Validation of the TracmorD Triaxial Accelerometer to Assess Physical Activity in Preschool Children


  • Relevant conflicts of interest/financial disclosures: Nothing to report. Full financial disclosures and author notes may be found in the online version of this article.

Correspondence: Anna Sijtsma (


Objectives: To assess validity evidence of TracmorD to determine energy used for physical activity in 3-4-year-old children.

Design and Methods: Participants were randomly selected from GECKO Drenthe cohort (n = 30, age 3.4 ± 0.3 years). Total energy expenditure (TEE) was measured using the doubly labeled water method. Sleeping metabolic rate (SMR) was measured by indirect calorimetry (Deltatrac). TEE and SMR were used to calculate physical activity level (PAL) and activity energy expenditure (AEE). Physical activity was monitored using a DirectLife triaxial accelerometer, TracmorD with activity counts per minute (ACM) and activity counts per day (ACD) as outcome measures.

Results: The best predictor for PAL was ACM with gender and weight, the best predictor for AEE was ACM alone (backward regression, R2 = 0.50, P = 0.010 and R2 = 0.31, P = 0.011, respectively). With ACD, the prediction model for PAL included ACD, height, gender, and sleep duration (R2 = 0.48, P = 0.033), the prediction model for AEE included ACD, gender and sleep duration (R2 = 0.39, P = 0.042). The accelerometer was worn for 5 days, but 3 days did not give a different estimated PAL.

Conclusion: TracmorD provides moderate-to-strong validity evidence that supports its use to evaluate energy used for physical activity in 3-4-year-old children.


Overweight and obesity are a considerable health problem in developed countries. Not only in adults but also in children the prevalence of obesity is growing [1, 2]. Children with overweight or obesity have a higher chance to be obese in adulthood than children without this condition [3]. To prevent obesity and related co morbidities, risk factors early in life need to be investigated. Low physical activity and high sedentary behavior are known risk factors for overweight and obesity in adults [4] and children [5]. Also in preschool children, it is suggested that physical activity is inversely related to percentage body fat, but more studies are needed to draw firm conclusions [6]. Besides the predictive value of physical activity on the development of obesity, obesity discourages physical activity, which promotes further weight gain [7]. Validated devices to evaluate physical activity in preschoolers are necessary for valid results. Physical activity can be defined as any bodily movement produced by the contraction of skeletal muscle that increases energy expenditure above a basal level. The energy used for physical activity, the activity energy expenditure (AEE), can be calculated from the total energy expenditure (TEE) minus the basal metabolic rate (BMR) (or sleeping metabolic rate (SMR)) and diet induced thermogenesis (about 10% of total energy intake). BMR and SMR can be measured by indirect calorimetry. Another method is to calculate the physical activity level (PAL) as TEE/BMR (or SMR) [8]. Measuring TEE by the doubly labeled water (DLW) method is generally accepted as “the gold standard.” Disadvantages of the DLW method are that it is expensive and both the DLW method and the indirect calorimetry are time consuming. A relative new device for measuring physical activity is the triaxial accelerometer. A triaxial accelerometer has shown to measure reliably physical activity during the day in adults [9, 10] and children [11]. Next to the total AEE and PAL, it provides information about duration, frequency, and intensity of the physical activity. Few studies have investigated the validity of triaxial accelerometers in young children aged 4-6 years [12, 13]. So far, the TracmorD was only validated in adults, the outcome of the TracmorD explained 46% of the variance in AEE and PAL [10]. From the perspective of prevention of obesity, it is important to validate the triaxial accelerometers in children aged younger than 5 years. Therefore, the aim of this study is to assess validity evidence of the TracmorD, a triaxial accelerometer, to determine energy used for free-living physical activity in 3-4-year-old children.

Methods and Procedures


Thirty preschool children aged 3-4 years old, living in Drenthe, one of the northern provinces of The Netherlands participated in this study. They were randomly selected from the GECKO-Drenthe birth cohort (n = 2,997 signed informed consent and n = 2,629 are still participating). GECKO-Drenthe is a population-based birth-cohort studying the risk factors for overweight in young children [14]. The inclusion criterion for this validation study was born in January, February, or March 2007. The exclusion criterion was having medical limitations which could affect physical activity. Written informed consent was obtained, and the study was approved by the Medical Ethics Committee of the University Medical Center Groningen.

