During the last 20 yr, physical activity (PA) in children has decreased in both developed and developing nations (40). While reduced PA is a key contributor (33) to the epidemic of childhood obesity (17), it is also associated with other known risk factors for the development of future cardiovascular disease (1). Accurate measures of PA are therefore needed, especially for the early detection of obesogenic lifestyles and for the evaluation of interventions aimed at promoting greater PA in children.
Because surrogate reports on children's activity by parents have limited validity (29) and self-reports from young children may be unreliable (19), researchers increasingly use accelerometers for objective PA measurement. Accelerometers are piezoelectric sensors that detect acceleration in one to three orthogonal planes (7), providing information on frequency, intensity, and duration of PA. Most accelerometers have a tamper-resistant casing, are water-resistant, are small and lightweight, and therefore interfere minimally with children's free-living PA behavior (15). Therefore, accelerometry (ACC) may be particularly useful in capturing everyday PA in settings such as preschools or schools.
Previous work suggests that the accuracy of PA measurement in older children and youth increases when ACC is combined with HR recordings (10-12). A recent advance in technology has resulted in the availability of devices that provide time-synchronized data of both types. Although this device makes combined measurement both more practicable and interpretable, its validity has been established in laboratory and free-living conditions against indirect calorimetry primarily in older children (11-12 yr old), youth, and adults (9,10,34).
Threshold values used to categorize activity levels (AL) by ACC in youth and adults might not be valid in younger children for several reasons, however. First, preschoolers' overall PA level seems to be much higher than the PA level of school-aged children (32), and therefore, PA cutoffs for discriminating high activity from low activity might vary. Second, from a qualitative perspective, preschoolers' exercise activities differ from those of other groups. Preschoolers are more likely to engage in exercise "play" rather than in the rough-and-tumble activities of older children. Third, preschoolers engage in briefer bouts of movement of vigorous intensity (35).
A separate issue in the validation of measures of PA in children is the choice of the gold standard against which measurements are compared. Established options, until now primarily used in laboratory settings, include the doubly labeled water technique and indirect calorimetry, both of which measure energy expenditure during PA. In field settings such as preschools, behaviorally based techniques that assess PA intensity through direct observation of movement may be more useful. Some have suggested that this is particularly important for younger children, in whom measurement and interpretation of energy expenditure data may otherwise be difficult (14,26).
Given key differences in the quantity and quality of PA in preschool-aged children and a rapid increase in the prevalence of obesity in this group (25), accurate measures of PA are increasingly important. The purpose of this study, therefore, was to assess the validity of combined measures of HR and ACC against directly observed PA in 3- to 6-yr-old children in the preschool setting.
A convenience sample of 33 healthy children between 3 and 6 yr was obtained from four preschools located in socioeconomically different neighborhoods (16% low socioeconomic status overall) in Mannheim, Baden-Württemberg, south Germany. Thirty-three percent of the sample children were immigrants, which is comparable to regional population figures for immigrants younger than 25 yr. Exclusion criteria included severe atopic dermatitis, serious acute infectious diseases (e.g., severe diarrhea or obstructive bronchitis), and physical malformations or disabilities. Written informed consent was obtained from each child's parents. The study was approved by the ethical committee of the Heidelberg University, Mannheim Medical Faculty.
We collected anthropometric data (weight and height) consisting of height measured to the nearest 0.1 cm (Seca Deutschland, Hamburg, Germany) and weight measured to the nearest 0.1 kg (Soehnle Pharo, Nassau, Germany).
ACC and HR measurements.
The Actiheart device (CamNtech, Cambridge, UK) recorded ACC and HR data using one-dimensional ACC in the vertical plane and a two-channel ECG. Although the technical validity and reliability of this device have been established previously (5), we confirmed measurement reliability during a trial of technical acceleration (Certomat S2; Sartorius AG, Göttingen, Germany). The minimal correlation coefficient between readings from each device and the test instrument was 0.99, suggesting a high degree of measurement reliability across all units used in our study.
The Actiheart can measure epochs of 15, 30, or 60 s in duration (9). We recorded HR and ACC data in our study using an epoch setting of 15 s to enable detection of rapid changes in movement intensity and short bursts of moderate-to-vigorous PA (MVPA) typically exhibited by young children (28,36).
