Volume 45 Number 4, 2008
Pages 611 — 618
Accuracy of uniaxial accelerometer in chronic obstructive pulmonary disease
Marilyn L. Moy, MD, MSc;1-4* Eric Garshick, MD, MOH;2,4-5 Kirby R. Matthess, BS;2,4 Robert Lew, PhD;6 John J. Reilly, MD3-4
Abstract — Wearable sensors have been developed and used in nondisabled adults. Little is known about their accuracy in patients with chronic disease. We tested the ActiHealth accelerometer, which measures step counts, in subjects with chronic obstructive pulmonary disease (COPD). We determined the intra- and interdevice coefficients of variation (CVs). We assessed the accuracy of the device in 15 nondisabled males and 46 subjects with COPD. Accuracy was defined as percent step capture, (device step count divided by manual step count) times 100. Predictors of percent step capture were identified using linear regression methods. The accelerometer has an intradevice CV ranging from 0.008 to 0.025 and an interdevice CV of 0.64. In nondisabled males, median percent step capture was 96% (interquartile range 81%-98%). In subjects with COPD, median percent step capture was 86% (interquartile range 72%-96%). Usual walking speed was the most important predictor of percent step capture (p = 0.004). The ActiHealth accelerometer has acceptable intra- and interdevice CVs. It is highly accurate in nondisabled subjects. The accuracy declines in subjects with COPD based on walking speed. Prior to using the ActiHealth accelerometer, researchers and clinicians should assess walking speed and percent step capture in each subject.
Key words: accelerometer, ambulation, chronic obstructive pulmonary disease, COPD, pedometer, physical activity, rehabilitation, step count, walking speed, wearable sensor.
Email: marilyn.moy@va.gov
Assessing ambulation and physical activity is important in patients with chronic obstructive pulmonary disease (COPD). In this patient population, decreased physical activity has been associated with acute exacerbations [1], hospitalizations, and greater mortality [2-3]. Moderate to high levels of regular physical activity have been shown to be associated with a reduction in smoking-related lung function decline and COPD risk among smokers [4]. The ability to extend monitoring of ambulation and physical activity from the supervised clinic setting to the free-living home environment may promote novel applications that allow maintenance of exercise benefits following a supervised pulmonary rehabilitation program, early detection of COPD exacerbations, and assessment of response to therapeutic interventions in COPD clinical trials [5-7].
The use of body-worn sensors to measure physical activity, such as walking, in free-living individuals in nondisabled and disease states has gained popularity. Wearable sensors have been developed, tested, and used in the nondisabled population. However, little is known about their accuracy in patients with chronic disease. Pedometers, or step counters, are typically worn on the belt or waistband and detect vertical acceleration of the hip during gait cycles. Although simple to use and inexpensive, pedometers are limited by their nonspecificity and inability to store long-term data, are affected by the amount of soft tissue at the waist, and require subjects to remember to put them on each time they are active [8-9]. Multiaxial accelerometers have been used to measure physical activity in addition to step counts. However, their output is highly variable and difficult to put into clinical context. Interpretation depends on calibration equations and cutoff points for defining light, moderate, vigorous, and total activity [10-15]. To date, these cutoff points have been defined for nondisabled, young adults in the supervised setting with correlations to maximum oxygen consumption on cardiopulmonary exercise tests, not during free-living conditions in the population being studied.
The ActiHealth uniaxial accelerometer (FitSense Technology, Inc; Southborough, Massachusetts) (Figure 1) overcomes many of the existing limitations of pedometers and accelerometers. The ActiHealth accelerometer is a lightweight and unobtrusive device that attaches to the shoe and continuously records step counts when a person walks. A sensor is used to capture the physiological waveforms associated with a foot stride [16]. Up to 30 days of step count data from the device can be wirelessly and securely transferred via the Internet to a database for storage and retrieval. The single coin-cell battery lasts for 1 to 2 years depending on use. The device remains secured to the shoe so subjects do not have to attach the device each time they walk. Subjects do not have to push buttons, recharge batteries, or record numbers.
In this study, we determined the intra- and interdevice coefficients of variation (CVs) of the ActiHealth accelerometer. We determined the accuracy of the accelerometer in nondisabled subjects and in subjects with COPD and identified the clinical variables that predict device performance.
To determine the intra- and interdevice CVs of the ActiHealth accelerometer, two of the investigators (MLM and KRM) both completed 5 walking studies (see "Accelerometer and Walking Studies" section, p. 613 for description) with each of nine ActiHealth accelerometers (total of 90 trials) over 2 days. Fifteen nondisabled males who had no known pulmonary disease performed a single walking study wearing an accelerometer on each foot to determine the accuracy of the device.
