Modifying Accelerometer Cut-Points Affects Criterion Validity in Simulated Free-Living for Adolescents and Adults

Abstract

PURPOSE:

To assess changes in criterion validity when modifying cut-points for use in different epoch lengths.

METHOD:

Simulated free-living data came from 42 adolescents (2-hr each) and 29 adults (6-hr each) wearing a hip-worn accelerometer and portable indirect calorimeter (Cosmed K4b²). K4b² data were classified as sedentary behavior (SB), light physical activity (LPA), or moderate-to-vigorous physical activity (MVPA), and compared to estimates from accelerometer data processed with three youth and three adult cut-points in six epoch lengths (1, 5, 10, 15, 30, and 60-s). A cut-point of 100 counts per minute was used for all SB estimates.

RESULTS:

For both adolescents and adults, SB estimates in all but 60-s epochs were significantly higher than the criterion, by 18.4%-78.4% (all p<0.02). CPS had varied effects on youth LPA, ranging from favorable effects for one cut-point (1.9% underestimation in 1-s epochs, versus 40.2% overestimation in the originally-calibrated epoch length; p<0.01 and p=0.91, respectively) to unfavorable effects for another (41.8% underestimation in 1-s epochs, versus 9.8% underestimation in the originally-calibrated epoch length; p<0.01 and p=0.39, respectively). Adult LPA estimates in 30-s or 60-s epochs were closest to the criterion (within 5.2%-37.3%, p=0.0001-0.49). Youth MVPA estimates in 60-s epochs were closest to the criterion (within 9.5%-53.2%, all p<0.05), whereas adult MVPA estimates in 1-s epochs were closest to the criterion (within 6.6%-34.2%, p=0.02-0.59).

CONCLUSION:

Cut-point modification is not universally beneficial, and thus it is not recommended.

Accelerometer cut-points are used to estimate time spent in different activity intensity categories, such as sedentary behavior (SB), light physical activity (LPA), and moderate-to-vigorous physical activity (MVPA). Cut-points can be developed using regression or receiver operating characteristic analysis (Welk, 2005), but in both cases the thresholds are applicable to only one epoch length, i.e., the calibration epoch length (often 60-s). To allow physical activity intensity to be measured over shorter time periods, researchers often scale cut-points to fit different epoch lengths. For example, the MVPA cut-point of 1500 counts per 30-s (Treuth et al., 2004) might be scaled to 500 counts per 10-s, via division by three. Hereafter, this practice is referred to as cut-point scaling (CPS).

The rationale for CPS is connected to the observation that sporadic activity patterns are common in certain populations, such as youth (Freedson, Pober, & Janz, 2005; Nilsson, Ekelund, Yngve, & Söström, 2002). In such populations, a long epoch could primarily consist of motionless data that would mask the presence of a short burst of activity (Mcclain, Abraham, Brusseau, & Tudor-Locke, 2008; Nilsson et al., 2002). Thus, using a shorter epoch length via CPS could provide a way to more effectively capture short activity bouts, ultimately leading to less artifact in the assessment. However, the assumed benefits of CPS have not been rigorously tested using a criterion measure.

Among studies with no criterion measure, CPS has consistently caused accelerometer-derived activity estimates to change. However, the patterns have not been consistent. In youth, the majority of studies have shown increased MVPA in shorter epochs (Banda et al., 2016; Kim, Beets, Pate, & Blair, 2013; Logan, Duncan, Harris, Hinckson, & Schofield, 2016; Nilsson et al., 2002; Obeid, Nguyen, Gabel, & Timmons, 2011; Vale, Santos, Silva, Soares-Miranda, & Mota, 2009; Vanderloo, Di Cristofaro, Proudfoot, Tucker, & Timmons, 2016). However, some studies have shown the opposite (Aibar & Chanal, 2015), or mixed results depending on levels of several independent variables (Edwardson & Gorely, 2010; Sanders, Cliff, & Lonsdale, 2014; Nettlefold et al., 2016). For adults, CPS has received considerably less attention, but two previous studies have both shown increased MVPA in shorter epochs (Gabriel et al., 2010; Orme et al., 2014). Taken together, previous research has clearly shown that CPS causes activity estimates to change, but it is impossible to tell whether CPS causes activity estimates to improve, due to the lack of data from criterion measures. It is crucial to address CPS from a criterion validity perspective, so that recommendations about its use can be made based on evidence, and not based on assumptions or preferences.

Only one study (Mcclain et al., 2008) has used a criterion measure (direct observation) to evaluate CPS, drawing from a sample of children who were assessed during physical education classes. Three sets of cut-points were tested, i.e., the cut-points of Treuth et al. (2004), Freedson et al. (2005), and Mattocks et al. (2007). Time spent in MVPA was the only outcome of the study, and the overall finding was that CPS led to improved individual-level estimates for all three cut-points that were tested, whereas group-level estimates were only improved (i.e., more MVPA in shorter epochs) for two of the three (Mcclain et al., 2008). Although physical education classes represent an important application for CPS (Aibar & Chanal, 2015; Mcclain et al., 2008; Sanders et al., 2014), it is also important to understand how CPS affects criterion validity when applied to data from extended protocols (e.g. one week in the National Health and Nutrition Examination Survey). To address the latter issue, free-living assessments are needed so that data from a broader range of behaviors can be examined. Furthermore, it is essential to look at how criterion validity changes in different intensity categories (i.e., SB, LPA, and MVPA), since improvement in one category may be compensated by poorer performance in another. Lastly, it should be explored whether CPS affects criterion validity for youth and adults in a consistent way.

Purpose

The purpose of this study was to demonstrate the practical impact of CPS on criterion validity in simulated free-living for adolescents and adults, using a criterion measure of indirect calorimetry.

Method

Data for the present study came from two previous investigations, one in adolescents (Crouter, Horton, & Bassett, 2013) and one in adults (Crouter, DellaValle, Haas, Frongillo, & Bassett, 2013). The present study was focused on discerning the applied consequences of using CPS. Thus, only simulated free-living data were used from either investigation. The specific methods for each study are summarized below, followed by a description of the analyses performed for the present study.

In the adolescent study (Crouter, Horton, et al., 2013), the participants were a convenience sample of 27 boys and 15 girls between 11 and 15 years old (mean ± SD 12.6 ± 0.8 years). More than half of the participants (55%) were overweight or obese. Written parental informed consent and adolescent assent were separately obtained prior to participation, and the Institutional Review Boards of the university and the public school system approved the study.

Testing took place during three separate sessions. The first session involved measuring anthropometrics and resting metabolic rate. The second and third sessions involved structured activity and simulated free-living activity, respectively. The present study does not include data from the second session, to maintain the previously-described focus on the applied consequences of CPS.