In the morning, at the beginning of the preschool day, the Actiheart device was securely affixed to the children's substernal thorax region (4) by two sticky electrodes (Kendall Arbo*ECG electrodes; Tyco Healthcare, Neustadt Donau, Germany) and additional tape. Children were then dressed and began their usual preschool activities. The Actiheart device was worn for approximately 24 h, allowing us to obtain HR and ACC recordings during an entire preschool day.
During the morning and the afternoon preschool hours, children's PA was directly observed by five trained observers and rated into five AL using the Children's Activity Rating Scale (CARS). We chose the CARS as the method of direct observation because it has been specially developed and validated in young children (24) and has been found to correlate with the full range of PA intensity as measured by ACC counts in preschoolers (13).
CARS levels 1 and 2 activities are sedentary, although the latter category includes movements of the limbs or torso without translocation. Levels 3-5 activities involve translocation, with levels 3-5 indicating slow, moderate, and vigorous translocations, respectively. Consistent with the procedures described by Puhl et al. (24), only AL lasting >3 s were recorded by the observers. In contrast to the original CARS method that averages observed AL during 1 min, we recorded every change from one AL to another, as it occurred in real time using handheld computers.
To develop measures of sedentary behavior (SB) and MVPA, we collapsed CARS ratings. AL during 15-s intervals rated by our observers as 1 or 2 comprised SB; 15-s intervals with AL 4 or 5 were considered MVPA. This approach, consistent with a previous work, defines MVPA as an AL generating at least 3 METs (37), which has been found to correspond to a CARS level ≥4. Because moderate levels of PA (i.e., those corresponding to a CARS level of at least 4) are associated with significant health benefits and because of our interest in generating cutoffs that could be easily applied in the evaluation of future interventions, we did not attempt to establish HR or ACC cutoffs that discriminated between moderate and vigorous PA as separate categories.
We limited the period of observation to 150 min per observer during the normal hours of operation of the preschool. To avoid disturbing daily routines at the study site, observers rated the activities of no more than two children per preschool day. Each child was monitored by a single observer from the study team.
Study observers underwent 10 h of standardized training during a minimum of 2 d, which consisted of an orientation to the CARS activity rating system and training videotapes for test-rating children in daily preschool activities and during gym lessons. To facilitate assessment of interrater reliability in training sessions, children's PA was videotaped. We synchronized the Actiheart device, the observer handheld computer, and the video recorder time to within ±1 s to yield time-synchronized data on children's AL using an accepted gold standard and the alternate methods under study. Interrater reliability compared two observers' ratings of PA in two children during a defined 20-min period using the contingency table method (overall mean agreement (Cramer V) of 0.74 and 0.68 for child 1 and 2, respectively).
Development of derivation and validation data sets.
Data were purposefully collected in the morning during preschool hours between 9.30 a.m. and noon and in the afternoon between 1 and 4 p.m. to generate derivation and validation data sets, respectively. These hours indicate the total time frame during which we were able to make our measurements. However, with some children leaving the afternoon session with the arrival of their parents, mean observation time in the validation data set was lower. Because previous work suggests that young children infrequently engage in level 4 or 5 activities for extended periods with more vigorous activities usually interspersed between those that are less intense (24), the morning session was purposefully organized to observe preschoolers' PA during both free play and a series of structured, high-intensity activities. This approach ensured that our subjects displayed a broader repertoire of activities at various levels of intensity, as has been previously recommended for accelerometer validation and calibration studies (16). In contrast, the afternoon session consisted of free play to reflect the predominant form of activity common among most preschool children.
The unit of analysis in this study was the 15-s epoch, to which all measurements were linked. Outlier values for HR (>250 min−1 and <40 min−1) were removed; the mean value for previous and subsequent HR readings replaced single missing values. We restricted our analysis of ACC counts to values that fell within 2 SD of the mean value for each child; missing values were not replaced.
