Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2021 Oct 26.
Published in final edited form as: J Meas Phys Behav. 2021 Feb 16;4(1):79–88. doi: 10.1123/jmpb.2020-0036

Agreement of sedentary behaviour metrics derived from hip-worn and thigh-worn accelerometers among older adults: with implications for studying physical and cognitive health

John Bellettiere 1, Fatima Tuz-Zahra 1, Jordan A Carlson 2, Nicola D Ridgers 3, Sandy Liles 4, Mikael Anne Greenwood-Hickman 5, Rod L Walker 5, Andrea Z LaCroix 6, Marta M Jankowska 7, Dori E Rosenberg 8,*, Loki Natarajan 9,*
PMCID: PMC8547742  NIHMSID: NIHMS1686095  PMID: 34708190

Abstract

Little is known about how sedentary behaviour (SB) metrics derived from hip-worn and thigh-worn accelerometers agree for older adults. Thigh-worn activPAL micro monitors were concurrently worn with hip-worn ActiGraph GT3X+ accelerometers (with SB measured using the 100 count-per-minute (cpm) cut-point; ActiGraph100cpm) by 953 older adults (age 77±6.6, 54% women) for 4-to-7 days. Device agreement for sedentary time and 5 SB pattern metrics was assessed using mean error and correlations. Logistic regression tested associations with 4 health outcomes using standardized (i.e., z-scores) and unstandardized SB metrics. Mean errors (activPAL-ActiGraph100cpm) and 95% limits of agreement were: sedentary time −54.7(−223.4,113.9) min/d; time in 30+ minute bouts 77.6(−74.8,230.1) min/d; mean bout duration 5.9(0.5,11.4) min; usual bout duration 15.2(0.4,30) min; breaks in sedentary time −35.4(−63.1,−7.6) breaks/d; and alpha −0.5(−0.6,−0.4). Respective Pearson correlations were: 0.66, 0.78, 0.73, 0.79, 0.51, 0.40. Concordance correlations were: 0.57, 0.67, 0.40, 0.50, 0.14, 0.02. The statistical significance and direction of associations was identical for ActiGraph100cpm and activPAL metrics in 46 of 48 tests, though significant differences in the magnitude of odds ratios were observed among 9 of 24 tests for unstandardized and 2 of 24 for standardized SB metrics. Caution is needed when interpreting SB metrics and associations with health from ActiGraph100cpm due to the tendency for it to overestimate breaks in sedentary time relative to activPAL. However, high correlations between activPAL and ActiGraph100cpm measures and similar standardized associations with health outcomes suggest that studies using ActiGraph100cpm are useful, though not ideal, for studying SB in older adults.

Keywords: Sedentary accumulation patterns, physical activity, frailty and physical function, sedentary behaviour patterns, measurement, sitting time

INTRODUCTION

Sedentary behaviour is increasing in modern society and among older adults it is the most prevalent behaviour among sleep, sedentary behaviour, and physical activity. (Diaz et al., 2016; Du et al., 2019; Jefferis et al., 2015; Matthews et al., 2012; Yang et al., 2019) By consensus, sedentary behaviour has been defined as all waking behaviours while in a seated or lying posture that result in an energy expenditure ≤ 1.5 metabolic equivalents (Tremblay et al., 2017). When measured using ActiGraph accelerometers worn around the participant’s hip, sedentary time is estimated based on lack of movement, and does not factor in posture. To obtain device-based measures of sedentary behaviour derived from both posture and movement, the most common approach is to attach an activPAL monitor—an inclinometer/accelerometer specifically designed as a thigh-worn device capable of assessing posture and thus capturing sitting/lying—to the participant’s thigh (Edwardson et al., 2016). This approach is often viewed as the device-based standard for the measurement of sedentary behaviour.

Accelerometers produce time-stamped data, making it possible to identify when during the day activity occurs (Glazer et al., 2013; Lord et al., 2011), as well as the patterns in which behaviours are accumulated (Chastin & Granat, 2010). These patterns include the timing, frequency, and duration of sedentary bouts and breaks throughout the day (Tremblay et al., 2017). While there has been an increase in the number of studies examining sedentary patterns and their associations with health [e.g., (Brocklebank, Falconer, Page, Perry, & Cooper, 2015; Jefferis et al., 2019)], the Physical Activity Guidelines for Americans, 2nd edition (U.S. Department of Health and Human Services, 2018) and the update (Katzmarzyk et al., 2019) both highlighted that further research using prospective cohorts to study sedentary patterns in relation to mortality and other health outcomes is needed.

One challenge to updating this literature is that most prospective cohort studies use hip-worn ActiGraph accelerometers rather than the thigh-worn activPAL monitor (Lee & Shiroma, 2014), and the measurement of sedentary bouts and breaks using hip-worn accelerometers and standard data processing techniques (Migueles et al., 2017) is not accurate (Barreira, Zderic, Schuna, Hamilton, & Tudor-Locke, 2015). For example, when compared to direct observation of sit-to-stand transitions during usual free-living conditions, one study of 13 adults reported 0.3% bias for thigh-worn activPAL-measured transitions and 98.6% bias for hip-worn ActiGraph-measured transitions (Lyden, Kozey Keadle, Staudenmayer, & Freedson, 2012). Despite the inaccuracy, the authors reported correlation coefficients compared to direct observation of 0.97 for the activPAL-measured transitions and 0.92 for the ActiGraph-measured transitions (Lyden et al., 2012). There has also been evidence of convergent validity in several labs and cohorts around the world that showed ActiGraph-derived measures of sedentary patterns were associated with various health outcomes in the expected direction (Bellettiere et al., 2019; Brocklebank et al., 2015; Diaz, Goldsmith, et al., 2017; Diaz, Howard, et al., 2017). Given the historical and continuing use of ActiGraph accelerometers in prospective cohorts, understanding the agreement between ActiGraph-based and posture-based measures for assessing sedentary behaviour, including accumulation patterns, is critical for advancing the field.

The aim of this study was to assess agreement between the most commonly used hip-worn ActiGraph measures of volumes and patterns of sedentary time and thigh-worn activPAL measures in a well characterized cohort of older adults and determine how any measurement error may bias associations of sedentary behaviour with physical or cognitive health.

