Journal of Physical Activity and Health, 2015, 12, 124 -131 http://dx.doi.org/10.1123/jpah.2013-0111 © 2015 Human Kinetics, Inc.
Official Journal of ISPAH www.JPAH-Journal.com ORIGINAL RESEARCH
Estimating Physical Activity in Children: Impact of Pedometer Wear Time and Metric Kelly R. Laurson, Gregory J. Welk, and Joey C. Eisenmann Background: The purpose of this study was to provide a practical demonstration of the impact of monitoring frame and metric when assessing pedometer-determined physical activity (PA) in youth. Methods: Children (N = 1111) were asked to wear pedometers over a 7-day period during which time worn and steps were recorded each day. Varying data-exclusion criteria were used to demonstrate changes in estimates of PA. Steps were expressed using several metrics and criteria, and construct validity was demonstrated via correlations with adiposity. Results: Meaningful fluctuations in average steps per day and percentage meeting PA recommendations were apparent when different criteria were used. Children who wore the pedometer longer appeared more active, with each minute the pedometer was worn each day accounting for an approximate increase of 11 and 8 steps for boys and girls, respectively (P < .05). Using more restrictive exclusion criteria led to stronger correlations between indices of steps per day, steps per minute, steps per leg length, steps per minute per leg length, and obesity. Conclusion: Wear time has a meaningful impact on estimates of PA. This should be considered when determining exclusion criteria and making comparisons between studies. Results also suggest that incorporating wear time per day and leg length into the metric may increase validity of PA estimates. Keywords: physical activity monitor, compliance, youth, exclusion criteria, inclusion criteria Obtaining accurate and reliable measures of habitual physical activity (PA) is paramount in research and practical applications (eg, recreation programs, physical education, clinical settings).1–3 Because of the limitations of self-report questionnaires and surveys in children (eg, cognitive ability, recall), many researchers turn to electronic motion sensors to objectively assess PA.4 Although accelerometry is a widely used method for objectively measuring PA, it may be a cost-prohibitive option for some research groups and practitioners. Pedometers are an often-used less expensive alternative to accelerometers for assessing PA. These step-counting tools offer some major advantages over accelerometers because of their simplicity, feasibility, and interpretability. However, many questions remain about how to most effectively use these devices to assess PA in free-living populations. Results have been mixed when determining the total number of monitoring days needed and which monitoring days to use to characterize habitual pedometer-determined PA in children. A review by Tudor-Locke et al5 outlined the reliability of pedometer-determined PA estimates using various combinations of weekdays and weekend days. The authors reported intraclass reliability coefficients ranging between .65 and .96 with monitoring periods from 2 to 8 days. No definitive acceptable number of monitoring days has been identified, but because of the variety of approaches and results found in the literature, this topic necessitates careful consideration.6 The number of hours that constitutes a complete monitoring day has yet to be standardized for pedometer usage, although it may
Laurson ([email protected]) is with the School of Kinesiology and Recreation, Illinois State University, Normal, IL. Welk is with the Dept of Health and Human Performance, Iowa State University, Ames, IA. Eisenmann is with the Dept of Radiology, Michigan State University, East Lansing, MI. 124
be more important than quantity of monitoring days. Participants are generally advised to wear the device at all times, except for activities such as sleeping, bathing, swimming, contact sports, and so forth. However, acceptable wear time criteria differ across the literature. For example, some studies have excluded participants for not wearing the pedometer ≥ 1 hour of the day,7,8 others have defined a day of wear as 8 hours9; our group included participants in data analyses if the pedometer was worn for at least 10 h/d.10 These issues are not unique to pedometers. Data reduction methodology for accelerometers, with regard to compliance and standardizing day length, have received considerable attention.11–13 Further investigation and standardization of data treatment methods are needed to improve the quality of information available with pedometers. Pedometer metric is also a central issue of data treatment after collection. For example, previous research has suggested that leg length is an important indicator of energy expenditure when using pedometry in children.14,15 Children with shorter legs (and a smaller stride length) require a greater number of steps to travel an equivalent distance compared with their taller peers. This could have implications for estimates of energy expenditure and may be at least partially controlled by incorporating leg length into the pedometer metric. Besides leg length, incorporating the amount of time the pedometer was worn into the metric (such as pedometer steps per minute) may also be of benefit by estimating PA intensity.5 The extent to which variations in wear time and data treatment may influence pedometer data is unknown. Because most pedometers simply record the raw volume of steps taken during a day, even small differences in monitoring time and/or subject compliance may have major implications for PA estimates. The purpose of this research is to demonstrate the impact of monitoring period and data cleaning practices on pedometer-determined PA estimates in children by comparing available sample sizes, estimates of activity, and the relationships between metrics and common health outcomes, such as adiposity.
Pedometer Exclusion Criteria 125
Methods
Downloaded by Heriot Watt Univ on 09/25/16, Volume 12, Article Number 1
Subjects The subjects were from 2 Midwestern communities (Lakeville, MN, and Cedar Rapids, IA) participating in a community-, school-, and family-based childhood obesity intervention. The details of the complete intervention can be found elsewhere.16 The data herein were collected before randomization into control or treatment groups for the intervention. Therefore, the current study can be considered a cross-sectional, observational design. The subjects were in grades 3 to 5 at 10 public elementary schools. We explained procedures and risks of the research to all subjects and primary caregivers, verbally and in writing, before participation. Written assent from each subject and consent from his or her primary caregiver was obtained. The study protocol was approved by the University of Minnesota Human Subjects Review Board and is in accordance with the Declaration of Helsinki. A total of 2114 elementary school children were invited to participate in the study, 1360 (64%) students accepted, and 1111 (53%) participated in the pedometer monitoring during the study period. A variety of pedometer exclusion criteria were compared, and some children had missing anthropometry data, so sample sizes vary by analysis (see below).
Anthropometry Standing height, sitting height, waist circumference, and body mass were measured by a school nurse according to standard procedures.17 Standing height and sitting height were measured using a portable stadiometer (Road Rod; Seca, Hanover, MD). Body mass was measured using a strain gauge scale (Lifesource MD; A&D, Milpitas, CA). Waist circumference (WC) was measured to the nearest 0.1 cm above the superior border of the iliac crest using a Gulick tape measure (CountryTechnology, Gays Mills, WI). All nurses participated in anthropometry measurement training sessions before data collection. Measurement error was determined during data collection by duplicate measures of every 25th subject. Overall, measurement error was small (SEM = 0.3 cm standing height, 0.1 kg body mass, 0.2 cm WC). Sitting height was subtracted from standing height to calculate leg length. Body mass index (BMI, kg/m2) was calculated from measurements of standing height and body mass. Overweight and obesity were determined on the basis of age- and sex-specific reference values developed by the International Obesity Task Force.18
Physical Activity Free-living PA was assessed with the widely accepted, researchquality Digiwalker SW-200 pedometer (YAMAX Health & Sports, San Antonio, TX).19 The subjects were given instructions on wearing the pedometer during the school day, and pedometers were checked using the shake test and 10-step test. The shake test was conducted by shaking the pedometer 10 times and verifying the recorded step count. The 10-step test was conducted by having the subject walk 10 steps and then checking the step count recorded. Pedometers that did not function properly during these validity checks were replaced. However, we did not record how many pedometers were replaced. Students were instructed to put on the pedometer (at the waist approximating the axillary line) upon waking and to take it off when going to bed. Students used the pedometer for 7 days. They were given a log sheet to record the times at which the pedometer was put on and taken off at any point in the day. Participants recorded the time on/off and number of steps accumulated each day during
the 7-day period. The classification of PA was based upon a normreferenced criterion of 13,000 steps/d for boys and 11,000 steps/d for girls, suggested by Vincent and Pangrazi.20 This pedometer-specific PA cutpoint has been noted to be associated with childhood obesity21 and is currently endorsed by the American Academy of Pediatrics within a policy statement on PA and childhood obesity.22
Statistical Analysis Descriptive statistics were calculated for the total sample and separately by sex. To demonstrate potential fluctuations in sample size and PA estimates, we calculated group means for pedometer steps per day and percentages of those meeting the Vincent and Pangrazi cutpoint using a variety of exclusion criteria. Exclusion groups were created on the basis of the number of days the pedometer was worn (≥ 1, ≥ 2, ≥ 4, and ≥ 6) and the number of hours the pedometer was worn on each day (≥ 1, ≥ 5, ≥ 8, ≥ 10, and ≥ 12). Days worn and times worn per day were selected to include a variety of criteria potentially used in the literature. Including subjects that wore the pedometer for 1 day for any amount of time would be considered equivalent to no exclusion criteria. However, including subjects that were required to wear the pedometer for at least 6 days for at least 12 hours each of those days would generally be considered restrictive. These criteria were used to cross-tabulate subjects into 20 groups that were not mutually exclusive. For example, the number of steps per day of a subject who wore the pedometer for 1 hour on 2 separate days of the week (totaling 2 hours of monitoring time) would be included in the “≥ 1 day and ≥ 1 hour each day” group and also in the “≥ 2 days and ≥ 1 hour each day” group. In contrast, the weekly stepsper-day estimate for a child who wore the pedometer for ≥ 6 days and ≥ 12 hours each day would be included in all group estimates. Three-way analysis of variance was used to determine whether sex, weight status (normal weight or overweight + obese), or age group (< 10 years or ≥ 10 years) accounted for potential differences in days worn and time worn per day (in minutes). Linear regression was used to estimate how average time worn per day influenced mean step count. Univariate regression models were created using only average minutes worn per day as the predictor. Then, additional models were created controlling for age, BMI, and leg length. Regressions were performed separately by sex. Finally, a series of partial correlations were used to investigate the influence of wear time and leg length when included in the pedometer metric. In theory, if requiring the subjects to wear the pedometer for a longer period of time to be included in the data analysis were to increase the validity of the PA measure, one would expect stronger associations between PA and PA correlates when using more restrictive criteria. To demonstrate this construct validity of PA, we correlated steps per day and steps per minute of time worn with BMI and WC while controlling for age using 2 different sets of exclusion criteria (≥ 1 day and ≥ 1 hour each day versus ≥ 6 days and ≥ 12 hours each day). Given the potential influence of leg length on pedometry, we also used metrics that incorporated leg length (such as steps per leg length) into the correlational analysis to determine whether the strength of the association between PA and these anthropometric indices of obesity would be further enhanced.