We assessed the mean ACC counts per 15-s epoch and mean HR per 15-s epoch for each observed CARS AL, separately for the derivation and the validation data sets. The Wilcoxon rank-sum test and linear regression were used to test for differences between AL and for linear trends as AL increased, respectively. As mentioned, we collapsed AL categories to form simple indicators of SB (AL 1 or 2) and MVPA (AL 4 or 5) and applied these indicators to each 15-s interval.
Using receiver operating characteristic (ROC) analysis, we determined cutoff points for both HR and ACC counts in classifying MVPA and SB during the morning period (derivation data set). For this analysis, only homogenous 15-s intervals (i.e., intervals during which only one AL was observed) from the derivation data set were used. The ROC curve captures the trade-off between sensitivity and specificity for a binary classification (e.g., discriminating MVPA from non-MVPA) because its threshold is varied (8). Cutoffs were calculated separately for boys and girls because ACC-based measurement as well as observation (3,22) of PA in preschoolers have revealed significant gender-related differences in total activity (18) and MVPA (16).
The cutoffs identified in data from the derivation data set were then tested in the validation data set. We assessed the overall accuracy of the cutoffs using classification rates for the correct identification of both MVPA and SB (true positives + true negatives/total number of 15-s intervals), area under the ROC curve (AUC), and sensitivity and specificity with 95% confidence intervals (CI). All analyses were performed with the SAS 9.2 (SAS Institute, Inc., Cary, NC) statistical package.
Of the children in our sample, 64% were boys; their mean body mass index was 15.2 ± 1.4 kg·m−2 compared with 15.4 ± 2.2 kg·m−2 for girls (P = 0.7; Table 1). Boys in the sample were older than girls (4.3 ± 1.1 vs 3.5 ± 0.8 yr, P = 0.04). ACC and HR recordings were made for an average of 2.5 ± 0.7 and 1.5 ± 0.3 h for the derivation and the validation data sets, respectively. This resulted in 12,989 homogenous 15-s epochs for analysis in the derivation data set (n = 6236 for boys and n = 6753 for girls) and 8146 15-s epochs for the validation data set (n = 5144 for boys and n = 3002 for girls).
Mean ACC counts and mean HR in the derivation data set were significantly higher in boys compared with those in girls (128 ± 25.5 vs 124 ± 23 bpm, P < 0.0001; 89 ± 135 vs 58 ± 113 counts per 15 s, P < 0.0001, Wilcoxon rank-sum test). As expected, increasing CARS AL were associated with higher mean ACC and mean HR in both boys and girls. The differences between AL were statistically significant (P < 0.0001, Wilcoxon rank-sum test) and seemed to follow a linear pattern for HR (P < 0.0001 for linear regression, with the explained variance (R2) rising modestly from 0.28 to 0.32 when a quadratic term was added). In contrast, differences across CARS levels for ACC seemed to be exponential (P < 0.0001 for linear and quadratic regression, with R2 rising from 0.45 to 0.63 with the addition of the quadratic term). When data from the derivation and the validation data sets were analyzed separately, there were significant differences between mean HR and ACC, with values being systematically higher at greater levels of PA in the derivation data set (Table 2).
Cutoffs at which correct classifications of MVPA were optimized included HR > 134 and ACC > 118 for boys and HR > 138 and ACC > 105 for girls. The extent to which these cutoffs correctly classified 15-s intervals as MVPA or SB was expressed as the AUC. Traditionally, AUC values between 0.7 and 0.8 are considered fair, values between 0.8 and 0.9 are good, and values between 0.9 and 1.0 are interpreted as excellent (20). For boys and girls together, the AUC was 0.86 for both HR and ACC (Fig. 1), with no significant gender differences in separate AUC analysis. The optimal cutoffs for correctly classifying SB were HR < 128 and ACC < 46 for boys and HR < 130 and ACC < 26 for girls. Values for the AUC in classifying SB differed somewhat for ACC and HR (0.79 and 0.71, respectively; no gender differences).