METHODS

In 1994, adults over 65 without dementia were randomly sampled from the King County membership of Group Health Cooperative of Puget Sound (now Kaiser Permanente Washington) to join the Adult Changes in Thought (ACT) study, a longitudinal cohort study of aging and incident dementia. An expansion cohort was enrolled starting in 2000. In 2005, a cohort refreshment protocol to maintain a constant 2000 participants was initiated by enrolling new members to replace those who died, were diagnosed with dementia, or dropped out of the study. Beginning in April 2016, enrolled participants were given the option to wear an ActiGraph GT3X+ accelerometer (ActiGraph LLC, Pensacola, FL, USA) and/or an activPAL micro (PAL Technologies, Glasgow, Scotland, UK). Of the 1688 eligible participants who were able to and asked to wear devices, 1151 wore the ActiGraph (1088 returned devices with 4 or more adherent days, which for both devices were defined as days having 10 or more hours of awake wear), and 1135 wore the activPAL (1039 returned devices with at least 4 adherent days). The 953 men and women who wore both devices concurrently for at least 4 adherent days comprised our study population. Details on the ACT cohort and accelerometer deployment are published elsewhere (Rosenberg et al., 2020). Ethics approval was obtained from the Kaiser Permanente Washington institutional review board and all participants provided written informed consent.

Accelerometers

Participants were asked to wear devices for the same 7 days and to complete sleep logs each night of wear to document their in-bed and out-of-bed times. Sleep-log data were double-entered to protect against transcription errors. Missing data were imputed using person-specific means if available and ACT sample means otherwise. Recorded in-bed and out-of-bed times were used to identify awake time for processing data from both devices.

The ActiGraph GT3X+ was worn 24-hours per day on an elastic waistband at the right hip region with data collected at 30 Hz. Using ActiLife v6.13.3, 15-second epoch data were generated using the normal filter then aggregated to 1-minute epochs. Device non-wear was determined using the Choi algorithm (Choi, Liu, Matthews, & Buchowski, 2011; Choi, Ward, Schnelle, & Buchowski, 2012) applied to vector magnitude counts per minute using a 90-minute window, 30-minute streamframe, and 2-minute tolerance. To classify awake-wear-time epochs as sedentary, the most common cut-point (100 counts per minute) was applied to the vertical axis (Migueles et al., 2017). Sedentary bouts were then derived as consecutive waking sedentary time epochs with no minimum and no tolerance.

The activPAL3 micro was placed in a waterproof casing and secured to the center of the right thigh using Tegaderm adhesive tape. Using the default setting in palBatch v7.2.32, data were converted to event-level files which were then visually inspected for anomalies using heat maps that had sleep-log and ActiGraph data superimposed. Sitting events that occurred while participants were awake (a.k.a., sitting bouts) were used to compute sedentary behaviour metrics.

Sedentary behaviour metrics

Sedentary behaviour metrics were computed using only adherent days on which both devices were worn. Total sedentary time was averaged across all adherent days for each device. Five of the most commonly used sedentary accumulation pattern metrics were evaluated. Time spent in 30+ minute bouts was computed as the average sedentary time per day that was accumulated in bouts that were 30 minutes in duration or greater. Mean bout duration was computed as the sum of all sedentary bout durations divided by the number of bouts. The usual bout duration and alpha were computed using the methods described by Chastin and colleagues (Chastin et al., 2015)(Chastin & Granat, 2010). The usual bout duration is the midpoint of the cumulative distribution of sedentary bout durations, computed using non-linear regression. Usual bout duration (UBD) indicates the bout duration above which half of all sedentary time is accumulated—higher values indicate a more prolonged accumulation pattern. Alpha characterizes the shape of the sedentary bout duration distribution for each person with lower alphas reflecting frequent long bouts with few short bouts, and higher alphas reflecting frequent short bouts with few long bouts (see Supplemental Figure 1 in our earlier publication for more details (Bellettiere et al., 2017)). The number of breaks was computed as the average number of sedentary bouts per day over all adherent days. In this regard, sedentary breaks represented the number of times per day there was a transition from a sedentary bout to movement above the 100cpm cut-point, while sitting breaks represented the number of times per day there was a transition from a sitting/lying posture to standing or stepping.

Outcomes

At the ACT study visit when accelerometers were distributed, participants’ height and weight were measured by trained staff using a tape measure and the clinic scale. Body mass index was computed as the weight (kg) divided by height2 (m2). Participants also completed questionnaires. From the RAND 36 questionnaire (J. E. Ware, 2000)(C. J. E. Ware & Sherbourn, 1992), self-rated health was assessed with a single item asking “In general, would you say your health is: excellent, very good, good, fair, poor?”; difficulty walking half a mile was assessed with a single item asking “Does your health now limit you in walking half a mile and if so, how much?” Responses were yes, limited a lot; yes, limited a little; and no, not limited at all. Global cognitive performance was measured using the Cognitive Abilities Screening Instrument (CASI) which consists of a short series of tests that assess 9 domains of cognitive impairment (Chiu, Yip, Woo, & Lin, 2019; De Oliveira et al., 2016; Teng et al., 1994). Resulting scores range from 0 to 100 (best functioning). Cognitive impairment was defined as having a score ≤86 on the CASI and/or referral for additional diagnostic workup. For analyses, BMI, self-rated health, difficulty walking a half mile, and cognitive function—selected for analysis because of known or expected associations with sedentary behaviour—were dichotomized as ≥30 vs. <30; good, poor, and very poor vs. very good and excellent; yes vs. no; and cognitive impairment, yes vs. no, respectively.

