TUDOR-LOCKE, CATRINE; JOHNSON, WILLIAM D.; KATZMARZYK, PETER T.
Physical activity assessment has benefited greatly from the swift growth and development of body-worn technologies, including accelerometers and pedometers, which allow precise and objective quantification of day-to-day physical activity patterns and volumes. Accelerometers in particular provide an output (e.g., activity counts), which is typically interpreted using established cut points to classify time in escalating intensities of physical activity (i.e., from light to vigorous intensity) (7). In turn, these outputs can be used to estimate levels of adherence to intensity-based physical activity recommendations (16). For example, the US National Health and Nutrition Examination Survey (NHANES) uses the ActiGraph AM-7164 (formerly distributed as CSA/MTI AM-7164; manufactured by ActiGraph, Fort Walton Beach, FL) to objectively capture free-living physical activity behaviors. On the basis of these data, Troiano et al. (21) reported that less than 5% of monitored adults obtained 30 min·d−1 of moderate-to-vigorous physical activity. More recently, Matthews et al. (13) used these data to present time spent in sedentary behaviors.
The accelerometer model used in NHANES also has a step counter function that provides another objective measure of physical activity. The assessment and the interpretation of the number of steps people take during the course of a day are becoming more acceptable to researchers and practitioners alike (3,17). Specifically, distilling step data as steps per day offers a simple means of expressing habitual daily volume of physical activity.
Previous studies have documented the validity of the NHANES accelerometer model in terms of both intensity (7) and step outputs (10); however, it is also known to be more sensitive to low threshold accelerations (e.g., slow walking) compared with highly revered research-quality pedometers (11,22). For example, in one study of free-living adults (22), the ActiGraph (then distributed as the CSA) detected approximately 1800 more steps per day than the Yamax (Yamax Corporation, Tokyo, Japan) pedometer, suggesting that values of steps from one instrument cannot be easily compared with the other. Unfortunately, current dialogues revolving around "how many steps are enough?" have been largely reflective of pedometer-determined steps per day, using research-quality pedometers like the Yamax model (27). In a proof of concept study, the ActiGraph detected approximately 17-fold more erroneous steps than the Yamax pedometer when both were worn concurrently during a 20-mile car drive (11). The actual manufacturer-released sensitivity threshold (over which a step is detected) of the ActiGraph is 0.30g compared with 0.35g of the Yamax (22). In a manner, the higher-intensity threshold of the Yamax pedometer censors out lower-intensity accelerations (i.e., does not count these as steps), effectively capturing only the faster walking paces. An exact translation of this sensitivity threshold in terms of the ActiGraph accelerometer steps output is not known.
The unique ability of the ActiGraph to capture both steps and intensity data simultaneously minute by minute allows exploration of the "quality" of steps taken (i.e., the number of steps taken at each intensity level). Further, it is possible that post hoc censoring of low-intensity steps (e.g., by using current cut points reflective of such intensity levels) in the NHANES data will produce an output more congruent with pedometer-assessed expected values previously collected in American samples (26,29). The 2005-2006 data release represents the first NHANES opportunity to examine these step data; previous releases did not make this particular variable available. Therefore, the purpose of this analysis of the 2005-2006 NHANES physical activity monitoring data is to provide the population- and sex-specific epidemiology of accelerometer-determined steps per day in the United States with and without censoring steps detected at the lower end of the activity spectrum (i.e., inactivity).
NHANES was originally designed as a periodic survey of the nation's health profile. Since 1999, it has been rolled out on a continuous basis with related data releases in 2-yr increments. NHANES is collected year-round and practices standardized data collection methods to minimize the potential for error. Initially, households are identified for inclusion and a NHANES interviewer visits the home. Once the household interview is completed, participants are asked to attend a mobile examination center (MEC) to receive a health examination. Like preceding surveys, 2005-2006 contains data from individuals selected under a complex, multistage probability design to be a nationally representative sample of the civilian, noninstitutionalized US population.