The duration of the simulated free-living session was two hours. During this period, all participants wore a Cosmed K4b² portable metabolic analyzer and an ActiGraph accelerometer (model GT3X+ or GT3X; ActiGraph, LLC, Pensacola, FL). The ActiGraph was worn on an elastic belt, and positioned on the right hip along the anterior axillary line at the level of the iliac crest. Participants were given free rein to select the activities they performed, except they could not leave the testing site. Diverse activities were available, ranging from SB (e.g. watching movies or reading books) to MVPA (e.g. active video games or sports). More detail is available in Crouter, Horton, et al. (2013). Participants were supervised at all times, but not interfered with.

In the adult study (Crouter, DellaValle, et al., 2013), the participants were a convenience sample of 12 males and 17 females, from 20-59 years of age (mean ± SD 37.7 ± 11.7 years). Nearly half of the participants (45%) were overweight or obese. Written informed consent was obtained prior to participation, and the appropriate Institutional Review Boards approved the protocol. Anthropometric measurements were taken, after which participants wore a Cosmed K4b² portable metabolic analyzer and an ActiGraph GT1M during 5-6 hours of simulated free-living activity. The GT1M was worn on an elastic belt, and positioned on the right hip along the anterior axillary line, at the level of the iliac crest. A total of 32 observations were completed in workplace (n = 23) or leisure time (n = 9) settings, with three participants being observed in both settings. Participants performed activities that ranged in intensity from SB to MVPA. Examples included seated office work and lifting and carrying boxes (in the workplace assessments), as well as driving a vehicle and yard work (in the leisure time assessments). More detail is available in Crouter, DellaValle, et al. (2013).

Criterion Measure of Activity Intensity

A Cosmed K4b² (Cosmed, Rome, Italy) was used in both studies for measurement of oxygen consumption (VO₂, in ml·kg⁻¹·min⁻¹) and carbon dioxide production. The K4b² was calibrated prior to each test, according to the manufacturer’s specifications that included a room air calibration, gas calibration using known gas concentrations, volume calibration, and a delay calibration. Participant food intake was not controlled prior to the test. Breath-by-breath data were downloaded from the K4b² and converted to one-minute averages of VO₂, which were then converted to metabolic equivalents (METs) for adolescents ($METs=\frac{{\mathrm{Measured}\mathrm{Activity}\mathrm{VO}}_2}{{\mathrm{Measured}\mathrm{Resting}\mathrm{VO}}_2}$) and adults ($METs=\frac{{\mathrm{Measured}\mathrm{Activity}\mathrm{VO}}_2}{3.5\mathrm{ml}\cdot k{g}^{-1}\cdot mi{n}^{-1}}$). Finally, each minute was classified as SB (≤ 1.5 METs), LPA (1.6-2.9 METs), or MVPA (METs ≥ 3.0). Total time in each category was calculated after excluding periods when the mask was not worn (e.g. for a water break).

Accelerometer Data Processing and Cut-Points

The GT3X+ monitors were initialized to collect raw data at a sampling rate of 30 Hz. After downloading the raw data, the low frequency extension was applied and the data were transformed to activity counts in one-second epochs. The GT3X and GT1M do not store raw data by default, and thus the data format was pre-programmed during initialization, such that data were stored in one-second epochs with the low frequency extension applied. Previously, Robusto and Trost (2012) compared count values and activity estimates from the GT1M, GT3X, and GT3X+, concluding that the three monitors can be used interchangeably. Furthermore, Cain, Conway, Adams, Husak, and Sallis (2013) showed that using the low frequency extension led to excellent backward compatibility between the GT3X+ and the uniaxial ActiGraph 7164, which was the original instrument used to calibrate all cut-points used in this study. Thus, the present methodology is not meaningfully influenced by differences between ActiGraph generations.

To obtain estimates from the ActiGraph, each participant’s data were evaluated using three sets of cut-points and six epoch lengths (the latter being obtained via reintegration of the one-second epoch data). Thus, there were a total of 18 predictions per intensity level for each participant. The cut-points were selected based on frequency of use among studies with child, adolescent, or adult participants. In order to provide the most relevant examples, the three most-used options (see Migueles et al., 2017) were selected for both age groups. For adolescents, the three sets of cut-points were developed by Treuth et al. (2004), Freedson et al. (2005), and Evenson, Catellier, Gill, Ondrak, & McMurray (2008). The three sets of cut-points for adults were developed by Freedson, Melanson, & Sirard (1998), Matthews (2005), and Troiano et al. (2008). The six epoch lengths were one, five, 10, 15, 30, and 60-s. All cut-points had a calibration epoch length of either 15, 30, or 60-s, meaning each cut-point was tested in its calibration epoch length, and scaled for use in the other five epoch lengths. All SB cut-points were based on the ubiquitous SB cut-point of 100 counts per minute (Matthews et al., 2008; Treuth et al., 2004). Total predicted time in SB, LPA, and MVPA was calculated for each cut-point and epoch length after excluding data from periods where the Cosmed was not worn. Notably, moderate and vigorous intensity were not assessed as separate categories, due to low adult engagement in vigorous activity, as well as the potential for estimates at high intensities to be distorted by the plateau phenomenon (John, Miller, Kozey-Keadle, Caldwell, & Freedson, 2012).

Table 1 shows each set of cut-points. Using the most common cut-points led to some differences in the applicability of each cut-point to this data, based on differences between the present methods (i.e., the participant characteristics and experimental protocol) and those in the original calibration studies. The differences are important to consider, and thus a brief description is given below for the calibration of each set of cut-points. However, it is important to emphasize that the goal of this study was not to validate the cut-points themselves, but rather to provide relevant examples of how criterion validity changes when CPS is used.

Table 1.

Cut-Points Applied to Data in the Present Study.

	Cut-point	CEL	EE Equation	Sedentary	Light	MVPA
Youth Study	Treuth (2004)	30	METs = 2.01 + 0.00171 (cp30s)	≤ 50	51-1499	≥ 1500
	Freedson (2005)	60	METs = 2.757 + (0.0015*cpm) – (0.08957*yr) – (0.000038*cpm*yr)	≤ 100	+	+
	Evenson (2008)	15	#	≤ 25	26-573	≥ 574
Adult Study	Freedson (1998)	60	METs = 1.439008 + (0.000795*cpm)	≤ 100	101-1951	≥ 1952
	Matthews (2005)	60	#	≤ 100	101-759	≥ 760
	Troiano (2008)	60	#	≤ 100	101-2019	≥ 2020

CEL, Calibration epoch length; EE, energy expenditure; METs, metabolic equivalents; cp30s, counts per 30 seconds; cpm, counts per minute; yr, age in years.

⁺Cut-points calculated on an individual basis by rearranging the EE equation to solve for age-specific moderate (3 METs) and vigorous (6 METs) cut-points, with light intensity defined as the range between the sedentary and moderate cut-points.

^#No EE equation presented in calibration study.

Note: All cut-points were developed using the ActiGraph 7164, and values correspond to vertical axis activity counts.