Statistical methods

Summary statistics described distributions of participant characteristics. Histograms were plotted to show distributions of the six sedentary behaviour metrics separately for activPAL and ActiGraph100cpm measures. Then, activPAL and ActiGraph100cpm measures of the six sedentary behaviour metrics were compared at the person-level by computing mean error (activPAL – ActiGraph100cpm) and mean absolute error. Agreement between the two devices was visually assessed using the Bland-Altman approach, with 95% limits of agreement computed using regression analysis to test for and account for relationships between bias and magnitude (Bland & Altman, 2007). Pearson and Spearman correlation coefficients were used to describe the between-device linear relationship of the sedentary behaviour metrics. Concordance correlation coefficients were used to assess the degree to which the between-device linear relationship aligns with the 45-degree line. To explore the epidemiologic implications of any potential measurement error, associations of the six sedentary behaviour metrics with four indicators of health (self-rated health, BMI, difficulty walking half a mile, and cognitive impairment) were estimated using separate multivariable logistic regression models. Values for all sedentary behaviour metrics were first converted to z-scores to standardize the resulting beta coefficients. This enabled a distribution-based “apples-to-apples” between-device comparison of the size of the odds ratios (ORs) since the underlying unit of change in each z-score-converted sedentary behaviour metric is identical (i.e., a 1 standard deviation increment). Models were adjusted for age, gender, race/ethnicity (Hispanic or non-white vs. non-Hispanic white), and education (less than high school, completed high school, some college, or completed college). The resulting standardized ORs from models using ActiGraph100cpm sedentary behaviour metrics were compared to the corresponding standardized ORs from models using activPAL metrics, according to the methods described by Horton et al (Horton & Fitzmaurice, 2004). Results from analyses of sedentary behaviour and health outcomes were also reported using unstandardized ORs, which were computed using sedentary behaviour metrics that were not converted to z-scores and instead remained in their original units of measure.

Following initial peer-review, we reran agreement analyses after classifying sedentary behaviour using alternate cutpoints of 200cpm applied to the vector magnitude and 25cpm applied to the vertical axis,(Aguilar-Farías, Brown, & Peeters, 2014) cutpoints that were similar to those reported by (Koster et al., 2016).

Analyses were conducted using R (R Foundation for Statistical Computing; Vienna, Austria) with two-tailed statistical tests and statistical significance set to p<0.05.

RESULTS

The 953 ACT participants included in this study were, on average, aged 77±7 years, 54.3% female, 87.0% non-Hispanic white, and 72.7% completed college (Table 1).

Table 1.

Participant Characteristics for Men and Women of the Adult Changes in Thought (ACT) Study Who Concurrently Wore ActiGraph and actiPAL Accelerometers; n = 953.

Quantitative variables, mean (sd)
 Age (years) 77 (6.6)
Categorical variables , n(%)
 Gender
  Men 421 (44.2%)
  Women 532 (55.8%)
 Race/ethnicity
  Hispanic or non-white 97 (10.2%)
  non-Hispanic white 853 (89.5%)
 Education
  Less than high school 15 (1.6%)
  Completed high school 74 (7.8%)
  Some college 152 (16.0%)
  Completed college 712 (74.7%)
 BMI
  BMI below 30 722 (75.8%)
  BMI 30 or above 212 (22.2%)
 Self-rated health
  Very good and excellent 599 (62.9%)
  Good, poor, or very poor 354 (37.1%)
 Difficulty in walking half a mile
  None 726 (76.2%)
  Some 227 (23.8%)
 Cognitive impairment
  Impaired 23 (2.4%)
  Not impaired 930 (97.6%)
Open in a new tab

Note. BMI = Body Mass Index. Percentages may not sum to 100% due to rounding.

Distributions for each sedentary behaviour metric separately for activPAL and ActiGraph100cpm measures are presented in Supplemental Figure 1. Mean total sedentary time as measured by ActiGraph100cpm (653 min/d) was 9.1% higher compared to the activPAL (598 min/d), resulting in a mean difference of 54.7 min/d (limit of agreement [LOA] = −223.4, 113.9; Table 2). There were nearly twice as many breaks in sedentary time detected by the ActiGraph100cpm (mean=79.9 breaks/d) than the activPAL (mean=44.5 breaks/d; mean difference = 35.4 breaks/d; LOA = −63.1, −7.6). The higher number of breaks recorded using ActiGraph100cpm resulted in 77.6 fewer min/d (LOA = −74.8, 230.1) by ActiGraph100cpm than activPAL in time spent in 30+ minute sedentary bouts. The mean difference in mean bout durations was 5.9 minutes (LOA=0.5, 11.4 minutes), resulting in a 39.9% shorter mean when measured by ActiGraph100cpm (8.9 minutes) than by activPAL (14.8 minutes). The mean difference in usual bout durations was 15.2 minutes (UBDactigraph=22.7 minutes, UBDactivPAL=37.9 minutes), resulting in a 40.1% shorter mean. The higher frequency of and generally shorter duration of bouts detected by ActiGraph100cpm resulted in a more rapid decay of the bout duration distribution and higher alpha values (alphaActiGraph=1.3, alphaactivPAL=1.8). Bland-Altman plots illustrating mean differences for each metric are shown in Supplemental Figure 2.

Table 2.

Summary and Agreement Results for Sedentary Time and Patterns of Sedentary Time Measured Using ActiGraph GT3X+ and activPAL Micro Among Adult Changes in Thought (ACT) Participants; n=953

Total sedentary time (min/d) Time spent in 30+ minute bouts (min/d) Mean bout duration (min) Usual bout duration (min) Number of breaks in sedentary time Alpha
activPAL, mean (SD) 598.4 (116.1) 351.2 (135.5) 14.8 (6.6) 37.9 (17.8) 44.5 (12.8) 1.30 (0.04)
ActiGraph, mean (SD) 653.2 (95.2) 273.6 (128.0) 8.9 (3.9) 22.7 (12.5) 79.9 (18.1) 1.79 (0.13)
Person Level Agreement
 Mean error (AP - AG) −54.7 77.6 5.9 15.2 −35.4 −0.49
 95% limits of agreement (−223.4, 113.9) (−74.8, 230.1) (0.5, 11.4) (0.4, 30.0) (−63.1, −7.6) (−0.6, −0.4)
 Mean absolute error 80.1 92.9 6.1 15.7 35.3 0.49
 Pearson correlation 0.66 (0.62,0.70) 0.78 (0.76,0.81) 0.73 (0.7,0.76) 0.79 (0.77,0.82) 0.51 (0.46,0.56) 0.40 (0.34,0.45)
 Concordance correlation 0.57 (0.53,0.61) 0.67 (0.64,0.70) 0.40 (0.37,0.43) 0.50 (0.47,0.53) 0.14 (0.12,0.15) 0.02 (0.01,0.02)
Open in a new tab

