary outcome), cardiovascular, autonomic, activity and sleep-related parameters were collected by a wearable biometric wristband and ring over 21 consecutive days in free-living outpatients experiencing and receiving treatment (≤3 days) for a current exacerbation from the home environment. The EXACT-PRO questionnaire served as the validated reference for daily symptom burden and to identify ‘recovered’ versus ‘persistent worsening’ participants. Unadjusted and adjusted (for age, sex, FEV₁) linear mixed-effects models were fitted to estimate associations between each physiological parameter with daily EXACT-PRO score (points, pts), in all, ‘recovered’, and ‘persistent worsening’ participants. Results are presented as point estimates with 95% CIs.

Findings

In 21 participants with COPD (43% female, mean age 66.8, BMI 27.7 kg/m², FEV₁ 36.3% predicted; 85.7% with GOLD 3–4 disease), significant associations in unadjusted models with daily EXACT-PRO score included RR variability (−1.45 [−2.84, −0.073] pts/breath/min) but not RR, daily step count (−0.56 [−0.82, −0.31] pts/1000 steps), and sleep efficiency (−0.12 [−0.20, −0.037] pts/%asleep). In ‘recovered’ participants (n = 10), significant associations included nighttime HR, movement intensity and nightly SpO₂. In ‘persistent worsening’ participants (n = 11), significant associations included HR variability, nightly RR variability, nightly SpO₂, sleep efficiency, and skin temperature. Similar results were found in adjusted models.

Interpretation

This study provides a prospective continuous characterisation of exacerbations of COPD using remotely collected, ambulatory/free-living data. The physiological patterns presented may contribute to the understanding of exacerbations and may enhance the development of effective remote monitoring solutions.

Funding

University hospital (MUHC-CAS) grant.

Research in context.

Evidence before this study

On July 29th, 2023, in preparation for a subsequently published review article, the ‘Medline’ database was searched for original research studies published since 2000 using the following search strategy (((((“exacerbation of COPD”) OR (“COPD exacerbation”)) OR (“AECOPD”)) OR (“ECOPD”)) AND ((“predict∗”) OR (“characterisation”) OR (“identification”))) AND (((((“remote”) OR (“home”)) OR (“telemonitoring”)) OR (“telehealth”)) OR (“wearable∗”)) AND (2000:2023[pdat]). This search yielded 105 results. On June 1st, 2024, the Medline database search was repeated using the identical search terms. This search yielded 4 additional results.

Of the 109 papers in total, identified across both literature searches, 100 were original research studies. 32 articles investigated prediction models for endpoints not focused on the exacerbations themselves (including mortality, hospital admission and length of stay). 22 articles focused on the management of exacerbations, by investigating the feasibility or outcomes of modalities such as home care models, remote pulmonary rehabilitation or exercises/tools for self-management at home. 15 articles studied the predictive value of physiological parameters, all of which utilised non-continuous (i.e. one-time or interval in-person visit) and/or in-person laboratory-based sampling methods (such as pulmonary function testing; bloodwork-derived lymphocyte count, pCO₂, hyponatraemia; and radiological testing) to ‘predict’ exacerbations. 13 articles investigated the predictive value of cough, dyspnoea, air pollution, psychosocial or demographic characteristics on exacerbation risk. 4 studies used data obtained from home non-invasive ventilation devices and 3 used handheld spirometry data. A total of 11 ‘remote monitoring’ articles focused on the variations of physiological parameters in patients with COPD. Four of these used portable devices (such as portable electronic auscultation, oximeter and spirometer) and collected data only once daily, and 5 of these used wearable devices. Of these 5 studies, one monitored daytime-only heart rate (HR), respiratory rate (RR), electrocardiogram and associated R–R interval and physical activity level through a chest-band biosensor; one monitored daytime-only RR, HR, skin temperature and physical activity with a biometric vest, and did not report the results of these measurements (only measurements of feasibility were presented); one used a chest strap to monitor R–R interval data and HRV during a one-time 7-min data collection period (single visit); one used waistband cardiorespiratory sensors to monitor HR, RR, continuous respiratory force, and activity for around 8 h per day; and one used a biometric wristband to monitor HR, step count, stairs climbed, distance, calorie consumption and sleep patterns. While this latter wristband-based study is the only article in the existing literature that appears to have collected continuous multiparameter data, these results are not presented in the article (this data was fed into a variety of machine learning-based prediction models) and therefore no characterisation of the exacerbations nor associations with either symptom burden or clinical outcome are available. Five articles used machine-learning prediction models for exacerbations of COPD, however none reported a strong evidence-based justification for the parameters (collected through portable or wearable devices) used in the models, nor gave an indication regarding which parameter(s) were most relevant in their modelling.

Added value of this study

In the present prospective observational study, respiratory, cardiac, autonomic, activity-specific, and sleep-specific physiological parameters were continuously collected from the home environment of 21 participants experiencing a current exacerbation over three consecutive weeks. Our results contribute to the existing literature regarding the understanding of exacerbations of COPD as clinical phenomena through detailed and prolonged physiological characterisation in an unperturbed setting. Patterns differed depending on the clinical trajectory and outcome (recovery versus persistent worsening) of the event. Beyond cardiac and respiratory parameters, daily physical activity (especially in those who ultimately recover) and autonomic function and sleep quality (especially in those who ultimately do not recover) appear to be important indicators not described in the latest exacerbation of COPD classification criteria.

Implications of all the available evidence

Our findings imply that specific combinations of physiological indicators exist and may depend on ultimate clinical outcome status. The patterns and associations described can inform remote patient exacerbation detection and monitoring platforms. A phased approach is necessary, requiring first the need to understand and to characterise exacerbation events in detail before being able to develop sophisticated, accurate, and ‘informed’ systems. These study findings provide foundational knowledge in support of this active and growing field of research.