Forty-six men with COPD were enrolled from a general pulmonary clinic, and each performed a single walking study wearing an accelerometer on each foot. COPD was defined as having a force expiratory volume in 1 second (FEV1)/forced vital capacity <0.70 and a smoking history of >10 pack-years or computed tomography evidence of emphysema. Subjects who wore slide-on shoes, could not walk, or required assistance for ambulation (cane, walker, or wheelchair) were excluded. Medical records were reviewed for diagnoses that could affect the lower limbs and ambulation. Diagnoses of osteoarthritis or degenerative joint disease, hip or knee replacements, rheumatoid arthritis, chronic low back pain, lumbar spine disease, peripheral vascular disease, or peripheral neuropathy were noted. The medical record was also reviewed for the most recent spirometry performed. Evidence of a COPD exacerbation, defined as subjects using antibiotics or oral corticosteroid therapy, at the time of the study was obtained by medical record review.
The protocol was approved by the Department of Veterans Affairs (VA) Boston Healthcare System Committee on Human Research, and informed consent was obtained from each subject.
During each walking study, participants walked a predetermined level course of 800 feet and were instructed to walk at their usual speed. Two accelerometers were chosen from a set of nine, and each subject wore one on the right and left foot. The accelerometer was mounted on the top of the shoe and secured with the laces or Velcro® straps. Every step taken by the right foot was manually counted by one of the investigators (MLM or KRM) and multiplied by 2 for the total manual step count. The time to walk the course was used to calculate usual walking speed. Supplemental oxygen, if prescribed for ambulation, and rests were permitted in subjects with COPD.
Accuracy was defined as the percent step capture, (accelerometer step counts/manual step counts) × 100. We calculated the intra- and interdevice CVs for the nine devices used in the 90 trials by MLM and KRM. For the 46 subjects with COPD, correlations between percent step capture and the continuous variables were assessed by Spearman correlation coefficients.
Linear regression methods (PROC MIXED, SAS, version 9.2, SAS Institute; Cary, North Carolina) identified the significant predictors of percent step capture [17]. Variables significant at the 0.10 level were assessed in multivariate models. The 92 trials (46 walking studies × 2 feet, each with a different accelerometer) in the 46 subjects with COPD were analyzed with SAS PROC MIXED, which adjusted for the correlated data from each subject wearing a different accelerometer on each foot during a single walking study.
In the 90 trials completed by MLM and KRM (9 devices × 5 trials × 2 investigators), the accuracy was 98 percent. The accelerometer had an intradevice CV that ranged from 0.008 to 0.025 and an interdevice CV of 0.64. The average speeds for walking trials performed by MLM and KRM were 3.1 ± 0.07 mph and 3.2 ± 0.09 mph, respectively (all data presented as mean ± standard deviation unless otherwise noted). For the 15 nondisabled male subjects (mean age 56 ± 12 years), the average walking speed was 2.8 ± 0.42 mph. The median percent step capture or accuracy was 96 percent (interquartile range 81%-98%).
The 46 men (mean age 71 ± 9 years) with predominantly moderate to severe COPD (Global Initiative for Chronic Obstructive Lung Disease stages II and III [18]) had a mean FEV1 of 1.73 ± 0.57 L (53% ± 18% of predicted values) (Table 1). The median percent step capture or accuracy was 86 percent (interquartile range 72%-96%). Their average walking speed was 2.2 ± 0.39 mph. Eighteen subjects (39%) had a concurrent medical diagnosis of osteoarthritis or degenerative joint disease, hip or knee replacements, rheumatoid arthritis, chronic low back pain, lumbar spine disease, peripheral vascular disease, or peripheral neuropathy. Three subjects were experiencing a COPD exacerbation at the time of the walking study, six subjects used supplemental oxygen, and four subjects briefly stopped to rest. For all 92 trials in the COPD subjects, the accelerometer undercounted the steps (mean accelerometer step count of 310 ± 100 vs mean manual step count of 411 ± 78) (Table 1). We defined acceptable accuracy as percent step capture of ≥90 percent. Plot of usual walking speed by percent step capture revealed that at speeds ≥2.2 mph, the average walking speed for the COPD group, subjects were more likely to have ≥90 percent step capture (Figure 2). For speeds <2.2 mph, 8 of the 42 trials (19%) had ≥90 percent step capture, and for speeds ≥2.2 mph, 33 of the 50 trials (66%) had ≥90 percent step capture.
