The present study is the first to investigate the feasibility and validity of GT1M ActiGraph accelerometer measurements in a convenience sample of 1- to 2-year old toddlers. Furthermore, this study is also the first to examine if ActiGraph cut points developed among preschool children (3- to 5-year olds) are appropriate to accurately identify directly observed physical activity intensities in a toddler population.
To test the feasibility of the GT1M ActiGraph accelerometer, 47 toddlers from 11 child care centers wore the device during waking hours for 6 consecutive days. It is of concern that many day care centers were not interested in participating in the present study (response rate of 61.4%). A possible explanation could be that the contacted child care centers were anxious that the results of the present study would be used to evaluate the quality of the child care center. There was also a very low parental response rate in this study (20.2%). Perhaps parents were reluctant to participate because the study included an observation of their child. Another possibility is that child care staff did not motivate the parents to participate in the present study as no incentives were provided for both the child care center and the parents. Based on the experience from this study it can be suggested to use a more active recruitment approach to recruit parents (e.g., parent information session) or to provide incentives for parents and/or child care centers to achieve a higher response rate in this age group.
Using the 70/80 rule, a minimum registration time of 7.5 h during weekdays and 7.7 h during weekend days was determined. Applying this criterion, resulted in 29 excluded weekdays and 36 excluded weekend days, and 17 out of 47 toddlers (36.2%) failing to meet the inclusion criteria. Compared to previous research in preschoolers, applying the 70/80 rule and requiring 3 eligible days for inclusion, the proportion of ineligible data is higher in the present study [21, 32]. In the study of Van Cauwenberghe et al.  applying these decision rules resulted in 97 excluded days and 40 out of 154 5-year old children (26.0%) providing non-eligible data and in the study of Verbestel et al.  this resulted in 48 out of 261 3- to 5-year old children (18.4%) providing non-eligible data. The rather high proportion of data not eligible for inclusion can probably be explained by the fact that wearing the accelerometer in the present study was a responsibility of both the parents and the child care staff. As child care staff were not instructed to fill in a diary, no information is available on the times the accelerometer was put on and taken off and the reasons for doing so when the child was at child care. One way to increase compliance could be to provide reminders (e.g., daily telephone calls) to parents and/or child care staff to put the accelerometers back on after sleeping and napping. Additionally, future studies could test the feasibility of wearing the accelerometer during day time napping as this would alleviate the issue of parents and child care staff having to remove and refit the monitors during the day. Further, it is possible that the inclusion criterion of 3 eligible days was too high in this population. If a criterion of 2 eligible days had been applied, another 8 toddlers (17.0%) could have been included for analyses. However, this suggestion could possibly decrease the reliability of the measurements as previous research in preschool-aged children indicated that the number of days of monitoring was more important to reliability than the number of hours . Future research should establish the minimum number of days accelerometers need to be worn in order to represent habitual physical activity in toddlers. Ultimately, mean registration time of the included days in the present study was 9.4 h during weekdays and 9.9 h during weekend days. Considering the parental reported sleeping patterns of the toddlers, namely 11 h sleeping per night and 2 h napping per day, these findings suggest that there was good compliance to the study protocol. Further, a threshold of 7.5 hours registration time per day appears to be a reasonable suggestion for future research in this age group. Nevertheless, further research is required to determine the variability in toddlers' physical activity behavior within days and the minimum number of minutes required to represent a typical day in toddlers.
To evaluate the validity of the GT1M ActiGraph accelerometer, observations during free play at child care were conducted. Results of the free play observations were similar to previous research at child care in 2- to 3-year old children with the majority of the observations classified as stationary and motionless to stationary with movement of limbs or trunk behavior and a minority as moderate to fast movement . The proportion of time spent in each activity level during free play was highly variable between toddlers, reflecting the different activities being undertaken by toddlers during free play. Descriptive statistics of the accelerometer output revealed that median ActiGraph activity counts increased in accordance with physical activity intensity but also demonstrated substantial variability. Results of a previous study, using a second-by-second coding protocol and the activity levels categories of the Children's Activity Rating Scale (CARS), indicate that mean accelerometer outputs for sedentary behavior, light physical activity, moderate physical activity and vigorous physical activity during free play are systematically higher in preschoolers, namely 448 ± 196, 734 ± 185, 823 ± 182 and 1115 ± 233 counts per 15 s, respectively . Several explanations are possible for this discrepancy in accelerometer output during free play, including differences in the observation system, the protocol and the activities undertaken during free play and age related changes in anthropometrics, movement patterns and walking biomechanics.