The physical activity monitor (PAM) component was added to NHANES in 2003, and 2005-2006 is the first release of accelerometer-determined step data in addition to the more commonly collected and reported intensity and duration data based on accelerometer-determined activity counts. Participants ≥6 yr were recruited for this component in the MEC. Those who had walking impairments or other limitations that prevented wearing an accelerometer were excluded. The device was worn on the right hip using an elasticized fabric belt. The accelerometers were programmed to record information in 1-min intervals (epochs). Participants were instructed to wear the accelerometer during waking hours for 7 d, to remove them during any water activities (e.g., swimming, showering, and bathing), and to return them by prepaid mail in return for a $40 remuneration.
The 2005-2006 raw data file was released in June, 2008, and consists of multiple records of sequential minute-by-minute activity count and step data for each participant. NHANES data processing and editing included review for outliers and unreasonable values based on published literature and expert judgment. In addition to flagging data from devices that were out of calibration upon return, a variable was added to indicate whether data were deemed reliable or not. Examples of data flagged as not reliable were records containing >10 min with 1) zero steps and >250 activity counts per minute, 2) >200 steps per minute, and 3) 32,767 (maximum value possible) activity counts per minute (Captain Richard P. Troiano, personal communication). The National Center for Health Statistics ethics review board approved the original survey protocols, and informed consent was obtained for all NHANES participants. The Pennington Biomedical Research Center's institutional review board approved of this secondary analysis.
This analysis is limited to NHANES adult participants ≥20yr (individuals older than 85 yr were top coded as 85) with designated reliable accelerometer data. Time worn (hours and minutes) was computed using a SAS macro provided by the National Cancer Institute at http://riskfactor.cancer.gov/tools/nhanes_pam/. Keeping with previous analyses (13,21), a valid day was defined as ≥10 h of wear. Previous analyses also required at least four valid days of monitoring for determining an average value of steps per day (21). Before accepting this threshold condition, we analyzed the data to ascertain potential for bias. Figure 1 displays obvious bias: mean steps per day vary with valid days worn. To emphasize, those with less than four valid days of monitoring (n = 731) took significantly fewer (8873± 205 vs 9857 ± 111 steps per day; independent t-test = 5.73, P < 0.0001) steps per day than those who had four or more valid days of monitoring (n = 3013); that is, the most active people tended to wear the accelerometer for more days than the least active people. We determined that by deleting those participants with less than four valid days, the sample mean would be inflated by approximately 1000 steps per day. We therefore decided that it was more appropriate to base this descriptive analysis on the 3744 participants who had at least one valid day of monitoring. Other analyses have also included those with at least one valid day (13).
FIGURE 1-Mean uncens...Image Tools
Intensity of steps taken was anchored using activity count cut points previously used to analyze NHANES 2003-2004 data (21). Specifically, each monitored day produced 1440 minute of records for each individual, and the activity counts recorded in each minute were accordingly classified based on 2020 activity counts per minute (indicating moderate intensity equivalent to 3-5.99 METs) and 5999 activity counts per minute (indicating vigorous intensity equivalent to ≥6 METs). For these analyses, we further stratified the lower-intensity categories (i.e., <2020 activity counts per minute) into inactive (0-499 activity counts per minute) and light (500-2019 activity counts per minute) intensities in agreement with earlier analyses (22). Finally, we adopted the Matthews et al. (13) cut point of <100 activity counts per minute to define sedentary behaviors; which also required that we adjust the inactive intensity to 100-499 for this analysis.
Step data were recorded concurrently minute-by-minute. Steps per minute were summed over the 1440-min day to produce a cumulative daily record. We linked the step data with the activity count data to identify steps taken at each accelerometer-defined intensity. Daily steps were summed and divided by the number of days the accelerometer was worn to derive average steps per day. We used established pedometer-determined physical activity cut points for healthy adults (23) to organize steps per day-defined activity levels: 1) <5000 steps per day ("sedentary"); 2) 5000-7499 steps per day ("low active"); 3) 7500-9999 steps per day ("somewhat active"); 4) ≥10,000-12,499 steps per day ("active"); and 5) ≥12,500 steps per day ("highly active"). These categories were reinforced in 2008 (27). For this analysis, we further segmented the sedentary category into<2500 steps per day ("basal physical activity") and 2500-4999 steps per day ("limited physical activity"). This strategy also helped to keep the pedometer-defined sedentary level (which encompasses basal and limited physical activity levels) distinct from the accelerometer-defined sedentary intensity (i.e., <100 activity counts per minute).