In the study by Treuth et al. (2004), diverse activities were performed by adolescent girls. The Treuth cut-points have frequently been used in mixed samples (boys and girls) to compare activity estimates from different cut-points (e.g. Crouter, Horton, et al., 2013; Kim et al., 2016; see Migueles et al., 2017 for a listing), including studies focused on epoch length (Banda et al., 2016; McClain et al., 2008). The cut-points were determined by testing values in manually-defined ranges (e.g. 1100-1600 counts per 30-s for the moderate PA threshold) and determining which candidate had the best sensitivity and specificity. The monitors were initialized to collect in 30-s epochs, but both 30-s and 60-s cut-points were given in the paper. For the present study, the 30-s cut-points were used for all calculations, except when processing data in 60-s epochs. In that case, the original 60-s cut-points were used instead of deriving new ones via CPS, although it made little difference (± one count) which approach was used.

In the study by Freedson et al. (2005) children and adolescents performed treadmill walking and running. Energy expenditure was regressed against activity counts, age, and an interaction term, as shown in Table 1. The regression equation is commonly used to develop age-specific cut-points based on common energy expenditure cutoffs (i.e., three metabolic equivalents for MVPA).

In the study by Evenson et al. (2008), children performed a diverse array of activities. Examples included quiet rest, bicycling, and jumping jacks. The cut-points were developed using receiver operating characteristic analysis.

The calibration protocol of Freedson et al. (1998) was similar to the one by Freedson et al. (2005), except with adult participants instead of children and adolescents. The participants completed three treadmill conditions, and the prediction equation regressed energy expenditure against activity counts. Cut-points for various intensities were given in the paper, based on rearranging the regression equation for common energy expenditure cutoffs (see Table 1).

In the study by Matthews (2005), secondary data analysis was applied to two data sets, with a goal of developing an improved cut-point for moderate PA (and thus MVPA). One data set was used to establish three candidate cut-points (575, 690, and 760 counts per minute) by manually examining activity count distributions during light and moderate activities. The other data set was used to assess which candidate performed the best, ultimately leading to the acceptance of the 760 counts per minute cut-point.

Finally, the cut-points of Troiano et al. (2008) were developed using a weighted average of previously developed cut-points, which were calibrated using only walking and running data. The Troiano cut-points were created for use on the accelerometer data from the National Health and Nutrition Examination Survey. Thus, they are particularly prominent cut-points that have been applied to a range of free-living data.

Statistical Analysis

The analyses in the present study compared measured and predicted time spent in SB, LPA, and MVPA. All valid minutes were used, with no bout length restrictions. Equivalence testing was initially used to compare group-level estimates from the criterion measure and the different epoch lengths. A key challenge was defining meaningful equivalence zones for SB, LPA, and MVPA, since there are currently no strong conventions. The selected zones (±6% for SB and LPA; ±8% for MVPA) were loosely based on daily time use recommendations, and there were no equivalent measures in any epoch length (see Appendix 1). Thus, the primary analyses were focused on characterizing the magnitude of the differences via one-way repeated measures ANOVAs (α = 0.05), with separate ANOVAs being run for each intensity category in adolescents and adults. Planned contrasts were used for follow-up testing (i.e., to compare each estimate to the criterion measure), and the false discovery rate was used to correct for multiple comparisons. To further assess group and individual level agreement, root mean square error (RMSE), mean bias, and 95% limits of agreement were examined. All statistical analysis was carried out in R.

Results

Predicted and measured minutes in SB, LPA, and MVPA are summarized in Figure 1 (group-level estimates) and Figure 2 (RMSE). Mean bias and limits of agreement are shown in Appendix 2 (Table A2.1). Below, results are first presented for SB, since all of the predictions used the same SB cut-point of 100 counts per minute (Matthews et al., 2008; Treuth et al., 2004). Afterwards, the results for LPA and MVPA are given sequentially for the various sets of cut-points, starting with adolescents and then adults.

Figure 1.

Predicted time spent in sedentary behavior (SB), light physical activity (LPA), and moderate-to-vigorous physical activity (MVPA) for adolescents (A-C, respectively) and adults (D-F, respectively), using different cut-points and epoch lengths. Horizontal dashed lines represent mean measured minutes from the Cosmed K4b², and asterisks denote significant difference from the measured values (p ? 0.05). Error bars are standard deviation, and downward arrows indicate the calibration epoch length.

Figure 2.

Root mean square error (RMSE) for predictions of sedentary behavior (SB), light physical activity (LPA), and moderate-to-vigorous physical activity (MVPA) for adolescents (A-C, respectively) and adults (D-F, respectively), using different cut-points and epoch lengths. Downward arrows indicate the calibration epoch length.

For SB in both adolescents and adults, the group-level estimate from 60-s epochs was closest to the K4b², while SB was significantly (all p < 0.02) overestimated in all other epoch lengths, by 9.0-23.6 minutes (or 29.7%−78.4% of the measured mean) in youth and by 30.8-94.5 minutes (or 18.4%−56.4% of the measured mean) in adults. At the individual level, 60-s epochs also had the lowest RMSE for adolescents (21.2 minutes) and adults (47.6 minutes). The highest RMSE occurred for 1-s epochs in both groups (31.7 minutes and 117.0 minutes for adolescents and adults, respectively).

Adolescent Cut-Points

When using the Treuth cut-points (calibration epoch length of 30 seconds), LPA was significantly overestimated in epochs longer than 15-s, by 10.9-13.5 minutes (or 40.2%-49.4% of the measured mean; all p < 0.01). MVPA was significantly underestimated in all epoch lengths, by 19.6-23.1 minutes (or 53.2%-62.7% of the measured mean; all p < 0.001). At the individual level, RMSE for LPA was similar for all epoch lengths (26.2-28.9 minutes), whereas for MVPA the RMSE was lowest for 60-second epochs (25.8 minutes) and highest for 1-s epochs (32.5 minutes).

When using the Freedson cut-points (calibration epoch length of 60 seconds), LPA was significantly underestimated in epochs shorter than 15 s, by 7.1-11.4 minutes (or 25.9%-41.8% of the measured mean; all p < 0.05). MVPA was significantly underestimated in all epoch lengths, by 3.5-12.2 minutes (or 9.5%-33.2% of the measured mean; all p < 0.05). At the individual level, the calibration epoch length had the lowest RMSE for LPA (18.8 minutes) and MVPA (10.7 minutes), and one-second epochs had the highest RMSE for both intensities (27.7 minutes and 21.3 minutes for LPA and MVPA, respectively).

When using the Evenson cut-points (calibration epoch length of 15 seconds) to predict LPA, there were no significant differences from the K4b² for any epoch length, with the smallest difference being underestimation in 5 s epochs (by 0.5 minutes or 1.8% of the measured mean; p = 0.90) and the largest being overestimation in 60 s epochs (by 6.7 minutes or 24.5% of the measured mean; p = 0.06). In contrast, MVPA was significantly underestimated in all epoch lengths, by 12.8-19.3 minutes (or 34.8%-52.5% of the measured mean; all p < 0.001). At the individual level, 60-second epochs had the lowest RMSE for LPA (21.7 minutes) and MVPA (17.2 minutes), and one-second epochs had the highest RMSE for both intensities (27.4 minutes and 28.0 minutes for LPA and MVPA, respectively).