Introduction

Chronic obstructive pulmonary disease (COPD) is a highly prevalent respiratory disease that is now the third-leading global cause of death.¹ In addition to chronic symptoms and gradual progressive decline, its natural history is also marked by acute, episodic attacks with increased respiratory symptoms (dyspnoea and/or cough and sputum) and abnormal physiological alterations known as exacerbations of COPD (ECOPDs).¹ Each occurrence is characterised by a pro-inflammatory burst capable of irreversible lung damage, leading to persistent worsening in lung function.¹ Non-respiratory sequelae of these pro-inflammatory events include increased cardiovascular event risk.² The overall rate of exacerbations and of COPD hospitalisations continues to increase, partly due to the increasing prevalence of severe COPD.^1,3 There is, however, substantial individual-level variation in exacerbation risk that is not necessarily predictable by airflow obstruction severity alone.¹ Exacerbation onset, duration, and recovery are also highly heterogeneous,⁴ and events often cluster together in time.⁵

The increasing global burden of COPD and of exacerbation admissions³ translates to increasingly challenging demands being placed on healthcare systems with limited capacities and resources.⁶ Notably, ECOPD admissions represent a top cause of hospitalisation across all adult chronic diseases³ and pose a significant economic strain.¹ The new exacerbation definition and classification criteria recently presented in the Rome Proposal⁷ and subsequently incorporated into international COPD reports¹ is a step towards more objective, pathophysiological nomenclature in that it addresses issues of previous frameworks that relied mainly on retrospective classification based on treatment(s) ultimately received and treatment setting. Six objectively measured variables including respiratory rate (RR), blood oxygen saturation (SpO₂), and heart rate (HR), in addition to a patient-reported symptom score, are now proposed as the criteria to classify exacerbation severity (‘mild’, ‘moderate’, or ‘severe’).⁷ Despite this, the present-day reliance on patient self-detection of new or worsening symptoms to initially identify exacerbations can cause late or even non-detection leading to worse outcomes,⁸ which highlights a significant and persisting care gap. The current understanding of exacerbation pathophysiology consists of an initial exposure-precipitated pro-inflammatory cascade (airway inflammatory ‘burst’) leading to downstream worsening expiratory flow limitation and manifesting clinically as sudden severe dyspnoea with physical activity limitation/avoidance.⁷ Prompt identification and treatment early in this cascading phenomenon is associated with substantial improvement in exacerbation severity trajectory and patient outcomes⁹ and therefore represents an important opportunity for monitoring and intervention in order to avoid hospitalisation.¹⁰

Novel approaches to improve the quality of early exacerbation detection such as the development and clinical deployment of physiology-based remote patient monitoring (RPM) platforms¹¹ could help to address important knowledge and care gaps. Previous work in the remote monitoring literature has demonstrated the potential for effective early exacerbation detection through the identification of subtle alterations in cardinal vital signs such as RR in the 3–5 days prior to the event itself and which, importantly, precede overt symptom onset.^12,13 Wearable biometric devices (‘wearables’), a recent innovation in non-invasive monitoring from the comfort of the home setting, are built to be worn on the body or clothing¹⁰ over extended periods and continuously gather and remotely transmit patient physiological data to a central hub/repository. Remote monitoring studies in the population of patients with COPD have established the capacity for prolonged wear⁶ and even to detect changes in RR both before^12,13 and after¹⁴ an exacerbation. While respiratory, cardiac, autonomic, activity-related, and sleep-related variables may be important candidate ‘sentinel’ variables relating to exacerbation-specific pathophysiological derangements,¹⁵ remarkably, despite the large and growing burden of exacerbations on health systems and patients alike the body of knowledge and understanding of the natural history of ECOPDs through the lens of these phenomena is still sorely lacking even in the modern era. This deficiency in a comprehensive multi-day, continuous (day-and-night) pathophysiological characterisation of exacerbations is particularly notable in ambulatory, free-living outpatients within the more ‘natural’ unperturbed home environment.

The objectives of the present study were to remotely measure and characterise respiratory, cardiac, autonomic, activity-specific and sleep-specific physiological patterns during a current exacerbation of COPD, over multiple weeks from the home environment using two wearable biometric devices, and to determine which physiological parameter(s) are associated with validated daily exacerbation symptom and recovery scores. We hypothesised that the continuous (daytime/nighttime) longitudinal physiological measurements would lead to a comprehensive multi-system characterisation of the pathophysiology of exacerbations, and that respiratory and activity parameters in particular would be closely associated with the validated exacerbation symptom and recovery scores.

Methods

Study design and eligibility criteria

This single-site, prospective, longitudinal cohort study was conducted at the Montreal Chest Institute of the McGill University Health Centre (MUHC) and included patients with COPD presenting with a current exacerbation at the urgent Day Hospital, ambulatory respiratory clinic or Emergency Department. Data collection took place between February 2023 and January 2024.

Further details about recruitment and the collection of demographic data can be found in the Supplement. Eligibility criteria were as follows: male or female; aged ≥40 years; current/former smokers ≥10 pack-years; diagnosed with COPD of any severity of airflow obstruction (GOLD 1–4)¹ as documented on spirometry (forced expiratory volume in 1 s (FEV₁)/forced vital capacity (FVC) < 0.7); currently experiencing and receiving treatment for a physician-diagnosed exacerbation with corticosteroids, antibiotics, or both; and able to provide informed consent. Participants without an existing COPD diagnosis, history of cardiac arrhythmia or presence of pacemaker/defibrillator, any medical/cognitive/functional condition rendering an inability to don/doff/recharge wearable devices or complete tablet-based daily symptom/recovery questionnaires, or who already received more than 3 days of consecutive corticosteroid and/or antibiotic therapy at the time of recruitment were excluded.

Ethics

This study was approved by the MUHC Research Ethics Board (study number: 2023–8851). All participants provided informed consent prior to participation. This study was registered on ClinicalTrials.gov (NCT05776654).