Results Of the children that agreed to participate in the study (n = 1360), approximately 82% wore the pedometer for any amount of time (n = 1111). Descriptive statistics for subjects that wore the pedometer are located in Table 1. There were no significant differences between
126 Laurson, Welk, and Eisenmann
Table 1 Descriptive Statistics Variable Age (y) Height (cm) Weight (kg) Leg length (cm) Body mass index (kg/m2) Overweight (%) Obese (%) Waist circumference (cm) Days pedometer worn Minutes pedometer worn (per day)
Boys (n = 510)
Girls (n = 601)
Combined (N = 1111)
9.6 (0.9) 138.6 (7.6) 35.6 (8.9) 65.7 (4.6) 18.4 (3.2) 19.0 6.3 63.6 (9.4) 6 (1) 737 (92)
9.6 (0.9) 137.8 (8.1) 35.3 (9.5) 65.5 (5.1) 18.4 (3.5) 20.0 8.5 63.1 (9.7) 6 (1) 740 (79)
9.6 (0.9) 138.1 (7.9) 35.4 (9.2) 65.5 (4.8) 18.4 (3.4) 19.5 7.5 63.4 (9.6) 6 (1) 738 (85)
Downloaded by Heriot Watt Univ on 09/25/16, Volume 12, Article Number 1
Note. Values are mean (SD) or percentage.
normal weight and overweight children in the number of days the pedometer was worn (P = .867). Girls wore the pedometer 0.15 days longer than boys, although this difference was not statistically significant (P = .069). Children ≥ 10 years old wore the pedometer 0.20 days longer than younger children (P = .018). Likewise, there were no differences between normal weight and overweight (P = .572) children or between boys and girls (P = .404) for time worn per day. However, children ≥ 10 years old wore the pedometer 19.3 minutes longer each day than younger children (P = .018). None of the interactions in the model were statistically significant (all P > .05). Over the course of the monitoring week, the pedometer was worn on average for 6 days and 738 min/d (12.3 h/d). The impact of using different exclusion criteria (various criteria for days worn and time worn per day) on PA outcomes is shown in Table 2 (boys) and Table 3 (girls). The use of more restrictive criteria, either by requiring more days of monitoring or more time worn per day, reduced the effective sample size and increased estimates of mean steps/day and the percentage of subjects that would be considered physically active. In the univariate regression analyses, each minute the pedometer was worn per day accounted for an increase of 10.6 steps/day in boys (n = 490) and 7.8 steps/day in girls (n = 578). Overall, 7.8% and 5.6% of the total variance in steps was accounted for by time worn per day in boys and girls, respectively. In the multiple regression models, time worn per day was a stronger predictor of steps per day than were age, leg length, or BMI in boys. For girls, only BMI provided more predictive utility than minutes worn per day. In the partial correlations (controlling for age), using more restrictive criteria and expressing PA as steps per minute or steps per leg length increased the strength of the association between steps and BMI or WC (Table 4). In general these correlations were stronger for girls, and when using WC, ranged from –0.242 to –0.406.
Discussion Overall, this study demonstrates that pedometer data cleaning and exclusion criteria have a meaningful impact on the resulting interpretation of the data. Specifically, the number of days worn and the time worn per day influence available sample size, estimates of steps per day, and categorization of children into active or inactive groups. These results are unique in that they are presented using
multiple exclusion criteria within the same sample, providing direct comparisons of various pedometer cleaning methods. Focusing on days worn restrictions, the largest subject loss occurred when we required subjects to have a minimum of 6 days of wear, rather than 4 days, from the 1-week monitoring period. However, meaningful differences in available sample were still evident when comparing a 2-day to a 4-day wear requirement. For example, the average mean sample sizes for girls were 586, 578, 527, and 384 for ≥ 1, ≥ 2, ≥ 4, and ≥ 6 days, respectively (Table 3). Increasing the days worn requirement decreased available sample size. This decrease appears to be nonlinear; there were larger decreases in sample size as additional days were required to meet the exclusion criteria. It should also be noted that the decreases in sample size caused by the increases in the days worn requirement appear to be modified by the time worn per day requirement, with greater losses in sample size across days worn when the wear time requirement was ≥ 10 h/d. Although many children wore the pedometer for ≥ 10 to 12 hours a few days during the monitoring week, less than 70% of the sample was able to provide ≥ 4 days of wear and ≥ 12 h/d of wear. The magnitude of subject loss because of unacceptable data, although not always reported, varies greatly in previous research. For example, 93% of 11,669 Canadian youth wore a pedometer for the entire 7-day protocol, but wear time each day was not considered.23 Another study using a 4-day protocol reported that 39% of 602 subjects failed to provide a full 4 days of monitoring when the sample was restricted to those who did not remove the pedometer for more than 1 h/d.24 Besides subject loss, time worn per day seems to have a larger impact on estimates of mean steps per day than does days worn. In Tables 2 and 3, the largest fluctuations in mean steps per day and percentage of subjects meeting step recommendations occurred across time-worn-per-day groups, not days worn. The longer per day the subjects were required to wear the pedometer, the larger the increase in estimates of steps per day. For example, boys’ estimates of steps per day were approximately 500 steps higher when the criterion for time worn was restricted to ≥ 12 h/d instead of ≥ 10 hours. This resulted in an increase of the estimate of boys achieving the steps-per-day recommendations by 6% to 10% of the available sample. A similar influence of time worn per day can be seen for girls. Using National Health and Nutrition Examination Survey accelerometer data, Herrmann et al11 compared the PA estimates from a 14-hour minimum-wear criterion to 13-, 12-, 11-, and 10-hour criteria. The mean absolute percentage error
127
≥ 1 Hours Sample size Mean steps per day (95% CI) Boys with > 13,000 steps per day when using selected exclusion criteria (95% CI) ≥ 5 Hours Sample size Mean steps per day (95% CI) Boys with > 13,000 steps per day when using selected exclusion criteria (95% CI) ≥ 8 Hours Sample size Mean steps per day (95% CI) Boys with > 13,000 steps per day when using selected exclusion criteria (95% CI) ≥ 10 Hours Sample size Mean steps per day (95% CI) Boys with > 13,000 steps per day when using selected exclusion criteria (95% CI) ≥ 12 Hours Sample size Mean steps per day (95% CI) Boys with > 13,000 steps per day when using selected exclusion criteria (95% CI)
Time worn per day
38.5% (34.3–43.1) n = 496 12,431 (12,131–12,732) 39.5% (35.2–43.8) n = 493 12,632 (12,331–12,933) 41.6% (37.3–46.0) n = 484 12,912 (12,606–13,218) 45.7% (41.2–50.1) n = 449 13,385 (13,056–13,714) 52.3% (47.7–57.0)
38.8% (34.6–43.1) n = 503 12,434 (12,134–12,735) 39.8% (35.5–44.0) n = 501 12,639 (12,340–12,938) 41.9% (37.7–46.3) n = 497 12,885 (12,582–13,188) 45.5% (41.1–49.9) n = 476 13,317 (12,991–13,643) 51.9% (47.4–56.4)
≥ 4 Days
53.7% (48.4–59.0)
n = 337 13,564 (13,187–13,941)
48.1% (43.4–52.9)
n = 428 13,119 (12,799–13,439)
43.5% (39.0–48.1)
n = 460 12,809 (12,502–13,117)
40.9% (36.5–45.4)
n = 474 12,555 (12,250–12,860)
39.2% (34.9–43.6)
n = 487 12,456 (12,147–12,766)
Days Worn
n = 506 12,367 (12,060–12,673)
≥ 2 Days
n = 510 12,375 (12,068–12,682)
≥ 1 Day
53.5% (45.3–61.7)
n = 142 13,581 (13,001–14,160)
43.3% (37.5–49.2)
n = 277 12,958 (12,560–13,357)
42.1% (37.0–47.4)
n = 349 12,816 (12,446–13,165)
41.3% (36.4–46.3)
n = 380 12,679 (12,344–13,015)
40.4% (35.7–45.2)
n = 403 12,622 (12,287–12,957)
≥ 6 Days
Table 2 Sample Size, Mean Steps per Day, and Percentage of Boys Meeting Steps-per-Day Recommendations by Exclusion Criteria Combinations of Days Worn and Time Worn per Day
Downloaded by Heriot Watt Univ on 09/25/16, Volume 12, Article Number 1
128
≥ 1 Hours Sample size Mean steps per day (95% CI) Girls with > 11,000 steps per day when using selected exclusion criteria (95% CI) ≥ 5 Hours Sample size Mean steps per day (95% CI) Girls with > 11,000 steps per day when using selected exclusion criteria (95% CI) ≥ 8 Hours Sample size Mean steps per day (95% CI) Girls with > 11,000 steps per day when using selected exclusion criteria (95% CI) ≥ 10 Hours Sample size Mean steps per day (95% CI) Girls with > 11,000 steps per day when using selected exclusion criteria (95% CI) ≥ 12 Hours Sample size Mean steps per day (95% CI) Girls with > 11,000 steps per day when using selected exclusion criteria (95% CI)
Time worn per day n = 600 10,688 (10,473–10,903) 44.5% (40.5–48.5) n = 587 10,750 (10,538–10,962) 45.7% (41.6–49.7) n = 584 10,946 (10,729–11,164) 48.6% (44.6–52.7) n = 577 11,171 (10,944–11,399) 52.3% (48.3–56.4) n = 542 11,530 (11,289–11,771) 59.2% (55.1–63.4)