All analyses were performed using procedures for sample survey data that are readily available in the SAS© System for Windows Version 9.1 (SAS Institute, Cary, NC, 2004) to account for the complex sampling design of NHANES. All analyses included sample weights to account for oversampling and nonresponse to provide nationally representative results. Sex-specific means, SE, and 95% confidence intervals (CI) for steps per day were computed using all steps (i.e., uncensored) and again after censoring out those steps taken at an intensity <500 activity counts per minute. We examined the proportion of uncensored and censored steps taken at each of the accelerometer-determined intensities by sex. Finally, we report the proportion of the population who were categorized at each of the step-defined activity levels described above, considering both uncensored and censored steps.
Sample characteristics (i.e., sex, age group, and race or ethnicity) of adult NHANES respondents who participated in the 2005-2006 PAM and who met the inclusion requirements for this analysis are presented in Table 1. The analyzed sample included 3744 (86%) of the eligible sample of 4372 adults. Three hundred and fifty-six were excluded based on NHANES reliability flags; 272 were excluded because they did not meet the threshold condition of at least 1 day with the minimum of 10 h of wear time. Table 2 shows the mean wear time and uncensored and censored steps per day. On average, the US sample took 9676 ± 107 uncensored steps per day or 6540 ± 106 censored steps per day. On average, males wore the device 30 min longer than females during the monitoring day and took 1696 more uncensored steps per day and 1675 more steps per day when lower-intensity steps were censored.
Figure 2 displays uncensored and censored steps per day for males and females, segmented to show the proportion of steps (relative to the sex-specific total) that are taken within each accelerometer-defined intensity category. The greatest proportion of daily steps are taken at light intensity for males (47% of uncensored and 66.9% of censored) and females (46.7% and 69.5%, respectively). As expected, extremely few steps are taken at sedentary intensity, and no steps are taken during nonwear time (which therefore does not appear in Fig. 2). Because both sedentary and inactive steps are not considered with censoring, these steps do not appear under this condition, increasing the relative percentages of all other intensities.
FIGURE 2-Uncensored ...Image Tools
Figures 3-5 present the proportion of the population, and by sex, who were categorized at each of the step-defined activity levels described above, considering both uncensored and censored steps. As anticipated, censoring shifts all of the curves to the left. The modal frequency in the total sample for uncensored steps is "somewhat active," whereas it is "limited activity" for the censored steps. Censoring also attenuates the predominant peak of highly active males that the uncensored data suggest.
The results indicate that US adults in this nationally representative sample took approximately 10,000 uncensored accelerometer-determined steps per day. This is considerably higher than at least two US samples: Colorado (≅6800 steps per day) (29) and South Carolina (≅5900 steps per day) (26). Further, it approximates a level of steps per day that is used to identify active individuals (23,27) and therefore suggests that physical activity intervention is unwarranted in this country. In reality, however, evidence continues to accumulate that the American obesity epidemic persists and is actually growing in magnitude unthwarted (4,15). Further, other NHANES accelerometer estimates (based solely on time in intensity, not steps taken) conclude that <5% of adults adhere to public health recommendations for physical activity (21). These incompatible results from the same instrument administered using highly controlled and systematic measurement protocols demand explanation.
