Ability of the Actiwatch Accelerometer to Predict ... - Wiley Online Library

Ability of the Actiwatch Accelerometer to Predict Free-Living Energy Expenditure in Young Children Mardya Lopez-Alarcon,* Jaime Merrifield,† David A. Fields,‡ Tena Hilario-Hailey,† Frank A. Franklin,†§ Richard M. Shewchuk,¶ Robert A. Oster,** and Barbara A. Gower†

Abstract LOPEZ-ALARCON, MARDYA, JAIME MERRIFIELD, DAVID A. FIELDS, TENA HILARIO-HAILEY, FRANK A. FRANKLIN, RICHARD M. SHEWCHUK, ROBERT A. OSTER, AND BARBARA A. GOWER. Ability of the Actiwatch accelerometer to predict free-living energy expenditure in young children. Obes Res. 2004;12: 1859 –1865. Objective: To determine whether activity counts obtained with the Actiwatch monitor are associated with total expenditure and body composition in young children. Research Methods and Procedures: Actiwatch activity monitors were tested in 29 children 4 to 6 years old under field conditions over eight days. Total energy expenditure (TEE) was assessed with the doubly labeled water (DLW) technique. Correlation analyses were used to identify variables related to energy expenditure and percentage body fat. Multiple linear regression analyses were used to examine the variance in TEE and percentage body fat explained by activity counts after adjusting for relevant covariates. Results: Both average total daily activity counts (658,816 ⫾ 201,657) and the pattern of activity were highly variable among subjects. TEE was significantly related to lean body mass (r ⫽ 0.45) and age (r ⫽ 0.48; p ⬍ 0.05 for both).

Received for review November 10, 2003. Accepted in final form September 1, 2004. The costs of publication of this article were defrayed, in part, by the payment of page charges. This article must, therefore, be hereby marked “advertisement” in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. *Nutrition Research Medical Unit, Pediatric Hospital, National Medical Center Siglo XXI, Mexican Institute of Social Security, Mexico City, Mexico; Departments of †Nutrition Sciences, §Pediatrics, and ¶Health Services Administration and the **General Clinical Research Center, University of Alabama, Birmingham, Alabama; and ‡Department of Health and Sport Sciences, University of Oklahoma, Norman, Oklahoma. Address correspondence to Barbara A. Gower, University of Alabama at Birmingham, Department of Nutrition Sciences, Division of Physiology and Metabolism, 429 Webb Building, 1675 University Boulevard, Birmingham, AL 35294-3360. E-mail: [email protected] Copyright © 2004 NAASO

Activity counts alone were not associated with TEE. In multiple linear regression analyses, TEE was independently associated with only lean body mass. Percentage fat mass was independently associated with body weight, being a girl, and being white, but not with average total activity counts. Discussion: Activity counts obtained with the Actiwatch under free-living conditions do not reflect TEE in 4- to 6-year-old children and are not correlated with percentage fat mass. Therefore, average total activity counts obtained with the Actiwatch may be of limited value in identifying children at risk for becoming obese. Key words: physical activity, percentage fat mass, body composition, stable isotopes, African American

Introduction The prevalence of overweight and obesity in the U.S. has increased significantly over the last 2 decades. As compared with past decades, obesity is a problem that is beginning to manifest itself at earlier ages (1,2). The first years of life constitute an important period in the development of an individual. Infants born overweight, and remaining so throughout childhood, are at high risk of being obese as adults (3–5). In addition, children with a high growth velocity during infancy are more likely to become obese as adults compared with children who grow at a slower pace (5). The causes of the increase in the prevalence of obesity are not fully understood; possible explanations include an increase in the availability of energy-dense food and a decline in the level of physical activity (PA)1 (6,7). Thus, it is important to analyze

1 Nonstandard abbreviations: PA, physical activity; DLW, doubly labeled water; TEE, total energy expenditure.

OBESITY RESEARCH Vol. 12 No. 11 November 2004

1859

Accelerometry and Energy Expenditure in Children, Lopez-Alarcon et al.

