Article Text
Abstract
Background Current research practice employs wide-ranging accelerometer wear time criteria to identify a valid day of physical activity (PA) measurement.
Objective To evaluate the effects of varying amounts of daily accelerometer wear time on PA data.
Methods A total of 1000 days of accelerometer data from 1000 participants (age=38.7±14.3 years; body mass index=28.2±6.7 kg/m2) were selected from the 2005–2006 National Health and Nutrition Examination Study data set. A reference data set was created using 200 random days with 14 h/day of wear time. Four additional samples of 200 days were randomly selected with a wear time of 10, 11, 12 and 13 h/day1. These data sets were used in day-to-day comparison to create four semisimulation data sets (10, 11, 12, 13 h/day) from the reference data set. Differences in step count and time spent in inactivity (<100 cts/min), light (100–1951 cts/min), moderate (1952–5724 cts/min) and vigorous (≥5725 cts/min) intensity PA were assessed using repeated-measures analysis of variance and absolute percent error (APE).
Results There were significant differences for moderate intensity PA between the reference data set and semisimulation data sets of 10 and 11 h/day. Differences were observed in 10–13 h/day1 for inactivity and light intensity PA, and 10–12 h/day for steps (all p values <0.05). APE increased with shorter wear time (13 h/day=3.9–14.1%; 12 h/day=9.9–15.2%, 11 h/day=17.1–35.5%; 10 h/day=24.6–40.3%).
Discussion These data suggest that using accelerometer wear time criteria of 12 h/day or less may underestimate step count and time spent in various PA levels.
- Physical activity measurement
- Physical activity and exercise methodology
- Measurement
Statistics from Altmetric.com
Introduction
Low physical activity (PA) and high inactivity and the associated health concerns (eg, cardiovascular disease, metabolic risk, obesity, etc.) have become increasingly important.1 ,2 Researchers and health professional have focused on assessing these behaviours to assess health and as an avenue for interventions. Objectively measuring PA with accelerometers has become a popular method for PA assessment. Standard measures and methodologies for accelerometer use are paramount to assess PA for appropriately classifying individuals and to accurately monitor changes in PA.3 For example, significant research has been done to understand PA behaviour reliability,4–7 replacing missing PA data8 ,9 and accelerometer epoch length.10 ,11 Few studies, however, have sought to identify the optimal number of hours per day an accelerometer should be worn to identify a valid day.12 ,13
The majority of studies do not require participants to wear accelerometers for 24 h/day, which creates a problem for researchers to determine how many hours of wear time represents a valid day. In current practice, there is a wide range of wear time criteria used to identify a valid day. Reports in the literature range from as few as 2 h14 ,15 to limiting the upper range to 16 h.16 Several studies have identified valid days by using different criteria for weekdays (10 h/day) versus weekend days (8 h/day).17 Some researchers report valid days based on a percentage of time awake ranging from 60%13 to 75%4 of awake time. Still others recommend using sample-specific criteria that may change based on the amount of time a specific sample wore the accelerometer.9 ,18 For example, Catellier et al9 proposed the 70/80 rule which requires 70% of the sample to have accelerometer data and 80% of that observed period becomes the valid day threshold. Another method that has been used is to ‘normalise’ the data to 12 h by inputting data (eg, 10 h/day of wear time was changed to 12 h/day increasing minutes in each intensity level proportionally).19 ,20 The criterion of 10 h/day of accelerometer wear time is regularly used to identify a valid day of accelerometer data.21–23 However, empirical evidence is lacking to support 10 h/day of wear time or that any other criterion is superior to another. Without data to support a consensus for daily wear time criteria, the validity of daily wear time and the comparability of studies using different wear time criteria must be questioned.
The purpose of this study was to evaluate the effects of varying amounts of daily accelerometer wear time on PA in a sample of adults participating in the 2005–2006 National Health and Nutrition Examination Study (NHANES).
Methods
Study design and participant selection
NHANES 2005–2006 employs a complex, multistage probability sampling method to obtain a representative sample of the US population. The purpose of NHANES is to assess the health and nutritional status of adults and children in the USA for use in understanding the prevalence and risk factors for diseases. NHANES participants undergo extensive evaluations that include interviews and health examinations. In 2003, as part of the evaluation process, all ambulatory participants older than 6 years were asked to wear a PA monitor (accelerometer). The full 2005–2006 accelerometer and demographic data sets were downloaded from the NHANES website (http://www.cdc.gov/nchs/nhanes/nhanes2005–2006/exam05_06.htm). Data from adults 18–65 years, with accelerometer data (approximately N=4000 participants), were selected for use in this study.