Body composition

Anthropometric measurements were carried out in the afternoon on day 0 at the child's home. Height to the nearest 0.1 cm and weight to the nearest 0.1 kg were measured without shoes in light clothing using a calibrated digital scale and a stadiometer, respectively. Body mass index (BMI) was calculated as weight (kg)/height squared (m2).

Sleeping metabolic rate

SMR was measured using a ventilated hood: the Deltatrac II MBM-200 metabolic monitor (Datex Ohmeda, Finland). The Deltatrac is an open system indirect calorimetry device that measures inline imageO2 and inline imageCO2 and, from these variables, calculates the respiratory quotient (RQ = inline imageCO2/ inline imageO2) and energy expenditure [15]. In the evening of day 0 at the child's home, during sleep, the child was covered with a ventilated hood for 30 min. The measurement of inline imageO2 and inline imageCO2 started after the gasses were stable in the hood (approximately after 5 min). When the child woke up, the hood was removed and the measurement was started again after the child falling back asleep. When the child woke up and would not sleep anymore, we tried to obtain the measurement on another evening within 2 weeks.

Total energy expenditure

TEE was measured with the DLW method. After a baseline saliva sample at day 0, participants drank a weighted amount (around 50,000 g) of DLW (Buchem, Apeldoorn, The Netherlands (2H2O: 6.02%, H218O: 12.05%); Campro Scientific, Berlin, Germany (2H2O: 6.63%, H218O: 11.50%)). Participants drank the DLW from a tube with a straw. Before and after drinking, the tube with the straw was weighted on a calibrated digital scale in milligrams. Saliva samples were collected by the parents at ∼4, 16, 72, and 120 h after administration of the DLW dose, who recorded the exact date and time of the collection. Saliva was sampled by swabbing a predried cotton rod (Sugi® Kettenbach, Eschenburg, Germany) in the child's mouth for 1-2 min and then putting the cotton rod in an airtight plastic container. Participants were not allowed to eat or drink for 30 min before saliva sampling. Parents stored the saliva sample in their refrigerator (0-7°C) till the researcher visited the child at day 6 or 7 to pick up the saliva samples and the accelerometer. From then, the saliva samples were stored frozen at −80°C before analysis. Enrichment of the 2H and 18O isotopes in the samples was measured in quintuplet by a high-temperature conversion elemental analyzer coupled with a Delta XP isotope-ratio mass spectrometer via a Conflo-III Interface (Thermo Fisher, Bremen, Germany) as described previously [16]. The results of every first 2 analyses were skipped and the main of the results of every last three analyses was used for calculation of the total body water (TBW). TBW was calculated by estimating the isotope dilution spaces (IDS) of 2H and 18O. Dilution space is calculated by determining the intercept of 2H and 18O, using the isotope dilution at three time points.

display math

The constants 1.007 and 1.041 were included to adjust for the differences between the IDS and TBW due to isotope exchange [17]. TBW was used to calculate rCO2 using the following equation, which is an adapted version by Racette et al. [17] of the original data by Schoeller et al. [18].

display math

where KO and KD are the 18O and 2H isotope disappearance rates, respectively, and RGf is the rate of water loss trough gaseous routes subject to isotope fractionation. The latter is estimated as 1.05TBW (1.007KO − 1.041KD) [19].

TEE was calculated by using the modified Weir's equation [20] from the measured mean daily carbon dioxide production rate (in mol/day):

display math

The rCO2 is expressed in L/day and can be converted from mol/day by multiplying by 22.4. The RQ is oxygen consumption/rCO2 [19]. It was assumed that children did not lose or gain weight during the measurement period. Without over or under eating, the food quotient most closely approximates the RQ during the day [21]. Food quotients were estimated according to the RIVM macro nutrient intake for age and gender in the Netherlands [22]. The values of the food quotients were 0.884 for the 3-year-old boys, 0.881 for the 4-year-old boys, 0.885 for the 3-year-old girls, and 0.882 for the 4-year-old girls. More detailed information about the calculations of TBW, rCO2, and TEE is described previously [19]. AEE was calculated as TEE − SMR. The PAL was calculated as TEE/SMR [8].


