Short Communication
An Objective Method for Measurement of Sedentary Behavior in 3- to 4-Year Olds John J. Reilly,* Jennifer Coyle,* Louise Kelly,* Genevieve Burke,* Stanley Grant,† and James Y. Paton*
Abstract REILLY, JOHN J., JENNIFER COYLE, LOUISE KELLY, GENEVIEVE BURKE, STANLEY GRANT, AND JAMES Y. PATON. An objective method for measurement of sedentary behavior in 3- to 4 year olds. Obes Res. 2003;11:1155-1158. Objective: To test the ability of accelerometry to quantify sedentary behavior in 3- to 4-year-old children. Research Methods and Procedures: We developed a cutoff for accelerometry output (validation study) in 30 healthy 3 to 4 year olds, which provided highest sensitivity and specificity for the detection of sedentary behavior relative to a criterion method of measurement, direct observation using the children’s physical activity form. We then cross-validated the cut-off in an independent sample of healthy 3 to 4 year olds (n ⫽ 52). Results: In the validation study, optimal sensitivity and specificity for the detection of sedentary behavior were obtained at an accelerometry output cut-off of ⬍1100 counts/min. In the cross-validation, sensitivity was 83%: 438/528 inactive minutes were correctly classified. Specificity was 82%: 1251/1526 noninactive minutes were correctly classified using this cut-off. Discussion: Sedentary behavior can be quantified objectively in young children using accelerometry. This new technique could be considered for a wide variety of applications in the etiology, prevention, and treatment of childhood obesity. Key words: motion sensors, accelerometers, computer science and applications actigraph, physical inactivity
Introduction Physical inactivity (sedentary behavior) has recently emerged as an important health construct, and it should be considered independently from physical activity (1). Objec-
Received for review September 11, 2002. Accepted in final form August 11, 2003. *Division of Developmental Medicine and †Institute of Biomedical and Life Sciences, University of Glasgow, Glasgow, Scotland. Address correspondence to: John J. Reilly, University of Glasgow, Division of Developmental Medicine, Yorkhill Hospitals, Dalnair Street, Glasgow G3 8SJ, Scotland. E-mail:
[email protected] Copyright © 2003 NAASO
tive measurement of inactivity is potentially useful for many clinical and public health applications, particularly in the prevention and treatment of childhood obesity (2– 4) and the early prevention of cardiovascular disease (5). For example, in studies of obesity prevention, targeting a reduction in inactivity may be a more effective means of increasing physical activity in children than the more traditional approach of targeting increases in physical activity (2,4). However, no simple, objective method has yet been validated for measurement of inactivity, and this is a serious barrier to the use of inactivity measurement in clinical practice and public health (6,7). Cochrane reviewers have recently concluded that randomized controlled trials in pediatric obesity prevention and treatment are greatly weakened by the absence of an objective measure of sedentary behavior (8,9), an important mediator of intervention success and a potentially valuable outcome measure. The assessment of television viewing time as a proxy for inactivity is useful but methodologically limited at present (1,2), and up to one-half of all time children spend inactive may involve behaviors other than television viewing (10). Simply attempting to measure television viewing is, therefore, unlikely to quantify sedentary behavior adequately. There is increasing evidence that sedentary behavior is now common even in preschool children (11). This is a particular concern because the preschool child is at a critical period for later obesity risk, the “adiposity rebound.” Factors that determine the timing of adiposity rebound are unclear at present (12), but a recent secular trend to earlier adiposity rebound in Europe may have been the result of increases in sedentary behavior in early childhood (12,13). Accelerometry is a practical means of measuring physical activity, even in young children (14), but we speculated that it might be even more suitable for measurement of physical inactivity, which is a simpler construct. Accelerometry output does not have biological meaning per se and must be validated against criterion measures (either direct observation of activity or energy expended on activity). The present study, therefore, aimed to assess the validity of accelerometry for measurement of physical inactivity in 3- to 4-yearold children by comparison against direct observation of behavior, a criterion method. OBESITY RESEARCH Vol. 11 No. 10 October 2003
1155
Measurement of Sedentary Behavior in 3- to 4-Year Olds, Reilly et al.