Recruitment and protocol

Using a convenience sampling approach, participants were recruited and trained at the bedside during one single in-person baseline visit on how to wear, recharge, and upload data from two wearable devices: a biometric wristband (EmbracePlus, Empatica Inc.) and a biometric ring (Oura Gen III, Oura Health Oy).¹⁶ A study tablet, for daily symptom/recovery questionnaire completion and for device-syncing/upload of physiological data, was also provided. Participants were asked to always wear the devices except when recharging and showering. Following the single recruitment/training visit, participants were discharged home and autonomously wore, recharged, collected, and regularly uploaded continuous (daytime/nighttime) data from their wearable devices from the home environment for 21 consecutive days. Participants completed a daily electronic Exacerbation of Chronic Pulmonary Disease Tool (EXACT) Patient-Reported Outcome (PRO) questionnaire for the duration of the study period (Days 1–21 inclusive) on the study tablet. A printed instruction sheet was provided, and periodic telephonic support with a study team member was available on an as-needed basis for troubleshooting assistance.

Parameters and outcomes

In total, wristband and ring wear yielded data on three respiratory parameters (RR-wristband, nighttime RR-ring, nighttime SpO₂-ring), four cardiac parameters (HR-wristband, HR-ring, nighttime HR-ring, nighttime nadir HR-ring), six autonomic parameters (HR variability (HRV)-wristband, HRV-ring, nighttime HRV-ring, nighttime RR variability (RRV)-ring, skin temperature-wristband, electrodermal activity-wristband), ten activity-related parameters (daily physical activity-ring, total daily calorie expenditure-ring, rest time-ring, daily METs-ring, active daily calorie expenditure-ring, daily time spent inactive-ring, daily step count-wristband, daily step count-ring, movement intensity-wristband, activity counts-wristband) and nine sleep-related parameters (time in deep sleep-ring, sleep disturbances-ring, sleep efficiency-ring, sleep onset latency-ring, time in rapid eye movement (REM) sleep-ring, awake time in sleep period-ring, total sleep period-ring, restless sleep-ring, and sleep duration-ring).¹⁶ The devices selected were observed to be associated with excellent wear time and very high usability scores.¹⁶

The EXACT-PRO is a 14-item symptom questionnaire developed in compliance with FDA PRO guidance.¹⁷ Raw scores are converted to a 0–100 scale, and higher scores indicate clinical worsening. It has been validated to quantify even subtle relative changes in clinical status during both peak and recovery phases of exacerbations18, 19, 20, 21 and can be used to determine exacerbation frequency, severity, magnitude and duration given the capacity to identify specific dates of ‘onset’ and ‘recovery’.²² The EXACT-PRO has been shown to detect exacerbation events that may have been otherwise unreported but with similar health consequences and treatment responses to ‘clinical’ (event-based) exacerbations,²³ and has been an important reference standard for identification of ECOPDs in outpatient remote monitoring research.¹² The “acute-treatment trial” EXACT-PRO was used in the present study.

Statistics

Effect size and sample size were estimated from two previous COPD studies that prospectively measured RR differences between stable-state and peak exacerbation phases (see Supplement).^12,13 Participant baseline characteristics are presented as means and standard deviations (SD) for continuous variables, or absolute frequencies (n) and percentages for categorical variables, unless otherwise indicated.

The validated definition of exacerbation ‘recovery’ (EXACT-PRO reduction ≥9 points (pts) from maximum observed value, sustained ≥7 days),²⁴ was used to label each participant as having either ‘recovered’ or as having demonstrated ‘persistent worsening’ during the 21-day observation period. Average EXACT-PRO scores and physiological measurements (respiratory, cardiac, autonomic, activity-related and sleep-related parameters) for all participants were determined at baseline (wristband: Study Day 1; ring: Study Day 2, due to device-specific calibration period),¹⁶ and then again on the mean exacerbation recovery study day of the ‘recovered’ group and on that same mean study day (‘iso-time’) in the ‘persistent worsening’ group for descriptive purposes. The distinct exacerbation period respiratory, cardiac, autonomic, activity and sleep-related profiles over time by group (recovered versus persistent worsening) were plotted graphically for a visual time-dependent detailed pathophysiological characterisation of each group. Only exacerbation (non-recovered) days were included in exacerbation characterisation plots and in statistical analyses.

In all regression models, daily EXACT-PRO score served as the dependent variable and the respiratory, cardiac, autonomic, activity-related, or sleep-related parameter served as the independent variable. In all study participants, the associations between each daytime/nighttime parameter and the daily EXACT-PRO score were estimated by fitting separate linear mixed-effects univariable and multivariable models. In group-specific analyses, these models were fitted separately for the recovered and persistent worsening groups. All models accounted for repeated subject measurements/clustering (random effects) and for study day (fixed effects),¹⁴ using an autoregressive correlation structure. Parameters with non-daily/nightly estimates (ex. q1-5 mins)¹⁶ were aggregated (averaged) to match corresponding daily EXACT-PRO scores. Multivariable models were additionally adjusted for age, sex, and FEV₁.

In sensitivity analyses, only ‘inactive’ periods of the wristband data (sampling increments with step count of zero) were included to remove any undue effect of ‘physiological’ (i.e., expected) changes secondary to movement/physical activity on the ‘pathophysiological’ exacerbation-specific phenomena under investigation.

All models generated point estimates with 95% confidence intervals (CIs). Associations are reported as the change in EXACT-PRO score (100-pt scale) for every 1-unit increase in the given parameter. A P-value <0.05 was considered statistically significant. Adjustments for multiple comparisons were not made. Statistical analyses were performed in R Statistical Software, version 4.3.2.²⁵

Role of funders

The funders played no role in study design, data collection, analysis, interpretation, or in the writing of the report.

