Physical Activity Assessment with the ActiGraph GT3X and Doubly Labeled Water : Medicine & Science in Sports & Exercise

Journal Logo


Physical Activity Assessment with the ActiGraph GT3X and Doubly Labeled Water


Author Information
Medicine & Science in Sports & Exercise 49(9):p 1935-1944, September 2017. | DOI: 10.1249/MSS.0000000000001299


The use of accelerometers in physical activity research has grown dramatically in recent years (3). Due to declining costs, accelerometers are even feasible for use in large-scale, longitudinal epidemiological studies seeking to relate this objective measure of physical activity to development of clinical endpoints (15). Although it has become common to use accelerometers in research, there is still debate on the best metrics to use to summarize the vast amount of data produced.

To assess physical activity level, data are typically obtained from the vertical axis of a waist-worn accelerometer and converted to number of minutes per day spent in various physical activity intensity categories with the goal of measuring minutes per day of moderate- to vigorous-intensity physical activity (MVPA) in 10-min bouts (3). Moreover, sometimes a “modified” 10-min bout is used that allows for 1–2 min of activity in a bout to drop below the defined threshold, reflecting events such as waiting to cross a street while jogging (33). Time spent in MVPA in 10-min bouts has been the status quo for most physical activity and public health researchers in the United States when using accelerometers, perhaps because the national physical activity recommendations have been expressed in this way. These recommendations (e.g., the 1996 U.S. Surgeon General’s Report and the 2008 Physical Activity Guidelines for Americans) are primarily based on studies that used questionnaires to assess physical activity; questionnaires assess MVPA performed in bouts more accurately than light-intensity or lifestyle activity (25,32,37). In studies using accelerometers, however, various investigator-defined cutpoints are currently being used to identify time spent in light-, moderate-, and vigorous-intensity activity. This is problematic as estimates of time spent in MVPA can vary considerably, depending on the chosen cutpoint (18). Furthermore, MVPA accumulated in 10-min bouts is only a small subset of all physical activity performed throughout a 24-h day (33). Emerging evidence indicates that other types of activity, in particular light-intensity physical activity, have significant health consequences as well (10,22).

Instead of focusing on a subset of physical activity, like MVPA, there may be situations where it is preferable to consider the total volume of physical activity performed. Physical activity is a complex behavior with multiple dimensions, including frequency, intensity, duration, and activity type (24). Nevertheless, for many applications, it is important to determine what single variable best accounts for all dimensions and represents total physical activity.

Total activity counts (TAC) per day provide an alternative to MVPA that represents total volume of physical activity performed (3). TAC accounts for minutes spent in sedentary, light, moderate, and vigorous physical activity and weights each minute according to intensity. Recently, TAC was found to have stronger associations to several cardiometabolic biomarkers when compared with MVPA accumulated in modified ≥ 10-min bouts (40). Thus, TAC may be an important measure of total physical activity volume and should be considered when using data collected with accelerometers.

Another metric that can be estimated from accelerometers and reflects the energy expended in physical activity is physical activity energy expenditure (PAEE). Several prediction equations exist to estimate energy expenditure based on data from accelerometers (19). These equations were developed using indirect calorimetry for comparison so that accelerometer activity counts can be converted into energy expenditure (8,19).

Finally, despite the robust information accelerometers can provide about intensity and duration of activity, steps per day is another measure of total activity volume that is simple, easy to interpret, and readily available through use of inexpensive devices like pedometers (36). In addition to recording body acceleration and deceleration, accelerometers can also capture step data.

As all physical activity results in energy expenditure, using energy expenditure measured by the doubly labeled water (DLW) method is a logical choice to validate an accelerometer measure of total volume of physical activity. Thus, the goal of this study was to answer the question: “Compared with considering MVPA alone, how much does accounting for total volume of physical activity improve the correlation between accelerometer-measured physical activity and PAEE measured by DLW?” To address this, we examined the degree to which four different accelerometer measures–TAC per day, steps per day, PAEE (kcal·kg−1·d−1), and minutes of MVPA per day– were correlated with PAEE measured by DLW. Additionally, we compared results using accelerometer activity counts from the vertical axis only to those obtained when using counts from all three axes (triaxial vector magnitude) because traditionally, accelerometers were uniaxial, but now it is more common to measure acceleration along all three axes.


Study Population

The original Nurses’ Health Study (NHS) cohort was established in 1976 and included 121,700 US female registered nurses age 30–55 yr (4). The NHS II is a similar cohort of 116,671 female nurses 25 to 42 yr of age at study initiation in 1989. The Health Professionals Follow-up Study (HPFS) is a cohort of 51,529 male US health professionals age 40–75 yr at study initiation in 1986 (9). The participants in all three cohorts respond to biennial questionnaires that assess updated lifestyle characteristics and medical history.