PA patterns and the amount of energy expended during PA in young children to detect those who are at high risk of becoming obese. A variety of methods exist for estimating PA and energy expenditure. Historically, PA has been measured by questionnaire, direct observation, heart rate monitoring, and, more recently, accelerometry. The doubly labeled water (DLW) method is one of the most accurate and valid tools used for assessing total energy expenditure (TEE) in free-living subjects (8). However, this method is not practical in the field because of the high cost of the isotopes (2H and 18O) and the need for sophisticated analytical equipment. The Actiwatch accelerometer (Minimitter Co., Inc., Bend, OR), a recently developed, small, lightweight activity monitor, may provide a practical means of assessing PA. However, to date, it has yet to be validated against a true criterion method (e.g., DLW) in children. Other accelerometers, such as the CSA and the TriTracR3D, have been reported to be valid and reliable for assessing energy expenditure during specific, prescribed exercises, such as walking (on and off a treadmill), cycling, jumping rope, and running (9 –11). In these studies, which primarily use indirect calorimetry as the criterion measure, the activity counts derived from the monitor are positively related to energy expenditure. Only two studies have evaluated the Actiwatch. Among preschool children, activity counts obtained with the Actiwatch in 3-minute intervals have been correlated with levels of PA obtained by direct observation (12). In the second study, PA measured by room calorimetry has been positively correlated with activity counts in children and adolescents. However, the association was observed under a tightly controlled laboratory setting using a structured protocol of physical activities over a 6-hour period (13). Therefore, it is unclear whether accelerometers, specifically the Actiwatch monitor, are sensitive to the low-intensity activities routinely practiced by young children under free-living conditions. The main objective of the present study was to determine whether activity monitor counts obtained with the Actiwatch were related to TEE as assessed with DLW in young children under free-living conditions. A secondary objective was to determine whether activity counts were associated with body composition and, therefore, might be a reliable means of identifying children at risk for obesity.

Research Methods and Procedures Study Design Actiwatch activity monitors were tested in children under free-living conditions for 8 days. On Day 1, the activity monitor was placed on the right ankle of the subject. The subsequent 24 hours served as a trial period to determine 1860


whether the child could tolerate wearing the unit, and data from this day were discarded. Children who successfully tolerated the Actiwatch wore the unit for the next 7 days, during which experimental data were collected. At the start of Day 2, the parents obtained a urine sample from all children who continued to wear the Actiwatch. DLW was subsequently administered, marking the beginning of the 7-day period of assessment for TEE. Subjects Thirty-one children participated in the study. Most children were recruited by word of mouth from the Birmingham, AL area. Six of the children were recruited directly through the Jefferson County Headstart program. Eligible children were those healthy boys and girls 4 to 6 years of age whose parents agreed to participate. Parents were asked to sign an informed consent after the methods, risks, and benefits of the study were carefully explained to both them and their children. The Institutional Review Board for Human Use (University of Alabama at Birmingham) approved the study. PA Monitors The Actiwatch activity monitor model AW16 (Minimitter Co., Inc.) was evaluated. The Actiwatch monitor contains an omnidirectional sensor capable of detecting acceleration in two planes. Sensitive to 0.01 gravity (0.098 m/s2), this type of sensor integrates the degree and speed of motion and produces an electrical current that varies in magnitude. An increased degree of speed and motion produces an increase in voltage. The monitor stores this information as activity counts. The maximum sampling frequency is 32 Hz (14). The monitor was placed on the right ankle of the children and worn for 8 days. It was programmed to store total activity counts per minute during the observation period. At the end of the data collection period, the activity counts were downloaded onto a spreadsheet for analysis. Activity counts obtained during the first day (i.e., the first 24 hours) were not included in the analyses. TEE by DLW TEE was calculated as previously described using a protocol with a theoretical error of ⬍5% (15). Carbon dioxide production was determined using the equation by Speakman et al. (16), assuming a fixed dilution space ratio of 1.0427. Energy expenditure was calculated using the de Weir equation (17). Baseline samples of urine were collected on day 2 of the study, immediately before isotope administration (1 g DLW/kg body weight). Subsequent urine samples were collected on days 3 and 8 (second and third urine of the day). In some cases, urine samples were obtained by the parents, refrigerated overnight, and collected by a member of the research team from the homes of the children. For