44.4% (40.5–48.4) n = 588 10,749 (10,538–10,960) 45.6% (41.6–49.6) n = 587 10,941 (10,725–11,158) 48.7% (44.7–52.8) n = 585 11,185 (10,955–11,415) 52.6% (48.6–56.7) n = 571 11,511 (11,269–11,752) 59.0% (55.0–63.1)
≥ 2 Days
n = 601 10,687 (10,473–10,902)
≥ 1 Day
≥ 4 Days
58.6% (53.9–63.3)
n = 418 11,532 (11,274–11,789)
52.3% (48.0–56.6)
n = 522 11,169 (10,943–11,396)
48.5% (44.3–52.6)
n = 551 10,954 (10,734–11,174)
46.1% (42.0–50.2)
n = 562 10,802 (10,585–11,018)
44.8% (40.8–48.9)
n = 582 10,716 (10,497–10,936)
Days worn
57.1% (50.0–64.3)
n = 182 11,611 (11,227–11,994)
51.6% (46.3–56.8)
n = 347 11,123 (10,863–11,383)
49.6% (44.9–54.4)
n = 425 11,051 (10,819–11,283)
45.8% (41.3–50.3)
n = 467 10,842 (10,619–11,065)
44.3% (39.9–48.6)
n = 499 10,676 (10,447–10,905)
≥ 6 Days
Table 3 Sample Size, Means Steps per Day, and Percentage of Girls Meeting Steps-per-Day Recommendations by Exclusion Criteria Combinations of Days Worn and Time Worn per Day
Downloaded by Heriot Watt Univ on 09/25/16, Volume 12, Article Number 1
Pedometer Exclusion Criteria 129
Table 4 Partial Correlations Between Pedometer Metrics and Obesity Indices by Data Exclusion Criteria Body mass index Exclusion criteria
Waist circumference
Pedometer metric
Boys
Girls
Boys
Girls
Steps per day Steps per minute worn per day Steps per leg length Steps per minute per leg length
–0.242 –0.248 –0.264 –0.268
–0.278 –0.284 –0.318 –0.341
–0.286 –0.270 –0.338 –0.328
–0.298 –0.306 –0.366 –0.392
Steps per day Steps per minute worn per day Steps per leg length Steps per minute per leg length
–0.261 –0.272 –0.296 –0.302
–0.283 –0.306 –0.325 –0.356
–0.294 –0.302 –0.357 –0.362
–0.304 –0.328 –0.374 –0.406
≥ 1 d and ≥ 1 h/da
Downloaded by Heriot Watt Univ on 09/25/16, Volume 12, Article Number 1
≥ 6 d and ≥ 10 h/db
Note. All correlations adjusted for age and are statistically significant (P < .05). a Available sample: boys = 489, girls = 591. b Available sample: boys = 267, girls = 341.
for steps per day in adults increased as wear time per day decreased (4%, 10%, 17%, and 25% versus the 14-hour day, respectively). Relaxing the exclusion criteria to include more subjects lowered estimates of PA, similar to what was found in the current study. The analysis of variance provided little insight into the reasons for compliance (or lack thereof) for hours worn per day or days worn per week. No differences were seen in wear time of the pedometer by sex or weight status. Older children did wear the pedometer slightly longer than younger children. Although this difference was statistically significant, it approximated a 20 minute difference in time worn per day and 0.2 days of wear over the week. Likewise, Schmidt et al25 found that mean pedometer wear time was not associated with age, sex, BMI, or education level in adults. This supports our findings that wear time is not a product of selection, because differences are unrelated to important sample characteristics. Schmidt et al25 also found that time worn (minutes per day) was correlated with the number of steps recorded (r = .20). Likewise, our regression analysis revealed that total time worn per day accounted for 6% to 8% of the variance in steps per day; each minute of daily wear time equated to approximately 9 recorded steps. In essence, the regression indicates that if the pedometer is removed for only 30 to 60 minutes during the day, this can potentially result in an underestimation error of 250 to 500 steps. These findings in children, along with those already described by Herrmann et al11 and Schmidt et al25 in adults, suggest that studies using stringent wear criteria (eg, pedometer removed for no more than 1 h/d) are still vulnerable to error from variation in wear time. Overall, with regard to sample size and the resulting estimates of activity, using the most restrictive criteria (≥ 6 d and ≥ 12 h/d) reduced the usable sample from 1111 to 324 subjects (29.2% compliance) but increased estimates of mean steps per day and percentage of subjects meeting steps-per-day recommendations. Although this is the most extreme example from the data, restricting usable data to ≥ 4 days and ≥ 10 h/d still resulted in a loss of approximately 15% of the total participating sample (usable n = 950 of 1111). In a review of pedometry methods in youth, Clemes and Biddle26 noted that compliance rates in previous research ranged from 46% to 99%. In addition, some of the studies reviewed also did not report the amount of wearing time that constituted a monitoring
day. Combined with the evidence provided in the current study, it is reasonable to assume that the estimates of PA provided throughout the literature are affected by time-worn-per-day criteria to varying degrees. More specifically, requiring subjects to wear the pedometer for larger portions of the day provides higher, more favorable estimates of PA. The dilemma resulting from this phenomenon is determining whether children with higher step counts are truly more physically active, merely wore the pedometer for longer than others in their cohort, or both. To demonstrate how exclusion criteria can impact construct validity, we computed partial correlations between steps per day and adiposity while controlling for age (Table 4). Using more restrictive exclusion criteria resulted in slightly stronger correlations between step metrics and BMI or WC. Correlations were further enhanced when expressing PA as steps per minute of wear time (steps per minute), regardless of the exclusion criteria used. This may indicate that statistically adjusting the data or metric on the basis of time worn per day may be an avenue to correct for differences in wear time per day between subjects (rather than subject exclusion). Differences in the correlations between steps per day and steps per minute were small but consistent. Using leg length in the metric has been suggested previously, because taller individuals require fewer steps to travel an equivalent distance as a shorter individual.14,27 Adding leg length to the metric (eg, steps per leg length) increased the strength of the correlations to a greater degree than using steps per minute alone, although the strongest correlations were generally found when using both minutes worn per day and leg length (steps per minute per leg length). Further consideration and refinement of these approaches to pedometer metric are recommended. The findings from this study and similar studies of pedometer compliance5,25,26 underscore the need for continued investigation on the topic of pedometer methodology. We recommend reporting pedometer wear duration per day as a proportion of the time during which the child was not asleep (eg, percentage of the waking day). Because sleep requirements decrease with age, older children may have a longer opportunity of time to wear the pedometer. Future investigations should also examine PA monitor compliance during different segments of the requested wear period. PA fluctuates throughout the day,28,29 so the effect of removing activity monitors
Downloaded by Heriot Watt Univ on 09/25/16, Volume 12, Article Number 1
130 Laurson, Welk, and Eisenmann
probably differs depending on when they are removed. Providing subjects with logs to record wake/sleep time and pedometer on/ off time could potentially reduce error,30 although this may also provide an avenue for self-report bias to influence estimates of PA. If children did not wear the monitor all day, statistical adjustments may provide more accurate estimates and suitable comparisons. Compliance might be enhanced by using incentives, detailed instructions, and more frequent contact with subjects to remind them of the importance of wearing the monitor. Sirard and Slater31 found that accelerometer wear compliance in adolescents was improved by using such strategies. The current study is not without limitations. Not all children in the school districts chose to participate in the original intervention and did not wear the pedometer, which could lead to bias. However, comparing the demographics of the sample to that of the school districts from which they were enrolled revealed no significant differences, indicating that the potential for bias is low. The pedometer model used in the study did not have a memory function, requiring that children and parents record daily step counts using logs, which could have led to lost data or inaccurate data through self-report bias or mistakes in data recording. However, logged steps have been found to be well correlated with alternative objective measures of physical activity.26,32 Regardless, neither of these limitations would be expected to detract from the analyses herein because the focus of the study was on the variance in stepping estimates created by the treatment of the data, rather than on the stepping data itself. Strengths include the use of a large, multisite sample of children and multiple combinations of wear criteria to demonstrate the variance in data interpretation and construct validity.