In 2010 to 2013, two studies were conducted within the NHS, NHS II, and HPFS to assess the validity of the food frequency questionnaire and physical activity questionnaire used in these cohorts. A subset of women from NHS/NHS II participated in the Women’s Lifestyle Validation Study (WLVS) in 2010 to 2012. Similarly, a subset of men from HPFS, along with additional men from the local community who were members of the Harvard Pilgrim Health Care insurance plan, participated in the Men’s Lifestyle Validation Study (MLVS) in 2011 to 2013. WLVS and MLVS participants were a random sample of NHS/NHS II participants and HPFS participants who completed the 2006/2007 FFQ and had previously provided blood in these cohorts. The Harvard Pilgrim participants were recruited at random from among men who have been enrolled in the program for at least 5 yr, to ensure stability for our study. We included men and women 40–85 yr of age and the sample was stratified by age (5-yr groups) to obtain a uniform distribution. Persons with a history of coronary heart disease, stroke, cancer, or major neurological disease were excluded. The availability of broad band internet was required as some of the study activities were to be completed online. Finally, no expected changes in habitual diet or physical activity patterns were also a requirement for eligibility. For the WLVS, 796 women consented to participate and were enrolled in the study. For the MLVS, 671 men consented to participate and were enrolled in the study. This study was approved by the Institutional Review Board of the Harvard T. H. Chan School of Public Health and Brigham and Women’s Hospital.

Data Collection Procedures

The data collected for each participant within the WLVS and MLVS included one DLW measurement (with n = 86 women and n = 102 men having a repeat measurement to assess within-person variability) and two accelerometer measurements with approximately 6 months between each measurement. The data collection period for each participant spanned 12 to 15 months, depending on study group (Fig. 1). The DLW measurement and one of the accelerometer measurements were designed to occur during the same 3-month time window (phase) during the study, but not during the same week to avoid artificially high correlations. For participants in study groups 1 and 2, the mean (± SD) number of days between DLW and accelerometer measurement 1 was 35.0 (±28.2) (median, 29; interquartile range [IQR], 22) in women and 45.1 (±43.2) (median, 41; IQR, 24) in men while the mean (± SD) number of days between DLW and accelerometer measurement 2 was 152.6 (±21.5) (median, 154; IQR, 6) in women and 150.9 (±23.1) (median, 154; IQR, 6) in men. For participants in study groups 3 and 4, the mean (± SD) number of days between DLW and accelerometer measurement 1 was 174.0 (±26.9) (median, 175; IQR, 9) in women and 169.1 (±41.0) (median, 175; IQR, 10) in men, whereas the mean (± SD) number of days between DLW and accelerometer measurement 2 was 32.4 (±20.5) (median, 25.5; IQR, 21) in women and 37.0 (±30.0) (median, 34; IQR, 22) in men. For the current analysis, we used the first DLW measurement and the average of the two accelerometer measurements for each participant (or a single accelerometer measurement for 25 women and 99 men who only had one valid measure). In addition, we used the repeat DLW measurements to adjust for random within-person variation in energy expenditure.

WLVS and MLVS timeline according to study group.


The participants were mailed an accelerometer (ActiGraph GT3X; Actigraph Corporation, Pensacola, FL), detailed instructions, and a wear time diary. Participants were instructed to wear the monitor on the hip for 7 d during all waking hours, except when bathing or swimming, and record the days the monitor was worn. Participants then shipped the accelerometer and completed wear time diary back to the study center where their data could be downloaded and reviewed.

The epoch was set to 1 s and the low-frequency extension was enabled. Enabling the low-frequency extension on the ActiGraph GT3X widens the bandpass filter that is applied to the raw acceleration signal after it has passed through a proprietary low-pass anti-aliasing filter (13). The low-frequency extension is recommended to increase the device sensitivity in the low-frequency range and make the device output (counts per minute) similar to an earlier model, the ActiGraph 7164 (38). Accelerometer data were screened for wear time using standard methods (33,35). For data based on vertical axis counts, nonwear time was defined as ≥ 60 consecutive minutes with zero accelerometer counts, allowing up to 2 min with limited movement (<100 counts per minute). For triaxial vector magnitude data, the algorithm was modified to allow for up to 2 min of counts < 200 counts per minute, which is the threshold for sedentary time when using triaxial counts (1). Daily wear time was determined by subtracting non-wear time from 24 h. Wear days were identified using the wear time diary (to avoid counting spurious physical activity while monitors were moving in mail transit). Participants with at least 4 d with ≥ 10 h of wear per day were included in the analysis.

Physical activity as measured by the accelerometer was examined in four ways: 1) TAC per day; 2) steps per day (steps per day); 3) PAEE (kcal·kg−1·d−1) and 4) minutes per day (min·d−1) spent in MVPA. Data based on both vertical axis counts only and triaxial counts were used. TAC are the accelerations captured by the device that were filtered, full-wave rectified, and integrated over time, representing the intensity of ambulatory activity. Vertical axis counts are activity counts measured in the vertical plane, whereas triaxial counts are a composite vector magnitude from three individual orthogonal planes (vertical, anteroposterior, and mediolateral). TAC per day based on vertical axis and triaxial counts were determined by averaging the TAC per day across valid wear days. ActiGraph uses a proprietary algorithm to count steps and no information about the specifics of this process is publicly available. Step counts are based only on accelerometer data collected on the vertical axis. Steps per day was determined by averaging the steps taken per day across valid wear days.