Table 1. Characteristics of the study sample White

Measure

All children (mean ⴞ SD) n ⴝ 29

Boys (mean ⴞ SD) n ⴝ 10

African-American Girls (mean ⴞ SD) nⴝ7

Boys (mean ⴞ SD) nⴝ7

Girls (mean ⴞ SD) nⴝ5

Age (years) 4.9 ⫾ 0.9 5.3 ⫾ 0.8 4.8 ⫾ 0.9 4.7 ⫾ 1.0 4.4 ⫾ 0.9 Body mass (kg) 20.3 ⫾ 4.1 20.0 ⫾ 3.0 20.7 ⫾ 5.1 19.7 ⫾ 3.2 21.3 ⫾ 6.4 Height (cm) 111.4 ⫾ 8.4 112.4 ⫾ 8.0 112.9 ⫾ 9.0 108.9 ⫾ 8.5 110.8 ⫾ 9.4 Total energy expenditure (kJ/d) 5982 ⫾ 1087 6068 ⫾ 1164 5656 ⫾ 1426 6201 ⫾ 938 5957 ⫾ 746 Lean body mass (kg) 15.93 ⫾ 2.56 16.1 ⫾ 1.9 15.0 ⫾ 3.1 16.2 ⫾ 1.6 16.6 ⫾ 4.1 Fat mass (kg) 4.51 ⫾ 2.00 4.3 ⫾ 1.1 5.7 ⫾ 2.4 3.5 ⫾ 2.0 4.7 ⫾ 2.4 Percentage fat mass 21.5 ⫾ 6.3 21.1 ⫾ 3.9 26.8 ⫾ 5.2 17.0 ⫾ 7.6 21.1 ⫾ 5.4 Activity counts (counts/d) 658,816 ⫾ 201,657 642,427 ⫾ 232,096 621,415 ⫾ 190,632 794,141 ⫾ 167,637 554,502 ⫾ 134,242

children enrolled in the Headstart Program, urine samples were collected at the facility by a member of the investigative team. Urine samples were analyzed for 2H and 18O using isotope ratio mass spectrometry. Body Composition Body composition was determined using stable isotope dilution (deuterium). Lean body mass was calculated as total body water in kilograms, divided by sex-specific hydration constants (0.75 for boys and 0.76 for girls) (18). Fat mass was calculated as body weight in kilograms minus lean body mass (19,20). Power Calculations The following power calculations were conducted for comparison of TEE, determined by DLW, with activity counts obtained with the Actiwatch accelerometer. These calculations assume use of a two-tailed statistical test and a 5% significance level and were based on the approximate normality of the transformed sample correlation coefficient as implemented in the nQuery Advisor (version 4.0) software package. For TEE, with a sample size of 29 children, a statistically significant correlation of 0.501 can be found with 80% power, whereas a statistically significant correlation of 0.562 can be found with 90% power. In a recent study, activity counts determined with the Actiwatch had correlated with TEE determined with indirect calorimetry (r ⫽ 0.80) in a sample of 26 children 6 to 16 years old (13).

The primary purpose of the study was to determine whether the Actiwatch activity monitor is a reasonable substitute method for DLW in determining energy expenditure. The Actiwatch activity monitor would be considered a good substitute method if the correlation between measurements obtained through its use and measurements obtained through the use of the DLW method were 0.90 or greater and would be considered a reasonable substitute method if the correlation were 0.80 or greater. Based on the prior paragraphs, it is clear than any correlation of 0.80 or greater will be statistically significant. Statistical Analysis Activity counts were expressed as the total counts per day for each child, averaged over the 7-day data collection period (average total counts per day). TEE was expressed as kilojoules per day. Statistical analyses were performed with SPSS version 11 (2001). Data are presented as mean ⫾ SD. Pearson correlation analyses were used to identify those variables related to TEE and percentage fat mass. Multiple linear regression analyses were used to examine the relationships between the dependent variables TEE and percentage fat mass and activity counts and other predictors likely related to energy expenditure and body composition. p ⬍ 0.05 was considered significant.