Conclusions In summary, the results herein demonstrate how criteria for pedometer days worn and time worn per day influence the available sample and subsequent outcomes (eg, relationship between pedometer PA and adiposity). The resulting differences in mean steps per day and percentage of subjects that are considered physically active is potentially important, as is the loss in sample size by enforcing more restrictive criteria. Care should be taken to ensure that monitoring is completed over enough days to provide a reliable estimate and also that the activity monitor is worn for enough time on those days to provide a valid estimate of PA. However, this research demonstrates how difficult it is to achieve such a balance. At the very least, future studies should provide more detailed information about time worn per day, days worn, and subject exclusion methodology because of the influence these factors have on the resulting data. Acknowledgments In Lakeville, Minnesota SWITCH is sponsored by Medica Foundation, Healthy and Active America Foundation, and Fairview Health Services. In Cedar Rapids, Iowa SWITCH is sponsored by Cargill and the Healthy and Active America Foundation. SWITCH is a trademark of the National Institute of Media and the Family MediaWise campaign.
References 1. Trost SG. Objective measurement of physical activity in youth: current issues, future directions. Exerc Sport Sci Rev. 2001;29(1):32–36. PubMed doi:10.1097/00003677-200101000-00007
2. Welk GJ, Corbin CB, Dale D. Measurement issues in the assessment of physical activity in children. Res Q Exerc Sport. 2000;71(2 Suppl):S59–S73. PubMed 3. Corder K, Ekelund U, Steele RM, Wareham NJ, Brage S. Assessment of physical activity in youth. J Appl Physiol. 2008;105(3):977–987. PubMed doi:10.1152/japplphysiol.00094.2008 4. Westerterp KR. Assessment of physical activity: a critical appraisal. Eur J Appl Physiol. 2009;105(6):823–828. PubMed doi:10.1007/ s00421-009-1000-2 5. Tudor-Locke C, McClain JJ, Hart TL, Sisson SB, Washington TL. Pedometry methods for assessing free-living youth. Res Q Exerc Sport. 2009;80(2):175–184. PubMed doi:10.1080/02701367.2009.1 0599551 6. Baranowski T, Mâsse LC, Ragan B, Welk G. How many days was that? We’re still not sure, but we’re asking the question better! Med Sci Sports Exerc. 2008;40(7 Suppl):S544–S549. PubMed doi:10.1249/ MSS.0b013e31817c6651 7. Duncan JS, Schofield G, Duncan EK. Pedometer-determined physical activity and body composition in New Zealand children. Med Sci Sports Exerc. 2006;38(8):1402–1409. PubMed doi:10.1249/01. mss.0000227535.36046.97 8. Duncan MJ, Al-Nakeeb Y, Woodfield L, Lyons M. Pedometer determined physical activity levels in primary school children from central England. Prev Med. 2007;44(5):416–420. PubMed doi:10.1016/j. ypmed.2006.11.019 9. Sigmund E, Sigmundová D, El Ansari W. Changes in physical activity in pre-schoolers and first-grade children: longitudinal study in the Czech Republic. Child Care Health Dev. 2009;35(3):376–382. PubMed doi:10.1111/j.1365-2214.2009.00945.x 10. Duncan JS, Schofield G, Duncan EK. Step count recommendations for children based on body fat. Prev Med. 2007;44(1):42–44. PubMed doi:10.1016/j.ypmed.2006.08.009 11. Herrmann SD, Barreira TV, Kang M, Ainsworth BE. Impact of accelerometer wear time on physical activity data: a NHANES semisimulation data approach. Br J Sports Med. 2014;48(3):278–282. PubMed doi:10.1136/bjsports-2012-091410 12. Ward DS, Evenson KR, Vaughn A, Rodgers AB, Troiano RP. Accelerometer use in physical activity: best practices and research recommendations. Med Sci Sports Exerc. 2005;37(11 Suppl):S582–S588. PubMed doi:10.1249/01.mss.0000185292.71933.91 13. Matthews CE, Hagströmer M, Pober DM, Bowles HR. Best practices for using physical activity monitors in population-based research. Med Sci Sports Exerc. 2012;44(1 Suppl 1):S68–S76. PubMed doi:10.1249/ MSS.0b013e3182399e5b 14. Eisenmann JC, Wickel EE. Moving on land: an explanation of pedometer counts in children. Eur J Appl Physiol. 2005;93(4):440–446. PubMed doi:10.1007/s00421-004-1227-x 15. Stone MR, Esliger DW, Tremblay MS. Comparative validity assessment of five activity monitors: does being a child matter? Pediatr Exerc Sci. 2007;19(3):291–309. PubMed 16. Eisenmann JC, Gentile DA, Welk GJ, et al. SWITCH: rationale, design, and implementation of a community, school, and family-based intervention to modify behaviors related to childhood obesity. BMC Public Health. 2008;8:223. PubMed doi:10.1186/1471-2458-8-223 17. Malina R. Anthropometry. In: Maud P, Foster C, eds. Physiological assessment of human fitness. Champaign, IL: Human Kinetics; 1995:205–219. 18. Cole TJ, Bellizzi MC, Flegal KM, Dietz WH. Establishing a standard definition for child overweight and obesity worldwide: international survey. BMJ. 2000;320(7244):1240–1243. PubMed doi:10.1136/ bmj.320.7244.1240
Downloaded by Heriot Watt Univ on 09/25/16, Volume 12, Article Number 1
Pedometer Exclusion Criteria 131 19. Schneider PL, Crouter SE, Bassett DR. Pedometer measures of free-living physical activity: comparison of 13 models. Med Sci Sports Exerc. 2004;36(2):331–335. PubMed doi:10.1249/01. MSS.0000113486.60548.E9 20. Vincent SD, Pangrazi RP. An examination of the activity patterns of elementary school children. Pediatr Exerc Sci. 2002;14:432–441. 21. Laurson KR, Eisenmann JC, Welk GJ, Wickel EE, Gentile DA, Walsh DA. Evaluation of youth pedometer-determined physical activity guidelines using receiver operator characteristic curves. Prev Med. 2008;46(5):419–424. PubMed doi:10.1016/j.ypmed.2007.12.017 22. Council on Sports Medicine and Fitness, Council on School Health. Active healthy living: prevention of childhood obesity through increased physical activity. Pediatrics. 2006;117(5):1834–1842. PubMed doi:10.1542/peds.2006-0472 23. Craig CL, Tudor-Locke C, Cragg S, Cameron C. Process and treatment of pedometer data collection for youth: the Canadian Physical Activity Levels among Youth study. Med Sci Sports Exerc. 2010;42(3):430– 435. PubMed doi:10.1249/MSS.0b013e3181b67544 24. Wilde B, Corbin C, Le Masurier G. Free-living pedometer step counts of high school students. Pediatr Exerc Sci. 2004;16(1):44–53. 25. Schmidt MD, Blizzard CL, Venn AJ, Cochrane JA, Dwyer T. Practical considerations when using pedometers to assess physical activity in population studies: lessons from the Burnie Take Heart Study. Res Q Exerc Sport. 2007;78(3):162–170. PubMed
26. Clemes SA, Biddle S. The use of pedometers for monitoring physical activity in children and adolescents: measurement considerations. J Phys Act Health. 2013;10(2):249–262. PubMed 27. Jago R, Watson K, Baranowski T, et al. Pedometer reliability, validity and daily activity targets among 10- to 15-year-old boys. J Sports Sci. 2006;24(3):241–251. PubMed doi:10.1080/02640410500141661 28. Fairclough SJ, Butcher ZH, Stratton G. Whole-day and segmented-day physical activity variability of northwest England school children. Prev Med. 2007;44(5):421–425. PubMed doi:10.1016/j.ypmed.2007.01.002 29. Van Cauwenberghe E, Jones RA, Hinkley T, Crawford D, Okely AD. Patterns of physical activity and sedentary behaviour in preschool children. Int J Behav Nutr Phys Act. 2012;9:138. PubMed doi:10.1186/1479-5868-9-138 30. Ottevaere C, Huybrechts I, De Meester F, De Bourdeaudhuij I, CuencaGarcia M, De Henauw S. The use of accelerometry in adolescents and its implementation with non-wear time activity diaries in free-living conditions. J Sports Sci. 2011;29(1):103–113. PubMed doi:10.1080/ 02640414.2010.521169 31. Sirard JR, Slater ME. Compliance with wearing physical activity accelerometers in high school students. J Phys Act Health. 2009;6(Suppl 1):S148–S155. PubMed 32. Tudor-Locke C, Williams JE, Reis JP, Pluto D. Utility of pedometers for assessing physical activity: convergent validity. Sports Med. 2002;32(12):795–808. PubMed doi:10.2165/00007256-200232120-00004
Official Journal of ISPAH www.JPAH-Journal.com ORIGINAL RESEARCH
Estimating Physical Activity in Children: Impact of Pedometer Wear Time and Metric Kelly R. Laurson, Gregory J. Welk, and Joey C. Eisenmann Background: The purpose of this study was to provide a practical demonstration of the impact of monitoring frame and metric when assessing pedometer-determined physical activity (PA) in youth. Methods: Children (N = 1111) were asked to wear pedometers over a 7-day period during which time worn and steps were recorded each day. Varying data-exclusion criteria were used to demonstrate changes in estimates of PA. Steps were expressed using several metrics and criteria, and construct validity was demonstrated via correlations with adiposity. Results: Meaningful fluctuations in average steps per day and percentage meeting PA recommendations were apparent when different criteria were used. Children who wore the pedometer longer appeared more active, with each minute the pedometer was worn each day accounting for an approximate increase of 11 and 8 steps for boys and girls, respectively (P < .05). Using more restrictive exclusion criteria led to stronger correlations between indices of steps per day, steps per minute, steps per leg length, steps per minute per leg length, and obesity. Conclusion: Wear time has a meaningful impact on estimates of PA. This should be considered when determining exclusion criteria and making comparisons between studies. Results also suggest that incorporating wear time per day and leg length into the metric may increase validity of PA estimates. Keywords: physical activity monitor, compliance, youth, exclusion criteria, inclusion criteria Obtaining accurate and reliable measures of habitual physical activity (PA) is paramount in research and practical applications (eg, recreation programs, physical education, clinical settings).1–3 Because of the limitations of self-report questionnaires and surveys in children (eg, cognitive ability, recall), many researchers turn to electronic motion sensors to objectively assess PA.4 Although accelerometry is a widely used method for objectively measuring PA, it may be a cost-prohibitive option for some research groups and practitioners. Pedometers are an often-used less expensive alternative to accelerometers for assessing PA. These step-counting tools offer some major advantages over accelerometers because of their simplicity, feasibility, and interpretability. However, many questions remain about how to most effectively use these devices to assess PA in free-living populations. Results have been mixed when determining the total number of monitoring days needed and which monitoring days to use to characterize habitual pedometer-determined PA in children. A review by Tudor-Locke et al5 outlined the reliability of pedometer-determined PA estimates using various combinations of weekdays and weekend days. The authors reported intraclass reliability coefficients ranging between .65 and .96 with monitoring periods from 2 to 8 days. No definitive acceptable number of monitoring days has been identified, but because of the variety of approaches and results found in the literature, this topic necessitates careful consideration.6 The number of hours that constitutes a complete monitoring day has yet to be standardized for pedometer usage, although it may