In applying these cutoff values for both HR and ACC to data from the afternoon (validation data set), we observed that 87% of the 15-s intervals rated as MVPA by the observers were correctly classified as MVPA in boys and 91% in girls when both ACC and HR cutoffs were used in combination (ACC + HR). If ACC cutoffs alone were used for classification, the correct classification rates decreased to 84% in boys and to 87% in girls. If HR cutoffs alone were used, the correct classification values were even lower (68% for boys and 80% for girls). Because we achieved the highest correct classification rates for MVPA with ACC + HR cutoffs, further test accuracy analysis was performed using this combination only. The combined approach led to a specificity of 91% (95% CI = 88.5-93.5) for boys and 93% (95% CI = 92.1-93.9) for girls. Sensitivity was lower with 27% (95% CI = 25.8-28.2) and 38% (95% CI = 36.2-39.8) for boys and girls, respectively.
When cutoffs for SB were applied to afternoon data using ACC + HR cutoffs, correct classification rates were lower than for MVPA (59% vs 67% in boys vs girls, respectively). Rates improved somewhat especially for boys when only the ACC cutoff was applied (correct classification rates: 67% and 69% for boys and girls, respectively). The specificity for detecting SB using only ACC was 52% and 61%; sensitivity was 78% and 75% for boys and girls, respectively.
The combination of HR and ACC recordings seems useful in capturing directly observed PA levels in preschoolers. We found that combining ACC and HR cutoffs (ACC + HR) resulted in higher rates of correct classification (boys = 87% and girls = 91%) for MVPA but not for SB. Moreover, our analysis supports the value of a gender-stratified assessment when identifying MVPA or SB.
With the rising prevalence of childhood obesity and recognition that the preschool period offers a valuable window of opportunity for PA promotion, PA interventions in preschools are becoming more common. To capture their effects, accurate measurement of PA in preschoolers is necessary. Recent research in older children has shown that the accuracy of PA measurement increases if ACC is combined with other physiological measures like HR recordings (10-12). Our study, conducted in a preschool setting, supports the utility of this approach when measuring MVPA. Interventionalists may therefore wish to consider this measurement strategy when developing and testing programs to increase higher levels of PA in this younger age group.
In our study, both ACC and HR were able to capture a full range of possible PA levels in preschoolers and reliably reflect their intensity: With increasing observed PA levels, both ACC counts and HR increased as well, reflecting the physiological response to activity that would be expected in preschoolers. These results are consistent with the results of previous studies that examined the association between ACC and observed AL in preschoolers (13,26).
In accordance with the previous work by Jackson et al. (18), who used ACC alone as an objective method to measure SB in preschool children, we found that SB was most correctly classified using ACC cutoffs rather than an approach on the basis of data from both ACC and HR (ACC + HR). The superiority of using ACC alone in the measurement of inactivity is also evident in the current study through higher AUC values for ACC compared with those for HR (AUC = 0.79 vs 0.71). These findings contrast with the advantage of ACC + HR for the measurement of MVPA. Previous research by Brage et al. (6), however, might explain this finding: Their branched model for estimating PA energy expenditure in young adults relies primarily on HR when predicting PA energy expenditure during higher activity states and primarily on ACC during less active states. HR thus gains explanatory power when PA intensity increases, for example, when vigorous PA is measured. This might also explain the higher accuracy of ACC cutoffs alone compared with ACC + HR for classifying SB. Generally, the limited value of HR measurement at low PA levels (27) might also be explained by confounding factors that assume greater influence during inactivity: Emotional stress will significantly elevate HR independent of any change of V˙O2 (31). In addition, studies have indicated that the state of fitness in childhood seems to have a greater effect on daily mean HR than the instantaneous level of activity (30). Also, it is known that HR tends to lag behind changes in movement, thus overestimating energy expenditure in inactive states (28). If an intervention in preschoolers specifically aims to decrease SB, the measurement of ACC alone and not the combination with HR seems to be more accurate.
Our results differ somewhat from other studies using direct observation of PA for validation (14,26) in that the sensitivity of our cutoffs in detecting inactivity was slightly lower. This difference might be explained by differences in the unit of analysis (15-s vs 1-min intervals), by the gold standard applied (CARS vs children's PA form), or by the device used to measure ACC (Actiheart vs WAM-7164 (MIT, Shalimar, FL)). As with previous work, however, our results support the use of ACC in evaluating interventions aimed at reducing SB.