Adult Cut-Points

When using the Freedson cut-points (calibration epoch length of 60 seconds), LPA was significantly underestimated in epoch lengths shorter than 30 seconds, by 26.1-77.7 minutes (or 22.3%-66.4% of the measured mean; all p < 0.01). MVPA was significantly underestimated in all epoch lengths, with differences ranging from 16.8-25.1 minutes (or 32.5%-48.6% of the measured mean; all p < 0.01). At the individual level, the RMSE trends for LPA and MVPA were opposite of one another. For LPA, the calibration epoch length had the lowest RMSE (43.3 minutes), and one-second epochs had the highest (98.4 minutes). For MVPA, one-second epochs had the lowest RMSE (40.7 minutes), and the calibration epoch length had the highest (45.3 minutes).

When using the Matthews cut-points (calibration epoch length of 60 seconds), LPA was significantly underestimated in all epoch lengths, by 30.2-97.9 minutes (or 25.8%-83.6% of the measured mean; all p < 0.01). MVPA was significantly overestimated in epoch lengths longer than 15 s, by 12.8-13.7 minutes (or 24.8%-26.5% of the measured mean; all p < 0.04). Notably, mean MVPA time for 1-s epochs was closest to the measured mean, and the estimates increased in longer epochs, the latter trend being opposite of the other adult cut-points. At the individual level, the calibration epoch length had the lowest RMSE for LPA (55.4 minutes), while one-second epochs had the highest (118.4 minutes). For MVPA, five-second epochs had the lowest RMSE (32.5 minutes), and the calibration epoch length had the highest (36.9 minutes).

Finally, when using the Troiano cut-points (calibration epoch length of 60 seconds), LPA was significantly underestimated in epoch lengths shorter than 30 s, by 24.9-76.9 minutes (or 21.3%-65.6% of the measured mean; all p < 0.02). MVPA was significantly underestimated in all epoch lengths, by 17.7-26.3 minutes (or 34.2%-50.9% of the measured mean; all p < 0.02). At the individual level, the RMSE trends for LPA and MVPA were opposite of one another. For LPA, the calibration epoch length had the lowest RMSE (43.8 minutes), and one-second epochs had the highest (97.6 minutes). For MVPA, one-second epochs had the lowest RMSE (41.3 minutes), and the calibration epoch length had the highest (46.4 minutes).

Discussion

In this study, the practical consequences of CPS were examined in simulated free-living, using a criterion measure of indirect calorimetry. There were disparate findings across lines of age (i.e., adolescents or adults), intensity, and cut-point method. At the group level, SB estimates (based on a 100 counts per minute cut-point) were closest to the criterion when using 60-s epochs, whereas the findings for LPA were mixed, with CPS having unfavorable effects on some cut-points (i.e., Freedson adolescent and Matthews) while yielding clear improvements for one cut-point (i.e., Treuth) and negligible improvements for the rest (i.e., Evenson, Freedson adult, and Troiano). For MVPA, group-level adolescent estimates were closest to the criterion measure when using 60-s epochs (regardless of which cut-points were used and what their calibration epoch length was), whereas adult estimates were closest when using 1-s epochs. At the individual level, RMSE tended to be lowest for 60-s epochs, regardless of the age group, intensity, cut-point, or calibration epoch length. In cases where 60-s epochs did not have the lowest RMSE, there were small differences (<1 minute for adolescents and <6 minutes for adults) between the lowest RMSE and the RMSE for 60-s epochs.

Even though CPS improved the estimates in some cases (most notably MVPA for adults), the improvements were always offset by worse performance in at least one other category. Thus, the value of CPS may be situational at best, and it is difficult to suggest specific instances in which CPS should be used. There has been similar confusion in previous research, particularly for MVPA. Over a dozen studies have compared activity estimates using CPS and different epoch lengths (Aibar & Chanal, 2015; Banda et al., 2016; Edwardson & Gorely, 2010; Gabriel et al., 2010; Kim et al., 2013; Logan et al., 2016; Mcclain et al., 2008; Nettlefold et al., 2016; Nilsson et al., 2002; Obeid et al., 2011; Orme et al., 2014; Sanders et al., 2014; Vale et al., 2009; Vanderloo et al., 2016), and although most have given at least partial support for the notion that CPS leads to higher estimates of MVPA as epoch length is shortened, a few have presented evidence to the contrary (Aibar & Chanal, 2015; Edwardson & Gorely, 2010; Mcclain et al., 2008; Nettlefold et al., 2016; Sanders et al., 2014). In fact, some studies have shown conflicting trends within a single study, depending on the cut-point (Mcclain et al., 2008), measurement context, i.e., in-class versus free-living (Sanders et al., 2014), age group, i.e., children versus adolescents (Edwardson & Gorely, 2010), or bout length stipulation (Nettlefold et al., 2016). Some of the confusion may arise from methodological differences among the studies, but it may also indicate that CPS is an unstable technique. Regardless, there has been almost no evidence concerning how CPS affects criterion validity, which was the key contribution of the present study.

CPS and Criterion Validity

To our knowledge, CPS has only been examined with a criterion measure in three previous studies (Mcclain et al., 2008; Montoye, Pfeiffer, Suton, & Trost, 2014; Oliver, Schofield, & Schluter, 2009), all of which were conducted in samples of youth. Of those studies, one did not use an ActiGraph accelerometer (Oliver et al., 2009), and another was focused on responsiveness to change rather than time spent at different intensities (Montoye et al., 2014). Thus, the study by McClain et al. (2008) is the only comparable study that has evaluated criterion validity, and there are notable contrasts between the findings and methods (including criterion methodology) in that study and those for the adolescent portion of the present study.

In the study by McClain et al. (2008), 32 fifth-grade students were assessed during physical education classes (30-min each). Participants wore an ActiGraph 7164 on the hip, and direct observation was the criterion measure of time spent in MVPA. The ActiGraph data were processed using the Treuth and Freedson cut-points, as well as the cut-points of Mattocks et al. (2007). MVPA estimates in six epoch lengths (5-s, 10-s, 15-s, 20-s, 30-s, and 60-s) were compared to the criterion. At the individual level, the findings of McClain et al. (2008) were opposite of the present study. Specifically, their findings showed that RMSE was lowest in 5-s epochs (29.6%-43.5% lower than for 60-s epochs, depending on the cut-point), whereas in this study RMSE was lowest in 60-s epochs (20.6%-77.2% lower than for 1-s epochs, depending on the cut-point).