To compute total daily energy expenditure (TDEE) in kilocalories per kilogram per day from the accelerometer, we first used the Freedson and Sasaki MET prediction equations to determine energy expenditure in METs for each minute (8,29). To predict METs from vertical axis counts, we used the equation METs = 1.439008 + (0.000795 × counts per minute) (8). To predict METs from triaxial counts, we used the equation METs = 0.000863(VM3) + 0.668876 where VM3 is the composite vector magnitude of activity counts from all three axes (29). In both cases, for minutes where the vertical axis or triaxial counts were below the threshold for sedentary behavior (100 and 200 counts per minute, respectively ([1,21])), a value of 1 MET was assigned. Each minute of non-wear time was also assigned a value of 1 MET. Additionally, for minutes where the vertical axis or triaxial counts were in the range for light-intensity activity (100–2020 counts per minute and 200–2690 counts per minute, respectively ([1,21])), two approaches were used: 1) Extrapolate the MET prediction equations to estimate METs in this range; and 2) Assign a constant 2.0 METs. Next, the computed METs were summed over all minutes to get TDEE in MET-minutes per day, which was then divided by 60 to get TDEE in MET-hours per day. TDEE in MET-hours per day is numerically equivalent to TDEE in kilocalories per kilogram per day based on the fact that 1.0 MET = 1.0 kcal·kg−1·h−1 (2). Finally, to get an estimate of PAEE in kilocalories per kilogram per day from the accelerometer, we multiplied TDEE by 0.90 (to correct for the thermic effect of food, which is assumed to be 10% of TDEE) and subtracted 24 kcal·kg−1·d−1 to account for resting energy expenditure (1 MET × 24 h) (5).

To determine estimates of time per day spent in different intensities of activity, we used two sets of thresholds. For data based on vertical axis counts only, we used thresholds described previously by Troiano et al. (33) as primarily reflecting locomotor activity: moderate intensity (3–5.99 METs) = 2020–5999 counts per minute; and vigorous intensity (≥6 METs) = ≥6000 counts per minute. Light-intensity physical activity was defined as the total number of minutes between 100 and 2020 counts per minute (21). We also examined associations for moderate-intensity activity when the threshold was lowered to the Matthews’ cut point of 760 counts per minute (MVPA-760), which may allow better capture of the full range of moderate-intensity activities encountered in daily living (20). For data based on triaxial counts, we used thresholds determined by Sasaki et al. for MVPA: moderate intensity (3–5.99 METs) = 2690–6166 counts per minute; and vigorous intensity (≥6 METs) = ≥6167 counts per minute (29). Light-intensity physical activity was defined as the total number of minutes between 200 and 2690 counts per minute (1). Because there is not yet a published threshold equivalent to the MVPA-760 based on triaxial counts, two of the authors (A.K.C. and D.R.B.) estimated a threshold of 1010 counts per minute by graphing the vertical axis and triaxial thresholds against each other (see Figure, Supplemental Digital Content 1,, Thresholds for vertical axis counts graphed against triaxial counts). A bout of sedentary behavior was defined as consecutive minutes in which the accelerometer registered less than 100 counts per minute for vertical axis counts and less than 200 counts per minute for triaxial counts (1,21).

Time spent in activity of a defined intensity (sedentary, light, moderate, vigorous, or MVPA) was determined by summing minutes in a day where the count met the criterion for that intensity. Time spent in physical activity is presented for every minute that meets the specific criterion as well as for modified 10-min activity bouts, as described previously (33). Mean daily time in intensity-specific categories was calculated across all valid wear days.


The DLW method was used to assess TDEE. Bottled DLW was mailed to participants a few days before they underwent the protocol. Participants provided four urine samples, two directly before and two after the administration of the DLW dose (i.e., 4.5 and 6 h postdose). The 4.5-h time point was not included, however, if the enrichment was less than the 6-h time point. After 10 to 14 d, participants provided two more urine samples on the same day, but at least 30 min apart, after the same procedures. Participants recorded the date and time that all samples were collected. All samples were sent to Dr. Jennifer Rood at the Mass Spectrometry Core at Pennington Biomedical Research Center to determine energy expenditure via mass spectroscopic analysis of urine specimens for deuterium and oxygen-18 (30,31). To estimate PAEE, we subtracted resting metabolic rate and the thermic effect of food from TDEE (5). Resting metabolic rate was calculated based on sex, weight, height, and age as described by Mifflin et al. (23) For thermic effect of food, we used a constant 10% of total energy (5).

For the analysis, TDEE and PAEE were expressed in kilocalories per kilogram per day to adjust for any differences in energy expenditure due to body mass. Additionally, multivariable models included self-reported body weight and fat free mass estimated from DLW.

Statistical Analysis

All analyses were performed using SAS statistical software, version 9.4 (SAS Institute Inc, Cary, NC). Analyses were run separately in WLVS and MLVS. Mean values and standard deviations (S.D.) were calculated for baseline characteristics, DLW measures, and accelerometer variables.