In univariate linear regression models, speed and lower-limb problems were significant predictors of percent step capture (Table 2). Speed was poorly correlated with FEV1 percent predicted (Spearman r = 0.10). Neither FEV1 percent predicted, COPD exacerbation, nor the person counting steps (MLM or KRM) was associated with percent step capture. In multivariate models, speed was the most significant determinant of percent step capture (p = 0.004, Table 3). For each increase in miles per hour, the percent step capture increased 24 points (95% confidence interval 8 to 40) (Table 3).
This study demonstrates that the ActiHealth accelerometer is accurate in a nondisabled cohort. However, its accuracy declines when it is used in subjects with COPD. Usual walking speed is the most important predictor of percent step capture and accuracy of the accelerometer. Before use of the ActiHealth accelerometer, walking speed and percent step capture should be verified in each subject.
Little is published in the literature about the accuracy of pedometers or accelerometers in the COPD population [19]. The present study is the first to assess the accuracy of an accelerometer device against the "gold standard" of manual step counts in the COPD population. Other devices used in the COPD population have been shown to have concurrent validity by correlations with lung function, exercise capacity, or dyspnea; however, the accuracy of these devices in the field is unknown. For example, a triaxial accelerometer, the RT3 activity monitor (Stayhealthy, Inc; Monrovia, California), has been used in subjects with COPD. Steele et al. showed moderate correlations between the vector magnitude units and existing measures of disease status, such as lung function, 6-minute walk distance, and dyspnea [20-22]. The DynaPort Activity Monitor, a triaxial accelerometer (375 g) (McRoberts; The Hague, the Netherlands), was shown to be accurate in assessing time spent walking, cycling, standing, sitting, and lying down compared with video recordings [23]. Time spent walking was shown to be moderately correlated with lung function, muscle function, and exercise capacity [24]. Similarly, the SenseWear® armband (Bodymedia, Inc; Pittsburgh, Pennsylvania), a biaxial accelerometer (80 g) worn on the arm, measures steps per day, which has been shown to correlate with FEV1 and 6-minute walk distance [25]. Our results indicate that assessing the accuracy of wearable sensors in persons with COPD is important. Subject characteristics, such as walking speed, may influence whether each step or physical activity is accurately detected. Our results suggest that sensors may have very different performance properties in subjects with COPD compared with nondisabled subjects.
In this cohort of subjects with predominantly moderate to severe COPD, walking speed was poorly correlated with FEV1 percent predicted. This finding suggests that excluding subjects with COPD who walk at slow speeds (in whom the accelerometer would be inaccurate) in future studies would not differentially exclude those with low FEV1 percent predicted. Conversely, FEV1 percent predicted cannot be used to predict walking speed or likelihood of accurate step capture by the ActiHealth accelerometer.
In this report, we have established the intra- and interdevice CVs of the ActiHealth accelerometer and have also assessed its accuracy in nondisabled subjects and in subjects with COPD. In the subset of patients with COPD who have acceptable accuracy (≥90% step capture), we plan to perform additional studies using the ActiHealth accelerometer to assess the day-to-day variability of step counts, its responsiveness to change in clinical status, the relationship between step counts and distance walked, and the relationship between step counts and clinical measures of COPD severity. Understanding these relationships will guide future clinical and research uses of the ActiHealth accelerometer in the COPD population. Furthermore, technical changes in the ActiHealth accelerometer need to be developed to improve its sensitivity and accuracy in the elderly population with chronic disease.
The ActiHealth accelerometer is accurate in nondisabled subjects and in a subset of patients with COPD. Walking speed is the most significant predictor of accelerometer accuracy. Before use of the ActiHealth accelerometer, researchers and clinicians should verify walking speed and percent step capture in each subject. In general, the accuracy of wearable sensors should be assessed in the population that is being studied, since they may have different performance properties compared with nondisabled subjects.
We thank Kelly Stolzmann for her assistance.
This material was based on work supported by the VA, Veterans Health Administration, Rehabilitation Research and Development Service, at the VA Boston Healthcare System, Boston, Massachusetts, through a VA Career Development Award to Dr. Moy and by the National Institutes of Health, National Institute of Child Health and Human Development, grant RO1 HD42141 to Dr. Garshick.
This study was initiated by the investigators, who do not receive any financial support from FitSense, Inc, and the results of the present study do not constitute endorsement of the product by the authors. FitSense, Inc, did not have any involvement in the study design; data collection, analysis, or interpretation; and writing or submission of this article.
The authors have declared that no competing interests exist.
Go to TOP
Go to the Table of Contents of Vol. 45 No. 4
Last Reviewed or Updated
Tuesday, September 1, 2009 7:55 AM