Laurson ([email protected]) is with the School of Kinesiology and Recreation, Illinois State University, Normal, IL. Welk is with the Dept of Health and Human Performance, Iowa State University, Ames, IA. Eisenmann is with the Dept of Radiology, Michigan State University, East Lansing, MI. 124
be more important than quantity of monitoring days. Participants are generally advised to wear the device at all times, except for activities such as sleeping, bathing, swimming, contact sports, and so forth. However, acceptable wear time criteria differ across the literature. For example, some studies have excluded participants for not wearing the pedometer ≥ 1 hour of the day,7,8 others have defined a day of wear as 8 hours9; our group included participants in data analyses if the pedometer was worn for at least 10 h/d.10 These issues are not unique to pedometers. Data reduction methodology for accelerometers, with regard to compliance and standardizing day length, have received considerable attention.11–13 Further investigation and standardization of data treatment methods are needed to improve the quality of information available with pedometers. Pedometer metric is also a central issue of data treatment after collection. For example, previous research has suggested that leg length is an important indicator of energy expenditure when using pedometry in children.14,15 Children with shorter legs (and a smaller stride length) require a greater number of steps to travel an equivalent distance compared with their taller peers. This could have implications for estimates of energy expenditure and may be at least partially controlled by incorporating leg length into the pedometer metric. Besides leg length, incorporating the amount of time the pedometer was worn into the metric (such as pedometer steps per minute) may also be of benefit by estimating PA intensity.5 The extent to which variations in wear time and data treatment may influence pedometer data is unknown. Because most pedometers simply record the raw volume of steps taken during a day, even small differences in monitoring time and/or subject compliance may have major implications for PA estimates. The purpose of this research is to demonstrate the impact of monitoring period and data cleaning practices on pedometer-determined PA estimates in children by comparing available sample sizes, estimates of activity, and the relationships between metrics and common health outcomes, such as adiposity.
Pedometer Exclusion Criteria 125
Methods
Downloaded by Heriot Watt Univ on 09/25/16, Volume 12, Article Number 1
Subjects The subjects were from 2 Midwestern communities (Lakeville, MN, and Cedar Rapids, IA) participating in a community-, school-, and family-based childhood obesity intervention. The details of the complete intervention can be found elsewhere.16 The data herein were collected before randomization into control or treatment groups for the intervention. Therefore, the current study can be considered a cross-sectional, observational design. The subjects were in grades 3 to 5 at 10 public elementary schools. We explained procedures and risks of the research to all subjects and primary caregivers, verbally and in writing, before participation. Written assent from each subject and consent from his or her primary caregiver was obtained. The study protocol was approved by the University of Minnesota Human Subjects Review Board and is in accordance with the Declaration of Helsinki. A total of 2114 elementary school children were invited to participate in the study, 1360 (64%) students accepted, and 1111 (53%) participated in the pedometer monitoring during the study period. A variety of pedometer exclusion criteria were compared, and some children had missing anthropometry data, so sample sizes vary by analysis (see below).
Anthropometry Standing height, sitting height, waist circumference, and body mass were measured by a school nurse according to standard procedures.17 Standing height and sitting height were measured using a portable stadiometer (Road Rod; Seca, Hanover, MD). Body mass was measured using a strain gauge scale (Lifesource MD; A&D, Milpitas, CA). Waist circumference (WC) was measured to the nearest 0.1 cm above the superior border of the iliac crest using a Gulick tape measure (CountryTechnology, Gays Mills, WI). All nurses participated in anthropometry measurement training sessions before data collection. Measurement error was determined during data collection by duplicate measures of every 25th subject. Overall, measurement error was small (SEM = 0.3 cm standing height, 0.1 kg body mass, 0.2 cm WC). Sitting height was subtracted from standing height to calculate leg length. Body mass index (BMI, kg/m2) was calculated from measurements of standing height and body mass. Overweight and obesity were determined on the basis of age- and sex-specific reference values developed by the International Obesity Task Force.18
Physical Activity Free-living PA was assessed with the widely accepted, researchquality Digiwalker SW-200 pedometer (YAMAX Health & Sports, San Antonio, TX).19 The subjects were given instructions on wearing the pedometer during the school day, and pedometers were checked using the shake test and 10-step test. The shake test was conducted by shaking the pedometer 10 times and verifying the recorded step count. The 10-step test was conducted by having the subject walk 10 steps and then checking the step count recorded. Pedometers that did not function properly during these validity checks were replaced. However, we did not record how many pedometers were replaced. Students were instructed to put on the pedometer (at the waist approximating the axillary line) upon waking and to take it off when going to bed. Students used the pedometer for 7 days. They were given a log sheet to record the times at which the pedometer was put on and taken off at any point in the day. Participants recorded the time on/off and number of steps accumulated each day during
the 7-day period. The classification of PA was based upon a normreferenced criterion of 13,000 steps/d for boys and 11,000 steps/d for girls, suggested by Vincent and Pangrazi.20 This pedometer-specific PA cutpoint has been noted to be associated with childhood obesity21 and is currently endorsed by the American Academy of Pediatrics within a policy statement on PA and childhood obesity.22
Statistical Analysis Descriptive statistics were calculated for the total sample and separately by sex. To demonstrate potential fluctuations in sample size and PA estimates, we calculated group means for pedometer steps per day and percentages of those meeting the Vincent and Pangrazi cutpoint using a variety of exclusion criteria. Exclusion groups were created on the basis of the number of days the pedometer was worn (≥ 1, ≥ 2, ≥ 4, and ≥ 6) and the number of hours the pedometer was worn on each day (≥ 1, ≥ 5, ≥ 8, ≥ 10, and ≥ 12). Days worn and times worn per day were selected to include a variety of criteria potentially used in the literature. Including subjects that wore the pedometer for 1 day for any amount of time would be considered equivalent to no exclusion criteria. However, including subjects that were required to wear the pedometer for at least 6 days for at least 12 hours each of those days would generally be considered restrictive. These criteria were used to cross-tabulate subjects into 20 groups that were not mutually exclusive. For example, the number of steps per day of a subject who wore the pedometer for 1 hour on 2 separate days of the week (totaling 2 hours of monitoring time) would be included in the “≥ 1 day and ≥ 1 hour each day” group and also in the “≥ 2 days and ≥ 1 hour each day” group. In contrast, the weekly stepsper-day estimate for a child who wore the pedometer for ≥ 6 days and ≥ 12 hours each day would be included in all group estimates. Three-way analysis of variance was used to determine whether sex, weight status (normal weight or overweight + obese), or age group (< 10 years or ≥ 10 years) accounted for potential differences in days worn and time worn per day (in minutes). Linear regression was used to estimate how average time worn per day influenced mean step count. Univariate regression models were created using only average minutes worn per day as the predictor. Then, additional models were created controlling for age, BMI, and leg length. Regressions were performed separately by sex. Finally, a series of partial correlations were used to investigate the influence of wear time and leg length when included in the pedometer metric. In theory, if requiring the subjects to wear the pedometer for a longer period of time to be included in the data analysis were to increase the validity of the PA measure, one would expect stronger associations between PA and PA correlates when using more restrictive criteria. To demonstrate this construct validity of PA, we correlated steps per day and steps per minute of time worn with BMI and WC while controlling for age using 2 different sets of exclusion criteria (≥ 1 day and ≥ 1 hour each day versus ≥ 6 days and ≥ 12 hours each day). Given the potential influence of leg length on pedometry, we also used metrics that incorporated leg length (such as steps per leg length) into the correlational analysis to determine whether the strength of the association between PA and these anthropometric indices of obesity would be further enhanced.