As the distributions of many accelerometer variables were skewed, they were log-transformed before analysis. Because DLW-measured PAEE in kilocalories per kilogram per day was not normally distributed despite transformation, Spearman correlation coefficients were calculated. Partial correlations adjusted for weight, age, accelerometer wear time, and fat free mass were calculated between TDEE and PAEE measured by DLW and TAC per day, steps per day, PAEE, minutes per day of MVPA, and minutes per day of MVPA using the lower threshold (MVPA-760/1010) measured by accelerometer. Additionally, correlations between TAC per day and the other accelerometer variables were examined. To adjust for random within-person variation in energy expenditure measured by DLW, we calculated deattenuated Spearman correlations using the methods of Rosner et al. for interclass correlations (27).


Among the 796 women who were enrolled in the WLVS, 775 women had a least one accelerometer measurement with at least 4 d with ≥10 h of wear per day. We excluded 31 women with negative values for accelerometer-estimated PAEE. We then excluded 23 women without a DLW measurement and 37 women with >40,000 steps per day, resulting in 684 women included in this analysis. Among 671 men enrolled in the MLVS, 667 men had a least one accelerometer measurement with at least 4 d with ≥10 h of wear per day. We excluded 27 men with negative values for accelerometer-estimated PAEE. We then excluded 17 men without a DLW measurement and 12 men with >40,000 steps per day, resulting in 611 men included in this analysis.

The mean (± S.D.) age of women included in this analysis was 62.8 (±9.4) and the mean (± SD) age of men was 67.9 (±7.7) (Table 1). Women had higher body mass index, and lower weight, than men. Men had higher mean TDEE (2778 kcal) and PAEE (916 kcal) than women (2198 and 722 kcal, respectively) as measured by DLW. The intraclass correlations for TDEE and PAEE are provided in Table, Supplemental Digital Content 2, Intraclass correlation coefficients (ICC) for DLW and accelerometer variables,

Baseline characteristics of participants in WLVS and MLVS.

Means and standard deviations for the accelerometer variables are presented in Table 2 and intraclass correlations in Supplemental Digital Content 2. Based on triaxial counts, women wore the accelerometer for a mean (± S.D.) of 15.1 (±1.0) h·d−1 over a mean (± SD) of 6.8 (±0.4) d; wear time for men was similar. Mean (± S.D.) wear time was less when using vertical axis counts: 14.8 (±1.1) h·d−1 for women and 14.7 (±1.1) h·d−1 for men. For women and men, mean TAC per day was higher using triaxial counts (597,545 and 602,923, respectively) compared to vertical axis counts (249,380 and 290,024, respectively), but the difference between men and women was much less when using triaxial counts. Amount of time spent in light, moderate, and vigorous activity was higher for women and men when triaxial counts were used compared with vertical axis counts, whereas sedentary time was lower. Using the 2020 counts per minute threshold and including every minute, women and men engaged in a mean (±SD) of 20.2 (±16.9) and 31.8 (±21.7) min·d−1 of MVPA based on vertical axis counts. When using triaxial counts and the 2690 counts per minute threshold, these means (±SD) nearly doubled to 41.6 (±24.9) min·d−1 for women and 51.6 (±28.3) min·d−1 for men.

Means (SD) for accelerometer variables using triaxial (3 axes) vs vertical axis only.

Using results presented in Tables 1 and 2, we can quantitatively compare TDEE and PAEE measured by DLW to TDEE and PAEE estimated by accelerometer. In Table 1, TDEE determined by DLW was 31.7 kcal·kg−1·d−1 for women and 34.5 kcal·kg−1·d−1 for men. In women, TDEE estimated by accelerometer ranged from 29.7 to 32.6 kcal·kg−1·d−1, which was similar to TDEE measured by DLW (Table 2). In men, accelerometer-estimated TDEE ranged from 30.2 to 32.6 kcal·kg−1·d−1, which was slightly lower than TDEE measured by DLW (Table 2). In contrast to TDEE, PAEE estimated by accelerometer was quite a bit lower than PAEE measured by DLW. PAEE determined by DLW was 10.5 kcal·kg−1·d−1 in women and 11.5 kcal·kg−1·d−1 in men (Table 1). PAEE estimated by accelerometer ranged from 2.8 to 5.3 kcal·kg−1·d−1 in women and 3.1 to 5.3 kcal·kg−1·d−1 in men (Table 2). The difference between PAEE determined by DLW and PAEE estimated by accelerometer is also shown in Bland-Altman plots (see Figures, Supplemental Digital Content 3, Bland–Altman plots comparing PAEE estimated by DLW to PAEE estimated by accelerometer, Based on the plots, it appears that the error in the accelerometer estimate of PAEE is minimal at lower levels of PAEE but is greater at higher levels.