At the group level, there were mixed results in the study of McClain et al. (2008). For the Treuth cut-points, they found that mean MVPA was closest to the criterion measure (within 39.0%) in 5-s epochs, and the same was true for the Mattocks cut-points (within 52.5% of the criterion measure). However, for the Freedson cut-points, mean MVPA time was closest to the criterion (within 0.6%) in 60-s epochs. Additionally, it is interesting to note that the change in MVPA time was not necessarily monotonic from epoch to epoch. For example, the mean MVPA estimates from the Freedson cut-points, from lowest to highest, were 488-s (using 5-s epochs), 495-s (using 30-s epochs), 504-s (using 15-s epochs), 506-s (using 10-s and 20-s epochs), and 525-s (using 60-s epochs).

It is difficult to fully reconcile the group-level findings of McClain et al. (2008) with those in the present study (i.e., mean MVPA closest to the criterion in 60-s epochs for all three cut-points). The conflicts could reflect an inherent instability of CPS, as described previously. Age and setting differences could also contribute to the disparate findings, as the participants in the former study were younger (10.5 ± 0.5 years old) than those in the present study, and were observed in physical education classes that were more structured than the simulated free-living setting used for the present study. The latter differences could have made intermittent activity bouts more common in one study than the other, which could influence the comparability of results. Another potential contributor would be differences between the criterion measures used in both studies, i.e., direct observation in the previous study (McClain et al., 2008) versus indirect calorimetry in the present study.

Direct observation captures behavior, whereas indirect calorimetry captures energy expenditure via oxygen consumption. Although both can be used to assess time spent in SB, LPA, and MVPA, they do so in different ways. Specifically, direct observation is uniquely able to capture transient behaviors that do not last long enough to affect aerobic metabolism (i.e., indirect calorimetry). Thus, it is possible that using direct observation would tend to favor shorter epochs, whereas indirect calorimetry would have the opposite tendency. However, those tendencies were not exhibited consistently (i.e., for all cut-points) in this study or the study of McClain et al. (2008). Nevertheless, the distinction between direct observation and indirect calorimetry does create a dilemma. On one hand, steady-state activity is rare in free-living contexts, which could limit the applicability of indirect calorimetry as a criterion measure. On the other hand, indirect calorimetry has virtually always been the criterion measure in calibration studies developing cut-points (Freedson et al., 2005; Matthews, 2005), which could suggest that cut-points themselves are designed for application to steady-state data. The difficulty is reflected in the lack of consensus regarding whether CPS should be used. Some have suggested CPS may be a viable technique if studied more thoroughly (Mcclain et al., 2008), while others have advocated the exclusive use of cut-points in their calibration epoch length (Banda et al., 2016). The equivocal findings of the present study give some support to the latter notion, but there may also be justification for making case-specific recommendations, e.g. for different age groups as advocated by Edwardson & Gorely (2010).

Apart from the previously-described differences between this study and the study by McClain et al. (2008), there were also notable limitations of the latter study that this study was able to address. Specifically, the latter study only assessed MVPA, which ignores the potential for compensatory effects in other categories. The present study addressed this by considering SB, LPA, and MVPA together, and CPS did not lead to improved prediction in all categories. Additional limitations addressed in this study were the lack of unstructured activity data, and the lack of data from adults. In this study, adolescents and adults were both assessed in simulated free-living environments, and some contrasts emerged between the findings for both groups.

Practical and Theoretical Considerations

A key objective of this investigation was to show the practical consequences of CPS by using methods (i.e., simulated free-living data and prominent cut-points) that reflected common practice. Thus, the findings of this study should be considered alongside issues with common practice itself (i.e., the use of cut-points). There are numerous problems with cut-points (Bassett, Rowlands, & Trost, 2012), which have prompted calls to discontinue the development of new cut-points and begin focusing on more sophisticated approaches (Strath, Pfeiffer, & Whitt-Glover, 2012). Accordingly, new calibration studies have increasingly used raw acceleration data rather than activity counts, and machine learning techniques rather than linear regression and receiver operating characteristic analysis (de Almeida Mendes et al., 2018). However, uptake of new methods (e.g. machine learning models) has been slow outside of the PA measurement research community, which may be due to the lack of user-friendly interfaces for applying them. A July 2019 PubMed search for “accelerometer cut points” matched 20-27 articles per year since 2014 (including partial data for 2019), compared to 14-22 articles between 2011 and 2013 and ≤8 every prior year. Thus, cut-point usage remains widespread, and will likely continue for some time.

As part of examining CPS, it is important to consider practical and theoretical concerns regarding its use. A potent practical concern is that CPS has a negative effect on the comparability of studies. When cut-points are not used in a standardized way, it becomes impossible to interpret differences between studies, because there is no way of telling how much of the difference is artifact caused by CPS. This has the potential to mislead clinicians, epidemiologists, and public health officials who rely on accelerometer-based research to inform their work. In the appendixes, Figure A2.1 uses 10 minutes of representative data to illustrate the issues caused by CPS. Data are presented in 1-s, 15-s, and 60-s epochs, and the MVPA estimates (using the Freedson youth cut-points, age 12) range from 4.6 minutes (1-s epochs) to 6.0 minutes (60-s epochs). The lack of agreement over a short time period has concerning implications for longer collection periods (e.g. one week), and the data also show how CPS can lead to lower estimates of MVPA, not higher.

The key theoretical concern with CPS is that models are always tied to the methods used to develop them. From that perspective, it is no more justifiable to use CPS than it is to apply wrist-specific cut-points to accelerometer data obtained on the hip. There is also a contradiction between the motivation for using CPS and the necessary assumptions for using it. That is, CPS is motivated in part by the observation that behavior may not be uniform throughout a longer epoch (Mcclain et al., 2008; Orme et al., 2014). But in order to use CPS, it is necessary to assume that dividing the cut-point will not affect the validity of the predictions, which is only possible if activity is uniform throughout the epoch. The case for using shorter epochs has been well-made, but it does not follow that CPS is an appropriate way to make predictions on data with shorter epochs. Instead, cut-points that were originally calibrated in a short epoch length should be used.

Strengths and Limitations

A strength of the present study was assessment in simulated free-living conditions, which brought out the consequences of CPS as practically as possible. Most previous studies have used similar data, although some have used data collected in a classroom setting (Aibar & Chanal, 2015; Kim et al., 2013; Mcclain et al., 2008; Vale et al., 2009), and one has used both classroom and free-living data (Sanders et al., 2014). It is important to consider the amount of structure in a protocol, because it impacts how generalizable the findings are. Unstructured protocols (e.g. involving simulated free living) allow more diverse movement patterns, thereby giving a more realistic picture of how activity accumulates in real life.

Although the use of free-living data was a strength of this study, it also brought out the previously-described limitations of indirect calorimetry. Short bouts of activity can be masked when using indirect calorimetry, because there is not sufficient time for aerobic metabolism to respond. Furthermore, steady-state is rarely achieved in free-living conditions, which means that indirect calorimetry captures only the oxygen cost of free-living activities, not the total energy expenditure. Despite its limitations, the use of indirect calorimetry as a criterion measure also holds some advantage over direct observation, particularly by eliminating subjective aspects of the assessment, and by using the same criterion measure as most calibration studies.