Research Methods and Procedures Study Design We compared output from the Computer Science and Applications WAM-7164 accelerometer (MTI, Shalimar, FL) against a validated direct observation technique, the children’s physical activity form (CPAF)1 (15). The WAM-7164 is small (5.0 ⫻ 3.8 ⫻ 1.5 cm) and lightweight (42 g). Children wore accelerometers, as previously described, on the right hip (14). The accelerometers were set to record activity in 1-minute intervals and were synchronized with direct observation of activity in nursery. The CPAF categorized activity on a scale of 1 to 4: 1, stationary, no movement; 2, stationary with limb movement but no trunk movement (e.g., drawing); 3, slow trunk movement (e.g., walking); and 4, rapid trunk movement (e.g., running). One-minute intervals spent solely in behavior categories 1 and 2 were defined as “inactive minutes” (true positives), and all others were defined as “noninactive minutes.” The design consisted of a validation study, to identify the optimum cut-off for accelerometry output, and a crossvalidation study in which the proposed cut-off was tested in an independent sample. Subjects, Statistical Analyses, and Power In the validation study, a convenience sample of 30 healthy 3 to 4 year olds (20 boys; mean age, 3.7 ⫾ 0.5 years) was observed for an average of 100 ⫾17 minutes. Sensitivity was defined as the percentage of inactive minutes correctly classified by any accelerometry output cut-off (counts per minute). Specificity was defined as the percentage of noninactive minutes (all other minutes) correctly classified. Sensitivity and specificity were calculated for each child for each cut-off point, and the resulting values were averaged for the sample. Receiver operator characteristic (ROC) analysis identified the optimal cut-off for accelerometry output. Sensitivity and specificity were then assessed in an independent convenience sample of 52 healthy children (21 boys; mean age, 3.5 ⫾ 0.5 years) observed for 40 ⫾ 2 minutes using the same methods. We also identified all minutes in which behaviors consisted exclusively of a single category of activity intensity by CPAF (1, 2, 3, or 4), and summary accelerometry output data were calculated for these minutes. Power of the study was difficult to assess at the outset, but we recruited a sample of children that was larger and more homogenous (in terms of age range) than all of the published pediatric studies that have proposed or tested accelerometry cut-off points to define levels of activity (16 –19). Informed written consent was obtained from parents/carers, and the study had the approval of the Yorkhill Hospitals Research Ethics Committee.
1
Nonstandard abbreviations: CPAF, children’s physical activity form; ROC, receiver operator characteristic.
1156
OBESITY RESEARCH Vol. 11 No. 10 October 2003
Figure 1: ROC curve for accelerometry cut-off points 500-1900 counts/min. Points from left to right indicate 500-1900 counts/min, in 200-counts/min increments.
Results
The validation study (n ⫽ 30) found optimal sensitivity and specificity at an accelerometry output cut-off of ⬍1100 counts/min. Mean sensitivity of this cut-off in the crossvalidation (n ⫽ 52) was 83 ⫾ 14% (95% CI, 78% to 86%; Figure 1). This meant that 438/528 inactive minutes were correctly classified as such using a cut-off point of ⬍1100 counts/min. Mean specificity was 82 ⫾ 11% (95% CI, 79% to 86%), i.e., 1251/1526 noninactive minutes were correctly classified using the cut-off of ⬍1100 counts/min (Figure 1). Data from boys and girls were combined after having established that there were no significant differences between the sexes in sensitivity (83% in boys, 95% CI, 75 to 90; 82% in girls, 95% CI, 76 to 86; p ⫽ 0.91 for difference between the sexes) and specificity (82% in boys, 95% CI, 76 to 87; 83% in girls, 95% CI, 78 to 87; p ⫽ 0.85 for difference between the sexes). Predictive values were not calculated because these are dependent on “prevalence” (in this case activity level) and therefore are study-specific. The impact of activity level on accuracy of classification of activity and inactivity is discussed below. Median accelerometer counts per minute differed significantly between CPAF behavior categories 1 to 4, as might have been expected (Kruskal Wallis test followed by MannWhitney tests, p ⬍ 0.001). However, accelerometer output (counts per minute) overlapped between categories (Figure 2).
Discussion A paradigm shift occurred some time ago, from emphasis on exercise to physical activity. Recent research on the etiology, prevention, and treatment of obesity has suggested that a further paradigm shift should occur, with a new
Measurement of Sedentary Behavior in 3- to 4-Year Olds, Reilly et al.