As the purpose of this paper is to validate a measure of physical activity, and correlations were similar when using TDEE or PAEE, we will focus on PAEE. In general, the Spearman correlations between accelerometer-assessed physical activity and PAEE measured by DLW were higher for triaxial counts than vertical axis counts for most of the accelerometer variables (Table 3). After adjusting for weight, age, wear time, and fat free mass, TAC per day, accelerometer-estimated PAEE (using a constant 2 METs for light activity), and time spent in MVPA-1010 based on triaxial counts and including every minute had the highest correlations with DLW-determined PAEE (ρ = 0.44 for each) in women. In men, steps per day had the highest correlation with PAEE (ρ = 0.42), while correlations with MVPA-1010, TAC per day, and accelerometer-estimated PAEE (constant) using triaxial counts were similar. The correlations with DLW-determined PAEE were lowest for MVPA using the higher (i.e., 2690 or 2020 counts per minute) threshold, particularly when only modified bouts of ≥ 10 min were included (ρ = 0.18 and 0.21 for women and men, respectively, using triaxial counts). After adjusting for random within-person error in DLW-determined PAEE, the Spearman correlations for time spent in MVPA-1010 including every minute, TAC per day, and accelerometer-estimated PAEE (constant) based on triaxial counts with DLW-determined PAEE increased to 0.64 (95% confidence interval [CI], 0.51–0.74), 0.61 (95% CI, 0.47–0.71), and 0.59 (95% CI, 0.46–0.69), respectively, in women. The deattenuated correlations for steps per day and accelerometer-estimated PAEE (constant) with DLW-determined PAEE increased to 0.53 (95% CI, 0.44–0.61) and 0.51 (95% CI, 0.41–0.59), respectively, in men.

Spearman correlation coefficients between physical activity measured by accelerometer and energy expenditure measured by DLW.

Table 4 presents the correlations between TAC per day and the other accelerometer metrics. TAC per day was highly correlated with steps per day, MVPA-760/1010, and accelerometer-estimated PAEE as well as MVPA when every minute is included with multivariable-adjusted correlations ranging from 0.77 to 0.99. Correlations between TAC per day and MVPA in modified bouts ≥10 min were lower, ranging from 0.56 to 0.73.

Spearman correlation coefficients between TAC and other accelerometer measures.


This study showed that TAC per day, steps per day, accelerometer-estimated PAEE, and minutes per day of MVPA based on the lower threshold (760 counts per minute for vertical axis or 1010 counts per minute for triaxial) had similar associations with PAEE determined by DLW. Incorporating accelerometer counts from all three axes by using triaxial vector magnitude resulted in stronger associations between each of the accelerometer variables and DLW-determined PAEE compared to only using counts from the vertical axis. Additionally, associations between the accelerometer measures and PAEE were stronger when every minute of activity was included versus only including minutes of activity that were accumulated in modified bouts of ≥10 min. Notably, minutes per day of MVPA based on the threshold of 2020 counts per minute (vertical axis counts) or 2690 counts per minute (triaxial counts) and accumulated in modified bouts of ≥10 min showed the weakest association with PAEE in the current study, which is not surprising as MVPA accumulated in 10-min bouts is only a small subset of all physical activity performed.

Previous studies have validated accelerometers against DLW. Recently, Plasqui and colleagues (26) systematically reviewed all accelerometer validation studies (2007–2011) using DLW as the reference. Although several studies included in the review validated the Actigraph, they were limited to the 7164 and GT1M models, both of which are uniaxial accelerometers, and the majority of the studies were done in children. Additionally, only three studies reported a partial correlation or R increase which is necessary for interpreting correlations between TAC and DLW-determined PAEE because body weight, age, and FFM are determinants of PAEE, even when expressed as kilocalories per kilogram per day (26). For the uniaxial Actigraph accelerometer, unadjusted correlations between TAC and DLW-determined PAEE ranged from 0.37 to 0.56 in studies that included adult men and women (5,26). For the triaxial accelerometers, unadjusted correlations ranged from 0.32 to 0.79 in adults (26). Partial correlations between TAC and PAEE adjusted for body mass or fat mass/fat free mass using triaxial accelerometers in men and women ranged from 0.27 to 0.55, which is comparable to the current study (26).

We found that values of PAEE estimated by accelerometer were 50% to 73% lower than PAEE determined by DLW. This was true whether vertical axis counts or triaxial counts were used in the MET prediction equations, although using triaxial counts produced estimates closer to DLW. These results were similar to those reported by Leenders et al. (17) who found that estimates of PAEE from the triaxial Tritrac accelerometer and uniaxial CSA (predecessor to Actigraph) accelerometer were 35% and 59% lower, respectively, than PAEE determined by DLW. A possible reason for this discrepancy is that waist-worn accelerometry underestimates energy expenditure because it does not register some types of activity, for example arm movements, walking uphill (versus on a flat surface), or additional work performed when carrying objects. In contrast, when we quantitatively compared DLW-determined TDEE and accelerometer-estimated TDEE, the values were much closer, with differences ranging from only −13% to +3%. Another study conducted by Leenders and colleagues (16) also found accelerometer-estimated TDEE to be closer to DLW than accelerometer-estimated PAEE. In their study, estimates of TDEE from published regression equations using data from both a uniaxial and triaxial accelerometer under- and overestimated DLW-determined TDEE, with percent differences ranging from −32% to +16% (with an additional equation overestimating TDEE by 101%). It makes sense that estimates of TDEE from the accelerometer and DLW are in closer agreement than estimates of PAEE because resting metabolic rate, which has fairly low day-to-day variability, accounts for 65%–70% of TDEE (23).