Conclusion

For adolescents and adults, CPS does not produce predictable and favorable changes in criterion validity, nor does it consistently lead to increased estimates of MVPA. This supports the exclusive use of cut-points in the calibration epoch length. Alternative accelerometry methods (e.g. using raw acceleration and machine learning models) provide an avenue for avoiding issues and confusion caused by CPS.

What Does This Article Add?

Previous studies have shown that accelerometer-derived activity estimates change when the epoch length is manipulated via CPS. This study builds on previous studies by examining whether the changes reflect improvements in criterion validity. In some cases, criterion validity was marginally improved by using CPS. However, criterion validity was generally affected negatively by CPS. Moreover, the effects of CPS were unpredictable. The findings of this study provide a new perspective on the issue of epoch length in accelerometry, which is important for researchers to bear in mind when carrying out sensor-based physical activity assessments.

Appendix 1: Using Equivalence Testing to Assess the Effects of Cut-Point Scaling

Background

When evaluating agreement between two measures, the traditional statistical approach has been to test for significant differences between the means. Recently, equivalence testing has received increased attention, because it directly tests for significant equivalence, rather than focusing on differences. There are subtle yet important distinctions between difference testing and equivalence testing, as explained in the recent work of Dixon et al. (2018). For the present study, it is worthwhile to use equivalence testing, since the research question concerns the criterion validity of accelerometer-derived physical activity predictions, i.e., the level of agreement between the predictions and the criterion measure.

Equivalence Zones

Equivalence testing hinges on a pre-specified zone of equivalence (ZOE, e.g. ±10%), which defines the level of tolerance below which two measures will be considered significantly equivalent. The ZOE can be defined in either absolute or relative terms, and different procedures can be used to perform the test depending on whether either measure is a true criterion. (See Dixon et al. (2018) for full details.)

The ZOE must be defined according to a practically meaningful tolerance, which can be difficult to establish. In the case of the present study, it is not appropriate to define a single ZOE for assessments of sedentary behavior (SB), light physical activity (LPA), and moderate-to-vigorous physical activity (MVPA). Rather, a unique ZOE is necessary for each category. There are currently no conventions for what the zones should be, so an original approach is needed. A sensible approach would be to start with published recommendations for daily time use.

The daily MVPA recommendation is 60 minutes for adolescents, which is similar to the ideal recommended amount for adults (i.e., 300 minutes per week; Piercy et al., 2018). Errors greater than five minutes (8%) would be unacceptable, and therefore MVPA will be tested with a ZOE of ±8%.

Assuming 60 minutes of MVPA and eight hours of sleep (see the sleep recommendations of Hirshkowitz et al., 2015), there are 15 remaining hours per day that can be allotted to SB and LPA. There are no specific recommendations for either category. Errors greater than 30 minutes in either category would be unacceptable, for a total allowable error of one hour out of 15 (6%). Thus, SB and LPA will each be tested with a ZOE of 6%.

Results

Equivalence testing results are shown in Tables A1.1 (SB), A1.2 (LPA), and A1.3 (MVPA). For adolescents and adults, none of the estimates were significantly equivalent to the criterion measure, regardless of intensity, cut-point, or epoch length. The tables also provide information concerning how wide the ZOE would have needed to be for the estimates to reach significant equivalence. The latter information provides a useful way of comparing the estimates to one another, as well as understanding the magnitude of their ineffectiveness.

SB estimates (using the 100 counts per minute cut-point) were closest to equivalence when using 60-s epochs in both adolescents (equivalent with 38.2% ZOE) and adults (equivalent with 18.0% ZOE). For other epoch lengths to reach equivalence, the zones needed to increase by factors of 1.25-2.53 (adolescents) and 1.52-3.83 (adults).

There were inconsistent findings for LPA estimates. In adolescents, the estimates were closest to equivalence when using 5-s epochs (Evenson cut-points, equivalent with 25.4% ZOE), 60-s epochs (Freedson cut-points, equivalent with 27.9% ZOE), and 1-s epochs (Treuth cut-points, equivalent with 30.1% ZOE). For other epoch lengths to reach equivalence, the ZOE needed to increase by factors of 1.08-2.42. For adults, the estimates were closest to equivalence when using 30-s epochs (Freedson cut-points, equivalent with 18.2% ZOE; Troiano cut-points, equivalent with 17.3% ZOE) and 60-s epochs (Matthews cut-points, equivalent with 37.9% ZOE). For other epoch lengths to reach equivalence, the ZOE needed to increase by factors of 1.01-4.70.

For MVPA, adolescent estimates were closest to equivalence when using 60-s epochs (equivalent with ZOE of 16.8%-65.4%, depending on cut-point), and adult estimates were closest when using 1-s epochs (equivalent with ZOE of 26.2%-56.4%, depending on cut-point). For other epoch lengths to reach equivalence, the ZOE needed to increase by factors of 1.02-2.73 (adolescents) and 1.07-1.78 (adults).

Table A1.1.

Results of equivalence testing for sedentary behavior. Bold text indicates the calibrated epoch length for each set of cut-points, and all estimates are based on a cut-point of 100 counts·min⁻¹.

Adolescents					Adult
Method	Epoch	Mean ± SD	Mean [90% CI] difference from criterion	Minimum significant ZOE	Method	Epoch	Mean ± SD	Mean [90% CI] difference from criterion	Minimum significant ZOE
Cosmed		30.1 ± 32.3			Cosmed		167.5 ± 99.4
Evenson	1	53.8 ± 33.6	23.6 [18.1, 29.2]	96.9%	Freedson	1	262.0 ± 52.2	94.5 [73.5, 115.5]	69.0%
	5	47.4 ± 33.5	17.2 [11.8, 22.6]	75.3%		5	237.2 ± 64.2	69.8 [51.2, 88.3]	52.8%
	10	44.1 ± 33.2	14.0 [8.6, 19.4]	64.4%		10	222.8 ± 70.6	55.3 [38.0, 72.6]	43.4%
	15	42.3 ± 33.0	12.2 [6.8, 17.5]	58.4%		15	214.5 ± 73.5	47.1 [30.6, 63.6]	38.0%
	30	39.1 ± 32.2	9.0 [3.6, 14.3]	47.8%		30	198.3 ± 79.0	30.8 [15.7, 46.0]	27.5%
	60	36.3 ± 31.7	6.2 [0.8, 11.5]	38.2%		60	184.0 ± 83.2	16.6 [3.0, 30.1]	18.0%
Freedson	1	53.8 ± 33.6	23.6 [18.1, 29.2]	96.9%	Matthews	1	262.0 ± 52.2	94.5 [73.5, 115.5]	69.0%
	5	47.4 ± 33.5	17.2 [11.8, 22.6]	75.3%		5	237.2 ± 64.2	69.8 [51.2, 88.3]	52.8%
	10	44.1 ± 33.2	14.0 [8.6, 19.4]	64.4%		10	222.8 ± 70.6	55.3 [38.0, 72.6]	43.4%
	15	42.3 ± 33.0	12.2 [6.8, 17.5]	58.4%		15	214.5 ± 73.5	47.1 [30.6, 63.6]	38.0%
	30	39.1 ± 32.2	9.0 [3.6, 14.3]	47.8%		30	198.3 ± 79.0	30.8 [15.7, 46.0]	27.5%
	60	36.3 ± 31.7	6.2 [0.8, 11.5]	38.2%		60	184.0 ± 83.2	16.6 [3.0, 30.1]	18.0%
Treuth	1	53.8 ± 33.6	23.6 [18.1, 29.2]	96.9%	Troiano	1	262.0 ± 52.2	94.5 [73.5, 115.5]	69.0%
	5	47.4 ± 33.5	17.2 [11.8, 22.6]	75.3%		5	237.2 ± 64.2	69.8 [51.2, 88.3]	52.8%
	10	44.1 ± 33.2	14.0 [8.6, 19.4]	64.4%		10	222.8 ± 70.6	55.3 [38.0, 72.6]	43.4%
	15	42.3 ± 33.0	12.2 [6.8, 17.5]	58.4%		15	214.5 ± 73.5	47.1 [30.6, 63.6]	38.0%
	30	39.1 ± 32.2	9.0 [3.6, 14.3]	47.8%		30	198.3 ± 79.0	30.8 [15.7, 46.0]	27.5%
	60	36.3 ± 31.7	6.2 [0.8, 11.5]	38.2%		60	184.0 ± 83.2	16.6 [3.0, 30.1]	18.0%