Results

Of 21 participants that completed the study (see Fig. 1), nine (43%) were female, with mean age 66.8 years, mean FEV₁ 1.3 L (36.3% predicted), and most with either ‘severe’ (GOLD 3: 52.4% of participants) or ‘very severe’ (GOLD 4: 33.3% of participants) COPD. Baseline characteristics are presented in Table 1. Median exacerbation treatment duration at enrolment was 1 day (minimum 0, maximum 3), thus the observation period captured nearly the full ‘clinical’ event. As observed previously,¹⁶ device wear-time adherence and end-of-study participant-reported device usability scores were excellent for both wristband (93.3% wear-time, mean System Usability Score (SUS) 82 ± 12/100) and ring (88.8% wear-time, mean SUS 89 ± 9/100). For all participants throughout the 21-day observation period, a daily estimate was available from the devices. For the parameters that were averaged in order to obtain a single daily estimate, the percentage of useable data originally obtained by the devices prior to averaging is presented in Table S3. The major source of data missingness was from unreliable/artefactual or low-quality data. Imputation was not performed.

Fig. 1.

Flow diagram of study participants. Twenty-one participants were included. COPD, chronic obstructive pulmonary disease; ECOPD, exacerbation of COPD.

Table 1.

Baseline participant demographics, lung function, symptom burden, and therapies.

	Value
Demographics
Participants, No.	21
Age, mean (SD): years	66.8 (6.4)
Female sex, No. (%)	9 (43%)
Female gender, No. (%)	9 (43%)
Male sex, No. (%)	12 (57%)
Male gender, No. (%)	12 (57%)
Ethnicity
Caucasian	18
African American	2
South Asian	1
BMI, mean (SD): kg/m2	27.7 (7.5)
Smoking PYHx, median (25th P, 75th P): years	40 (30, 57.5)
Lung function & symptom burden
FEV1, median (25th P, 75th P): litres	1 (0.8, 1.4)
FEV1, mean (SD): % predicted	36.3 (12.2)
FVC, median (25th P, 75th P): litres	1.9 (1.6, 3.1)
FVC, mean (SD): % predicted	61.5 (16.2)
Ratio, mean (SD)	0.5 (0.1)
GOLD 1, No. (%)	0 (0%)
GOLD 2, No. (%)	3 (14.3%)
GOLD 3, No. (%)	11 (52.4%)
GOLD 4, No. (%)	7 (33.3%)
mMRC, mean (SD):/4	3.4 (0.6)
Chronic COPD therapies & exacerbation treatment duration
LTOT use, No. (%)	1 (4.8%)
LAMA use, No. (%)	21 (100%)
LABA use, No. (%)	20 (95.2%)
ICS use, No. (%)	20 (95.2%)
Macrolide use, No. (%)	10 (47.6%)
Treatment duration at enrolment, median: days (25th P, 75th P)	1 (0, 1)

Abbreviations: BMI, body mass index; FEV₁, forced expiratory volume in 1 s; FVC, forced vital capacity; GOLD, Global Initiative for Chronic Obstructive Lung Disease; kg, kilogram; LABA, long-acting beta-agonist; LAMA, long-acting muscarinic antagonist; LTOT, long-term oxygen therapy; m², metre squared; mMRC, modified Medical Research Council scale; No., number; P, percentile; PYHx, pack-year history; ICS, inhaled corticosteroid; IQR, interquartile range; SD, standard deviation; %, percent.

Ten participants fully met criteria for ‘recovery’ from their exacerbation within the 21-day follow-up period (recovered group), whereas eleven did not (persistent worsening group). No participant required a change in medication during the observation period. Mean exacerbation duration from enrolment to the date of recovery in the recovered group was 6 days (see Table 2), with observed minimum and maximum of 3 and 12 days, respectively (see Fig. 2). Average EXACT-PRO score on study Day 1 (baseline) was 53/100 pts in all participants, 45.4/100 pts on the mean exacerbation recovery date in the recovered group, and 52.3/100 pts on the iso-time mean date (corresponding to study Day 6) in the persistent worsening group. While SpO₂ values were similar between baseline and mean recovery date (recovered group) and iso-time (persistent worsening group), parameters generally improved from baseline to the recovery/iso-time date, with a general trend of a larger magnitude of improvement in the recovered than in the persistent worsening group.

Table 2.

Validated symptom scores and physiological measurements at baseline (all participants), on the day of recovery (recovered group), and at iso-time (persistent worsening group).

	Baseline	Recovery	Iso-Time
Participants, No.	21	10	11
Date of baseline data & duration to recovery (and Iso-Time)
Baseline wristband study day	1	NA	NA
Baseline ring study day	2	NA	NA
Baseline EXACT-PRO score study day	1	NA	NA
Study day No., median (25th P, 75th P) and mean	1	4.5 (3.3, 7.8); 6	6
Symptom score
EXACT-PRO score, mean pts (SD):/100	53.3 (8.3)	45.4 (6.7)	52.3 (8)
Heart rate measurements
Average heart rate (wristband), beats/min (SD)	92.7 (11.6)	86.9 (10.3)	80.2 (5.7)
Average nighttime heart rate (ring), beats/min (SD)	79 (10.6)	76.4 (13.3)	73.1 (11.4)
Respiratory rate measurements
Average respiratory rate (wristband), breaths/min (SD)	23.7 (4.3)	20.3 (3.4)	20.7 (2.5)
Average nighttime respiratory rate (ring), breaths/min (SD)	18.1 (2.2)	16.5 (1.9)	17.3 (1.9)
Oxygen saturation measurements
Average nighttime oxygen saturation (ring), SpO2% (SD)	96.5 (0.6)	96.7 (1.1)	96.5 (0.8)
Activity
Daily step count (ring), median (25th P, 75th P): steps/day	1980 (1494, 3255)	3785 (2404.8, 6847.5)	2228 (1901, 3654.5)
Daily energy expenditure (ring), median (25th P, 75th P): kcal/day	1919 (1795, 2027)	2113.5 (1888, 2292.3)	1930 (1820.5, 2113)
Sleep quality
Total sleep (ring), hours/night (SD)	6.2 (1.6)	6.3 (1.1)	5.6 (2.0)
REM sleep (ring), hours/night (SD)	0.9 (0.7)	1.3 (0.7)	1.2 (0.9)

Abbreviations: beats/min, beats per minute; EXACT-PRO, Exacerbation of Chronic Pulmonary Disease Tool Patient-Reported Outcome; NA, not applicable; P, percentile; pts, points; %, percent; SD, standard deviation; SpO₂%, peripheral blood oxygen saturation percentage.