The DirectLife triaxial accelerometer for movement registration, TracmorD (Philips DirectLife, Amsterdam, The Netherlands) was used to estimate the free-living physical activity for five consecutive days during waking hours except for water activities. The accelerometer was worn at the same days as when the TEE was measured by the DLW method. Parents reported sleeping and wake times and times of wearing the accelerometer. The TracmorD is a triaxial accelerometer, equipped with a piezo-capacitive accelerometer, which allows the detection of both dynamic and static accelerations. It measures 3.2 × 3.2 × 0.5 cm and weights 12.5 g (battery included). Because of its small size and light weight the TracmorD does not interfere with daily activities. The accelerometer was worn in the middle of the lower back in a small case on a belt. Detailed information about the TracmorD has been described before [10]. Accelerometer output (counts/minute) is based upon a 20 Hz measurement of the x-axis, y-axis, and z-axis. Output of the accelerometer, as mean of all three axis of measurement, is defined as mean activity counts per minute (ACM) and as mean total activity counts per day (ACD). ACM is calculated as the sum of all counts/min during the wear period divided by the total amount of wear time in minutes. ACD is calculated as the sum of all counts/min of wear time of the valid days divided by the amount of valid days. A day was invalid when the time during which they wore the accelerometers and the time spent sleeping did not add up to at least 19 h.


Pearson correlations and linear regression analyses were made with TEE, AEE, or PAL as dependent variable and ACD or ACM as independent variables. Prediction models of AEE and PAL were calculated using linear backward regression analyses with weight, gender, sleep duration, and ACM or ACD as prediction factors (entry criteria: 0.20, removal criteria: 0.25). Prediction models of TEE were calculated using linear backward regression analyses with SMR, weight, gender, sleep duration, and ACM or ACD as prediction factors (entry criteria: 0.20, removal criteria: 0.25). All regression models were checked on multicollinearity (<0.7), and the residuals had to be normally distributed. To see if 3 instead of 5 days are enough to make a valid estimation of the PAL, a Bland-Altman plot was made with estimated PAL calculated with 5 days of physical activity measurement as reference variable and estimated PAL with the first 3 days of physical activity measurement as the alternative variable. Mean difference (bias) between the two methods was calculated and tested against zero using a one-sample t-test. Statistical analyses were performed using SPSS 16.0 for Windows (SPSS, Chicago, IL). Difference between the children who were excluded and included for analyses and difference between boys and girls were tested with a Chi square for variables at interval/ratio level, an independent t-test for the normal distributed variables at nominal level and a Mann-Whitney test for the nonparametric variables at nominal level. Graphs were made using GraphPad Prism 5.04 for Windows (GraphPad Software, San Diego, CA, The significance level was set to P < 0.05 (2-tailed).


Thirty children were included in this study. In Table 1, the participant characteristics are shown. Mean accelerometer wear time was 10.0 ± 1.2 h/day, mean SMR was 765 ± 88 kcal/24 h, and mean TEE was 1,301 ± 193 kcal/24 h (Table 1). Two children were overweight and 1 child was obese according to the BMI cut-off points of Cole et al. [23]. No significant differences were found between boys and girls. In 5 children (4 boys, 1 girl), it was not possible to measure the SMR, because they woke up every time we tried. Four children (3 boys, 1 girl) had no valid TEE data. One girl had no accelerometer data because of battery problems. Twenty-five children had complete data to perform the analyses for the prediction model of TEE. Twenty children had complete data to perform the analyses for the prediction model of PAL and AEE. Of those 20 children, mean TEE = 1,260 ± 166, AEE = 484 ± 169, SMR = 776 ± 85, and PAL = 1.6 ± 0.3. The children not included in the analyses consisted of significantly more boys than girls (χ2 = 5.6, P = 0.02), for the other characteristics from Table 1 no significant differences were found between the included and not included children. For the prediction of TEE from the ACD, the measurement was based on 3 valid wear days in 1 child, on 4 valid days in 4 children, and on 5 valid days in 20 children. For the prediction of AEE and PAL from the ACD, the measurement was based on 3 valid wear days in 1 child, on 4 valid days in 2 children, and on 5 valid days in 17 children.