Results Thirty-one children participated in the study. However, the activity monitor failed in one child, and the parents of OBESITY RESEARCH Vol. 12 No. 11 November 2004

1861


Figure 1: Actigrams for PA counts from 2 weekend days (A and B) and from 1 weekday (C). Patterns and magnitude of PA were highly variable among children, with long intervals of inactivity in some children (C) and very low levels of PA in others (A) during a complete day.

another child failed to collect urine samples; therefore, data from these two children were not included in the analyses. Body composition and energy expenditure variables are presented in Table 1. None of the study variables differed with respect to gender or race; thus, data were combined for the analyses. Representative activity charts (actigrams) for 2 weekend days and 1 weekday are shown in Figure 1. Both average activity counts and the pattern of activity were highly variable among subjects. The mean ⫾ SD for average total activity counts was 658,816 ⫾ 201,657 (Table 1). Analyses based on activity pattern were not conducted. 1862


Simple Pearson correlation coefficients for energy expenditure and body composition variables are presented in Table 2. TEE was significantly related to lean body mass and age and tended to be related to body mass (p ⫽ 0.06). Percentage fat mass was associated with fat mass, age, and body mass. Activity counts were not significantly associated with TEE (r ⫽ 0.27, p ⫽ 0.15, Figure 2) or percentage fat mass (r ⫽ ⫺0.03, p ⫽ 0.86). Several regression models were tested for the dependent variable TEE, adjusting for those variables associated with TEE. In all cases, activity counts were not significantly related to TEE. The model that best predicted TEE was that in


Table 2. Pearson correlation coefficients 关r (P)兴 TEE (kJ/d) TEE (kJ/d)

Fat mass (kg) Percentage fat mass Body mass (kg) Age (years)

Lean body mass (kg)

0.27 (0.153)

Activity counts (counts/d) Lean body mass (kg)

Activity Counts (counts/d)

0.27 (0.153) 0.44 (0.015) 0.14 (0.472) ⫺0.01 (0.969) 0.35 (0.063) 0.48 (0.008)

0.03 (0.869) ⫺0.01 (0.953) ⫺0.03 (0.866) 0.05 (0.781) 0.04 (0.835)

which the association was adjusted for lean body mass (Table 3). In multiple linear regression analyses for the dependent variable percentage fat mass, body mass emerged as a significant independent variable, as did gender and ethnicity. (Percentage fat mass was greater among girls and whites; Table 4.) Percentage fat mass was not associated with average total activity counts. The association with activity counts remained nonsignificant (p ⫽ 0.31) after adjusting for age, body weight, gender, ethnicity, and TEE.

0.44 (0.015) 0.03 (0.869)

0.54 (0.003) 0.17 (0.387) 0.87 (⬍0.001) 0.52 (0.004)

Fat mass (kg)

Percentage fat mass

Body mass (kg)

Age (years)

0.14 (0.472) ⫺0.01 (0.953) 0.54 (0.003)

⫺0.01 (0.969) ⫺0.03 (0.866) 0.17 (0.387) 0.90 (⬍0.001)

0.35 (0.063) 0.05 (0.781) 0.87 (⬍0.001) 0.86 (⬍0.001) 0.58 (⬍0.001)

0.48 (0.008) 0.04 (0.835) 0.52 (0.004) 0.48 (0.008) 0.37 (0.047) 0.59 (⬍0.001)

0.90 (⬍0.001) 0.86 (⬍0.001) 0.48 (0.008)

0.58 (⬍0.001) 0.37 (0.047)

0.59 (⬍0.001)

related with activity counts obtained during performance of several types of exercise (15-minute intervals; r ⫽ 0.56) (10). In these studies, data were collected during short periods of prescribed exercise. When data were collected over 24 hours, the association between activity counts and energy expenditure depended on the intensity of activity. Although activity counts collected from school children were highly correlated with energy expenditure during PA periods, activity counts underestimated energy expenditure during sedentary periods (11).