Fig. 2.

Daily EXACT-PRO score. Daily scores are presented in the recovered (solid-and-dashed line, n = 10) and persistent worsening (solid line, n = 11) groups. Dashed segment corresponds to period after which all recovered group participants recovered from their exacerbation (no persistent worsening group participants met this criteria). Grey surrounding each line denotes 95% CIs.

The daily EXACT-PRO score over time for each group is presented in Fig. 2, and the distinct physiological patterns (cardiorespiratory; activity/sleep; autonomic) for each group are presented in Fig. 3, Fig. 4, Fig. 5. Downshifted (y-axis) ring-obtained cardio-respiratory data illustrates the effect of measurement at rest (sleep/inactivity) during the nighttime period (Fig. 3). The recovered group, in contrast to the persistent worsening group, demonstrated progressive EXACT-PRO symptom score improvements, RR and HR trends toward normal ranges (Fig. 3), notable early increases in physical activity (Fig. 4a and b) and sustained sleep quality improvements (Fig. 4c and d), and variable autonomic (temperature and HR/RR variability) responses (Fig. 5). Interestingly, the multiparametric ‘objectively’ (passively) collected cardio-respiratory and sleep physiological trends (Fig. 3, Fig. 4, Fig. 5) closely matched the validated ‘subjective’ (actively) collected daily self-reported EXACT-PRO symptom score trends (Fig. 2) in both groups. While the persistent worsening group demonstrated fluctuant but generally persistently abnormal cardio-respiratory (high HR and RR) and sleep (low total sleep and sleep efficiency) patterns, the recovered group demonstrated gradual return towards normal ranges over time. Activity and autonomic values, in contrast, were more variable: daily step count improved substantially, quite early in the recovered group, and then trended low again, whereas temperature and cardio-respiratory variabilities (HRV/RRV) appeared to be more fluctuant/variable in the recovered group.

Fig. 3.

Respiratory and cardiac exacerbation patterns. Overall (wristband-obtained, solid lines) and nighttime (ring-obtained, dashed lines) respiratory rate (a) and heart rate (b) in the recovered group (light colours, n = 10) and the persistent worsening group (dark colours, n = 10). Grey surrounding each line denotes 95% CIs.

Fig. 4.

Activity-related and sleep-related exacerbation patterns. Daily step count (a), active calorie expenditure (b), nighttime total sleep duration (c) and sleep efficiency (d) (all ring-obtained, dashed lines) in the recovered group (light colours, n = 10) and the persistent worsening group (dark colours, n = 11). Grey surrounding each line denotes 95% CIs.

Fig. 5.

Autonomic exacerbation patterns. Nighttime respiratory rate variability (ring-obtained, dashed lines) (a), nighttime heart rate variability (ring-obtained, dashed lines) (b), and overall skin temperature (wristband-obtained, solid lines) (c), in the recovered group (light green) and the persistent worsening group (dark green). Grey surrounding each line denotes 95% CIs.

Linear mixed-effects univariable analyses in all participants (Table 3) did not reveal an association between daily EXACT-PRO and RR however did reveal a significant negative association with nighttime RRV. Significant activity-related negative associations included ring-obtained and wristband-obtained daily step counts, total and active daily calorie expenditure, daily rest, daily metabolic equivalents (METs), and movement intensity (acceleration). Finally, significant sleep-related parameter negative associations were observed with total sleep, sleep efficiency and deep sleep, and a negative trend with Rapid Eye Movement (REM) sleep that was not statistically significant.

Table 3.

Univariable linear mixed-effects regression associations with daily validated symptom scores in all participants, recovered group, and persistent worsening group.