Table 1. Participant characteristics
 NMean ± SDRange (min-max)
  1. ACD, activity counts per day; ACM, activity counts per minute; AEE, activity energy expenditure; PA, physical activity; PAL, physical activity level; SD, standard deviation; SMR, sleeping metabolic rate; TEE, total energy expenditure.
  2. aFor valid days only.
Boys/girls, n3012/18 
Age, y303.5 ± 0.33.1-4.1
Height, m30101.0 ± 5.486.0-113.1
Weight, kg3016.3 ± 1.912.1-21.2
BMI, kg/m23015.9 ± 1.214.1-19.3
Sleep duration, h/day3012.4 ± 0.910.0-14.6
Wear time, h/day2910.0 ± 1.27.3-11.6
Sleep and wear time, h/daya2922.4 ± 1.020.0-23.9
Daily PA (ACM)293,666 ± 5792,791-4,980
Daily PA (ACD), × 1052923 ± 4.716-35
Valid wear days294.8 ± 0.53-5
SMR, kcal/day25765 ± 88610-920
TEE, kcal/day261,301 ± 1931,017-1,734
AEE, kcal/day21485 ± 165186-781
PAL, TEE × SMR−1211.6 ± 0.21.2-2.1

Mean ACM and mean ACD were significantly correlated with PAL (r = 0.61; P = 0.004 and r = 0.46; P = 0.042, respectively). Mean ACM was significantly correlated with AEE (r = 0.56, P = 0.011), but no correlation between mean ACD and AEE was found (r = 0.38, P = 0.098). Mean ACM and mean ACD were not correlated with TEE (r = 0.34; P = 0.094 and r = 0.21; P = 0.326, respectively). The univariate regression analysis of PAL and AEE with ACM or ACD as prediction factors are shown in Table 2.

Table 2. Prediction models of PAL, AEE, and TEE
 ACM as PA outcome ACD as PA outcome
 B95% CIR2 B95% CIR2
  1. P value model *P < 0.05, **P < 0.01.
  2. ACD, activity counts per day; ACM, activity counts per minute; AEE, activity energy expenditure; CI, confidence interval; PA, physical activity; PAL, physical activity level; SMR, sleeping metabolic rate; TEE, total energy expenditure.
  3. an = 20.
  4. bBackward regression analyses with SMR (only for the prediction model of TEE), weight (kg), height (cm), gender (1: boys, 2: girls), sleep duration (hours/day), and ACM or ACD as prediction factors. Entry criteria: 0.20, removal criteria: 0.25.
  5. cn = 25.
Univariate model
Intercept0.55  Intercept0.97  
ACM × 10−42.961.06-4.87 ACD × 10−72.930.12-5.75 
   0.37**   0.21*
Multivariate modelb
Intercept1.04  Intercept0.265  
ACM × 10−42.861.04-4.69 ACD × 10−74.241.16-7.33 
Weight−0.044−0.102 to 0.014 Height−0.011−0.030 to 0.008 
Gender0.15−0.06 to 0.36 Sleep duration0.096−0.022 to 0.213 
   0.50**Gender0.195−0.033 to 0.424 
Univariate model
Intercept−189.8  Intercept110.6  
ACM0.1830.048 to 0.319 ACD × 10−41.650.34 to 3.64 
   0.31*   0.15
Multivariate modelb
Intercept−189.8  Intercept−1,411.6  
ACM0.1830.048-0.319 ACD × 10−42.930.82-5.04 
   0.31*Gender122.3−38.0 to 282.5 
    Sleep duration82.81.9-163.8 
Multivariate modelb
Intercept−19.9  Intercept−1,169.6  
ACM0.1730.022-0.323 ACD × 10−42.710.37-5.05 
SMR0.831−0.0.85 to 1.746 Sleep duration82.48−0.83 to 165.80 
   0.29SMR0.77−0.17 to 1.70 
    Gender117.57−48.41 to 283.54 