Note. SD = Standard Deviation; AP = activPAL; AG = ActiGraph

Pearson correlations between ActiGraph100cpm and activPAL for all sedentary behaviour variables ranged between 0.40 and 0.79. Sedentary breaks and alpha had lower correlations (r=0.40 and r=0.51, respectively) than total sedentary time (r=0.66), time in long bouts (r=0.78), mean bout duration (r=0.73), and usual bout duration (r=0.79). Spearman correlations produced similar results (data not shown). Concordance correlation coefficients followed a similar pattern ranging from 0.02 to 0.67. Alpha and sedentary breaks had lower concordance correlations (0.02 and 0.14, respectively) compared to total sedentary time (0.57), time in long bouts (0.67), mean bout duration (0.40), and usual bout duration (0.50).

Measures of total sedentary time using the Aguilar-Farias et al. cutpoints were, on average, lower than activPAL-measured sedentary time, with 95% limits of agreement contained the value of zero, and a slightly lower mean error (Supplemental Table 1). Total sedentary time correlation coefficients for ActiGraph25cpm (r=0.70 and concordance correlation=0.64) and ActiGraph200cpm (r=0.80 and concordance correlation=0.77) were higher than those for ActiGraph100cpm (r=0.66 and concordance correlation=0.57). Agreement for sedentary accumulation pattern metrics were not appreciably different when processing data using the Aguilar-Farias et al. cutpoints vs the 100cpm cutpoint, though more similarities were observed for the 200cpm vector magnitude cutpoint than for the 25cpm vertical axis cutpoint.

Figure 1 shows standardized ORs and 95% confidence intervals (CIs) for associations between sedentary behaviour and health. Lower self-rated health (i.e., good, poor, or very poor) was significantly related to all sedentary behaviour metrics regardless of the device used, and there were no statistically significant differences between the computed ORs for ActiGraph100cpm and activPAL measures. While not significantly different in general, point estimates for ORs were slightly stronger when computed using activPAL measures than when using ActiGraph100cpm measures of total sedentary time, 30+ minute bouts, mean bout duration, and usual bout duration; OR point estimates were slightly larger when computed using ActiGraph100cpm measures of alpha and sedentary breaks. The same general pattern was observed for associations with BMI, with the exception that differences in ORs between ActiGraph and activPAL were significant for total sedentary time (p=0.03) and approached significance for time in 30+ minute bouts (p=0.05). For example, a 1 standard deviation increment in total sedentary time measured by activPAL was associated with 1.94 times higher odds of BMI ≥ 30 compared to BMI < 30 (95% CI=1.63, 2.32), whereas a 1 standard deviation increment in ActiGraph100cpm total sedentary time resulted in OR (95% CI) of 1.56 (1.32, 1.85). Difficulty walking half a mile was associated with all 6 sedentary behaviour metrics. The only significant difference between ActiGraph100cpm and activPAL OR point estimates was observed for alpha (p<.001), with a greater magnitude OR for ActiGraph (OR=0.55; 95% CI=0.46,0.67) than for activPAL (OR=0.77; 95% CI=0.65,0.91). For cognitive impairment, there were no statistically significant associations with any of the sedentary behaviour metrics and no significant differences in ORs between ActiGraph100cpm and activPAL.

Figure 1.

Figure 1

Open in a new tab

Forest plots for standardized (based on a one standard deviation unit change in each metric) associations of ActiGraph100cpm and activPAL sedentary behaviour metrics with self-rated health (panel A), body mass index (panel B), difficulty walking half mile (panel C), and cognitive impairment (panel D).

Note. ORAP indicates the odds ratio and 95% confidence interval for associations with activPAL-measured variables, ORAG indicates the odds ratio and 95% confidence interval for associations with ActiGraph-measured variables. All models are adjusted for age, sex, race/ethnicity, and education. The p-valueHorton is from models testing the hypothesis that ORAG and ORAP are different.

When metrics were not standardized, significant differences in ORs between ActiGraph100cpm and activPAL measures were more common (Figure 2). Except for breaks in sedentary time, associations of all 6 sedentary behaviour metrics with BMI differed significantly between devices. For all four health outcomes, the largest differences were observed for mean bout duration, with higher ORs for ActiGraph100cpm measures than activPAL measures. Alpha was consistently differentially associated with three outcomes, with stronger ORs for activPAL than ActiGraph100cpm.

Figure 2.

Figure 2

Open in a new tab

Forest plots for unstandardized* associations between sedentary behaviour measured using ActiGraph100cpm and activPAL and self-rated health (panel A), body mass index (panel B), difficulty walking half mile (panel C), and cognitive impairment (panel D).

Note. ORAP indicates the odds ratio and 95% CI for associations with activPAL-measured variables, ORAG indicates the odds ratio and 95% CI for associations with ActiGraph-measured variables. All models are adjusted for age, sex, race/ethnicity, and education. The p-valueHorton is from models testing the hypothesis that ORAG and ORAP are different. *Odds ratios are unstandardized, meaning that for activPAL and ActiGraph100cpm, they are presented for the same increments for each sedentary behaviour metric: 116 min/day for total sedentary time; 136 min/day for time in 30+ min bouts; 6.6 min for mean bout duration; 17.8 for usual bout duration; 12.8 for breaks in sedentary behaviour; and 0.04 for alpha.

DISCUSSION

In this investigation of sedentary behaviour metrics computed using data from concurrently worn ActiGraph and activPAL devices among older adults, agreement was generally highest for the two metrics that relied most heavily on the volume of sedentary behaviour: total sedentary time and time in 30+ minute bouts. Our data showed that the ActiGraph100cpm detected significantly more breaks in sedentary time than did the activPAL, which resulted in shorter sedentary bouts, on average. Estimates of ActiGraph100cpm metrics thus reflected more frequently interrupted sedentary time accumulation than activPAL estimates, as shown by significantly shorter mean bout duration and usual bout duration, and higher alpha. Studies that rely on ActiGraph100cpm estimates of sedentary behaviour patterns (e.g., descriptive epidemiology studies, dose response analyses) should take these between-device differences into account, recognizing that ActiGraph100cpm-derived metrics systematically misestimate sedentary behaviour patterns. This will be particularly important for researchers working to establish sedentary behaviour guidelines since the values that might be considered “too sedentary” or “too prolonged of a sedentary behaviour pattern” will differ depending on the device used. More studies are needed to quantify the between-device differences and their potential impact on future recommendations. Notably, the approximate 1-hour per day difference in total sedentary time and time in 30+ minute bouts observed in this study highlight that the between-device differences are large enough that they could have important public health implications.