Instrument
The NHANES uses the ActiGraph model 7164 accelerometer (formerly the CSA/MTI AM-7164, manufactured by ActiGraph, Ft. Walton Beach, Florida, USA) to assess PA. The ActiGraph 7164 (5.08×3.81×1.27 cm) uses a single-axis accelerometer that measures vertical accelerations. The output frequency is 0.25–2.5 Hz, digitised by an analog-to-digital converter, and stored in 1 min epochs (sampling intervals).24 The accelerometer is powered by a 3 V coin-cell lithium battery.
NHANES participants were provided verbal and written instructions to wear the accelerometer for 7 days during all waking hours. The accelerometer was secured to the waist over the right hip by an elastic belt, which included a Velcro pouch for the accelerometer. Exclusions for the NHANES accelerometer assessment include waist girths that are too large for the belt, individuals in wheelchairs and those with recent abdominal surgery.
Data preparation
A SAS statistical program (SAS V. 9.2) was used to identify outlier values due to accelerometer malfunction, non-wear periods and outcome variables, including steps and time spent in each intensity level (inactivity, light, moderate and vigorous). For this study, non-wear periods were identified as 60 consecutive minutes with no movement data (zero counts) allowing up to two consecutive minutes of 1–100 cts/min.25 Non-wear periods were ended with >100 cts/min or with three consecutive 1–100 cts/min.25 ActiGraph monitors were scored to assess steps and time spent in each intensity level. Time spent in inactivity was identified by cut-points <100 cts/min,26 ,27 and Freedson's cut points were used to determine time spent in light intensity (100–1951 cts/min), moderate intensity (1952–5724 cts/min) and vigorous intensity (5725+ cts/min).24
Semisimulation approach
A semisimulation design8 ,28 was used to create new data sets with different amounts of wear time from an original data set. These semisimulated data sets can then be compared to the reference data set to understand the amount of error associated with varying amounts of wear time. The basic premise of the semisimulation approach is that it considers data characteristics (eg, real accelerometer wear pattern) to remove data instead of a random data removal approach.
The reference wear time value was set at 14 h/day, which was based upon the average wear time of participants from several large studies using accelerometers.23 ,29 A random sample of 200 days from different individuals was selected from NHANES participants who wore the accelerometer for 14 valid hours (ie, the accelerometer was worn for at least 40 min for each hour).30 Four additional 200-day samples were randomly selected from individuals who wore the accelerometer for 13, 12, 11 and 10 h/day. These comparison data sets were used as a model for their missing data pattern in the semisimulation approach.
The semisimulation approach matches the wear time data pattern in a one day-to-one day comparison to remove data from the reference 14 h/day data set (eg, 14 h/day matched with 13 h/day, 14 h/day matched with 12 h/day, etc). For example, a reference day with a 14 h wear pattern indicates the accelerometer was worn from 07:00 until 21:00 hours and a comparison day with a 12 h pattern shows it was worn from 07:00 until 19:00 hours. Then the data from 19:00 to 21:00 hours from the 14 h day would be removed creating a semisimulated 12 h day. This one day-to-one day matching and removing data provide four new 200-day semisimulated data sets of 13, 12, 11 and 10 h/day with real-world missing data patterns. These new semisimulated data sets can be compared with the reference 14 h/day data set to identify differences in step count and min/day spent in varying activity intensity levels. Figure 1 displays the data management procedure for the semisimulation approach.
Statistical analysis
Descriptive statistics were computed for all variables and data for the study sample. Repeated-measures analyses of variance (ANOVAs) with the least significant difference post-hoc method was performed to assess differences in daily steps and minutes spent in inactivity, light, moderate and vigorous PA intensity levels between semisimulation data sets (13, 12, 11 and 10 h/day) and the reference data set (14 h/day). Step count and PA intensity levels were dependent variables, and separate repeated-measure ANOVAs were performed for each dependent variable. The duration of wear time is considered a within-subject factor in the repeated-measures ANOVA. To check the assumption of sphericity, Mauchly's sphericity test was first examined. If data violated the assumption, adjustment to the degrees of freedom and F value was made using the Greenhouse–Geisser estimate of sphericity. Absolute percent error (APE)=(|Observed Value−Reference Value|/Reference Value×100) was computed between the reference value and each of the semisimulation data sets for step count and PA intensity level (min/day) with a lower APE desired.31 APE has been used in similar analyses to identify the number of days needed to assess habitual PA with a pedometer.31 Proportion of time spent in each PA intensity level was calculated for each semisimulated data set and the reference data set.