SD- standard deviation; CI- confidence interval; ZOE- zone of equivalence.

Table A1.2.

Results of equivalence testing for light physical activity. Bold text indicates the calibrated epoch length for each set of cut-points.

Adolescents					Adult
Method	Epoch	Mean ± SD	Mean [90% CI] difference from criterion	Minimum significant ZOE	Method	Epoch	Mean ± SD	Mean [90% CI] difference from criterion	Minimum significant ZOE
Cosmed		27.2 ± 24.9			Cosmed		117.1 ± 74.0
Evenson	1	22.9 ± 12.5	−4.3 [−11.4, 2.8]	41.9%	Freedson	1	39.4 ± 27.5	−77.7 [−96.1, −59.4]	82.1%
	5	26.7 ± 13.4	−0.5 [−6.9, 5.9]	25.4%		5	66.2 ± 41.6	−50.9 [−67.4, −34.4]	57.6%
	10	28.6 ± 14.2	1.4 [−4.6, 7.4]	27.6%		10	82.2 ± 49.3	−34.9 [−50.6, −19.3]	43.2%
	15	29.6 ± 15.2	2.4 [−3.4, 8.3]	30.5%		15	91.0 ± 52.8	−26.1 [−40.9, −11.2]	35.0%
	30	31.9 ± 17.0	4.7 [−0.9, 10.2]	37.8%		30	109.8 ± 60.7	−7.3 [−21.2, 6.6]	18.2%
	60	33.9 ± 19.5	6.7 [1.2, 12.1]	44.5%		60	125.7 ± 68.0	8.5 [−4.4, 21.4]	18.4%
Freedson	1	15.8 ± 8.6	−11.4 [−18.0, −4.8]	66.5%	Matthews	1	19.2 ± 13.9	−97.9 [−118.2, −77.6]	100.9%
	5	18.7 ± 10.0	−8.5 [−14.4, −2.6]	52.9%		5	39.5 ± 24.0	−77.6 [−96.0, −59.3]	82.0%
	10	20.2 ± 11.7	−7.1 [−12.5, −1.6]	46.3%		10	51.8 ± 28.9	−65.3 [−82.7, −48.0]	70.6%
	15	21.0 ± 12.9	−6.2 [−11.5, −1.0]	42.3%		15	58.7 ± 30.9	−58.4 [−75.3, −41.5]	64.4%
	30	23.0 ± 15.5	−4.2 [−9.2, 0.8]	34.2%		30	73.5 ± 35.6	−43.6 [−59.2, −28.1]	50.6%
	60	24.5 ± 18.4	−2.7 [−7.6, 2.2]	27.9%		60	86.9 ± 41.1	−30.2 [−44.4, −16.1]	37.9%
Treuth	1	26.7 ± 15.3	−0.5 [−8.1, 7.1]	30.1%	Troiano	1	40.3 ± 28.1	−76.9 [−95.2, −58.6]	81.3%
	5	31.3 ± 15.8	4.0 [−2.9, 11.0]	40.4%		5	67.3 ± 42.3	−49.8 [−66.3, −33.4]	56.6%
	10	33.7 ± 16.4	6.5 [−0.2, 13.2]	48.5%		10	83.2 ± 50.0	−33.9 [−49.5, −18.3]	42.3%
	15	35.3 ± 17.0	8.1 [1.6, 14.7]	54.0%		15	92.2 ± 53.6	−24.9 [−39.8, −10.1]	34.1%
	30	38.2 ± 18.4	10.9 [4.6, 17.3]	63.6%		30	111.0 ± 61.8	−6.1 [−20.2, 8.0]	17.3%
	60	40.7 ± 20.4	13.5 [7.2, 19.7]	72.8%		60	126.8 ± 68.8	9.7 [−3.3, 22.7]	19.5%

SD- standard deviation; CI- confidence interval; ZOE- zone of equivalence.

Table A1.3.

Results of equivalence testing for moderate-to-vigorous physical activity. Bold text indicates the calibrated epoch length for each set of cut-points.

