24-hour accelerometry in COPD: Exploring physical activity, sedentary behavior, sleep and clinical characteristics

Abstract

Background

The constructs and interdependency of physical behaviors are not well described and the complexity of physical activity (PA) data analysis remains unexplored in COPD. This study examined the interrelationships of 24-hour physical behaviors and investigated their associations with participant characteristics for individuals with mild–moderate airflow obstruction and healthy control subjects.

Patients and methods

Vigorous PA (VPA), moderate-to-vigorous PA (MVPA), light PA (LPA), stationary time (ST), average movement intensity (vector magnitude counts per minute), and sleep duration for 109 individuals with COPD and 135 healthy controls were obtained by wrist-worn accelerometry. Principal components analysis (PCA) examined interrelationships of physical behaviors to identify distinct behavioral constructs. Using the PCA component loadings, linear regressions examined associations with participant (+, positive correlation; -, negative correlation), and were compared between COPD and healthy control groups.

Results

For both groups PCA revealed ST, LPA, and average movement intensity as distinct behavioral constructs to MVPA and VPA, labeled “low-intensity movement” and “high-intensity movement,” respectively. Sleep was also found to be its own distinct behavioral construct. Results from linear regressions supported the identification of distinct behavioral constructs from PCA. In COPD, low-intensity movement was associated with limitations with mobility (−), daily activities (−), health status (+), and body mass index (BMI) (−) independent of high-intensity movement and sleep. High-intensity movement was associated with age (−) and self-care limitations (−) independent of low-intensity movement and sleep. Sleep was associated with gender (0= female, 1= male; [−]), lung function (−), and percentage body fat (+) independent of low-intensity and high-intensity movement.

Conclusion

Distinct behavioral constructs comprising the 24-hour day were identified as “low-intensity movement,” “high-intensity movement,” and “sleep” with each construct independently associated with different participant characteristics. Future research should determine whether modifying these behaviors improves health outcomes in COPD.

Keywords:

• accelerometry
• COPD
• physical activity
• principal components analysis
• sedentary behavior

Supplementary materials

Reliability testing

Accelerometers are commonplace in the field of PA research, in part due to their small size, light weight, ability to measure human movement (acceleration), and store data over many days.¹ Despite the capability of these devices to quantify acceleration with high sensitivity (eg, acceleration data can be recorded 100 times every second; 100 Hz) it is a good practice to check that accelerometers are working within an acceptable measurement error before deployment. This is particularly important when large number of devices are being deployed in a single study due to the increased likelihood of inter-device variability. There are plethora of examples of studies examining the validity of accelerometers in both controlled and free-living conditions using human participants.^2,3,11,12 However, variations in the participants themselves even when a single person wears multiple devices,³ introduce inherent variability in the assessment of monitor accuracy. An alternative approach for examining the accuracy of accelerometers has been through the use of mechanical shakers.^4,5 The advantages of using shakers include the large number of accelerations that can be produced, the ability to assess many accelerometers at once, and the reliable and precise oscillations that can be produced.¹ The importance of limiting inter-device variation and using devices with acceptable measurement errors is pivotal for accurate and reliable behavior quantification as greater magnitudes of acceleration occur at the wrist compared to locations closer to the center of mass.⁶

Mechanical shaking and inter-device variability

One hundred and fifty-five ActiGraph wGT3x-BT accelerometers (ActiGraph, LLC) were assessed using an orbital shaker table to provide the researcher full control of the magnitude of the acceleration and the frequency of the oscillation the devices were exposed to. Five different conditions were selected to produce a range of physiologically relevant accelerometer counts within the confines of the shaker capacity; these were 100, 125, 150, 175, and 200 revolutions per minute. Each condition was time-stamped and lasted 2.5 minutes with 1.25 minutes between each condition to allow time for the orbital shaker to adjust the oscillation frequency. Care was taken to secure the monitors, being firmly fixed in a vertical position along their sensitive axis in order to maximize and standardize the output. Once all accelerometers were in position the orbital shaker was switched on and allowed to warm up in order to facilitate the optimal execution of the conditions.

In order to identify accelerometers working outside acceptable limits ie, ±10% as per manufacturer guidelines, the mean difference percent (Equation S1$\begin{array}{l}\text{Mean difference percent}=\\ \frac{\text{Unit specific mean}-\text{Condition grand mean}}{\text{Condition grand mean}}\times 100\end{array}$(S1) ) was calculated for each unit and visualized using Bland–Altman plots for each condition. Units which exceeded this tolerance were deemed “out of calibration” and returned to the manufacturer. Twelve accelerometers (7.7%) were returned to the manufacturer and 113 devices (72.9%) were used as part of the study.