The measurement of "step" is inexact and unregulated by any authoritative body with a single exception: pedometer quality is regulated by Japanese industry standards to a maximum permissible miscounting error of 3% (9). It has been previously noted that Japanese-manufactured pedometers are among the most accurate instruments available to measure steps per day (5). Outside of Japan and lacking industry standards to guide instrument performance, manufacturers are independent to operationally define a step for their own purposes. Unfortunately, enduring continued lowering of sensitivity thresholds in a dogged pursuit of "every step" results in a predictable sensitivity/specificity trade-off that blurs the more important focus on health-related physical activity. Further, manufacturers' attempts at standardization are inevitably constrained by patents protecting their unique designs. There are also concerns that models within the same brand are discrepant in measurement mechanism; new evidence derived from mechanical oscillations suggests that a newer generation of the ActiGraph (GT1M model) than that used in NHANES is actually less sensitive to lower force activities (18). This implies that fewer low-intensity steps would be detected and that population estimates of steps per day would also be lower if this newer model were to be adopted. A head-to-head comparison of the different generations of the ActiGraph under free-living conditions has not yet been conducted, so again, no conversion factor is available at this time. A solution to these measurement issues is not straightforward and also not within the scope of this manuscript. However, because current steps per day recommendations (23,27) and cut points(6,24,28) are based on the output of similarly performing research-quality pedometers, it behooves us to attempt to interpret the NHANES accelerometer-determined steps per day output accordingly.
Although instrument sensitivity thresholds are known to differ between the ActiGraph 7164 and the Yamax pedometer, for example, there is not an exact method of converting the output of one to the other. Although we previously demonstrated that the ActiGraph (then distributed as the CSA) detected approximately 1800 more steps per day than the Yamax pedometer in a small free-living sample (22), we felt it was too simplistic to apply this single strategy uniformly to all NHANES PAM participants in abroad-handed attempt to make their accelerometer-determined steps per day appear more congruent with that of research-quality pedometers. A more acceptable strategy would consider each individual's unique activity patterns and would attempt to censor out excessive low-force accelerations that are more likely to be picked up by this accelerometer (11,22). Obviously, our decision to censor those steps taken <500 activity counts per minute was arbitrary. However, faced with no exact conversion factor, we believe that it was a logical choice. Extrapolating from work done by Barnett and Cerin (2), 500 activity counts per minute is approximately equal to walking at 2.7 km·h−1 or 1.7 miles·h−1. Interpreting a figure provided by Matthews (12) gives a good indication of the types of ambulatory activities that would fall below 500 activity counts per minute, for example, steps undertaken while cooking, ironing, washing dishes, grocery shopping, laundry, and light cleaning. We did consider censoring those steps taken at <100 activity counts per minute, following the study of sedentary time by Matthews et al. (13). However, using this cut point reduced total estimates by only 557 steps per day (because very few steps are detected when sitting is implied). We also considered censoring out all those steps taken at less than moderate intensity (i.e., at <2020 activity counts per minute); however, this drastically reduced steps per day estimates to 1828 ± 64, which was not plausible either. Until a more valid conversion factor is ascertained for translating accelerometer and pedometer-determined steps, our use of the <500 activity count per minute threshold holds considerable merit.
It is important to disclose here that the previous accelerometer analysis focused on time in moderate to vigorous intensity (21), indicating that <5% of US adults were achieving recommended amounts of physical activity. As indicated above, even our conservative use of censored steps leaves us with an estimate of 31% taking ≥7500 steps per day ("somewhat active" to "highly active"). It is apparent that the decision rules that count only minimal 10-min bouts above set activity counts per minute are much more restrictive than the allowance for accumulation of a volume of steps taken over the course of a day. Perhaps the literal translation of a recommendation primarily based on self-report (21) is too narrow to reflect what may truly be healthful physical activity. This is, of course, pure speculation in an attempt to reconcile vastly discrepant conclusions. Continued research using multiple instruments and/or cross-tabulated examination of different outputs (e.g., step and activity count) within the same instrument will likely provide additional insight into these measurement conundrums.
Several limitations must be acknowledged. First, like other waist-mounted motion sensors, the ActiGraph is most sensitive to ambulatory activity (i.e., walking), missing upper body movements, load carrying, and water activities (the latter of which requires removal of the instrument). However, walking is the most prevalent form of leisure time physical activity in the United States (20) and is a functional part of daily life (25). Further, those activities most likely to be missed (e.g., swimming, aqua-fitness, bicycling) or underestimated (e.g., weight-training, yoga) by waist-mounted motion sensors are infrequently performed (8). Attempts to adjust for nonambulatory steps taken for these types of activity are unwarranted in terms of population estimates (14).