The current study provides evidence that using accelerometer data from all three axes provides a better measure of physical activity volume than only using data from the vertical axis. Prior research comparing uniaxial and triaxial accelerometers has yielded conflicting results. Some previous studies suggest that triaxial accelerometers provide a more accurate estimate of DLW-determined PAEE than uniaxial devices (11,28). Hendelman et al. (11) reported stronger correlations between energy cost and activity counts from the triaxial Tritrac accelerometer compared to the uniaxial CSA accelerometer. A study by Howe et al. (12), however, used the RT3 triaxial accelerometer in 212 individuals and found triaxial counts did not significantly improve the relationship between TAC and AEE compared with vertical axis counts. The study by Howe et al. compared vertical axis and triaxial counts from the same device, whereas the other studies compared counts from two different devices which can be complicated by different monitor technology and data processing (3). Nevertheless, in the current study, we compared metrics measured with the same monitor (Actigraph GT3X) and found metrics using triaxial counts had a higher correlation with PAEE measured by DLW compared to metrics using vertical counts. This indicates that movements that result in horizontal accelerations in the anterior–posterior and medial–lateral directions expend energy; thus, accounting for this by using triaxial counts increases the correlation between accelerometer metrics and energy expenditure measured by DLW.

This study further demonstrates the effects of using different investigator-defined cutpoints for MVPA. The estimates of time spent in MVPA varied greatly when the threshold of 2020 counts per minute was used compared to the threshold of 760 counts per minute. Additional factors that affected the estimates of time spent in MVPA were: (a) whether vertical axis or triaxial counts were used, and (b) whether every minute of activity was included or only modified bouts of ≥10 min. In the current analysis, correlations with DLW-estimated PAEE were highest for MVPA when triaxial counts and the lower threshold (1010 counts per minute) were used and every minute of activity was included.

The current study is important because we directly compared various summary accelerometer metrics against DLW, the accepted gold standard for energy expenditure. Previous studies examining TAC per day versus minutes per day of MVPA have used an indirect validation approach in that they considered correlations between the accelerometer summary variables and biomarkers known to be correlated with physical activity, HDL cholesterol and triglycerides (40). Strengths of the current study include the fact that it included a much larger sample of men and women than previous studies that compared accelerometers to DLW. In addition, the study sample included middle-age and older adults with a wide age range. As we measured DLW twice in a subset of study participants, we were also able to adjust for the within-person variation in energy expenditure measured by DLW.

Our study also has several important limitations that should be mentioned. First, the study population, consisting of predominantly white women and men of higher socioeconomic status, is not representative of the general population. Another limitation is that TDEE and PAEE may have been underestimated as the study participants included older adults for whom urinary retention can be a substantial problem. If the bladder is not completely voided before consuming the DLW dose, the postdose measurements are diluted and the peak isotopic enrichments are underestimated.

Additionally, neither of the two accelerometer measurements occurred at the same time as the DLW measurement; the DLW and accelerometer measurement that were done in the same 3-month time window were measured on average 32 to 45 d apart. Thus, the true correlations between short term energy expenditure measured by DLW and the accelerometer measures of physical activity are almost certainly higher than those reported in the current study (14). However, collecting both measures concurrently overestimates the true long term correlation between these methods, which is usually the focus of epidemiologic studies with disease outcomes (39). Furthermore, resting energy expenditure was not measured, but was estimated using the Mifflin–St. Jeor equation for the DLW data and 1 kcal·kg−1·h−1 for the accelerometer data. These additional sources of error may have attenuated the correlations between the accelerometer metrics and PAEE estimated by DLW.

Another limitation was that we chose to enable the low-frequency extension on the ActiGraph GT3X accelerometer. While this has the benefit of making the ActiGraph GT3X counts per minute values resemble the previous Actigraph 7164, it nearly doubles the steps per day compared to use of the default filter (34), and it vastly overestimates steps per day compared to a criterion device, the ankle-worn StepWatch. In contrast, the default filter produces 24-h step counts that are much more reasonable (7). Unfortunately, it was not possible for us to reanalyze the data with the extension turned off because we no longer have the raw.gt3x files. Additionally, we elected not to use “censored steps,” a method in which any steps taken during minutes where the Actigraph counts per minute were <500 are not counted, in the current analysis (36). This approach can potentially eliminate a lot of light-intensity physical activity and intermittent MVPA. Furthermore, previous work has shown in free-living adults that the Actigraph 7164 steps per day (uncensored) corresponded well with the StepWatch (6).