On the other hand, even with the between-device differences in estimates of sedentary behaviour volume and patterns, we observed moderate-to-high linear relationships between sedentary behaviour metrics measured by ActiGraph100cpm and activPAL (except for breaks in sedentary time and alpha) which suggest that metrics from either device similarly ranked participants according to high vs. low sedentary time, and according to prolonged vs. frequently interrupted sedentary behaviour patterns. And when the metrics were first standardized, inferences of associations between sedentary behaviour and health outcomes were not appreciably different for ActiGraph100cpm vs. activPAL measures. Associations have three main components: direction, magnitude, and statistical significance. ActiGraph100cpm and activPAL similarities in the magnitude of associations were most apparent when sedentary behaviour metrics were standardized (2 associations out of 24 were significantly different) compared to when they were unstandardized (9 associations out of 24 were significantly different). Accuracy of the magnitude of associations are important because they enable more accurate computation of population attributable risk (or population attributable fraction), which are needed for efficiently allocating public health and research-related resources. Additionally and importantly, in all but one test, the statistical significance and the direction of associations were consistent between ActiGraph100cpm and activPAL measures. Taken together, when interpreting associations using standardized (not unstandardized) metrics, the two devices resulted in generally similar inferences about the relationship of sedentary behaviour with physical and cognitive health.

Similar findings from other studies

As far as we know, the only other studies that evaluate agreement of ActiGraph100cpm and activPAL measures of sedentary behaviour patterns among adults focus on the metric ‘breaks in sedentary time’. Breaks in sedentary time are the main component for measuring how sitting and sedentary time are accumulated because they define the beginning and end of each sedentary bout, and sedentary bouts are the underlying input for all sedentary behaviour pattern metrics. In the present study, we observed that significantly more breaks were counted by ActiGraph100cpm compared to activPAL devices, leading to a high mean error and low concordance in the number of breaks measured per day, but a modest Pearson correlation coefficient of 0.51. These first findings among older adults, who accumulate sitting time differently than younger adults, (Diaz et al., 2016) corroborate results from the previous studies discussed below that were conducted among adults (Barreira et al., 2015; Lyden et al., 2012) and children (Carlson et al., 2019).

Lyden et al directly observed 13 participants between 20 and 60 years old (mean±SD age 25±5 years) on two 10-hour occasions, one during normal living, and another during a time when they were asked to reduce their sitting and to break up sedentary time more frequently (Lyden et al., 2012). The authors found that activPAL data provided accurate estimates of the number of breaks compared to direct observation. However, ActiGraph data, processed using the 100 counts per minute (cpm) cut-point, overestimated breaks by more than two-fold. Despite the differences in accuracy, the correlation between ActiGraph100cpm and direct observation was 0.86 during normal living and 0.64 during the treatment condition.

Barreira and colleagues compared breaks measured by activPAL to those measured by ActiGraph among 15 participants aged 28±3 years. The number of ActiGraph100cpm breaks were, on average, almost two times (90%) higher than activPAL breaks (Barreira et al., 2015). Most of the additional ActiGraph100cpm breaks (52%) occurred when, according to the activPAL, participants were sitting. This demonstrates one source of error when measuring breaks in sedentary time using ActiGraph100cpm: long sitting bouts are artificially broken into several shorter bouts when enough movement occurs (e.g., fidgeting or wiggling) that the 100cpm threshold is breached. Barreira et al. also reported that 40% of the additional ActiGraph100cpm breaks occurred while participants were standing. This demonstrates a second source of error when measuring breaks in sedentary time using ActiGraph100cpm. ActiGraph100cpm breaks reflect a transition from low levels of movement (i.e., movement below 100cpm) to relatively more movement, and does not capture transitions in posture from sitting to standing. Therefore, ActiGraph100cpm breaks can occur when a participant is standing with low level of movement then transitions to a higher level of movement (e.g., walking), or when they are engaging in a high level of movement and then stand sufficiently still before going back to a high level of movement. Since this source of error will occur more often among people who stand more and standing has been associated with beneficial health outcomes (Katzmarzyk, 2014; Winkler et al., 2018), this might explain why, in the present study, we observed marginally stronger associations with physical health outcomes for standardized ActiGraph100cpm–measured breaks in sedentary time and alpha.

Among children, the most common accelerometer data processing protocol is to classify sedentary time using a 25 counts/15-second cut-point. Using that protocol among 195 children aged 10.5 (SD = 0.7) years, Carlson and colleagues reported that ActiGraph25counts detected a 307% mean absolute increase in the number of breaks compared to activPAL (Carlson et al., 2019). While they reported that the respective mean absolute increase was just 25% when using ActiGraph100cpm, the interclass correlation between ActiGraph100cpm and activPAL breaks was just 0.31. A deep dive into data revealed that despite similar averages (ActiGraph100cpm = 89.9 per day; activPAL = 82.8 per day) the timing of the breaks was very different between the devices, with the activPAL detecting breaks that went undetected by ActiGraph100cpm, and ActiGraph100cpm detecting breaks that went undetected by the activPAL. Deviation in the timing of transitions resulted in a 57% and a 59% mean difference in usual bout duration and alpha, respectively, and similar mean differences were observed in the present study.

Similar to two smaller studies of agreement in measuring total sedentary time, ActiGraph100cpm overestimated measures of daily sedentary time—by 55 min/d in the present study compared to 107 min/d among 37 adults aged 74±7 years (Aguilar-Farías et al., 2014) and to 114 min/d among 62 adults aged 78±6 years (Koster et al., 2016). Like in the two previously mentioned studies, when alternate cutpoints were used, agreement with activPAL in total sedentary time improved and was underestimated by 45 min/d for ActiGraph200cpm and 27 min/d ActiGraph25cpm.