Adolescents					Adult
Method	Epoch	Mean ± SD	Mean [90% CI] difference from criterion	Minimum significant ZOE	Method	Epoch	Mean ± SD	Mean [90% CI] difference from criterion	Minimum significant ZOE
Cosmed		36.9 ± 37.7			Cosmed		51.6 ± 56.3
Evenson	1	17.5 ± 18.8	−19.3 [−24.7, −14]	67.0%	Freedson	1	34.8 ± 29.8	−16.8 [−28.1, −5.5]	54.4%
	5	20.1 ± 23.4	−16.7 [−21.1, −12.4]	57.2%		5	32.8 ± 30.4	−18.8 [−29.9, −7.7]	58.1%
	10	21.5 ± 26.2	−15.4 [−19.2, −11.6]	52.4%		10	31.3 ± 31.3	−20.3 [−31.3, −9.3]	60.8%
	15	22.3 ± 27.6	−14.6 [−18.2, −11.0]	49.4%		15	30.6 ± 32.3	−21.0 [−31.9, −10.1]	61.8%
	30	23.2 ± 30.0	−13.6 [−16.9, −10.4]	46.1%		30	28.1 ± 33.5	−23.5 [−34.5, −12.5]	67.0%
	60	24.0 ± 31.7	−12.8 [−15.8, −9.8]	43.1%		60	26.5 ± 35.8	−25.1 [−36.6, −13.6]	70.9%
Freedson	1	24.6 ± 23.0	−12.2 [−16.8, −7.6]	45.9%	Matthews	1	55.0 ± 39.7	3.4 [−6.7, 13.5]	26.2%
	5	28.1 ± 27.6	−8.7 [−12.3, −5.1]	33.6%		5	59.5 ± 44.2	7.9 [−1.8, 17.5]	33.9%
	10	29.9 ± 30.1	−6.9 [−10.1, −3.7]	27.7%		10	61.7 ± 47.4	10.0 [0.3, 19.7]	38.4%
	15	30.9 ± 31.6	−5.9 [−8.9, −2.9]	24.1%		15	63.0 ± 49.7	11.3 [1.4, 21.3]	41.5%
	30	32.1 ± 33.5	−4.7 [−7.5, −2.0]	20.3%		30	64.4 ± 54.0	12.8 [2.9, 22.7]	44.0%
	60	33.4 ± 35.3	−3.5 [−6.2, −0.8]	16.8%		60	65.3 ± 56.9	13.7 [3.2, 24.1]	46.7%
Treuth	1	13.8 ± 15.8	−23.1 [−29.1, −17.1]	79.0%	Troiano	1	34.0 ± 29.5	−17.7 [−29, −6.3]	56.4%
	5	15.6 ± 19.8	−21.3 [−26.4, −16.1]	71.9%		5	31.7 ± 30.0	−19.9 [−31.2, −8.7]	60.4%
	10	16.4 ± 21.8	−20.5 [−25.3, −15.7]	68.9%		10	30.2 ± 30.7	−21.4 [−32.5, −10.2]	63.1%
	15	16.6 ± 22.8	−20.3 [−24.9, −15.6]	67.8%		15	29.5 ± 31.7	−22.1 [−33.2, −11.1]	64.3%
	30	17.0 ± 24.4	−19.9 [−24.4, −15.4]	66.5%		30	26.9 ± 32.5	−24.7 [−36.0, −13.4]	69.9%
	60	17.2 ± 25.9	−19.6 [−24.0, −15.2]	65.4%		60	25.3 ± 35.2	−26.3 [−37.9, −14.6]	73.6%

SD- standard deviation; CI- confidence interval; ZOE- zone of equivalence.

Appendix 2: Other Supplemental Materials

Table A2.1.

Agreement of the predictions with the criterion measure (Cosmed K4b²). Bolded epoch lengths indicate the calibrated epoch length for each cut-point, i.e. the epoch length used in the original calibration study. See footnote for units.

		Sedentary			Light			MVPA
Cut Point	Epoch Length	Mean Bias	Lower LOA	Upper LOA	Mean Bias	Lower LOA	Upper LOA	Mean Bias	Lower LOA	Upper LOA
Treuth (2004)	1	−23.6	−65.5	18.2	0.5	−56.8	57.8	23.1	−22.1	68.3
	5	−17.2	−58.0	23.6	−4.0	−56.5	48.4	21.3	−17.5	60.0
	10	−14.0	−54.5	26.5	−6.5	−56.9	43.9	20.5	−15.9	56.9
	15	−12.2	−52.7	28.4	−8.1	−57.6	41.4	20.3	−14.8	55.4
	30	−9.0	−49.3	31.4	−10.9	−58.8	36.9	19.9	−14.0	53.9
	60	−6.2	−46.4	34.1	−13.5	−60.7	33.8	19.6	−13.7	53.0
Freedson (2005)	1	−23.6	−65.5	18.2	11.4	−38.7	61.5	12.2	−22.4	46.9
	5	−17.2	−58.0	23.6	8.5	−35.9	52.9	8.7	−18.5	36.0
	10	−14.0	−54.5	26.5	7.1	−34.4	48.5	6.9	−17.4	31.2
	15	−12.2	−52.7	28.4	6.3	−33.3	45.8	5.9	−16.6	28.4
	30	−9.0	−49.3	31.4	4.2	−33.5	42.0	4.7	−16.1	25.6
	60	−6.2	−46.4	34.1	2.7	−34.3	39.6	3.5	−16.6	23.6
Evenson (2008)	1	−23.6	−65.5	18.2	4.3	−49.3	57.9	19.3	−20.8	59.5
	5	−17.2	−58.0	23.6	0.5	−47.7	48.7	16.7	−16.0	49.5
	10	−14.0	−54.5	26.5	−1.4	−46.8	44.0	15.4	−13.4	44.2
	15	−12.2	−52.7	28.4	−2.4	−46.7	41.9	14.6	−12.4	41.6
	30	−9.0	−49.3	31.4	−4.7	−46.7	37.3	13.6	−11.1	38.4
	60	−6.2	−46.4	34.1	−6.7	−47.6	34.2	12.8	−9.8	35.5
Freedson (1998)	1	−94.5	−231.8	42.7	77.7	−42.4	197.9	16.8	−57.0	90.5
	5	−69.8	−191.1	51.6	50.9	−57.0	158.9	18.8	−53.8	91.5
	10	−55.3	−168.3	57.7	34.9	−67.1	137.0	20.3	−51.6	92.3
	15	−47.1	−154.9	60.8	26.1	−71.1	123.2	21.0	−50.4	92.3
	30	−30.8	−129.8	68.2	7.3	−83.7	98.3	23.5	−48.7	95.7
	60	−16.6	−105.4	72.2	−8.5	−93.0	75.9	25.1	−50.0	100.2
Matthews (2005)	1	−94.5	−231.8	42.7	97.9	−34.7	230.5	−3.4	−69.4	62.6
	5	−69.8	−191.1	51.6	77.6	−42.5	197.8	−7.9	−70.7	55.0
	10	−55.3	−168.3	57.7	65.3	−48.1	178.7	−10.0	−73.4	53.3
	15	−47.1	−154.9	60.8	58.4	−52.2	169.0	−11.3	−76.6	53.9
	30	−30.8	−129.8	68.2	43.6	−58.2	145.4	−12.8	−77.3	51.7
	60	−16.6	−105.4	72.2	30.2	−62.3	122.8	−13.7	−81.8	54.5
Troiano (2008)	1	−94.5	−231.8	42.7	76.9	−42.9	196.6	17.7	−56.7	92.0
	5	−69.8	−191.1	51.6	49.8	−57.8	157.5	19.9	−53.5	93.4
	10	−55.3	−168.3	57.7	33.9	−68.1	135.8	21.4	−51.6	94.3
	15	−47.1	−154.9	60.8	25.0	−72.3	122.2	22.1	−50.2	94.4
	30	−30.8	−129.8	68.2	6.1	−86.0	98.2	24.7	−49.2	98.6
	60	−16.6	−105.4	72.2	−9.7	−94.7	75.3	26.3	−49.8	102.4

MVPA, moderate-to-vigorous physical activity; LOA, limits of agreement. Epoch lengths are in seconds; mean bias and LOA are in minutes.

Figure A2.1.

Depiction of 10 minutes of data in 1-s, 15-s, and 60-s epochs. Horizontal dashed lines represent MVPA cut-points (Freedson, 2005; age 12) scaled to each epoch length.