Next to the univariate regression analysis, the multivariate backward regression outcome is shown in Table 2. ACM, gender, and weight explained 50% of the variance in PAL (P = 0.010), ACM without other variables included explained 31% of the variance in AEE (P = 0.011). The ACD, height, gender, and sleep duration explained 48% of the variance in PAL (P = 0.033), ACD, gender, and sleep duration explained 39% of the variance in AEE (P = 0.042). The multivariate models predicting TEE with ACM and SMR or with ACD and sleep duration were not significant (P = 0.053 and P = 0.103). In Figure 1, the regression lines of the backward regression models to predict PAL and AEE are shown.

Figure 1.

Measured versus predicted values for physical activity level (PAL) and activity energy expenditure (AEE). A: PALpredicted = 1.04 + 2.86 × ACM × 10−4 − 0.044 × weight + 0.15 × gender. B: PALpredicted = 0.265 + 4.24 × ACD × 10−7 − 0.011 × height + 0.096 × sleep duration + 0.195 × gender. C: AEEpredicted = −189.8 +0.183 × ACM. D: AEEpredicted = −1,411.6 + 2.93 × ACD × 10−4 + 82.8 × sleep duration + 122.3 × gender. ACD, activity counts per day; ACM, activity counts per minute; gender (1: boys, 2: girls); height in cm; sleep duration in minutes; weight in kg.

We also evaluated whether TBW as an indicator of lean body mass was correlated with TEE. We found that TBW explained 36% of the variance in TEE. Adding the ACM, as measured with the activity monitor to the model, improved the explained variance to 43%.

Level of agreement in PAL between 5 and 3 days of predicted physical activity measures by the TracmorD was assessed by using a Bland-Altman plot (Figure 2). PAL was predicted according to the equation PAL = 1.04 + 2.86 × ACM × 10−4 − 0.044 × weight (kg) + 0.15 × gender from Table 2. The difference in calculated PAL between 5 and 3 days wearing the accelerometer was not significant.

Figure 2.

Bland-Altman plot for the level of agreement in predicted physical activity level (PAL) between 5 or 3 days of physical activity measurement by the TracmorD. Mean difference (bias): −0.02 ± 0.07 (−1.2% of mean PAL) and 95% limits of agreement of −0.16; 0.11 (−9.8; 6.8% of mean PAL).


The aim of this study was to test the validity evidence of the TracmorD to estimate energy expenditure used for free-living physical activity in preschoolers. This study showed that the TracmorD provides moderate-to-strong validity evidence that supports its use to evaluate physical activity in this age group. Output of the TracmorD in ACM together with weight and gender gives a valid estimation of PAL, and ACM individually, gives a valid estimation of AEE. Output of the TracmorD in ACD together with height, gender, and sleep duration gives a valid estimation of PAL and ACD together with gender and sleep duration of AEE. In our study, the accelerometer was worn for 5 days, but we found 3 days of wearing the accelerometer did not gave a different estimated PAL.

In the univariate model, ACM gave a higher explained variance than ACD, whereas in the multivariate model we found the opposite for the prediction of AEE. One of the differences in ACM and ACD is sleep duration. A higher sleep duration results in less wear time. Therefore, two children with comparable ACM, but with different sleep duration could have different ACD. For the multivariate prediction models of PAL and AEE less prediction variables are needed when the ACM as outcome of the TracmorD is used instead of the ACD. Above that, the use of ACM needs no correction for nonvalid wear days. Therefore, for the prediction of PAL or AEE, ACM instead of ACD as output of the accelerometer might be more practical.