Strengths and Limitations

This study has several strengths. Data were collected from a large community-based cohort study of older men and women. We included 5 different measures of sedentary behaviour patterns since there remains a lack of consensus concerning which metric might be best. Additionally, our criterion measure (activPAL) has been shown to be as good as direct observation for identifying breaks in sedentary behaviour, which is requisite for measuring sedentary behaviour patterns.

As in all studies, our results should be interpreted within the context of study limitations. Generalizability would be enhanced by introducing more racial/ethnic diversity and by including more geographic variability. Our ActiGraph sedentary behaviour metrics were derived from data processed with the normal filter within ActiLife, as we do not yet have data processed using the low frequency extension filter. However, at least one previous study among adults showed that agreement between sedentary breaks measured using ActiGraph100cpm and direct observation is nearly the same when data were processed using the low frequency extension filter or the normal frequency filter (Lyden et al., 2012). The Choi algorithm classifies periods of 90 minutes or more with zero movement (and requiring a 30-minute streamframe and 2-minute tolerance) as non-wear time. The 90-minute duration and streamframe criteria were developed specifically for older adults and help prevent misclassifying sedentary time as non-wear time (Choi et al., 2012). However, it is possible that differences in sedentary behaviour metrics between devices could have been impacted by using this data processing protocol, which is the most commonly used protocol among studies of older adults (Migueles et al., 2017). To test this, we repeated all analyses among the 596 adults who had at 4 or more days without Choi-identified non-wear time and the results were not appreciably changed (data not shown). While the ActiGraph GT3X+ is no longer sold by ActiGraph, the cross-generational accuracy with the currently available model (w-GT3X-BT) is exceptionally high (Miller, 2015). Finally, our study used data from the activPAL which has very low error in comparison with gold standard direct observation (Lyden et al., 2012), but is not without error.

Conclusion

This study showed that using ActiGraph100cpm led to an overestimation of breaks in sedentary time and systematically more interrupted sedentary behaviour patterns than were observed using activPAL measures. For example, mean bout duration as measured by activPAL was 66% longer, on average, than duration as measured by ActiGraph100cpm and breaks in sedentary time measured by ActiGraph100cpm was nearly twice as large. As a result, caution should be used when interpreting estimates of sedentary behaviour patterns from ActiGraph100cpm and any associations that rely on those pattern metric estimates. When our metrics were first standardized based on their underlying distribution—we used the standard deviation, but an interquartile range or comparing quartiles of the exposure are also distribution-based methods that would work—associations with physical and cognitive health using ActiGraph100cpm and activPAL were similar when the associations were interpreted using the distribution-based unit of analysis but not when the distribution based units were converted back to the absolute unit of analysis. For example, we strongly recommend reporting results for a 1-standard deviation increment in mean bout duration instead of an x-minute increment in mean bout duration, where x is the standard deviation of the mean bout duration. Whenever estimates of associations are presented using absolute units of sedentary behaviour pattern measures (e.g., minute, minute/day, n/day) the magnitude of the association will differ according to which device is used in determining those pattern measures. Consequently, if physical activity guidelines were developed using research conducted only by ActiGraph100cpm, the guidelines could be meaningfully different than if the same research were conducted using only activPAL. Future development of sedentary behaviour guidelines will rely on accelerometer data from longitudinal studies. Since many studies have already collected hip-worn ActiGraph data and many new studies will likely continue to employ the ActiGraph, research is needed to develop accelerometer data processing techniques (Kerr et al., 2018) or measurement error correction models (Sampson, Matthews, Freedman, Carroll, & Kipnis, 2016) to increase the accuracy of sedentary behaviour pattern metrics derived from hip-worn devices. Until then, for studies assessing associations of sedentary behaviour volumes and patterns with health using ActiGraph100cpm, we recommend presenting results for standardized and unstandardized point estimates in ActiGraph100cpm based studies and interpreting associations using units of the standardized metric rather than absolute units of sedentary behaviour measures (e.g., minute, minute/day, n/day). This will increase accuracy of the relative magnitude of associations between sedentary behaviour and health and might help harmonize results from prospective cohort studies that use different activity monitors. Ultimately, such an approach will help to answer the 2018 Physical Activity Guidelines Advisory Committee’s call to provide more data about associations of sedentary behaviour patterns, including bouts and breaks, with health outcomes (2018 Physical Activity Guidelines Advisory Committee, 2018).

Supplementary Material

Supplemental Table 1
Supplemental Figure 1
Supplemental Figure 2

Acknowledgments

We have immense gratitude for the volunteers who took part in the Adult Changes in Act Study. This work was funded by the National Institute on Aging (U01 AG006781; DR and P01 AG052352; AZL) and the National Institute of Diabetes and Digestive and Kidney Diseases (R01 DK114945; LN). NR is supported by a Future Leader Fellowship from the National Heart Foundation of Australia (ID101895). The funders had no role in the design, conduct, analysis, and decision to publish results from this study.

Contributor Information

Jordan A. Carlson, Center for Children’s Healthy Lifestyles and Nutrition; Children’s Mercy Hospital; Kansas City, MO, USA

Nicola D. Ridgers, Deakin University, Geelong, Australia, Institute for Physical Activity and Nutrition, School of Exercise and Nutrition Sciences

Sandy Liles, Department of Family Medicine and Public Health, University of California San Diego.

Andrea Z. LaCroix, Department of Family Medicine and Public Health, University of California San Diego

Marta M. Jankowska, Qualcomm Institute/Calit2, University of California San Diego

Dori E. Rosenberg, Kaiser Permanente Washington Health Research Institute.

Loki Natarajan, Department of Family Medicine and Public Health, University of California San Diego.