Second, there are many different activity count cut-point values that have been used to ascertain moderate and/or vigorous physical activity using this particular accelerometer. For example, the choice of cut point will affect classification of moderate-intensity steps; a lower cut point (e.g., 1952 (7) vs 2020 used herein) will shift more light-intensity steps into the moderate-intensity category. We decided it was most appropriate to use the same cut points that had been previously used with the NHANES data (21).
Third, the question of "how many days" should a motion sensor be worn to achieve a valid and reliable estimate of habitual activity continues to be debated (1). Before accepting a previously implemented threshold of 4 d (21), we examined the relationship between steps per day and days monitored and discovered a pattern suggesting a self-selection bias. As indicated above, the most active people tended to wear the accelerometer for more days, and the least active people wore it for fewer days. Although it is tempting to eliminate those at the lower end, basing the argument on presumed lack of compliance, this act inflates physical activity estimates by screening out the most sedentary individuals. Wearing a physical activity monitor is not a blind experience. We speculate that it is plausible that those who are most active enjoy showing this and are therefore readily compliant with the regimen. In contrast, the most sedentary can easily realize that this is a physical activity monitor, and because they do not typically do any physical activity, they may grow weary of the whole point of wearing the instrument at all and therefore give up earlier. Put another way, fewer days of wear may be a reasonable proxy estimate for a sedentary lifestyle.
Finally and along these lines also falls the question concerning minimal time worn: how long must a PAM be worn to capture a valid day of physical activity? The worry is that we will miss physical activity performed when the motion sensor is not attached. Schmidt et al. (19) performed a thorough examination of this question in regards to pedometer time worn. They assessed alternative methods for adjusting for wear time and also evaluated the relationship between steps and associated biological measures (e.g., BMI, waist circumference, and systolic blood pressure) to determine whether any manner of adjustment improved correlations observed using the raw data. None of the adjustment methods used produced substantially stronger associations. Ultimately, the researchers concluded that error related to wear time is dependent on participants' activity level when not wearing the pedometer; if those people who wear it for a short time also do little physical activity when it is off, then the magnitude of error is minimal. They also asserted that adjusting for wear time using a stepping rate (e.g., steps per hour) might actually overestimate physically activity in those individuals with a short wear time who are not active after they take off the pedometer. It is our speculation that those who are physically active are reluctant to remove a motion sensor that is able to capture their lifestyle. Further, those who are more physically inactive are likely to remove the device early if they believe that it has nothing more to assess (i.e., it is not relevant) in their typically sedentary day.
In summary, the ActiGraph AM-7164 used in the 2005-2006 NHANES offers both accelerometer-defined step data in addition to time in intensity (defined by activity counts). Because this accelerometer is known to detect more low-force movements than accepted research-quality pedometers, this analysis examined population- and sex-specific descriptive epidemiology of accelerometer-determined steps per day with and without censoring steps detected at the lowest end of the activity spectrum (i.e., inactivity). The uncensored values obtained are almost 3000 steps per day more than what might be expected compared with other pedometer-based studies of free-living physical activity in the United States. Censoring these accelerometer data using a defensible activity count per minute cut point provides a distribution that is more in line with current understanding. Further cross-examination of these robust objectively monitored data is warranted to advance understanding of patterns of physical activity and inactivity in the United States.
The authors would like to thank Meghan McGlone for her assistance with data analysis and presentation. P.T. Katzmarzyk is supported, in part, by the Louisiana Public Facilities Authority Endowed Chair in Nutrition.
Conflict of interest: None of the authors have conflicts of interest to report. The results of the present study do not constitute endorsement by the American College of Sports Medicine.