Results Of the children that agreed to participate in the study (n = 1360), approximately 82% wore the pedometer for any amount of time (n = 1111). Descriptive statistics for subjects that wore the pedometer are located in Table 1. There were no significant differences between
126 Laurson, Welk, and Eisenmann
Table 1 Descriptive Statistics Variable Age (y) Height (cm) Weight (kg) Leg length (cm) Body mass index (kg/m2) Overweight (%) Obese (%) Waist circumference (cm) Days pedometer worn Minutes pedometer worn (per day)
Boys (n = 510)
Girls (n = 601)
Combined (N = 1111)
9.6 (0.9) 138.6 (7.6) 35.6 (8.9) 65.7 (4.6) 18.4 (3.2) 19.0 6.3 63.6 (9.4) 6 (1) 737 (92)
9.6 (0.9) 137.8 (8.1) 35.3 (9.5) 65.5 (5.1) 18.4 (3.5) 20.0 8.5 63.1 (9.7) 6 (1) 740 (79)
9.6 (0.9) 138.1 (7.9) 35.4 (9.2) 65.5 (4.8) 18.4 (3.4) 19.5 7.5 63.4 (9.6) 6 (1) 738 (85)
Downloaded by Heriot Watt Univ on 09/25/16, Volume 12, Article Number 1
Note. Values are mean (SD) or percentage.
normal weight and overweight children in the number of days the pedometer was worn (P = .867). Girls wore the pedometer 0.15 days longer than boys, although this difference was not statistically significant (P = .069). Children ≥ 10 years old wore the pedometer 0.20 days longer than younger children (P = .018). Likewise, there were no differences between normal weight and overweight (P = .572) children or between boys and girls (P = .404) for time worn per day. However, children ≥ 10 years old wore the pedometer 19.3 minutes longer each day than younger children (P = .018). None of the interactions in the model were statistically significant (all P > .05). Over the course of the monitoring week, the pedometer was worn on average for 6 days and 738 min/d (12.3 h/d). The impact of using different exclusion criteria (various criteria for days worn and time worn per day) on PA outcomes is shown in Table 2 (boys) and Table 3 (girls). The use of more restrictive criteria, either by requiring more days of monitoring or more time worn per day, reduced the effective sample size and increased estimates of mean steps/day and the percentage of subjects that would be considered physically active. In the univariate regression analyses, each minute the pedometer was worn per day accounted for an increase of 10.6 steps/day in boys (n = 490) and 7.8 steps/day in girls (n = 578). Overall, 7.8% and 5.6% of the total variance in steps was accounted for by time worn per day in boys and girls, respectively. In the multiple regression models, time worn per day was a stronger predictor of steps per day than were age, leg length, or BMI in boys. For girls, only BMI provided more predictive utility than minutes worn per day. In the partial correlations (controlling for age), using more restrictive criteria and expressing PA as steps per minute or steps per leg length increased the strength of the association between steps and BMI or WC (Table 4). In general these correlations were stronger for girls, and when using WC, ranged from –0.242 to –0.406.
Discussion Overall, this study demonstrates that pedometer data cleaning and exclusion criteria have a meaningful impact on the resulting interpretation of the data. Specifically, the number of days worn and the time worn per day influence available sample size, estimates of steps per day, and categorization of children into active or inactive groups. These results are unique in that they are presented using
multiple exclusion criteria within the same sample, providing direct comparisons of various pedometer cleaning methods. Focusing on days worn restrictions, the largest subject loss occurred when we required subjects to have a minimum of 6 days of wear, rather than 4 days, from the 1-week monitoring period. However, meaningful differences in available sample were still evident when comparing a 2-day to a 4-day wear requirement. For example, the average mean sample sizes for girls were 586, 578, 527, and 384 for ≥ 1, ≥ 2, ≥ 4, and ≥ 6 days, respectively (Table 3). Increasing the days worn requirement decreased available sample size. This decrease appears to be nonlinear; there were larger decreases in sample size as additional days were required to meet the exclusion criteria. It should also be noted that the decreases in sample size caused by the increases in the days worn requirement appear to be modified by the time worn per day requirement, with greater losses in sample size across days worn when the wear time requirement was ≥ 10 h/d. Although many children wore the pedometer for ≥ 10 to 12 hours a few days during the monitoring week, less than 70% of the sample was able to provide ≥ 4 days of wear and ≥ 12 h/d of wear. The magnitude of subject loss because of unacceptable data, although not always reported, varies greatly in previous research. For example, 93% of 11,669 Canadian youth wore a pedometer for the entire 7-day protocol, but wear time each day was not considered.23 Another study using a 4-day protocol reported that 39% of 602 subjects failed to provide a full 4 days of monitoring when the sample was restricted to those who did not remove the pedometer for more than 1 h/d.24 Besides subject loss, time worn per day seems to have a larger impact on estimates of mean steps per day than does days worn. In Tables 2 and 3, the largest fluctuations in mean steps per day and percentage of subjects meeting step recommendations occurred across time-worn-per-day groups, not days worn. The longer per day the subjects were required to wear the pedometer, the larger the increase in estimates of steps per day. For example, boys’ estimates of steps per day were approximately 500 steps higher when the criterion for time worn was restricted to ≥ 12 h/d instead of ≥ 10 hours. This resulted in an increase of the estimate of boys achieving the steps-per-day recommendations by 6% to 10% of the available sample. A similar influence of time worn per day can be seen for girls. Using National Health and Nutrition Examination Survey accelerometer data, Herrmann et al11 compared the PA estimates from a 14-hour minimum-wear criterion to 13-, 12-, 11-, and 10-hour criteria. The mean absolute percentage error
127
≥ 1 Hours Sample size Mean steps per day (95% CI) Boys with > 13,000 steps per day when using selected exclusion criteria (95% CI) ≥ 5 Hours Sample size Mean steps per day (95% CI) Boys with > 13,000 steps per day when using selected exclusion criteria (95% CI) ≥ 8 Hours Sample size Mean steps per day (95% CI) Boys with > 13,000 steps per day when using selected exclusion criteria (95% CI) ≥ 10 Hours Sample size Mean steps per day (95% CI) Boys with > 13,000 steps per day when using selected exclusion criteria (95% CI) ≥ 12 Hours Sample size Mean steps per day (95% CI) Boys with > 13,000 steps per day when using selected exclusion criteria (95% CI)
Time worn per day
38.5% (34.3–43.1) n = 496 12,431 (12,131–12,732) 39.5% (35.2–43.8) n = 493 12,632 (12,331–12,933) 41.6% (37.3–46.0) n = 484 12,912 (12,606–13,218) 45.7% (41.2–50.1) n = 449 13,385 (13,056–13,714) 52.3% (47.7–57.0)
38.8% (34.6–43.1) n = 503 12,434 (12,134–12,735) 39.8% (35.5–44.0) n = 501 12,639 (12,340–12,938) 41.9% (37.7–46.3) n = 497 12,885 (12,582–13,188) 45.5% (41.1–49.9) n = 476 13,317 (12,991–13,643) 51.9% (47.4–56.4)
≥ 4 Days
53.7% (48.4–59.0)
n = 337 13,564 (13,187–13,941)
48.1% (43.4–52.9)
n = 428 13,119 (12,799–13,439)
43.5% (39.0–48.1)
n = 460 12,809 (12,502–13,117)
40.9% (36.5–45.4)
n = 474 12,555 (12,250–12,860)
39.2% (34.9–43.6)
n = 487 12,456 (12,147–12,766)
Days Worn
n = 506 12,367 (12,060–12,673)
≥ 2 Days
n = 510 12,375 (12,068–12,682)
≥ 1 Day
53.5% (45.3–61.7)
n = 142 13,581 (13,001–14,160)
43.3% (37.5–49.2)
n = 277 12,958 (12,560–13,357)
42.1% (37.0–47.4)
n = 349 12,816 (12,446–13,165)
41.3% (36.4–46.3)
n = 380 12,679 (12,344–13,015)
40.4% (35.7–45.2)
n = 403 12,622 (12,287–12,957)
≥ 6 Days
Table 2 Sample Size, Mean Steps per Day, and Percentage of Boys Meeting Steps-per-Day Recommendations by Exclusion Criteria Combinations of Days Worn and Time Worn per Day
Downloaded by Heriot Watt Univ on 09/25/16, Volume 12, Article Number 1
128
≥ 1 Hours Sample size Mean steps per day (95% CI) Girls with > 11,000 steps per day when using selected exclusion criteria (95% CI) ≥ 5 Hours Sample size Mean steps per day (95% CI) Girls with > 11,000 steps per day when using selected exclusion criteria (95% CI) ≥ 8 Hours Sample size Mean steps per day (95% CI) Girls with > 11,000 steps per day when using selected exclusion criteria (95% CI) ≥ 10 Hours Sample size Mean steps per day (95% CI) Girls with > 11,000 steps per day when using selected exclusion criteria (95% CI) ≥ 12 Hours Sample size Mean steps per day (95% CI) Girls with > 11,000 steps per day when using selected exclusion criteria (95% CI)
Time worn per day n = 600 10,688 (10,473–10,903) 44.5% (40.5–48.5) n = 587 10,750 (10,538–10,962) 45.7% (41.6–49.7) n = 584 10,946 (10,729–11,164) 48.6% (44.6–52.7) n = 577 11,171 (10,944–11,399) 52.3% (48.3–56.4) n = 542 11,530 (11,289–11,771) 59.2% (55.1–63.4)