	All Participants (n = 21)		Recovered (n = 10)		Persistent Worsening (n = 11)
	Point estimates [95% CIs]	P value	Point estimates [95% CIs]	P value	Point estimates [95% CIs]	P value
Respiratory
RR (wristband)	−0.16 [−0.40, 0.080] pts per breath/min	0.19	0.25 [−0.52, 1.02] pts per breath/min	0.52	−0.23 [−0.47, 0.013] pts per breath/min	0.064
Nighttime RR (ring)	0.22 [−0.37, 0.82] pts per breath/min	0.48	1.42 [−0.93, 3.73] pts per breath/min	0.23	−0.056 [−0.66, 0.56] pts per breath/min	0.86
Nighttime SpO2 (ring)	0.22 [−0.99, 1.39] pts per %	0.71	−7.11 [−10.79, −3.40] pts per %	0.0012∗	1.53 [0.074, 2.99] pts per %	0.044∗
Cardiac
HR (wristband)	0.0026 [−0.10, 0.10] pts per beat/min	0.96	0.23 [−0.049, 0.51] pts per beat/min	0.12	0.083 [−0.031, 0.19] pts per beat/min	0.15
Daily HR (ring)	−0.058 [−0.19, 0.071] pts per beat/min	0.38	0.32 [−0.020, 0.65] pts per beat/min	0.079	−0.0023 [−0.13, 0.13] pts per beat/min	0.97
Nighttime HR (ring)	0.00080 [−0.13, 0.13] pts per beat/min	0.99	0.46 [0.11, 0.82] pts per beat/min	0.018∗	−0.015 [−0.14, 0.11] pts per beat/min	0.82
Nighttime nadir HR (ring)	0.061 [−0.074, 0.20] pts per beat/min	0.37	0.43 [0.061, 0.81] pts per beat/min	0.032∗	−0.0050 [−0.14, 0.13] pts per beat/min	0.94
Autonomic
HRV (wristband)	−0.011 [−0.0063, 0.029] pts per beat/min	0.20	−0.037 [−0.090, 0.016] pts per beat/min	0.17	0.017 [0.0023, 0.032] pts per beat/min	0.026∗
HRV (ring)	0.00079 [−0.098, 0.098] pts per beat/min	0.99	−0.13 [−0.39, 0.14] pts per beat/min	0.33	−0.025 [−0.13, 0.076] pts per beat/min	0.63
Nighttime HRV (ring)	−0.0083 [−0.11, 0.090] pts per beat/min	0.87	−0.20 [−0.45, 0.052] pts per beat/min	0.13	−0.034 [−0.14, 0.070] pts per beat/min	0.53
Nighttime RRV (ring)	−1.45 [−2.84, −0.073] pts per breath/min	0.041∗	−3.34 [−8.14, 2.16] pts per breath/min	0.12	−1.45 [−2.79, −0.12] pts per breath/min	0.036∗
Skin temperature (wristband)	−0.071 [−0.88, 0.74] pts per degree Celsius	0.86	−1.18 [−4.75, 2.11] pts per degree Celsius	0.46	0.87 [0.18, 1.58] pts per degree Celsius	0.014∗
Electrodermal activity (wristband)	−0.22 [−0.52, 0.072] pts per μS	0.14	−0.78 [−0.42, 1.93] pts per μS	0.19	0.031 [−0.26, 0.32] pts per μS	0.83
Activity
Daily physical activity (converted to approximate distance) (ring)	−0.00056 [−0.00083, −0.00030] pts per m	<0.0001∗	−0.00069 [−0.0014, 0.000048] pts per m	0.070	−0.00021 [−0.00052, 0.00011] pts per m	0.20
Total daily calorie expenditure (ring)	−0.0072 [−0.011, −0.0034] pts per kcal	0.00024∗	−0.0080 [−0.019, 0.0026] pts per kcal	0.15	−0.0014 [−0.0054, 0.0027] pts per kcal	0.50
Rest time (ring)	−0.30 [−0.57, −0.037] pts per hour	0.026∗	0.0028 [−0.72, 0.72] pts per hour	0.99	−0.26 [−0.56, 0.042] pts per hour	0.094
Daily METs (ring)	−9.30 [−15.55, −3.08] pts per MET	0.0037∗	−13.45 [−31.20, 4.64] pts per MET	0.15	−2.65 [−9.35, 4.01] pts per MET	0.44
Active daily calorie expenditure (ring)	−0.012 [−0.017, −0.0063] pts per kcal	<0.0001∗	−0.014 [−0.029, 0.0015] pts per kcal	0.082	−0.0036 [−0.0097, 0.0024] pts per kcal	0.25
Daily time spent inactive (ring)	0.18 [−0.12, 0.48] pts per hour	0.25	0.27 [−0.87, 1.40] pts per hour	0.65	−0.11 [−0.17, 0.39] pts per hour	0.44
Daily step count (ring)	−0.56 [−0.82, −0.31] pts per 1000 steps	<0.0001∗	−0.67 [−1.38, 0.051] pts per 1000 steps	0.071	−0.18 [−0.47, 0.10] pts per 1000 steps	0.21
Daily step count (wristband)	−0.23 [−0.37, −0.086] pts per 1000 steps	0.0019∗	−0.46 [−0.95, 0.030] pts per 1000 steps	0.069	−0.020 [−0.16, 0.13] pts per 1000 steps	0.79
Movement intensity (acceleration) (wristband)	−0.069 [−0.120, −0.019] pts per ADC-U	0.0078∗	−0.21 [−0.37, −0.040] pts per ADC-U	0.018∗	−0.014 [−0.065, 0.036] pts per ADC-U	0.57
Activity counts (movement counts/min) (wristband)	−0.0086 [−0.059, 0.041] pts per a.U.	0.74	−0.12 [−0.27, 0.041] pts per a.U.	0.15	−0.019 [−0.030, 0.068] pts per a.U.	0.44
Sleep
Time in deep sleep (ring)	−2.15 [−3.58, −0.72] pts per hour	0.0035∗	−2.24 [−6.88, 2.40] pts per hour	0.37	−1.39 [−2.81, −0.021] pts per hour	0.049∗
Sleep disturbances (ring)	−0.030 [−0.082, 0.022] pts per disturbance point	0.25	−0.066 [−0.26, 0.14] pts per disturbance point	0.52	−0.011 [−0.059, 0.036] pts per disturbance point	0.66
Sleep efficiency (ring)	−0.12 [−0.20, −0.037] pts per % of sleep period spent asleep	0.0044∗	−0.12 [−0.41, 0.17] pts per % of sleep period spent asleep	0.42	−0.10 [−0.18, −0.028] pts per % of sleep period spent asleep	0.0074∗
Sleep onset latency (ring)	−0.0022 [−0.059, 0.054] pts per min	0.94	−0.0027 [−0.061, 0.052] pts per min	0.74	−0.037 [−0.24, 0.17] pts per min	0.27
Time in REM sleep (ring)	−1.14 [−2.29, 0.020] pts per hour	0.054	−0.41 [−5.30, 4.79] pts per hour	0.87	−0.18 [−1.38, 1.01] pts per hour	0.76
Awake time in sleep period (ring)	−0.37 [−0.35, 1.09] pts per hour	0.32	1.67 [−0.67, 3.97] pts per hour	0.17	−0.0064 [−0.79, 0.82] pts per hour	0.99
Total sleep period (ring)	−0.27 [−0.62, 0.078] pts per hour	0.13	1.22 [−0.061, 2.57] pts per hour	0.072∗	−0.20 [−0.54, 0.14] pts per hour	0.25
Restless sleep (ring)	−0.055 [−0.033, 0.14] pts per % of sleep time spent moving	0.22	−0.24 [−0.51, 0.48] pts per % of sleep time spent moving	0.92	0.088 [0.021, 0.16] pts per % of sleep time spent moving	0.011∗
Sleep duration (ring)	−0.61 [−1.06, −0.16] pts per hour	0.0089∗	1.46 [−0.43, 3.53] pts per hour	0.14	−0.34 [−0.78, 0.10] pts per hour	0.13

Abbreviations: ADC, Analog-to-Digital Converter; A.U., arbitrary units; beat/min, beat per minute; CIs, confidence intervals; HR, heart rate; HRV, heart rate variability; kcal, kilocalories; %, percent; MET, metabolic equivalent; m, meter; min, minute; pts, points; REM, rapid eye movement; RR, respiratory rate; RRV, respiratory rate variability; s, second; SpO₂, peripheral blood oxygen saturation percentage; μS, microSiemens.