1. Baranowski T, Masse 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):S544-9.
2. Barnett A, Cerin E. Individual calibration for estimating free-living walking speed using the MTI monitor. Med Sci Sports Exerc
3. Bravata DM, Smith-Spangler C, Sundaram V, Gienger AL, Lin N, Lewis R, et al. Using pedometers to increase physical activity and improve health: a systematic review. JAMA
4. Centers for Disease Control (CDC). State-specific prevalence of obesity among adults-United States, 2007. MMWR
5. Crouter SE, Schneider PL, Karabulut M, Bassett DR Jr. Validity of 10 electronic pedometers for measuring steps, distance, and energy cost. Med Sci Sports Exerc
6. Duncan JS, Schofield G, Duncan EK. Pedometer-determined physical activity and body composition in New Zealand children. Med Sci Sports Exerc
7. Freedson PS, Melanson E, Sirard J. Calibration of the Computer Science and Applications, Inc. accelerometer. Med Sci Sports Exerc
8. Ham SA, Kruger J, Tudor-Locke C. Participation by US adults in sports, exercise, and recreational physical activities. J Phys Act Health
9. Hatano Y. Prevalence and use of the pedometer. Res J Walking
10. Le Masurier GC, Lee SM, Tudor-Locke C. Motion sensor accuracy under controlled and free-living conditions. Med Sci Sports Exerc
11. Le Masurier GC, Tudor-Locke C. Comparison of pedometer and accelerometer accuracy under controlled conditions. Med Sci Sports Exerc
12. Matthews CE. Calibration of accelerometer output for adults. Med Sci Sports Exerc
. 2005;37(11 suppl):S512-22.
13. Matthews CE, Chen KY, Freedson PS, Buchowski MS, Beech BM, Pate RR, et al. Amount of time spent in sedentary behaviors in the United States, 2003-2004. Am J Epidemiol
14. Miller R, Brown W, Tudor-Locke C. But what about swimming and cycling? How to 'count' non-ambulatory activity when using pedometers to assess physical activity. J Phys Act Health
15. Ogden CL, Carroll MD, Curtin LR, McDowell MA, Tabak CJ, Flegal KM. Prevalence of overweight and obesity in the United States, 1999-2004. JAMA
16. Physical Activity Guidelines Advisory Committee. Physical Activity Guidelines Advisory Committee Report, 2008
. Washington (DC): US Department of Health and Human Services; 2008. p.D-3.
17. Richardson CR, Newton TL, Abraham JJ, Sen A, Jimbo M, Swartz AM. A meta-analysis of pedometer-based walking interventions and weight loss. Ann Fam Med
18. Rothney MP, Apker GA, Song Y, Chen KY. Comparing the performance of three generations of ActiGraph accelerometers. JAppl Physiol
19. 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
20. Siegel PZ, Brackbill RM, Heath GW. The epidemiology of walking for exercise: implications for promoting activity among sedentary groups. Am J Public Health
21. Troiano RP, Berrigan D, Dodd KW, Masse LC, Tilert T, McDowell M. Physical activity in the United States measured by accelerometer. Med Sci Sports Exerc
22. Tudor-Locke C, Ainsworth BE, Thompson RW, Matthews CE. Comparison of pedometer and accelerometer measures of free-living physical activity. Med Sci Sports Exerc
23. Tudor-Locke C, Bassett DR Jr. How many steps/day are enough? Preliminary pedometer indices for public health. Sports Med
24. Tudor-Locke C, Bassett DR Jr, Rutherford WJ, Ainsworth BE, Chan CB, Croteau K, et al. BMI-referenced cut points for pedometer-determined steps per day in adults. J Phys Act Health
. 2008;5(1 suppl):S126-39.
25. Tudor-Locke C, Ham SA. Walking behaviors reported in the American Time Use Survey 2003-2005. J Phys Act Health
26. Tudor-Locke C, Ham SA, Macera CA, Ainsworth BE, Kirtland KA, Reis JP, et al. Descriptive epidemiology of pedometer-determined physical activity. Med Sci Sports Exerc
27. Tudor-Locke C, Hatano Y, Pangrazi RP, Kang M. Revisiting "how many steps are enough?" Med Sci Sports Exerc
28. Tudor-Locke C, Pangrazi RP, Corbin CB, Rutherford WJ, Vincent SD, Raustorp A, et al. BMI-referenced standards for recommended pedometer-determined steps/day in children. Prev Med
29. Wyatt HR, Peters JC, Reed GW, Barry M, Hill JO. A Colorado statewide survey of walking and its relation to excessive weight. Med Sci Sports Exerc