In conclusion, we found that measures of total activity volume assessed by an accelerometer were reasonably correlated with PAEE measured by DLW after accounting for within-person variation in the DLW. This study is the first to compare the triaxial Actigraph GT3X accelerometer with PAEE measured by DLW and indicates that metrics using triaxial counts are more highly correlated with PAEE compared to metrics using vertical axis counts. TAC per day had higher correlations with DLW-determined PAEE than minutes per day of MVPA using the 2020 counts per minute threshold, consistent with prior research showing stronger associations between TAC per day and cardiometabolic biomarkers compared to MVPA (39). Additionally, steps per day, minutes per day of MVPA based on the lower threshold (760 counts per minute for vertical axis or 1010 counts per minute for triaxial), and accelerometer-estimated PAEE had similar correlations with DLW-measured PAEE and were highly correlated with TAC per day. One of the major advantages of accelerometers over physical activity questionnaires is that accelerometers can capture ubiquitous, light-intensity activity performed throughout the day. As TAC or other measures of total activity volume account for time spent in activities of all intensity levels, it may be preferable over minutes per day of MVPA for researchers seeking to reduce accelerometer data down to a single metric. Nonetheless, it is important to note that TAC per day is not a perfect measure of PAEE, but with the development of more advanced pattern recognition methods there is potential for improved approaches to capture all movement that contributes to energy expended in physical activity. Finally, we recognize that physical activity has multiple dimensions and researchers will continue to be interested in questions related to frequency, intensity, and duration of activity bouts. Thus, measures of total activity volume should complement, not replace, other physical activity metrics.

The authors thank Kate Clowry and Sean Sinnott for their help in coordination and data management of the WLVS and MLVS as well as the study staff and participants for their devotion to detail and data quality.

This study was supported National Institute of Health grants UM1-CA186107, UM1-CA176726, UM1-CA167552, P01-CA055075-18S1, U01-CA152904, and CA55075.

Disclosures: David R. Bassett, Jr., is a member of the ActiGraph Scientific Advisory Board. For the remaining authors, there are no disclosures.

The results of this study do not constitute endorsement by the American College of Sports Medicine. The results of the study are presented clearly, honestly, and without fabrication, falsification, or inappropriate data manipulation.