In our study, correct classification rates for MVPA were consistently higher than those for SB. An explanation for this discrepancy could be that the ACC device we used in the study is less accurate in capturing activities that do not include a vertical component, e.g., crawling. Crawling, for example, would be rated as CARS level 3 by the observers, but by ACC, data might be incorrectly categorized as more sedentary because the activity lacked a vertical component. In contrast, miscategorization of more vigorous activity would be less likely as the vertical component increases.
Compared with the high overall accuracy and specificity of the cutoffs for the classification of MVPA, their sensitivity was substantially lower. Physiologically, this finding could be explained by the higher interindividual variability in movement patterns during MVPA compared with those during non-MVPA. Furthermore, in the derivation data set, MVPA was predominantly observed during the structured PA lesson. In contrast, the validation data set predominantly consisted of free play. The decreased sensitivity thus could also be related to qualitative differences between MVPA during structured gym lessons and unstructured free play. Another reason for the low sensitivity could be that our criterion measure, CARS ratings of directly observed movement, contained some error (23). Future work should test the combination of HR and ACC in preschoolers against other gold standards applicable in the field, such as the doubly labeled water.
Statistically, the low sensitivity for detecting MVPA could be explained by the low prevalence of MVPA (only 6.8% and 4.5% of all observed 15-s intervals for boys and girls, respectively), despite efforts to ensure the representation of higher levels of PA intensity during the morning observation period. Previous work suggests, for example, that sensitivity, like positive predictive values, can be affected when the prevalence of the state to be detected (e.g., MVPA-observed intervals) is lower (8,21). Yet, in other studies on children, the percentage of time spent in MVPA is comparably low (22). Even with an observation period of 5-6 h, only 5% of all time was spent in AL 3 and 1% in AL 4 or 5 in Finn and Specker's study (13). This consistent finding might suggest that the low sensitivity for the detection of MVPA could be a problem in all measurements during free play in preschool-aged children.
Although there is growing evidence for gender-related differences in preschoolers' PA (3,16,18,22), a gender-stratified measurement approach has not been reported to date. Our analysis, for example, confirmed significantly higher ACC and HR means in boys versus girls. Although these findings may have been confounded by age (boys in our sample were older than girls), we nonetheless observed higher sensitivity and specificity for discriminating MVPA or SB in our sample when ACC and HR cutoff values were determined separately for boys and girls.
Moreover, classification rates for MVPA were consistently lower for preschool boys than for girls. A tentative explanation for this gender difference might be that the accuracy of cutoffs, leading to a binary classification, might be lowered by a higher variability in ACC counts and HR evident within the group of the boys. Given these findings, it seems prudent to recommend gender-stratified analyses in the evaluation of future PA interventions in this age group.
This study had several strengths and weaknesses. It focused on 3- to 6-yr-old children, an understudied but potentially important age group, particularly given its proximity in time to the adiposity rebound that occurs in most children (39). To our knowledge, this is the first study to establish cutoffs for the combination of ACC and HR in the field in this age group. Because cutoffs may differ across populations, future work in larger, more diverse settings and with longer observation times is therefore needed to confirm cutoffs with the broadest applicability and utility.
Our study uses direct observation and the CARS as our criterion measure. This approach has several advantages in younger children in comparison to energy expenditure measures (14,26). Most importantly for the current study, we were able to use CARS to measure PA levels in smaller time units and in a way that did not interfere with children's free-living activities. The value of indirect calorimetry is limited when these are important considerations. Moreover, we recorded ACC and HR in intervals sufficiently short to capture rapid changes in activity type and intensity that is characteristic of play among preschoolers (2,28,36). This approach was further strengthened by rating children's short bursts of activity in real time. This allowed us to capture each change in AL as it occurred (minimal duration = 3 s), thereby yielding potentially higher measurement accuracy compared with an approach that uses longer epochs of observed PA levels (24,38).
In our study, the validation and derivation data sets were not well matched with respect to HR and ACC at moderate and high AL (AL 4 and 5). This results most probably from qualitative differences in the types of PA taking place during the morning and afternoon periods (structured + free play vs free play alone). Although we attempted to account for brief periods of intense activity using a 15-s epoch length, we observed that AL changed rapidly. This may have resulted in the systematic misclassification of MVPA, when observers categorized a heterogeneous epoch as homogenously filled by MVPA. Despite the potential for misclassification, the validity of our proposed MVPA cutoffs may not have been affected significantly because we chose to use broader categories that simply distinguished PA levels greater than AL 3 from more SB.