Results
Details of the entire NHANES accelerometer sample can be found elsewhere.23 The analysed sample in this study used 1000 days from 1000 participants, which came from over 4000 eligible participants that wore an accelerometer for over 25 000 days in total. Demographic data and anthropometric characteristics of the study population are presented in table 1.
Table 2 shows the proportion of time spent in each activity category by h/day of wear time. The proportion of time spent in each activity level was not different across different hours of wear time.
Table 3 shows the accelerometer measured step count and minutes of activity by intensity and wear time duration (10, 11, 12, 13, and 14 h/day). Compared to the reference value of 14 h/day, minutes tended to decrease across activity categories as the wear time decreased. Results from the repeated-measures ANOVA showed significant differences in daily minutes between the 14 h/day reference value and the semisimulated data sets of 10, 11, 12 and 13 h/day for inactivity and light intensity (all p values<0.05). There were significant differences for minutes of moderate intensity between the 14 h/day reference value and the 10 and 11 h/day semisimulated data sets (all p values<0.05). No difference was found for vigorous intensity activity, F (1.09, 1083.38)=0.31, p = 0.87). Daily step count followed a similar pattern as step count decreased as wear time decreased. Significant differences were observed between the 14 h/day reference value and the semisimulated data sets of 10, 11 and 12 h/day (all p values<0.05).
Table 4 displays the APE values for accelerometer measured activity by wear time (h/day), step count and PA categories. APE values ranged from 3.9% for steps with 13 h/day of wear time to 40.3% for vigorous intensity activity with 10 h/day. APE values increased with the shorter accelerometer wear time (13 h/day=3.9–14.1%; 12 h/day=9.9–15.2%, 11 h/day=17.1–35.5%; 10 h/day=24.6–40.3%). There were no h/day categories with APE values below 10% for all dependent levels.
Discussion
This study offers insight into the effects of accelerometer wear time on estimates of PA for time spent in varying intensities and step count. In general, longer wear times provided significantly greater amounts of steps and time in inactivity, light intensity and moderate intensity PA. Furthermore, the amount of error was greater with less wear time for all variables (step count, physical inactivity and PA). Few individuals in this sample participated regularly in vigorous intensity activity providing limited information about a highly active sample.
These findings provide a similar pattern to results from a previous study that investigated wear time differences in a sample of adults participating in a worksite health promotion study to increase walking to 10 000 steps/day.32 This study demonstrated that with increased accelerometer wear time there were significantly more minutes recorded in inactivity and light to moderate intensity PA. As with the current sample, the duration spent in vigorous intensity activity was insufficient to be affected by accelerometer wear time. However, one prominent difference in the previous study compared to our current analyses was that the previous sample did more moderate intensity activity (45 min/day) as a result of participating in the walking intervention compared to this sample of NHANES participants (26 min/day). Despite this large difference, the error patterns remained strikingly similar. Furthermore, the previous study did not investigate differences in step count across wear time. Excluding vigorous intensity PA, the APE reported for time spent in various activity intensity levels for 12 h/day ranged from 10.2% to 15.2% for this study and 13.5% to 14.3% from the walking intervention study.32 Both of these studies demonstrate that by reducing the wear time criteria by more than 1 h from the 14 h/day referent value, it would result in an error greater than 10%.