Discussion In this study, we found that the Actiwatch accelerometers did not accurately predict TEE in 4- to 6-year-old children. In addition, activity counts were not associated with percentage fat mass. Several other studies have reported that activity counts from various models of accelerometer are associated with energy expenditure in adults and children (9 –13). In a group of adults who completed four 5-minute bouts of walking at self-selected speeds, high correlations were observed between activity counts and energy expenditure, measured simultaneously with a portable metabolic unit (r ⫽ 0.59 to 0.78) (9). Five-minute bouts of other less intense activities, such as window washing and vacuuming, also were tested, and when the analysis was repeated with all activities combined (walking ⫹ window washing ⫹ vacuuming), the correlation coefficients decreased (r ⫽ 0.35 to 0.39). Also in adults, energy expenditure measured with indirect calorimetry was well-cor-

Figure 2: Correlation between PA counts and TEE (r ⫽ 0.27, p ⫽ 0.15).


1863


Table 3. Multiple linear regression analysis for the dependent variable TEE (kJ/d) Variable

Coefficient

Constant Activity counts/d Lean body mass (kg)

2100.53 0.0014 185.96

SE 1280.9 0.0009 71.4

P

Model R2

Model P

0.113 0.136 0.015

0.27

0.018

boys, as assessed by both BMI and the sum of five skinfolds, and also were less active than boys. In our study, activity counts were not related to either energy expenditure or percentage of fat mass, suggesting either that the monitors are not able to detect physiologically relevant differences in PA or that PA is not related to body composition in children. Although the sample size of our study was comparable with that of Puyau et al. (13), who reported an association between activity counts and TEE in children, we are aware that it may not have been sufficient to detect an association between PA and energy expenditure over longer periods of time that included active, sedentary, and sleeping periods and multiple types of activity. However, our results suggested a trend for an association between activity counts and TEE (p ⫽ 0.136). Therefore, with a larger sample size, the relationship may have achieved statistical significance. In summary, we conclude that, under free-living conditions, activity counts obtained with the Actiwatch did not reflect TEE in 4- to 6-year-old children and were not correlated with percentage fat mass. Therefore, average total activity counts obtained with the Actiwatch may be of limited value in identifying children at risk for becoming obese. Future studies might consider determining whether

In our study, we did not observe an association between the mean daily activity counts (obtained over a 7-day study period) and TEE assessed over the same interval. During this relatively long period of collection, the subjects engaged in a diverse suite of activities, ranging from sports and playing tag to watching television and sleeping. Taken together, studies conducted to date suggest that activity monitors may be valid for predicting energy expenditure during intense activity but not during sedentary periods. In addition, the monitors may not be capable of reflecting energy expenditure over a period of time during which multiple types of activity are performed. To our knowledge, there are no studies that have reported the ability of the activity monitors to accurately identify inactive children or children at risk for becoming obese. The only study to address this issue was conducted in school children for whom activity counts were obtained over a 7-day interval (21). The authors compared activity counts between weekdays and weekend days and between boys and girls. Although the authors concluded that activity monitors are a valid means of identifying children at risk, they did not measure TEE or other variables that indicate risk of obesity. They based their conclusion on the observations that girls were fatter than

Table 4. Multiple regression analysis for the dependent variable percentage fat mass Variable

Coefficient

SE

P

Model R2

Model P

Constant Activity counts/d TEE (kJ/d) Age (years) Body mass (kg) Gender (women)* Ethnicity (white)†

9.31 4.48⫻10⫺6 ⫺1.17⫻10⫺3 0.92 0.80 4.64 4.36

5.75 4.38⫻10⫺6 9.15⫻10⫺4 1.43 0.27 1.84 1.83

0.120 0.318 0.213 0.525 0.007 0.019 0.026

0.64

0.0005

* Compared with male subjects. † Compared with African-American subjects.