Our results suggest an advantage to an approach that combines two physiological measures such as HR and ACC for quantifying PA of middle and high intensity (MVPA) in preschool-aged children. Because this approach was not as useful in correctly classifying SB, future work is needed to develop, test, and validate alternate measurement methods for detecting low levels of activity, especially in younger children. Our data show significant differences in PA between the genders, thereby confirming previously published findings but at the same time implying the potential need of a gender-stratified approach for measuring and quantifying PA intensity in preschoolers. In conclusion, interventionalists who primarily aim at increasing MVPA in preschoolers might wish to combine the measurement of HR and ACC; those who primarily aim to reduce inactivity should consider relying on ACC as their primary outcome measure.
This work was supported by grants from the Landesstiftung Baden-Württemberg and the Veterans Administration Health Services Research and Development Service (IIR-06-091).
The authors report no conflicts of interest.
The authors have not received support from any company whose product is mentioned in the article. Neither the funding bodies nor any company played a role in the design of the study, analysis or interpretation of the results, the decision to publish, or the contents of the report.
The authors thank the parents, families, and preschools for their cooperation. The authors also thank the trained observers and Dr. Stefan Hey (Research Group hiper.campus, KIT Karlsruhe, Germany) and his team, who programmed the handheld computer system for real-time direct observation.
Comments made by the authors do not necessarily reflect the opinions of the American College of Sports Medicine or the Department of Veterans Affairs.
1. Andersen LB, Harro M, Sardinha LB, et al. Physical activity and clustered cardiovascular risk in children: a cross-sectional study (The European Youth Heart Study). Lancet
2. Bailey RC, Olson J, Pepper SL, Porszasz J, Barstow TJ, Cooper DM. The level and tempo of children's physical activities: an observational study. Med Sci Sports Exerc
3. Baranowski T, Thompson WO, DuRant RH, Baranowski J, Puhl J. Observations on physical activity in physical locations: age, gender, ethnicity, and month effects. Res Q Exerc Sport
4. Brage S, Brage N, Ekelund U, et al. Effect of combined movement and heart rate monitor placement on physical activity estimates during treadmill locomotion and free-living. Eur J Appl Physiol
5. Brage S, Brage N, Franks PW, Ekelund U, Wareham NJ. Reliability and validity of the combined heart rate and movement sensor Actiheart
. Eur J Clin Nutr
6. Brage S, Brage N, Franks PW, et al. Branched equation modeling of simultaneous accelerometry and heart rate monitoring improves estimate of directly measured physical activity energy expenditure. J Appl Physiol
7. Chen KY, Bassett DR Jr. The technology of accelerometry-based activity monitors: current and future. Med Sci Sports Exerc
. 2005;37(11 suppl):490-500.
8. Cook NR. Statistical evaluation of prognostic versus diagnostic models: beyond the ROC curve. Clin Chem
9. Corder K, Brage S, Mattocks C, et al. Comparison of two methods to assess PAEE during six activities in children. Med Sci Sports Exerc
10. Corder K, Brage S, Wareham NJ, Ekelund U. Comparison of PAEE from combined and separate heart rate and movement models in children. Med Sci Sports Exerc
11. Corder K, Ekelund U, Steele RM, Wareham NJ, Brage S. Assessment of physical activity in youth. J Appl Physiol
12. Eston RG, Rowlands AV, Ingledew DK. Validity of heart rate, pedometry, and accelerometry for predicting the energy cost of children's activities. J Appl Physiol
13. Finn KJ, Specker B. Comparison of Actiwatch activity monitor and Children's Activity Rating Scale
in children. Med Sci Sports Exerc
14. Freedson P, Pober D, Janz KF. Calibration of accelerometer output for children. Med Sci Sports Exerc
. 2005;37(11 suppl):523-30.
15. Freedson PS, Miller K. Objective monitoring of physical activity using motion sensors and heart rate. Res Q Exerc Sport
. 2000;71(2 suppl):21-9.