Figure 2: Accelerometry output within each behavioral category from the CPAF. Median (quartiles and range) of accelerometry output (counts per minute) for minutes exclusively in CPAF activity categories 1 (n ⫽ 96 minutes; median, 131); 2 (n ⫽ 205; median, 735); 3 (n ⫽ 374; median, 2385); and 4 (n ⫽ 85; median, 4171). Differences among categories were significant (MannWhitney U tests, p ⬍ 0.001).
emphasis on physical inactivity or sedentary behavior (1– 4). The need to consider physical inactivity as a distinct construct is being increasingly widely recognized, but no simple, validated means of measurement has been available to date. As a result, the potential of inactivity measurement highlighted by Dietz (1) has not yet been realized. This study found that a novel application of accelerometry can provide objective assessment of engagement in physical inactivity in young children. We have also shown in this study that variation in accelerometry output within observer defined activity categories is substantial (Figure 2), demonstrating a need for the ROC approach to analysis (Figure 1) to establish the validity of any proposed accelerometry output cut-off points. This study had a number of limitations. First, our focus was on the preschool child. Whereas this degree of focus provided a relatively large and homogenous sample, increasing the confidence in our conclusions for this age group, extrapolation of our results to other populations should be considered with caution. The cut-off proposed may be age-specific, and the appropriate cut-off might differ between populations. However, carrying out the necessary cross-validation studies in older subjects should be relatively straightforward. We used direct observation of behavior as our criterion method. Energy expended on activity is an equally valid criterion, but we decided against this approach for several reasons. Measuring the energy cost of unrestricted activity in young children is difficult, and results obtained can be difficult to interpret. Extrapolation of laboratory or treadmill exercise to free-living behaviors may introduce errors (16), and describing activity intensity using metabolic equivalents is problematic in children because the
appropriate energetic value for a metabolic equivalent is very different from that used in adults and is age-specific (16). If future studies can address these issues, then energy expended on activity will be appropriate as a criterion method for preschool children. Finally, limitations in the direct observation method we used probably led to underestimation of the sensitivity of accelerometry for assessment of inactivity. This bias is likely to have arisen because the CPAF required that we categorize each minute of observation, and minutes that included both inactive (categories 1 and 2) and active (categories 3 and 4) behaviors had to be considered as noninactive for the purposes of our ROC analysis. This produced a bias in our assessment of sensitivity, which we felt was acceptable because it was a conservative approach: the performance of accelerometry may be higher than our estimates indicate. Potential users of accelerometry should consider the evidence presented in Figure 1 when deciding on which cutoff to use; this will depend in part on whether sensitivity or specificity is the main concern and whether sedentary behavior or physical activity is the variable of primary interest. The optimal balance between sensitivity and specificity will depend on the application, and future users should consider the potential trade-offs carefully. The accuracy of accelerometry in quantifying behavior will also depend heavily on the levels of inactivity and activity in the children being studied. For the current sample of 3 to 4 year olds in this particular setting, where ⬃25% of minutes monitored were spent entirely in sedentary behavior (in levels 1 to 2 of the CPAF), 38% of minutes classified as inactive by accelerometry were misclassified, and 7% of active minutes (all other minutes) were misclassified. In a sample and setting in which 75% of minutes were spent inactive (the converse of what was observed in this study), predictive values would be approximately switched, with 6% of inactive minutes misclassified and 39% of active minutes misclassified. In a sample and setting with numbers of inactive and active minutes equal, ⬃17% of active minutes and 17% of inactive minutes would be misclassified, with a total misclassification of around one-third of total minutes monitored by accelerometry. Measurement of inactivity could have many clinical and public health applications. These include epidemiological surveys of inactivity (1,6); inactivity as an outcome measure in obesity or cardiovascular disease prevention/treatment trials (7–9); assessment of the determinants of inactivity (1); and assessment of risk factors for obesity (20). The ease of use of accelerometry might even permit self-monitoring of sedentary behavior in clinical practice, a useful element of successful treatment (4). The accelerometers used in this study, like other accelerometers, are small, easy to use, well tolerated by children, and fairly inexpensive. Accelerometry can also simultaneously measure physical activity accurately and practically, even in young children (14,16 –19). OBESITY RESEARCH Vol. 11 No. 10 October 2003
1157
Measurement of Sedentary Behavior in 3- to 4-Year Olds, Reilly et al.