1864



activity patterns, activity counts obtained only from periods of relatively high activity, or activity counts above a certain threshold are better predictors of energy expenditure and obesity.

Acknowledgments This work was supported by NIH Grant R01 DK51684, by General Clinical Research Center Grant M01-RR00032, and by Clinical Nutrition Research Unit Grant P30DK56336. Actiwatches were provided by Minimitter. References 1. Booth M. The worldwide epidemic of obesity in adolescents. Adolesc Med. 2003;14:1–9. 2. Strauss RS, Pollack HA. Epidemic increase in childhood overweight, 1986-1998. J Am Med Assoc. 2001;286: 2845– 8. 3. Rogers I, EURO-BLCS Study Group. The influence of birthweight and intrauterine environment on adiposity and fat distribution in later life. Int J Obes Relat Metab Disord. 2003;27: 755–77. 4. Cameron N, Demerath EW. Critical periods in human growth and their relationship to diseases of aging. Am J Phys Anthropol. 2002;S35:159 – 84. 5. Martorell R, Stein AD, Schroeder DG. Early nutrition and later adiposity. J Nutr. 2001;131:874 – 80S. 6. US Department of Health and Human Services. Physical Activity and Health: A Report of the Surgeon General. Atlanta, GA: Center for Disease Control and Prevention, National Center for Chronic Disease Prevention and Health Promotion; 1996. 7. Liebman M, Pelican S, Moore SA, et al. Dietary intake, eating behavior, and physical activity-related determinants of high body mass index in rural communities in Wyoming, Montana, and Idaho. Int J Obes Relat Metab Disord. 2003; 27:684 –92. 8. Schoeller DA. Recent advances from application of doubly labeled water to measurement of human energy expenditure. J Nutr. 1999;129:1765– 8.

9. 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:S442–9. 10. Swartz AM, Strath SJ, Bassett DR, O’Brie WL, King GA, Ainsworth BE. Estimation of energy expenditure using CSA accelerometers at hip and wrist sites. Med Sci Sports Exerc. 2000;32:S450 – 6. 11. Rodriguez G, Beghin L, Michaud L, Moreno LA, Turck D, Gottrand F. Comparison of the TriTrac-R3D accelerometer and self-report activity diary with heart-rate monitoring for the assessment of energy expenditure in children. Br J Nutr. 2002;87:623–31. 12. 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. 13. Puyau MR, Adolph AL, Vohra FA, Butte NF. Validation and calibration of physical activity monitors in children. Obes Res. 2002;10:150 –7. 14. Minimitter Company, Inc. Actiwatch 16/Actiwatch 64/Actiwatch-L Activity Monitors: Instruction Manual. Bend, OR: Minimitter Company, Inc.; 1999. 15. Goran MI, Carpenter WH, McGloin A, Johnson R, Hardin JM, Weinsier RL. Energy expenditure in children of lean and obese parents. Am J Physiol. 1995;268:E917–24. 16. Speakman JR. The history and theory of the doubly labeled water technique. Am J Clin Nutr. 1998;68:S932– 8. 17. de Weir JB. New methods for calculating metabolic rate with special reference to protein metabolism. J Physiol. 1949;109: 1–9. 18. Lohman TG. Assessment of body composition in children. Pediat Exerc Sci. 1989;1:19 –30. 19. Shoeller DA. Hydrometry. In: Roche AF, Heymsfeld SB, Lohman TG, eds. Human Body Composition. Champaign, IL: Human Kinetics; 1996, pp. 25– 43. 20. Wang Z, Deurenberg P, Wang W, Pietrobelli A, Baumgartner RN, Heymsfeld SB. Hydration of fat-free body mass: review and critique of a classic body-composition constant. Am J Clin Nutr. 1999;69:833– 41. 21. Metcalf BS, Voss LD, Wilkin TJ. Accelerometers identify inactive and potentially obese children (Early Bird 3). Arch Dis Child. 2002;87:166 –7.


1865