These findings have important clinical and research implications for studies using accelerometers. Studies requiring a wear time less than 13 h/day may be underestimating the true amount of PA and inactivity performed. National PA studies conducted in the USA and Canada have required a minimum of 10 h/day of wear time.21 ,23 Using this 10 h/day minimum wear time duration, yielded 18–24 min/day of moderate intensity activity for Canadians21 and 18–33 min/day for US adults.23 Even though the mean accelerometer wear times were approximately 14 h/day, data with fewer hours of wear are included in the analyses, which may have attenuated the activity and inactivity values. Our results showed that allowing PA data with 10 h/day of wear time may be missing 25–30% of time spent in inactivity, light and moderate intensity activity compared to 14 h/day of wear time. This could result in inaccurate estimates of the proportion of adults who meet national guidelines of 30 min/day or 150 min/week of moderate intensity activity and also underestimate the time spent in inactive behaviours.33 ,34
Recently, Colley et al35 investigated a variety of accelerometer data reduction methods. One method was comparing 6, 8, 10, 12 and 14 h/day of wear time. While only looking at the percentage of the sample that would be included or excluded from analysis with these wear time criteria, the authors noted that lowering the wear time criteria from 14 to 10 h/day resulted in a substantial increase in valid days, whereas lowering from 10 to 6 h/day only minimally affected the number of valid days for analysis. Their study did not examine the error in min/day of activity and inactivity associated with these different wear time values. Colley et al35 showed that accepting a shorter wear time criteria, such as 10 h/day, increases the sample size for analysis. However, doing this also increases the error associated with estimates of time spent in PA and inactivity and may introduce a sizeable amount of underestimation in the results. For example, our results indicate that when compared with 14 h/day, individuals with 10 h/day of wear time may be missing roughly 135 min/day of inactivity, 95 min/day of light, 7 min/day of moderate and 0.5 min/day of vigorous activity.
Jerome et al36 examined the difference in a 6 h/day vs a 10 h/day of accelerometer wear time criteria in a sample of obese individuals. They found similar estimates of time spent in moderate-to-vigorous intensity PA between 6 and 10 h/day and consequently suggest that 6 h/day of wear time is adequate in identifying valid days.36 These researchers then used the Spearman-Brown prophecy formula to predict the number of days that need to be measured to reach an acceptable level of reliability. Using their suggested wear time criteria of 6 h/day, the ICC values indicated that 16–35 days of monitoring would be necessary to achieve a reliability of 0.80 for measuring time spent in moderate-to-vigorous PA.36 This amount of monitoring would significantly increase the burden on research participants and increase the cost of conducting a study. Conversely, using these criteria and too few days of monitoring would cause low reliability and compromise the results of studies. In comparison, a study designed to identify sources of variance in daily PA, Matthews et al4 analysed data with more than 12 h/day of wear time and found that only 3–4 days were necessary to obtain a reliability of ICC >0.80. It is unclear as to whether this same result would be found if the wear time criteria were reduced to 10 h/day of wear time. On the other hand, requiring more than 12 h/day of wear time as suggested from our results to minimise error, it may be possible to further reduce the number of days needed to monitor.
If a longer wear time is superior for a more accurate assessment of time spent in inactivity and PA intensities, then it is reasonable to have a greater focus on increasing accelerometer wear compliance to ensure that the devices are worn during all waking hours or encourage 24 h wear. A few researchers have incorporated 24 h/day of accelerometer wear time, suggesting removal only for bathing and water activities.37–39 Little is known if such a practice would yield more accurate measures of time spent in inactivity or in PA.
Limitations
This study illustrates the difference in activity estimates if an entire sample has exactly 10, 11, 12, 13 or 14 h/day of accelerometer wear time. Therefore, the absolute differences in minutes of PA intensities and inactivity for an entire sample may not be as large because participants are included with a range of accelerometer wear time, even though the minimum criteria may be as low as 10 h/day. Nonetheless, this study demonstrates that individuals with less wear time may be adding to an underestimation of the true amount of PA that is actually being performed.
Summary
These data illustrate the effect of accelerometer wear time on PA data. Allowing data with less wear time may significantly reduce estimates of time spent in activity and inactivity and potentially affect estimates of individuals meeting activity recommendations. This may also influence the results of PA interventions which might show that individuals were not successful in changing their PA or inactivity behaviours when in fact they merely had insufficient accelerometer wear time needed to be accurately assessed and detect significant differences. Sacrificing the quality of PA assessment by reducing wear time criteria to achieve a greater quantity of participants for analysis may adversely impact study results. This study supports longer accelerometer wear time recommendations of greater than 12 h/day to ensure accurate estimates of daily PA.
What this study adds
-
Allowing physical activity (PA) data with less wear time may underestimate the amount of activity performed.
-
Caution should be used when comparing data from studies using different wear time criteria.
-
Additional work is needed to understand the influence of wear time on PA reliability and to understand wear time effects in different populations.
How might it impact on clinical practice in the near future?
Researchers and clinicians should strive to ensure that accelerometers are worn long enough to provide accurate assessments of daily PA and be cognizant of the trade-off between less wear time and underestimation of PA data.
References
Footnotes
-
Competing interests None.
-
Provenance and peer review Not commissioned; externally peer reviewed.