44.4% (40.5–48.4) n = 588 10,749 (10,538–10,960) 45.6% (41.6–49.6) n = 587 10,941 (10,725–11,158) 48.7% (44.7–52.8) n = 585 11,185 (10,955–11,415) 52.6% (48.6–56.7) n = 571 11,511 (11,269–11,752) 59.0% (55.0–63.1)
≥ 2 Days
n = 601 10,687 (10,473–10,902)
≥ 1 Day
≥ 4 Days
58.6% (53.9–63.3)
n = 418 11,532 (11,274–11,789)
52.3% (48.0–56.6)
n = 522 11,169 (10,943–11,396)
48.5% (44.3–52.6)
n = 551 10,954 (10,734–11,174)
46.1% (42.0–50.2)
n = 562 10,802 (10,585–11,018)
44.8% (40.8–48.9)
n = 582 10,716 (10,497–10,936)
Days worn
57.1% (50.0–64.3)
n = 182 11,611 (11,227–11,994)
51.6% (46.3–56.8)
n = 347 11,123 (10,863–11,383)
49.6% (44.9–54.4)
n = 425 11,051 (10,819–11,283)
45.8% (41.3–50.3)
n = 467 10,842 (10,619–11,065)
44.3% (39.9–48.6)
n = 499 10,676 (10,447–10,905)
≥ 6 Days
Table 3 Sample Size, Means Steps per Day, and Percentage of Girls Meeting Steps-per-Day Recommendations by Exclusion Criteria Combinations of Days Worn and Time Worn per Day
Downloaded by Heriot Watt Univ on 09/25/16, Volume 12, Article Number 1
Pedometer Exclusion Criteria 129
Table 4 Partial Correlations Between Pedometer Metrics and Obesity Indices by Data Exclusion Criteria Body mass index Exclusion criteria
Waist circumference
Pedometer metric
Boys
Girls
Boys
Girls
Steps per day Steps per minute worn per day Steps per leg length Steps per minute per leg length
–0.242 –0.248 –0.264 –0.268
–0.278 –0.284 –0.318 –0.341
–0.286 –0.270 –0.338 –0.328
–0.298 –0.306 –0.366 –0.392
Steps per day Steps per minute worn per day Steps per leg length Steps per minute per leg length
–0.261 –0.272 –0.296 –0.302
–0.283 –0.306 –0.325 –0.356
–0.294 –0.302 –0.357 –0.362
–0.304 –0.328 –0.374 –0.406
≥ 1 d and ≥ 1 h/da
Downloaded by Heriot Watt Univ on 09/25/16, Volume 12, Article Number 1
≥ 6 d and ≥ 10 h/db
Note. All correlations adjusted for age and are statistically significant (P < .05). a Available sample: boys = 489, girls = 591. b Available sample: boys = 267, girls = 341.
for steps per day in adults increased as wear time per day decreased (4%, 10%, 17%, and 25% versus the 14-hour day, respectively). Relaxing the exclusion criteria to include more subjects lowered estimates of PA, similar to what was found in the current study. The analysis of variance provided little insight into the reasons for compliance (or lack thereof) for hours worn per day or days worn per week. No differences were seen in wear time of the pedometer by sex or weight status. Older children did wear the pedometer slightly longer than younger children. Although this difference was statistically significant, it approximated a 20 minute difference in time worn per day and 0.2 days of wear over the week. Likewise, Schmidt et al25 found that mean pedometer wear time was not associated with age, sex, BMI, or education level in adults. This supports our findings that wear time is not a product of selection, because differences are unrelated to important sample characteristics. Schmidt et al25 also found that time worn (minutes per day) was correlated with the number of steps recorded (r = .20). Likewise, our regression analysis revealed that total time worn per day accounted for 6% to 8% of the variance in steps per day; each minute of daily wear time equated to approximately 9 recorded steps. In essence, the regression indicates that if the pedometer is removed for only 30 to 60 minutes during the day, this can potentially result in an underestimation error of 250 to 500 steps. These findings in children, along with those already described by Herrmann et al11 and Schmidt et al25 in adults, suggest that studies using stringent wear criteria (eg, pedometer removed for no more than 1 h/d) are still vulnerable to error from variation in wear time. Overall, with regard to sample size and the resulting estimates of activity, using the most restrictive criteria (≥ 6 d and ≥ 12 h/d) reduced the usable sample from 1111 to 324 subjects (29.2% compliance) but increased estimates of mean steps per day and percentage of subjects meeting steps-per-day recommendations. Although this is the most extreme example from the data, restricting usable data to ≥ 4 days and ≥ 10 h/d still resulted in a loss of approximately 15% of the total participating sample (usable n = 950 of 1111). In a review of pedometry methods in youth, Clemes and Biddle26 noted that compliance rates in previous research ranged from 46% to 99%. In addition, some of the studies reviewed also did not report the amount of wearing time that constituted a monitoring
day. Combined with the evidence provided in the current study, it is reasonable to assume that the estimates of PA provided throughout the literature are affected by time-worn-per-day criteria to varying degrees. More specifically, requiring subjects to wear the pedometer for larger portions of the day provides higher, more favorable estimates of PA. The dilemma resulting from this phenomenon is determining whether children with higher step counts are truly more physically active, merely wore the pedometer for longer than others in their cohort, or both. To demonstrate how exclusion criteria can impact construct validity, we computed partial correlations between steps per day and adiposity while controlling for age (Table 4). Using more restrictive exclusion criteria resulted in slightly stronger correlations between step metrics and BMI or WC. Correlations were further enhanced when expressing PA as steps per minute of wear time (steps per minute), regardless of the exclusion criteria used. This may indicate that statistically adjusting the data or metric on the basis of time worn per day may be an avenue to correct for differences in wear time per day between subjects (rather than subject exclusion). Differences in the correlations between steps per day and steps per minute were small but consistent. Using leg length in the metric has been suggested previously, because taller individuals require fewer steps to travel an equivalent distance as a shorter individual.14,27 Adding leg length to the metric (eg, steps per leg length) increased the strength of the correlations to a greater degree than using steps per minute alone, although the strongest correlations were generally found when using both minutes worn per day and leg length (steps per minute per leg length). Further consideration and refinement of these approaches to pedometer metric are recommended. The findings from this study and similar studies of pedometer compliance5,25,26 underscore the need for continued investigation on the topic of pedometer methodology. We recommend reporting pedometer wear duration per day as a proportion of the time during which the child was not asleep (eg, percentage of the waking day). Because sleep requirements decrease with age, older children may have a longer opportunity of time to wear the pedometer. Future investigations should also examine PA monitor compliance during different segments of the requested wear period. PA fluctuates throughout the day,28,29 so the effect of removing activity monitors
Downloaded by Heriot Watt Univ on 09/25/16, Volume 12, Article Number 1
130 Laurson, Welk, and Eisenmann
probably differs depending on when they are removed. Providing subjects with logs to record wake/sleep time and pedometer on/ off time could potentially reduce error,30 although this may also provide an avenue for self-report bias to influence estimates of PA. If children did not wear the monitor all day, statistical adjustments may provide more accurate estimates and suitable comparisons. Compliance might be enhanced by using incentives, detailed instructions, and more frequent contact with subjects to remind them of the importance of wearing the monitor. Sirard and Slater31 found that accelerometer wear compliance in adolescents was improved by using such strategies. The current study is not without limitations. Not all children in the school districts chose to participate in the original intervention and did not wear the pedometer, which could lead to bias. However, comparing the demographics of the sample to that of the school districts from which they were enrolled revealed no significant differences, indicating that the potential for bias is low. The pedometer model used in the study did not have a memory function, requiring that children and parents record daily step counts using logs, which could have led to lost data or inaccurate data through self-report bias or mistakes in data recording. However, logged steps have been found to be well correlated with alternative objective measures of physical activity.26,32 Regardless, neither of these limitations would be expected to detract from the analyses herein because the focus of the study was on the variance in stepping estimates created by the treatment of the data, rather than on the stepping data itself. Strengths include the use of a large, multisite sample of children and multiple combinations of wear criteria to demonstrate the variance in data interpretation and construct validity.