1. Aguilar-Farías N, Brown WJ, Peeters GM. ActiGraph GT3X+ cut-points for identifying sedentary behaviour in older adults in free-living environments. J Sci Med Sport. 2014;17(3):293–9.
2. Ainsworth BE, Haskell WL, Whitt MC, et al. Compendium of physical activities: an update of activity codes and MET intensities. Med Sci Sports Exerc. 2000;32(9 Suppl):S498–504.
3. Bassett DR, Troiano RP, McClain JJ, Wolff DL. Accelerometer-based physical activity: total volume per day and standardized measures. Med Sci Sports Exerc. 2015;47(4):833–8.
4. Belanger CF, Hennekens CH, Rosner B, Speizer FE. The nurses' health study. Am J Nurs. 1978;78(6):1039–40.
5. Colbert LH, Matthews CE, Havighurst TC, Kim K, Schoeller DA. Comparative validity of physical activity measures in older adults. Med Sci Sports Exerc. 2011;43(5):867–76.
6. Feito Y, Bassett DR, Thompson DL. Evaluation of activity monitors in controlled and free-living environments. Med Sci Sports Exerc. 2012;44(4):733–41.
7. Feito Y, Garner HR, Bassett DR. Evaluation of ActiGraph's low-frequency filter in laboratory and free-living environments. Med Sci Sports Exerc. 2015;47(1):211–7.
8. Freedson PS, Melanson E, Sirard J. Calibration of the Computer Science and Applications, Inc. accelerometer. Med Sci Sports Exerc. 1998;30(5):777–81.
9. Grobbee DE, Rimm EB, Giovannucci E, Colditz G, Stampfer M, Willett W. Coffee, caffeine, and cardiovascular disease in men. N Engl J Med. 1990;323(15):1026–32.
10. Hamilton MT, Hamilton DG, Zderic TW. Role of low energy expenditure and sitting in obesity, metabolic syndrome, type 2 diabetes, and cardiovascular disease. Diabetes. 2007;56(11): 2655–67.
11. Hendelman D, Miller K, Baggett C, Debold E, Freedson P. Validity of accelerometry for the assessment of moderate intensity physical activity in the field. Med Sci Sports Exerc. 2000;32(9 Suppl):S442–9.
12. Howe CA, Staudenmayer JW, Freedson PS. Accelerometer prediction of energy expenditure: vector magnitude versus vertical axis. Med Sci Sports Exerc. 2009;41(12):2199–206.
13. John D, Sasaki J, Staudenmayer J, Mavilia M, Freedson PS. Comparison of raw acceleration from the GENEA and ActiGraph GT3X+ activity monitors. Sensors (Basel). 2013;13(11):14754–63.
14. Lee IM. Epidemiologic Methods in Physical Activity Studies. Vol. xii. New York: Oxford University Press; 2009. pp. 328.
15. Lee IM, Shiroma EJ. Using accelerometers to measure physical activity in large-scale epidemiological studies: issues and challenges. Br J Sports Med. 2014;48(3):197–201.
16. Leenders NY, Sherman WM, Nagaraja HN. Energy expenditure estimated by accelerometry and doubly labeled water: do they agree? Med Sci Sports Exerc. 2006;38(12):2165–72.
17. Leenders NY, Sherman WM, Nagaraja HN, Kien CL. Evaluation of methods to assess physical activity in free-living conditions. Med Sci Sports Exerc. 2001;33(7):1233–40.
18. Loprinzi PD, Lee H, Cardinal BJ, Crespo CJ, Andersen RE, Smit E. The relationship of actigraph accelerometer cut-points for estimating physical activity with selected health outcomes: results from NHANES 2003–06. Res Q Exerc Sport. 2012;83(3):422–30.
19. Lyden K, Kozey SL, Staudenmeyer JW, Freedson PS. A comprehensive evaluation of commonly used accelerometer energy expenditure and MET prediction equations. Eur J Appl Physiol. 2011;111(2):187–201.
20. Matthews CE. Calibration of accelerometer output for adults. Med Sci Sports Exerc. 2005;37(11 Suppl):S512–22.
21. Matthews CE, George SM, Moore SC, et al. Amount of time spent in sedentary behaviors and cause-specific mortality in US adults. Am J Clin Nutr. 2012;95(2):437–45.
22. Matthews CE, Moore SC, Sampson J, et al. Mortality benefits for replacing sitting time with different physical activities. Med Sci Sports Exerc. 2015;47(9):1833–40.
23. Mifflin MD, St Jeor ST, Hill LA, Scott BJ, Daugherty SA, Koh YO. A new predictive equation for resting energy expenditure in healthy individuals. Am J Clin Nutr. 1990;51(2):241–7.
24. Pettee Gabriel KK, Morrow JR Jr, Woolsey AL. Framework for physical activity as a complex and multidimensional behavior. J Phys Act Health. 2012;9(1 Suppl):S11–8.
25. Physical Activity Guidelines Advisory Committee. Physical Activity Guidelines Committee Report, 2008. Washington, D.C.: U.S. Department of Health and Human Services; 2008. 683 p.
26. Plasqui G, Bonomi AG, Westerterp KR. Daily physical activity assessment with accelerometers: new insights and validation studies. Obes Rev. 2013;14(6):451–62.
27. Rosner B, Glynn RJ. Interval estimation for rank correlation coefficients based on the probit transformation with extension to measurement error correction of correlated ranked data. Stat Med. 2007;26(3):633–46.
28. Rothney MP, Schaefer EV, Neumann MM, Choi L, Chen KY. Validity of physical activity intensity predictions by ActiGraph, Actical, and RT3 accelerometers. Obesity (Silver Spring). 2008;16(8): 1946–52.
29. Sasaki JE, John D, Freedson PS. Validation and comparison of ActiGraph activity monitors. J Sci Med Sport/Sports Med Australia. 2011;14(5):411–6.
30. Schoeller DA, Colligan AS, Shriver T, Avak H, Bartok-Olson C. Use of an automated chromium reduction system for hydrogen isotope ratio analysis of physiological fluids applied to doubly labeled water analysis. J Mass Spectrom. 2000;35(9):1128–32.
31. Schoeller DA, Luke AH. Rapid 18O analysis of CO2 samples by continuous-flow isotope ratio mass spectrometry. J Mass Spectrom. 1997;32(12):1332–6.
32. Shephard RJ. Limits to the measurement of habitual physical activity by questionnaires. Br J Sports Med. 2003;37(3):197–206.
33. Troiano RP, Berrigan D, Dodd KW, Mâsse LC, Tilert T, McDowell M. Physical activity in the United States measured by accelerometer. Med Sci Sports Exerc. 2008;40(1):181–8.
34. Tudor-Locke C, Barreira TV, Schuna JM Jr. Comparison of step outputs for waist and wrist accelerometer attachment sites. Med Sci Sports Exerc. 2015;47(4):839–42.
35. Tudor-Locke C, Camhi SM, Troiano RP. A catalog of rules, variables, and definitions applied to accelerometer data in the National Health and Nutrition Examination Survey, 2003–2006. Prev Chronic Dis. 2012;9:E113.
36. Tudor-Locke C, Johnson WD, Katzmarzyk PT. Accelerometer-determined steps per day in US adults. Med Sci Sports Exerc. 2009;41(7):1384–91.
37. US Department of Health and Human Services. Physical activity and health: a report of the Surgeon General. Atlanta, GA: US Department of Health and Human Services, Public Health Service, CDC, National Center for Chronic Disease Prevention and Health Promotion; 1996.
38. Wanner M, Martin BW, Meier F, Probst-Hensch N, Kriemler S. Effects of filter choice in GT3X accelerometer assessments of free-living activity. Med Sci Sports Exerc. 2013;45(1):170–7.
39. Willett WC, Lenart EB. Chapter 6: Reproducibility and Validity of Food Frequency Questionnaires. In: Willett WC, editor. Nutritional Epidemiology. Third ed. New York: Oxford University Press; 2013. pp. 96–141.
40. Wolff DL, Fitzhugh EC, Bassett DR, Churilla JR. Total Activity counts and bouted minutes of moderate-to-vigorous physical activity: relationships with cardiometabolic biomarkers using 2003–2006 NHANES. J Phys Act Health. 2015;12(5):694–700.


Supplemental Digital Content

© 2017 American College of Sports Medicine