∗, p < 0.05. Bold font signifies a P value of <0.05.

In recovered participants (Table 3), a significant positive association between daily EXACT-PRO score and nighttime HR and nighttime nadir HR, and a non-significant positive trend with average HR, were observed. Furthermore, significant negative associations with nightly SpO₂, and with movement intensity (acceleration), were observed. Non-significant negative trends with several other activity-related parameters included active daily calorie expenditure and ring-obtained and wristband-obtained daily step count. No associations were observed with sleep-related parameters.

In participants with persistent worsening (Table 3), significant associations between daily EXACT-PRO score with nightly SpO₂, with HRV, and with nighttime RRV were observed. There was a non-significant negative trend in the association with RR that was statistically significant in the multivariable model following adjustment for age, sex and FEV₁ (P = 0.044 [linear mixed-effects regression], see Supplement). Significant associations with sleep-related parameters included negative associations with sleep efficiency and deep sleep, as well as a positive association with restless sleep. Finally, uniquely in the persistent worsening group, a significant positive association was observed with skin temperature.

Multivariable model results and sensitivity analyses of respiratory, cardiac, autonomic, activity-related, and sleep-related parameters are presented in the Supplement. Overall, adjustment for age, sex and FEV₁ yielded similar associations when compared with univariable model results, and likewise repeating analyses with inactive data only yielded similar associations to models with all data included.

Discussion

This study has characterised the respiratory, cardiac, autonomic, activity, and sleep profiles of exacerbations of COPD in free-living outpatients with mostly severe/very severe COPD, as well as their associations with daily symptom burden and recovery status. A detailed, continuous, multi-week, multiparameter description of the pathophysiological profile of exacerbations of COPD in the natural home setting is presented. In analyses including all participants, the most consistent associations with observed reductions (improvements) in the validated and sensitive daily patient-reported EXACT-PRO symptom score were with increased activity level and improved sleep quality metrics. In contrast to our hypothesis, there was a significant association only with the autonomic component of respiration (RRV) rather than with respiratory rate. In participants who ultimately recovered from their exacerbation, the most consistent associations were observed with improved cardiac parameters, increased activity level, and increased SpO₂, whereas in participants with persistent worsening during the study period, the most notable associations were observed with respiratory parameters, sleep quality, and interestingly, autonomic parameters. The overall pattern of associations was consistent across univariable, multivariable, and sensitivity analyses.

With the emerging shift towards more ‘objective’ exacerbation physiological classification criteria⁷ adopted in the latest international COPD reports,¹ a better understanding of which pathophysiological derangements (and by extension, which physiological ‘biomarkers’) are most closely associated with evolving symptom burden and ultimate event trajectory/outcome carries considerable clinical significance. Distinct patterns were observed across all participants versus in those who ultimately did, and did not, recover during the follow-up period, providing group-specific patterns. Foundational RPM work in patients with COPD demonstrated the proof-of-concept of remotely capturing cardio-respiratory alterations (specifically, in RR) associated with exacerbations, using a single parameter and fixed/tethered apparatus.^12,13 Building on this essential knowledge, subsequent work adopted the use of biometric wearable devices and reported on either one physiological variable¹⁴ or on data collected during one visit^26,27 or during the daytime.^6,14,28,29 In order to capture phenomena at rest and/or during sleep,⁶ which may also represent an opportunity to obtain high-quality data with lower likelihood of motion artefact, activity influences, and other confounders, data during both daytime and nighttime over several weeks was collected in the present study.

In all participants, the notable associations with daily EXACT-PRO scores were principally activity-related and sleep-related. Exacerbations are systemic, pro-inflammatory phenomena that are not limited to the lungs.⁷ Physical activity is known to be decreased in the exacerbation period given multi-system compromise and substantial symptom burden including acute-on-chronic breathlessness. The capacity and willingness of patients experiencing an exacerbation to perform even minimal daily activity become compromised.³⁰ These factors also appear to substantially affect sleep quality,³¹ supporting the present findings and providing a rationale for the relevance of clinical sleep measurement/monitoring in addition to physical activity monitoring in the exacerbation period.

In participants who ultimately recovered from their exacerbation, the notable associations with daily EXACT-PRO scores were principally cardiac and activity-related, whilst an association with overnight SpO₂ was also observed. Exacerbation-related tachycardia is multifactorial and reflects responses to the multi-systemic inflammatory burst, stress hormone release, increased work of breathing from narrowed airways, variable hypoxaemia/hypercapnia, as well as the potential use and/or over-use of short-acting rescue beta-agonist bronchodilator medication.³² In one large RPM study using a Bluetooth-enabled pulse oximeter (worn ∼100 sec. at a time), symptom diary, and sophisticated back-end processing, SpO₂ exhibited strong associations with exacerbation symptoms.²⁹ These associations align with the Rome Proposal given that, in addition to RR, HR and SpO₂ are emphasised as the principal physiological parameters to classify exacerbations.⁷

In participants who did not recover from their exacerbation within the observation period, notable associations with daily EXACT-PRO scores were principally respiratory (including RR and overnight SpO₂), sleep-related, and interestingly, autonomic (RRV, HRV and skin temperature). RR has long been considered a key biomarker for exacerbation monitoring³³ given the intrinsic link to the pathophysiology of exacerbations and reflects the RR-centric approach pursued to date in the early exacerbation detection literature.^12,13 Exacerbation-related tachypnoea reflects compensatory responses to increased airway resistance, expiratory flow limitation, hyperinflation and ventilation-perfusion mismatching, which gradually return towards baseline in a detectable manner.¹⁴ Regarding the autonomic associations observed, COPD is characterised by increased sympathetic tone even at rest (sympathetic-parasympathetic disequilibrium), contributing to decreased sensitivity in autonomic variations.³⁴ The interesting autonomic associations, most consistently observed in the persistent worsening group, may reflect continued sympathetic disruption. If these findings are replicated in prospective studies that investigate the prodromal period of exacerbations, additional candidate biomarkers not currently included in the Rome Proposal may be considered.