TracmorD is validated before in adults [10]. In that study, weight and mean ACD explained 46% of the variance in AEE and mean ACD explained 46% of the variance in PAL. In a small group of children (n = 11) with a wide age range (3-11 years), an earlier version, the Tracmor2, was validated [11]. This earlier version is of a larger size and weight and contains other sensors [9]. Mean ACD predicted 62% of the variance in PAL [11]. Few studies validated accelerometers in preschool children. Only Actical, Actiwatch, and Actigraph were validated in children aged 2-5 years [24]. In this age group, only the CSA/MTI accelerometer (Actigraph) was validated against the DLW method to predict the AEE or the PAL. REE was estimated from the Schofield equation to calculate the PAL. In 104 children, mean age 5.5 years (range 2.6-6.9), accelerometer output (ACM) explained significantly 11% of the variance in PAL [25]. In our study, we found an explained variance of 31%. Different factors could contribute to this higher explained variance. First, we used a triaxial instead of uniaxial measurements. Second, we measured on exactly the same days TEE by the DLW method and physical activity by the accelerometer, whereas Montgomery et al. did not [25]. Finally, REE was measured by the SMR from the ventilated hood instead of using the Schofield equation.

Important for the validation of the TracmorD is that we used the gold standard for TEE, the DLW method, as reference method for TEE, and we measured SMR by a ventilated hood instead of estimation of the REE according to the Schofield equation.

A first limitation of our study is the relatively small sample size with complete data. Second, we measured SMR instead of BMR. BMR is the official variable used to calculate PAL from TEE. BMR is the resting, awake, thermo neutral, fasting energy expenditure. Children at this age are not willing to lay down for half an hour under a ventilated hood and fasting is a problem. Therefore, we replaced the BMR measurement by the “evening” SMR measurement which was the best and most feasible option in children of this age category. Treuth et al. [26] found a small, but not significant difference between the overnight SMR and the BMR, measured in the morning in 7-10-year-old girls. We measured the SMR rather soon after a meal; therefore, it included also the diet induced thermogenesis. The SMR, as measured by us, might, therefore, be 5-10% higher than the BMR. Using the SMR in the early evening instead of the BMR might have given a small underestimation, but no overestimation of the PAL.

According to the FAO/WHO/UNU, the PAL value in 3-4-year-old children is 1.44. In our study, PAL was 1.64. The FAO/WHO/UNU calculated PAL as TEE/BMRestimated. The BMR estimation was based on the predictive equations on body weight according to Schofield (1985) (27). Calculation of the BMR using the Schofield equation in our study gave a mean BMR of 860 kcal (results not shown). This is higher than the measured evening SMR, 765 kcal. When we calculated the PAL using the BMR estimated from the Schofield equation, the PAL is 1.46, what is almost equal to the PAL value given by the WHO: 1.44. Because of the lower value for the SMR, a higher value for the PAL was found. The most likely explanation is that the Schofield equation overestimates the BMR. When comparing the BMR estimated from the equation and our results, we found that the equation specially overestimates in the infants with a lower SMR. When looking at the Bland-Altman plot (results not shown) of the measured versus the estimated SMR/BMR, we see that agreement is good at the higher levels of BMR/SMR. These children were the oldest and largest, and most close to the lower limits of age for which Schofield equation was developed. For younger children, we found that the lower is the SMR, the larger is the deviation between methods. Therefore, we believe that the discrepancy is due to the fact that BMR estimated according to the equation used by the FAO/WHO/UNU in this young age group may lead to errors, it may be less suitable for 3-4-year-old children. It is a well-known phenomenon that the error increases at the boundaries of the population for which the prediction equation was developed.

The analyses for the prediction model of PAL and AEE contained significantly more girls than boys. We found no differences in the characteristics from Table 1 between boys and girls; therefore, we do not expect that it influenced our results.

In conclusion, TracmorD provides moderate-to-strong validity evidence that supports its use to evaluate PAL and AEE in preschool children. Wearing the accelerometer for 3 days is comparable with 5 days in this age group.


We thank the participants and their parents who participated in this study, Philips (Philips, DirectLife, The Netherlands) for providing the TracmorD instruments and Rotterdam Erasmus MC for performing the DLW analyses of the saliva samples.

A.S. participated in the study design, carried out measurements, analyzed data, and wrote the manuscript. H.S. analyzed the DLW samples and provided writing assistance. A.G. and K.J. provided instruments, technical support, and writing assistance. I.K. participated in the study design. E.C. and P.S. participated in the study design and in the interpretation of the data and critically supervised writing of the manuscript.