In the present study, the criterion validity of the GT1M ActiGraph accelerometer for measuring physical activity in toddlers was considered acceptable (r = 0.66 and ρ = 0.52). A recent review, summarizing the evidence on the validity of the ActiGraph accelerometer to assess physical activity in older children and adolescents, suggests that the results of the present study are in line with previous validation studies where ActiGraph activity counts were moderately to highly correlated with observed activity (r = 0.52 - 0.77).
Accelerometer activity counts are a dimensionless unit and researchers have attempted to calibrate these counts into biologically meaningful and interpretable data, such as time spent in activity levels of different intensities [10, 18, 19]. Calibration studies in toddlers are lacking, but numerous investigations involving preschool children have attempted to calibrate ActiGraph activity counts . Therefore, the present study aimed to explore whether previously developed ActiGraph cut points for 3- to 5-year old children [21, 23, 25] allow for the accurate categorization of directly observed physical activity intensities in toddlers. It is critical to understand which cut points are able to accurately classify physical activity intensity in young children as others demonstrated that lack of consensus on this issue results in an inability to estimate population prevalence levels of physical activity in young children [21, 33].
The Bland-Altman plots illustrate that the mean bias between directly observed and predicted time spent in each activity intensity was the lowest when the Pate cut points were used. Yet, wide limits of agreement were found, indicating that the time in the directly observed physical activity intensities was not accurately classified. Large mean differences and wide limits of agreement were established for the other two sets of cut points, suggesting that they were unable to accurately identify time spent in each observed activity intensity level in toddlers. To evaluate the predictive validity of the cut points thoroughly, sensitivity, specificity and ROC-AUC were calculated. The Pate sedentary behavior cut point performed fairly well to classify activity counts as stationary and motionless to stationary with movement of limbs or trunk behavior while the Sirard and Van Cauwenberghe sedentary behavior cut points exhibited an unacceptably high false positive rate. These findings do support the use of the Pate cut point to define sedentary time and non-sedentary time (a combination of light, moderate and vigorous physical activity) in toddlers. With respect to detecting slow, easy movement, all three cut points performed poorly. These findings endorse the development of toddler specific light physical activity cut points. Finally, for the purpose of categorizing activity counts as moderate to fast movement, all three moderate-to-vigorous physical activity cut points performed poorly as a function of a low true positive rate, indicating the need for toddler specific cut points to classify moderate-to-vigorous physical activity. Most importantly, the cut points currently used, appeared to be too high to accurately identify the time toddlers spent in moderate to fast movement. A very important consideration is that the GT1M ActiGraph is a hip-mounted accelerometer and measures accelerations in the vertical plane. Consequently, the accelerometer registers a reduced amount of vertical acceleration when non-ambulatory activities with limited trunk movement occur (e.g., climbing, pulling, pushing, peddling on a tricycle), resulting in misclassification of light physical activity as sedentary behavior or moderate-to-vigorous physical activity as light physical activity. Moreover, results from the present study illustrate that toddlers often engage in such activities during free play at child care. Therefore, combining accelerometers with monitors capable of detecting posture, using multiple monitors to measure movement of the trunk and the limbs simultaneously or applying pattern recognition may provide more accurate information beyond the capability of the GT1M ActiGraph when defining physical activity intensities in toddlers as well as in older children [10, 17, 18, 34]. Research in these areas is urgently needed.
Some limitations of the present study need to be acknowledged. To measure physical activity intensity during free play, a 15 seconds measurement interval was used for both the accelerometers and the OSRAC-P. A 15 seconds measurement interval has been put forward to measure the spontaneous activities in young children [9, 10]. However, there is a possibility that the 15 seconds measurement interval does not allow for the accurate detection of intermittent changes in physical activity intensity and fidgeting might be missed [10, 35]. Furthermore, the OSRAC-P protocol requires to code the highest level of activity and the corresponding activity type during the 15 seconds observation interval which may mask other activity intensities and types during the observation interval. Especially for the classification of slow, easy movement and moderate to fast movement, it can be expected that levels of agreement were reduced because of this coding system. A continuous coding protocol may have been more appropriate to capture physical activity intensity and type during free play. Further, although the OSRAC-P decision rules to classify physical activity intensities in young children are well-established [7, 26], the classification of standing as sedentary behavior is questionable . Finally, the present study was limited by the small convenience sample used. Larger and more variable samples are needed to determine if individual factors, such as body size and motor development, modify the findings.
Several strengths of the present study are also noteworthy. First, the ActiGraph accelerometer was validated against directly observed free play activities with excellent inter-observer reliability. Additionally, by using direct observation, in-depth information on the physical activity types of toddlers was gathered. Second, the criterion validity and the predictive validity were evaluated using appropriate statistical approaches [7, 10].