$\begin{array}{l}\text{Mean difference percent}=\\ \frac{\text{Unit specific mean}-\text{Condition grand mean}}{\text{Condition grand mean}}\times 100\end{array}$(S1)

Data processing

Wrist-worn accelerometry is in its infancy within the field of PA and sedentary behavior measurement, but there is general consensus and initial evidence to suggest that this location will permit improvements in wear time compliance;⁷ an advantage for capturing data representative of the wearer’s usual activities both within and between days. The main reason for this is likely the added comfort for the participant, which enables them to wear the device without disturbing sleep. As a result, participants were only asked to take off the monitor for water-based activities such as showering. However, with this comes the challenge of differentiating time in waking and non-waking behaviors. Traditional approaches have utilized participant diaries, whereby individuals record the time they went to bed and the time they got up each day, but this is plagued by recall inaccuracies and adds to the burden of study participation. Data-driven approaches are needed to objectively identify sleep onset and end without the additional burden to participants.

Location and device setup

Objectively derived PA and sedentary time were collected using the ActiGraph wGT3X-BT accelerometer worn on the nondominant wrist (non-writing hand) continuously except for water-based activities at a sample rate of 100 Hz. Monitors were deployed in delay mode on day 0 and commenced logging on day 1 at 00:00 with a 7-day stop time indicated. Each accelerometer was returned via mail after seven full days of wear. Monitors were initialized and downloaded using ActiLife software (ActiGraph, LLC) version 6.13.2 and were analyzed using KineSoft (KineSoft, Loughborough, UK) version 3.3.80. Data were processed in 60-second epochs.

Preprocessing accelerometry analysis

Sixty-second, agd files were processed through KineSoft using Choi wear-time criteria^8,9 to identify periods of non-wear. Individual files were exported in “processed mode” using the File Inspector function in KineSoft. The processed data (ie, with non-wear coded) were then inserted into an automated sleep detection system.

Identifying time in bed and out of bed

Sleep detection was determined using sustained periods of (in)activity from vector magnitude count values based on the work of Carney et al.¹⁰ To identify the time when participants went to sleep (INBED), the algorithm identified consecutive dips in activity, specifically a 90% reduction from the previous epoch for 15 minutes between the hours of 21:00 and 23:59.¹⁰ Once the INBED criteria were met, the original epoch containing the 90% reduction in counts was used to signify the start of sleep. To identify the time when participants were awake (OUTBED), the algorithm detected consecutive rises in activity level of at least 75% from the previous epoch for 5 minutes between 06:00 and 09:00.¹⁰ Once the OUTBED criteria were met, the original epoch containing the 75% increase in counts was used to signify the end of sleep.

Window identification

In order to facilitate the aforementioned algorithm, a sub-sample of 80 files (comprising 20 apparently healthy males, 20 apparently healthy females, 20 male COPD patients, and 20 female COPD patients) was used to visually inspect the suitability of using the 06:00–09:00 and 21:00–23:59 windows as part of the sleep detection verification process. Minute-by-minute vector magnitude was plotted for each of the 7 days of wear and subjected to visual inspection for spikes in activity between 06:00 and 09:00 and dips in activity between 21:00 and 00:00. Whilst these patterns were consistently observed, for the 06:00–09:00 window it was noticed that the activity was still relatively high before this window for some individuals, therefore, additional criteria were included to identify OUTBED occurrences prior to 06:00 and after 09:00. Similarly, between 21:00 and 00:00 it was noticed that the activity was still relatively high after this window for some individuals, therefore, additional criteria were included to identify INBED occurrences after 00:00. Consequently, additional criteria were put in place to account for variation in sleep/wake cycles between participants.

Postprocessing data checking

If no INBED occurrence from 21:00 to 23:59 was identified, the algorithm used a default time stamp of 23:59 and the file was flagged for visual inspection to determine the exact INBED occurrence. For INBED time stamps that complied with the 21:00–23:59 window, visual inspection was required if additional time stamps were present. The default time stamp (first occurrence) was altered based on visual inspection if subsequent spikes in activity lasted at least 2 minutes at light or moderate intensity, or 5 minutes if sedentary intensity was present. For all OUTBED time stamps, an automated time-stamped detection of sustained spikes in vector magnitude was conducted. Files were flagged for visual inspection if a spike in activity occurred within 1 hour of the algorithm-derived time stamp.

Accelerometry algorithm alterations

Of the 436 total accelerometry files, 435 (99.8%) files were visually inspected for at least 1 day for either INBED or OUTBED classification. The whole sample of 436 files provided a total number of 3,052 potential days of wear for the PhARaoH participants. Of these, 2,437 (79.8%) required visual inspection for INBED detection of which 1,515 (62.2%) required an alteration to the algorithm time stamp (originally windowed between 21:00 and 23:59). For OUTBED detection, 694 (22.7%) days required visual inspection of which 598 (86.0%) required an alteration to the algorithm time stamp (originally windowed after 06:00).

For the 1,515 INBED detections requiring alterations from the original algorithm-derived time stamps, 596 (39.3%) were due to participants going to sleep after midnight, 870 (57.4%) were from adjustments made to the first time stamp after 21:00, and 49 (3.2%) were from visual inspection alone. Of the remaining 922 days the algorithm was not altered with 203 (22.0%) because periods of non-wear were detected, 63 (6.8%) were for day 7 defaulting to 23:59, and 10 (1.1%) were not altered following visual inspection.