We conclude that accelerometry could be used much more widely now that validity is established for measurement of both activity and inactivity.
Acknowledgments We thank the parents and families for their cooperation and Glasgow City Council for access to nurseries. We also thank Dr. Lilian Murray for statistical advice. This work was funded by Sport Aiding Medical Research for Kids (SPARKS) and the Rank Prize Funds. The funding bodies had no role in the design of the study, analysis or interpretation of results, or the decision to publish or in the content of the report. References 1. Dietz WH. The role of lifestyle in health: the epidemiology and consequences of inactivity. Proc Nutr Soc. 1996;55:829 – 40. 2. Robinson TN. Reducing children’s television viewing to prevent obesity: a randomized controlled trial. JAMA. 1999;282: 1561–7. 3. Reilly JJ, Wilson M, Summerbell CD, Wilson D. Obesity diagnosis, prevention, and treatment: evidence based answers to common questions. Arch Dis Child. 2002;86:392–5. 4. Epstein LH, Paluch RA, Gordy CC, Dorn J. Decreasing sedentary behaviours in treating pediatric obesity. Arch Pediatr Adolesc Med. 2000;154:220 – 6. 5. Williams CL, Hayman LL, Daniels SR, et al. Cardiovascular health in childhood. Circulation. 2002;106:143– 60. 6. Livingstone MBE. How active are we? Levels of routine physical activity in children and adults. Proc Nutr Soc. 2003; 62:681–701. 7. Reilly JJ, McDowell ZC. Physical activity interventions in the prevention and treatment of paediatric obesity: systematic review and critical appraisal. Proc Nutr Soc. 2003;62:611–9. 8. Campbell K, Waters E, O Meara S, Summerbell CD. Interventions for preventing obesity in children. Cochrane Library 2002;2(issue).
1158
OBESITY RESEARCH Vol. 11 No. 10 October 2003
9. Summerbell C, Kelly S, Waters E, et al. Interventions for treating obesity in children. Cochrane Database Systematic Review 2002;2(issue). 10. Strauss RS, Rodzilsky D, Burack G, Colin M. Psychosocial correlates of physical activity in healthy children. Arch Pediatr Adolesc Med. 2001;155:897–902. 11. Certain LK, Kahn RS. Prevalence, correlates, and trajectory of television viewing among infants and toddlers. Pediatrics. 2002;109:634 – 42. 12. Dorosty AR, Rogers IS, Emmett PM, Reilly JJ, ALSPAC Study Team. Factors associated with early adiposity rebound. Pediatrics. 2000;105:1115– 8. 13. Reilly JJ, Kelly A, Ness P, et al. Premature adiposity rebound in children treated for acute lymphoblastic leukemia. J Clin Endocrinol Metab. 2001;86:2775– 8. 14. Fairweather SC, Reilly JJ, Grant S, Whittaker A, Paton JY. Using the CSA activity monitor in pre-school children. Pediatr Exerc Sci. 1999;11:414 –21. 15. O’Hara NM, Baranowski T, Simons-Morton BG, Wilson BS, Parcel GS. Validation of the observation of children’s physical activity. Res Q Exerc Sport. 1989;60:42–7. 16. Puyau MR, Adolph AL, Vohra FA, Butte NF. Validation and calibration of activity monitors in children. Obes Res. 2002;10:150 –7. 17. Trost SG, Ward DS, Moorehead SM, Watson PD, Riner W, Burke JR. Validity of the CSA activity monitor in children. Med Sci Sports Exerc. 1998;30:629 –33. 18. Eston RG, Rowlands AV, Ingledew DK. Validity of heart rate, pedometry, and accelerometry for predicting energy cost of childrens activities. J Appl Physiol. 1998;84:361– 71. 19. Finn KJ, Specker B. Comparison of Actiwatch activity monitor and Children’s Activity Rating Scale in children. Med Sci Sports Exerc. 2000;32:1794 –7. 20. Gortmaker SL, Must A, Sobol AM, Peterson K, Colditz GA, Dietz WH. Television viewing as a cause of increasing obesity in the United States, 1986 –1990. Arch Pediatr Adolesc Med. 1996;150:356 – 62.