16. Grontved A, Pedersen GS, Andersen LB, Kristensen PL, Moller NC, Froberg K. Personal characteristics and demographic factors associated with objectively measured physical activity in children attending preschool. Pediatr Exerc Sci
17. Hedley AA, Ogden CL, Johnson CL, Carroll MD, Curtin LR, Flegal KM. Prevalence of overweight and obesity among US children, adolescents, and adults, 1999-2002. JAMA
18. Jackson DM, Reilly JJ, Kelly LA, Montgomery C, Grant S, Paton JY. Objectively measured physical activity in a representative sample of 3- to 4-year-old children. Obes Res
19. Kohl HW, Fulton JE, Caspersen CJ. Assessment of physical activity among children and adolescents: a review and synthesis. Prev Med
20. Metz CE. Basic principles of ROC analysis. Semin Nucl Med
21. Moons KG, van Es GA, Deckers JW, Habbema JD, Grobbee DE. Limitations of sensitivity, specificity, likelihood ratio, and Bayes' theorem in assessing diagnostic probabilities: a clinical example. Epidemiology
22. Pate RR, McIver K, Dowda M, Brown WH, Addy C. Directly observed physical activity levels in preschool children. J Sch Health
23. Phelps CE, Hutson A. Estimating diagnostic test accuracy using a "fuzzy gold standard." Med Decis Making
24. Puhl J, Greaves K, Hoyt M, Baranowski T. Children's Activity Rating Scale
(CARS): description and calibration. Res Q Exerc Sport
25. Reilly JJ. Physical activity, sedentary behaviour and energy balance in the preschool child: opportunities for early obesity prevention. Proc Nutr Soc
26. Reilly JJ, Coyle J, Kelly L, Burke G, Grant S, Paton JY. An objective method for measurement of sedentary behavior
in 3- to 4-year olds. Obes Res
27. Riddoch CJ, Boreham CA. The health-related physical activity of children. Sports Med
28. Rowlands AV, Eston RG, Ingledew DK. Measurement of physical activity in children with particular reference to the use of heart rate and pedometry. Sports Med
29. Sallis JF, Saelens BE. Assessment of physical activity by self-report: status, limitations, and future directions. Res Q Exerc Sport
. 2000;71(2 suppl):1-14.
30. Saris WH. The relationship between working performance, daily physical activity, fatness, blood lipids and nutrition in schoolchildren. In: Berg K, Ericsson BO, editors. Children and Exercise
. Baltimore (MD): University Park Press; 1980. p. 166-74.
31. Saris WH. Habitual physical activity in children: methodology and findings in health and disease. Med Sci Sports Exerc
32. Sigmund E, De Ste Croix M, Miklankova L, Fromel K. Physical activity patterns of kindergarten children in comparison to teenagers and young adults. Eur J Public Health
33. Stevens J, Murray DM, Baggett CD, et al. Objectively assessed associations between physical activity and body composition in middle-school girls: the Trial of Activity for Adolescent Girls. Am J Epidemiol
34. Thompson D, Batterham AM, Bock S, Robson C, Stokes K. Assessment of low-to-moderate intensity physical activity thermogenesis in young adults using synchronized heart rate and accelerometry with branched-equation modeling. J Nutr
35. Timmons BW, Naylor PJ, Pfeiffer KA. Physical activity for preschool children-how much and how? Can J Public Health
. 2007;98(2 suppl):122-34.
36. Trost SG, McIver KL, Pate RR. Conducting accelerometer-based activity assessments in field-based research. Med Sci Sports Exerc
. 2005;37(11 suppl):531-43.
37. Trost SG, Pate RR, Sallis JF, et al. Age and gender differences in objectively measured physical activity in youth. Med Sci Sports Exerc
38. Welk GJ, Corbin CB, Dale D. Measurement issues in the assessment of physical activity in children. Res Q Exerc Sport
. 2000;71(2 suppl):59-73.
39. Whitaker RC, Pepe MS, Wright JA, Seidel KD, Dietz WH. Early adiposity rebound and the risk of adult obesity. Pediatrics
40. Wilkinson KM. Increasing obesity in children and adolescents: an alarming epidemic. JAAPA
. 2008;21(12):31-6, 8.