The present study has important limitations. This study was conducted at a single site, an academic tertiary centre, and baseline and follow-up data of non-consenting potentially eligible patients were not systematically collected. Both of these limitations may introduce bias and affect generalisability. The precipitating exacerbation trigger, cause, or phenotype was not investigated. Daily short-acting beta-agonist medication use was not systematically recorded during the outpatient observation period. Despite the conservative 21-day observation period, chosen to maximise the likelihood of capturing even prolonged exacerbations,⁴ roughly half of included participants still did not actually demonstrate ‘recovery’ as defined by the validated symptom/recovery score. Because of the previously well-described risk of a second episode shortly following an initial exacerbation,⁵ the 21-day observation period may have either been too long such that exacerbation ‘clusters’ were captured (i.e. two consecutive episodes), not long enough to have captured a protracted recovery of the first/sentinel episode, or even representative of a very symptomatic (advanced disease) baseline. To account for sizeable trajectory heterogeneity,⁴ in addition to analyses in all participants, ‘recovered’ and ‘persistent worsening’ groups were identified and analysed separately, leading to the discovery of unique and previously largely unreported group-specific patterns. These findings, however, may be limited by the small sample sizes of the two groups. Study tablet use (data transfer/questionnaire completions) required reasonable dexterity, eyesight and literacy, which may have unduly contributed to selection bias. Missingness of data, particularly for wristband-obtained RR and HRV, may have introduced bias in parameter daily average estimates. Finally, there are important confounders to consider which include supplemental oxygen use and activity level during parameter measurement, as well as age, sex and baseline lung function, all of which can affect the associations under investigation. No adjustment was made for domiciliary/supplemental oxygen use, potentially influencing all SpO₂-EXACT-PRO associations estimated. Activity level and age, sex and baseline lung function were adjusted for in sensitivity analyses and multivariable analyses, respectively, however these may have also been limited by sample size.

In conclusion, this study reports the continuous multi-parameter characterisation of exacerbations among free-living outpatients with COPD with mostly severe or very severe disease and highlights significant associations between daily symptom scores obtained from a validated questionnaire and physiological biomarkers collected from two wearable devices, providing insight into exacerbation pathophysiology and helping to address persisting knowledge gaps between pilot/feasibility studies and future real-world implementation of COPD remote monitoring/detection platforms. This active field of research has the potential to harness advances in hardware, software, and ‘big data’ to revolutionise COPD management through personalised diagnostic monitoring and more timely, appropriate intervention.

Contributors

BAR conceptualised and designed the methodology of the study, obtained funding, and drafted the study protocol. AH, DJ, SP, and JX reviewed the study protocol. FAC and OCI managed the project and data collection. BAR, FAC and OCI curated and analysed the data and wrote the original draft of the manuscript. BAR, FAC, OCI and SN have directly accessed and verified the underlying data reported in the manuscript. All authors read and approved the final version of the manuscript. FAC and BAR were responsible for the decision to submit the manuscript.

Data sharing statement

Due to data privacy regulations, individual participant data collected during this study is not publicly accessible. However, access to anonymised data may be granted upon evaluation by the trial management group. Additional documents will also be available upon inquiry. All requests should be directed to the corresponding author (BAR) at bryan.ross@mcgill.ca.

Declaration of interests

FAC, OCI, and SN declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. AH reports the following conflicts of interest: Co-Founder & CEO, SensifAI Health Inc./SensifAI Sante Inc, Montreal, Quebec, Canada; Inflammatory response indicator (2023). US Patent, Pending. DJ reports the following conflicts of interest: Co-Founder & Chief Scientific Officer, SensifAI Health Inc./SensifAI Sante Inc, Montreal, Quebec, Canada; Inflammatory response indicator (2023). US Patent, Pending. SP reports the following conflicts of interest: Chair of Planning Committee, American Thoracic Society, Assembly of Sleep, Respiratory and Neurobiology; grants from the Canadian Institutes of Health Research (CIHR), Fonds de Recherche du Québec and Chest Foundation Grant; Sleep and Breathing Conference in New Zealand 2023– travel paid for (flight and hotel). JX declares that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. BAR reports the following conflicts of interest: Honoraria from the Canadian Thoracic Society (CTS–COPD Educational Event (Speaker) and Content Creator (educational materials)), CHEST (Content Creator (educational materials)), Respiplus non-profit (Content Creator (educational materials)), Alberta Kinesiology Association (AKA–Content Creator (educational materials)), Association des Pneumologues de la Province du Québec (APPQ–presenter), McGill University Continuing Professional Development (CPD–COPD Educational Event (Speaker)), GSK (Speaker and Moderator), AZ (Speaker and Moderator), and COVIS (COPD Educational Event (Speaker)); Research funding as Principal Investigator from the McGill University Health Centre (MUHC) Department of Medicine Contract Academic Staff (CAS) Research Award, the Quebec Respiratory Health Network (QHRN), the Ministère de l’Éducation et le Ministère de l’Enseignement Supérieur Innovation and Partnership Program (McGill University and Thorasys, Inc.), and the MUHC Foundation/MCI Respiratory Research Campaign Innovation Grant; and reception of in-kind support (placebo and intervention) for research from Amazentis, and reception of in-kind support (diagnostic device(s)) for research from Thorasys Inc., Spire Health, and Restech.