References

  1. Aguilar-Farías N, Brown WJ, & Peeters GMEEG (2014). ActiGraph GT3X+ cut-points for identifying sedentary behaviour in older adults in free-living environments. Journal of Science and Medicine in Sport, 17(3), 293–299. 10.1016/j.jsams.2013.07.002 [DOI] [PubMed] [Google Scholar]
  2. Barreira TV, Zderic TW, Schuna JM, Hamilton MT, & Tudor-Locke C (2015). Free-living activity counts-derived breaks in sedentary time: Are they real transitions from sitting to standing? Gait & Posture, 42(1), 70–72. 10.1016/j.gaitpost.2015.04.008 [DOI] [PubMed] [Google Scholar]
  3. Bellettiere J, LaMonte MJ, Evenson KR, Rillamas-Sun E, Kerr J, Lee I-M, … LaCroix AZ (2019). Sedentary Behavior and Cardiovascular Disease in Older Women: The OPACH Study. Circulation, 139(8), 1036–1046. 10.1161/CIRCULATIONAHA.118.035312 [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Bellettiere J, Winkler EAH, Chastin SFM, Kerr J, Owen N, Dunstan DW, & Healy GN (2017). Associations of sitting accumulation patterns with cardio-metabolic risk biomarkers in Australian adults. PLoS ONE, 12(6), 1–17. 10.1371/journal.pone.0180119 [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Bland JM, & Altman DG (2007). Agreement between methods of measurement with multiple observations per individual. Journal of Biopharmaceutical Statistics, 17(4), 571–582. 10.1080/10543400701329422 [DOI] [PubMed] [Google Scholar]
  6. Brocklebank LA, Falconer CL, Page AS, Perry R, & Cooper AR (2015). Accelerometer-measured sedentary time and cardiometabolic biomarkers: A systematic review. Preventive Medicine, 76, 92–102. 10.1016/j.ypmed.2015.04.013 [DOI] [PubMed] [Google Scholar]
  7. Carlson JA, Bellettiere J, Kerr J, Salmon J, Timperio A, Verswijveren SJJM, & Ridgers ND (2019). Day-level sedentary pattern estimates derived from hip-worn accelerometer cut-points in 8–12-year-olds: Do they reflect postural transitions? Journal of Sports Sciences, 37(16), 1899–1909. 10.1080/02640414.2019.1605646 [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Chastin SFM, & Granat MH (2010). Methods for objective measure, quantification and analysis of sedentary behaviour and inactivity. Gait and Posture, 31(1), 82–86. [DOI] [PubMed] [Google Scholar]
  9. Chastin SFM, Winkler EAH, Eakin EG, Gardiner PA, Dunstan DW, Owen N, & Healy GN (2015). Sensitivity to change of objectively-derived measures of sedentary behavior. Measurement in Physical Education and Exercise Science, 19(3), 138–147. 10.1080/1091367X.2015.1050592 [DOI] [Google Scholar]
  10. Chiu EC, Yip PK, Woo P, & Lin Y Te. (2019). Test-retest reliability and minimal detectable change of the Cognitive Abilities Screening Instrument in patients with dementia. PLoS ONE, 14(5), e0216450. 10.1371/JOURNAL.PONE.0216450 [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Choi L, Liu Z, Matthews CE, & Buchowski MS (2011). Validation of accelerometer wear and nonwear time classification algorithm. Medicine and Science in Sports and Exercise, 43(2), 357–364. 10.1249/MSS.0b013e3181ed61a3.Validation [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Choi L, Ward SC, Schnelle JF, & Buchowski MS (2012). Assessment of wear/nonwear time classification algorithms for triaxial accelerometer. Medicine and Science in Sports and Exercise, 44(10), 2009–2016. 10.1249/MSS.0b013e318258cb36 [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. De Oliveira GM, Yokomizo JE, E Silva LDSV, Saran LF, Bottino CMC, & Yassuda MS (2016). The applicability of the cognitive abilities screening instrument-short (CASI-S) in primary care in Brazil. International Psychogeriatrics, 28(1), 93–99. 10.1017/S1041610215000642 [DOI] [PubMed] [Google Scholar]
  14. Diaz KM, Goldsmith J, Greenlee H, Strizich G, Qi Q, Mossavar-Rahmani Y, … Kaplan R (2017). Prolonged, uninterrupted sedentary behavior and glycemic biomarkers among US Hispanic/Latino adults : The Hispanic Community Health Study/Study of Latinos (HCHS/SOL). Circulation, 136(15), 1362–1373. 10.1161/CIRCULATIONAHA.116.026858 [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Diaz KM, Howard VJ, Hutto B, Colabianchi N, Vena JE, Blair SN, & Hooker SP (2016). Patterns of Sedentary Behavior in US Middle-Age and Older Adults: The REGARDS Study. Medicine and Science in Sports and Exercise, 48(3), 430–438. 10.1249/MSS.0000000000000792 [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Diaz KM, Howard VJ, Hutto B, Colabianchi N, Vena JE, Safford MM, … Hooker SP (2017). Patterns of sedentary behavior and mortality in U.S. middle-aged and older adults a national cohort study. Annals of Internal Medicine, 167(7), 465–475. 10.7326/M17-0212 [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Du Y, Liu B, Sun Y, Snetselaar LG, Wallace RB, & Bao W (2019). Trends in Adherence to the Physical Activity Guidelines for Americans for Aerobic Activity and Time Spent on Sedentary Behavior among US Adults, 2007 to 2016. JAMA Network Open, 2(7), 1–11. 10.1001/jamanetworkopen.2019.7597 [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Edwardson CL, Winkler EAH, Bodicoat DH, Yates T, Davies MJ, Dunstan DW, & Healy GN (2016). Considerations when using the activPAL monitor in field based research with adult populations. Journal of Sport and Health Science, (February). 10.1016/j.jshs.2016.02.002 [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Glazer NL, Lyass A, Esliger DW, Blease SJ, Freedson PS, Massaro JM, … Vasan RS (2013). Sustained and shorter bouts of physical activity are related to cardiovascular health. Medicine and Science in Sports and Exercise, 45(1 NV-1), 109–115. [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Horton NJ, & Fitzmaurice GM (2004). Regression analysis of multiple source and multiple informant data from complex survey samples. Statistics in Medicine, 23(18), 2911–2933. 10.1002/sim.1879 [DOI] [PubMed] [Google Scholar]