For the 598 OUTBED detections requiring amendment from the original algorithm output, 540 (90.3%) were due to participants waking up before 06:00 and 58 (9.7%) were from visual inspection alone. Of the remaining 96 days the algorithm was not altered with 57 (59.4%) due to periods of non-wear and 39 (40.6%) were not altered following visual inspection.

Accelerometry processing

After establishing the INBED and OUTBED times for each day, sleep was coded as 0 counts (equivalent to non-wear) in order to be removed by non-wear algorithm during data processing. As a result, time spent in activity intensities was derived from waking wear time. Time spent sleeping was calculated from the time between INBED and OUTBED occurrences. VMCPM was calculated by dividing average total counts per day by average waking wear time. PA intensities were defined according to published cut points for sedentary time, light intensity activity, and MVPA. ST was defined as <2,000 VMCPM, LPA as 2,000–7,499 VMCPM, MVPA as ≥7,500 VMCPM, and VPA as ≥8,250 VMCPM.⁶

Figure S1 Scatterplot of component scores for “low-intensity movement” and “high-intensity movement.”

Notes: Solid circles: COPD; blank circles: controls.

Abbreviation: PCA, principal components analysis.

References

• RothneyMPApkerGASongYChenKYComparing the performance of three generations of ActiGraph accelerometers. J Appl Physiol(1985)2008105410911097
   PubMed Web of Science ®Google Scholar
• WelkGJPrinciples of design and analyses for the calibration of accelerometry-based activity monitorsMed Sci Sports Exerc20053711 SupplS501S51116294113
   PubMed Web of Science ®Google Scholar
• NicholsJFMorganCGChabotLESallisJFCalfasKJAssessment of physical activity with the Computer Science and Applications, Inc., accelerometer: laboratory versus field validationRes Q Exerc Sport2000711364310763519
   PubMed Web of Science ®Google Scholar
• MetcalfBSCurnowJSEvansCVossLDWilkinTJTechnical reliability of the CSA activity monitor: The Early Bird StudyMed Sci Sports Exerc20023491533153712218751
   PubMed Web of Science ®Google Scholar
• EsligerDWTremblayMSTechnical reliability assessment of three accelerometer models in a mechanical setupMed Sci Sports Exerc200638122173218117146326
   PubMed Web of Science ®Google Scholar
• KamadaMShiromaEJHarrisTBLeeIComparison of physical activity assessed using hip- and waist-worn accelerometersGait Posture201644232827004628
   PubMed Web of Science ®Google Scholar
• van HeesVTRenströmFWrightAEstimation of daily energy expenditure in pregnant and non-pregnant women using a wrist-worn tri-axial accelerometerPLoS One201167e2292221829556
   PubMed Web of Science ®Google Scholar
• ChoiLLiuZMatthewsCEBuchowskiMSValidation of accelerometer wear and nonwear time classification algorithmMed Sci Sports Exerc20114335736420581716
   PubMed Web of Science ®Google Scholar
• ChoiLWardSCSchnelleJFBuchowskiMSAssessment of wear/nonwear time classification algorithms for triaxial accelerometerMed Sci Sports Exerc2012442009201622525772
   PubMed Web of Science ®Google Scholar
• CarneyCELajosLEWatersWFWrist ActiGraph versus self-report in normal sleepers: sleep schedule adherence and self-report validityBehav Sleep Med20042313414315600229
   PubMedGoogle Scholar
• EsligerDWRowlandsAVHurstTLCattMMurrayPEstonRGValidation of the GENEA AccelerometerMed Sci Sports Exerc20114361085109321088628
   PubMed Web of Science ®Google Scholar
• HendelmanDMillerKBaggettCDeboldEFreedsonPValidity of accelerometry for the assessment of moderate intensity physical activity in the fieldMed Sci Sports Exerc2000329 SupplS442S44910993413
   PubMed Web of Science ®Google Scholar

Acknowledgments

The study was funded by facilitation funds from the NHS England. The authors acknowledge support from the National Institute for Health Research (NIHR) Leicester Biomedical Research Center, which is a partnership between University Hospitals of Leicester NHS Trust, Loughborough University, and the University of Leicester and acknowledge support from the NIHR Collaboration for Leadership in Applied Health Research and Care – East Midlands and The Primary Care Research Network (PCRN). The views expressed are those of the authors and not necessarily those of the NHS, the NIHR, Loughborough University, or the Department of Health.

Author contributions

MWO had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. MWO, MCS, MDM, DWE, APK, LBS, and SJS contributed substantially to the study design, data analysis and interpretation; took part in drafting the article or revising it critically for important intellectual content; gave final approval of the version to be published; and agree to be accountable for all aspects of the work.

Disclosure

The authors report no conflicts of interest in this work.