Conclusions In summary, the results herein demonstrate how criteria for pedometer days worn and time worn per day influence the available sample and subsequent outcomes (eg, relationship between pedometer PA and adiposity). The resulting differences in mean steps per day and percentage of subjects that are considered physically active is potentially important, as is the loss in sample size by enforcing more restrictive criteria. Care should be taken to ensure that monitoring is completed over enough days to provide a reliable estimate and also that the activity monitor is worn for enough time on those days to provide a valid estimate of PA. However, this research demonstrates how difficult it is to achieve such a balance. At the very least, future studies should provide more detailed information about time worn per day, days worn, and subject exclusion methodology because of the influence these factors have on the resulting data. Acknowledgments In Lakeville, Minnesota SWITCH is sponsored by Medica Foundation, Healthy and Active America Foundation, and Fairview Health Services. In Cedar Rapids, Iowa SWITCH is sponsored by Cargill and the Healthy and Active America Foundation. SWITCH is a trademark of the National Institute of Media and the Family MediaWise campaign.
References 1. Trost SG. Objective measurement of physical activity in youth: current issues, future directions. Exerc Sport Sci Rev. 2001;29(1):32–36. PubMed doi:10.1097/00003677-200101000-00007
2. Welk GJ, Corbin CB, Dale D. Measurement issues in the assessment of physical activity in children. Res Q Exerc Sport. 2000;71(2 Suppl):S59–S73. PubMed 3. Corder K, Ekelund U, Steele RM, Wareham NJ, Brage S. Assessment of physical activity in youth. J Appl Physiol. 2008;105(3):977–987. PubMed doi:10.1152/japplphysiol.00094.2008 4. Westerterp KR. Assessment of physical activity: a critical appraisal. Eur J Appl Physiol. 2009;105(6):823–828. PubMed doi:10.1007/ s00421-009-1000-2 5. Tudor-Locke C, McClain JJ, Hart TL, Sisson SB, Washington TL. Pedometry methods for assessing free-living youth. Res Q Exerc Sport. 2009;80(2):175–184. PubMed doi:10.1080/02701367.2009.1 0599551 6. Baranowski T, Mâsse LC, Ragan B, Welk G. How many days was that? We’re still not sure, but we’re asking the question better! Med Sci Sports Exerc. 2008;40(7 Suppl):S544–S549. PubMed doi:10.1249/ MSS.0b013e31817c6651 7. Duncan JS, Schofield G, Duncan EK. Pedometer-determined physical activity and body composition in New Zealand children. Med Sci Sports Exerc. 2006;38(8):1402–1409. PubMed doi:10.1249/01. mss.0000227535.36046.97 8. Duncan MJ, Al-Nakeeb Y, Woodfield L, Lyons M. Pedometer determined physical activity levels in primary school children from central England. Prev Med. 2007;44(5):416–420. PubMed doi:10.1016/j. ypmed.2006.11.019 9. Sigmund E, Sigmundová D, El Ansari W. Changes in physical activity in pre-schoolers and first-grade children: longitudinal study in the Czech Republic. Child Care Health Dev. 2009;35(3):376–382. PubMed doi:10.1111/j.1365-2214.2009.00945.x 10. Duncan JS, Schofield G, Duncan EK. Step count recommendations for children based on body fat. Prev Med. 2007;44(1):42–44. PubMed doi:10.1016/j.ypmed.2006.08.009 11. Herrmann SD, Barreira TV, Kang M, Ainsworth BE. Impact of accelerometer wear time on physical activity data: a NHANES semisimulation data approach. Br J Sports Med. 2014;48(3):278–282. PubMed doi:10.1136/bjsports-2012-091410 12. Ward DS, Evenson KR, Vaughn A, Rodgers AB, Troiano RP. Accelerometer use in physical activity: best practices and research recommendations. Med Sci Sports Exerc. 2005;37(11 Suppl):S582–S588. PubMed doi:10.1249/01.mss.0000185292.71933.91 13. Matthews CE, Hagströmer M, Pober DM, Bowles HR. Best practices for using physical activity monitors in population-based research. Med Sci Sports Exerc. 2012;44(1 Suppl 1):S68–S76. PubMed doi:10.1249/ MSS.0b013e3182399e5b 14. Eisenmann JC, Wickel EE. Moving on land: an explanation of pedometer counts in children. Eur J Appl Physiol. 2005;93(4):440–446. PubMed doi:10.1007/s00421-004-1227-x 15. Stone MR, Esliger DW, Tremblay MS. Comparative validity assessment of five activity monitors: does being a child matter? Pediatr Exerc Sci. 2007;19(3):291–309. PubMed 16. Eisenmann JC, Gentile DA, Welk GJ, et al. SWITCH: rationale, design, and implementation of a community, school, and family-based intervention to modify behaviors related to childhood obesity. BMC Public Health. 2008;8:223. PubMed doi:10.1186/1471-2458-8-223 17. Malina R. Anthropometry. In: Maud P, Foster C, eds. Physiological assessment of human fitness. Champaign, IL: Human Kinetics; 1995:205–219. 18. Cole TJ, Bellizzi MC, Flegal KM, Dietz WH. Establishing a standard definition for child overweight and obesity worldwide: international survey. BMJ. 2000;320(7244):1240–1243. PubMed doi:10.1136/ bmj.320.7244.1240
Downloaded by Heriot Watt Univ on 09/25/16, Volume 12, Article Number 1
Pedometer Exclusion Criteria 131 19. Schneider PL, Crouter SE, Bassett DR. Pedometer measures of free-living physical activity: comparison of 13 models. Med Sci Sports Exerc. 2004;36(2):331–335. PubMed doi:10.1249/01. MSS.0000113486.60548.E9 20. Vincent SD, Pangrazi RP. An examination of the activity patterns of elementary school children. Pediatr Exerc Sci. 2002;14:432–441. 21. Laurson KR, Eisenmann JC, Welk GJ, Wickel EE, Gentile DA, Walsh DA. Evaluation of youth pedometer-determined physical activity guidelines using receiver operator characteristic curves. Prev Med. 2008;46(5):419–424. PubMed doi:10.1016/j.ypmed.2007.12.017 22. Council on Sports Medicine and Fitness, Council on School Health. Active healthy living: prevention of childhood obesity through increased physical activity. Pediatrics. 2006;117(5):1834–1842. PubMed doi:10.1542/peds.2006-0472 23. Craig CL, Tudor-Locke C, Cragg S, Cameron C. Process and treatment of pedometer data collection for youth: the Canadian Physical Activity Levels among Youth study. Med Sci Sports Exerc. 2010;42(3):430– 435. PubMed doi:10.1249/MSS.0b013e3181b67544 24. Wilde B, Corbin C, Le Masurier G. Free-living pedometer step counts of high school students. Pediatr Exerc Sci. 2004;16(1):44–53. 25. Schmidt MD, Blizzard CL, Venn AJ, Cochrane JA, Dwyer T. Practical considerations when using pedometers to assess physical activity in population studies: lessons from the Burnie Take Heart Study. Res Q Exerc Sport. 2007;78(3):162–170. PubMed
26. Clemes SA, Biddle S. The use of pedometers for monitoring physical activity in children and adolescents: measurement considerations. J Phys Act Health. 2013;10(2):249–262. PubMed 27. Jago R, Watson K, Baranowski T, et al. Pedometer reliability, validity and daily activity targets among 10- to 15-year-old boys. J Sports Sci. 2006;24(3):241–251. PubMed doi:10.1080/02640410500141661 28. Fairclough SJ, Butcher ZH, Stratton G. Whole-day and segmented-day physical activity variability of northwest England school children. Prev Med. 2007;44(5):421–425. PubMed doi:10.1016/j.ypmed.2007.01.002 29. Van Cauwenberghe E, Jones RA, Hinkley T, Crawford D, Okely AD. Patterns of physical activity and sedentary behaviour in preschool children. Int J Behav Nutr Phys Act. 2012;9:138. PubMed doi:10.1186/1479-5868-9-138 30. Ottevaere C, Huybrechts I, De Meester F, De Bourdeaudhuij I, CuencaGarcia M, De Henauw S. The use of accelerometry in adolescents and its implementation with non-wear time activity diaries in free-living conditions. J Sports Sci. 2011;29(1):103–113. PubMed doi:10.1080/ 02640414.2010.521169 31. Sirard JR, Slater ME. Compliance with wearing physical activity accelerometers in high school students. J Phys Act Health. 2009;6(Suppl 1):S148–S155. PubMed 32. Tudor-Locke C, Williams JE, Reis JP, Pluto D. Utility of pedometers for assessing physical activity: convergent validity. Sports Med. 2002;32(12):795–808. PubMed doi:10.2165/00007256-200232120-00004