  21. Jefferis BJ, Parsons TJ, Sartini C, Ash S, Lennon LT, Papacosta O, … Whincup PH (2019). Objectively measured physical activity, sedentary behaviour and all-cause mortality in older men: does volume of activity matter more than pattern of accumulation? British Journal of Sports Medicine, 53(16), 1013–1020. [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Jefferis BJ, Sartini C, Shiroma E, Whincup PH, Wannamethee SG, & Lee IM (2015). Duration and breaks in sedentary behaviour: Accelerometer data from 1566 community-dwelling older men (British Regional Heart Study). British Journal of Sports Medicine, 49(24), 1591–1594. 10.1136/bjsports-2014-093514 [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Katzmarzyk PT (2014). Standing and mortality in a prospective cohort of canadian Adults. Medicine and Science in Sports and Exercise, 46(5), 940–946. 10.1249/MSS.0000000000000198 [DOI] [PubMed] [Google Scholar]
  24. Katzmarzyk PT, Powell KE, Jakicic JM, Troiano RP, Piercy K, & Tennant B (2019). Sedentary Behavior and Health: Update from the 2018 Physical Activity Guidelines Advisory Committee. Medicine and Science in Sports and Exercise, 51(6), 1227–1241. 10.1249/MSS.0000000000001935 [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. Koster A, Shiroma E, Caserotti P, Mathews C, Chen K, Glynn N, & Harris T (2016). Comparison of Sedentary Estimates between activPAL and Hip- and Wrist-worn ActiGraph. In Medicine & Science in Sports & Exercise (Vol. 48). 10.1249/MSS.0000000000000924 [DOI] [PMC free article] [PubMed] [Google Scholar]
  26. Lee I-M, & Shiroma EJ (2014). Using accelerometers to measure physical activity in large-scale epidemiological studies: issues and challenges. British Journal of Sports Medicine, 48(3), 197–201. [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. Lord S, Chastin SFM, McInnes L, Little L, Briggs P, & Rochester L (2011). Exploring patterns of daily physical activity and sedentary behaviour in community-dwelling older adults. Age and Ageing, 40, 205–210. [DOI] [PubMed] [Google Scholar]
  28. Lyden K, Kozey Keadle SL, Staudenmayer JW, & Freedson PS (2012). Validity of two wearable monitors to estimate breaks from sedentary time. Medicine and Science in Sports and Exercise, 44(11), 2243–2252. 10.1249/MSS.0b013e318260c477 [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Matthews CE, George SM, Moore SC, Bowles HR, Blair A, Park Y, … Schatzkin A (2012). Amount of time spent in sedentary behaviors and cause-specific mortality in US adults. American Journal of Clinical Nutrition, 95(2), 437–445. 10.3945/ajcn.111.019620 [DOI] [PMC free article] [PubMed] [Google Scholar]
  30. Migueles JH, Cadenas-Sanchez C, Ekelund U, Nyström CD, Mora-Gonzalez J, Löf M, … Ruiz Jonatan R. Ortega FB (2017). Accelerometer Data Collection and Processing Criteria to Assess Physical Activity and Other Outcomes: A Systematic Review and Practical Considerations. Sports Medicine, 47(9), 1821–1845. 10.1007/s40279-017-0716-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Miller J (2015). ActiGraph Cross-Generation Accuracy Confirmation. Retrieved November 12, 2020, from https://s3.amazonaws.com/actigraphcorp.com/wp-content/uploads/2018/03/11163743/cross-generation-accuracy.pdf
  32. Rosenberg D, Walker R, Greenwood-Hickman MA, Bellettiere J, Xiang Y, Richmire K, … Lacroix AZ (2020). Device-assessed physical activity and sedentary behavior in a community-based cohort of older adults. BMC Public Health, Under Revi. [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. Teng EL, Hasegawa K, Homma A, Imai Y, Larson E, Graves A, … White LR (1994). The Cognitive Abilities Screening Instrument (CASI): A Practical Test for Cross-Cultural Epidemiological Studies of Dementia. International Psychogeriatrics, 6(1), 45–58. 10.1017/S1041610294001602 [DOI] [PubMed] [Google Scholar]
  34. Tremblay MS, Aubert S, Barnes JD, Saunders TJ, Carson V, Latimer-Cheung AE, … Participants, on behalf of S. T. C. P. (2017). Sedentary Behavior Research Network (SBRN) – Terminology Consensus Project process and outcome. International Journal of Behavioral Nutrition and Physical Activity, 14(75). 10.1186/s12966-017-0525-8 [DOI] [PMC free article] [PubMed] [Google Scholar]
  35. U.S. Department of Health and Human Services. (2018). Physical Activity Guidelines for Americans, 2nd edition. Retrieved from U.S. Department of Health and Human Services website: https://health.gov/paguidelines/second-edition/report/ [Google Scholar]
  36. Ware CJE, & Sherbourn D (1992). The MOS 36-Item Short-Form Health Survey (SF-36): I. Conceptual Framework and Item Selection. Medical Care, 30(6), 473–483. [PubMed] [Google Scholar]
  37. Ware JE (2000). SF-36 Health Survey update. Spine, 25(24), 3130–3139. 10.1097/00007632-200012150-00008 [DOI] [PubMed] [Google Scholar]
  38. Winkler EAH, Chastin S, Eakin EG, Owen N, Lamontagne AD, Moodie M, … Healy GN (2018). Cardiometabolic Impact of Changing Sitting, Standing, and Stepping in the Workplace. In Medicine and Science in Sports and Exercise (Vol. 50). 10.1249/MSS.0000000000001453 [DOI] [PubMed] [Google Scholar]
  39. Yang L, Cao C, Kantor ED, Nguyen LH, Zheng X, Park Y, … Cao Y (2019). Trends in Sedentary Behavior Among the US Population, 2001–2016. Jama, 321(16), 1587. 10.1001/jama.2019.3636 [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplemental Table 1
NIHMS1686095-supplement-Supplemental_Table_1.pdf (202.1KB, pdf)
Supplemental Figure 1
NIHMS1686095-supplement-Supplemental_Figure_1.pdf (625.9KB, pdf)
Supplemental Figure 2
NIHMS1686095-supplement-Supplemental_Figure_2.pdf (799.1KB, pdf)

RESOURCES