A Catalog of Rules, Variables, and Definitions Applied to Accelerometer Data in the National Health and Nutrition Examination Survey, 2003–2006

1The National Health and Nutrition Examination Survey (NHANES) is a publicly available data resource that provides information from self- or proxy reports of health conditions and behaviors and biomedical data for a sample representing the US civilian noninstitutionalized population (www.cdc.gov/nchs/nhanes.htm). NHANES is administered in 2-year data collection cycles; the Physical Activity Monitor (PAM) component was introduced in the 2003–2004 and 2005–2006 cycles to collect accelerometer-based measures of physical activity among participants aged 6 years or older. During these 2 cycles, an ActiGraph model 7164 accelerometer (ActiGraph, LLC, Pensacola, Florida) was provided to ambulatory participants, representing the first time that a surveillance study collected accelerometer measures on a US representative sample.

2The uniaxial accelerometer measured and recorded vertical acceleration as “activity counts.” The device also recorded “steps” by using a proprietary signal-filtering algorithm. These 2 related quantities measure physical activity movement associated primarily with locomotion. A 1-minute time interval, or “epoch,” was used in NHANES. Data for activity counts and steps were recorded during each epoch for up to 1 week. Both activity count and step data were released for the 2005–2006 NHANES cycle, but because of missing step data on a portion of the sample in the 2003–2004 cycle, only activity count data were released for that cycle. The data are available from http://www.cdc.gov/nchs/nhanes.htm; the National Cancer Institute (NCI) offers SAS (SAS Institute, Inc, Cary, North Carolina) syntax for analyzing the data at http://riskfactor.cancer.gov/tools/nhanes_pam/. The syntax facilitates the editing of invalid and unreliable intensity values (defined by NCI) and summarizes derived variables that describe the duration of nonwear periods and activity bouts of moderate, vigorous, and moderate-to-vigorous intensities. The NCI website acknowledges that “users can modify these programs to examine other issues, such as alternate definitions of valid data, monitor wear periods, or activity bouts.”

3The release of PAM data in 2007 provided researchers a unique opportunity to study objectively measured physical activity on a large and representative US sample and relate it to a range of other health-related variables. Numerous studies using the data have been published, but these studies have treated, analyzed, and reported the data by using myriad accelerometer decision rules, derived variables, and cut point definitions. A catalog of these rules, variables, and definitions is needed so that researchers can begin to work toward more standardized and comparable data. The objective of this study was to catalogue the accelerometer decision rules, derived variables, and cut point definitions used in studies on PAM data published since 2007.

4We conducted an advanced English-only literature search of original research articles in PubMed by using the key terms “activity monitor” or “ActiGraph” or the wildcard term “acceleromet*” in addition to “NHANES” or “National Health and Nutrition Examination Survey.” We searched articles published from January 1, 2007, through December 31, 2011. We used the following search strategy: (“activity monitor” OR ActiGraph OR acceleromet*) AND (NHANES OR “National Health and Nutrition Examination Survey”) AND English[Language] AND (“2007/01/01”[Date of publication]: “2011/12/31”[Date of publication]). We included forthcoming and “epub ahead of print” articles and updated the search on February 10, 2012. We found 74 articles that met our search criteria. One author (R.P.T.) identified 1 other published study, prepared for a special conference, not indexed in PubMed bringing the initial search total to 75 articles.

5Twenty-one articles (28%) did not directly analyze PAM data, and they were eliminated; the remaining 54 articles (72%) were included in this review.

6The first author read and abstracted the following details from each identified article: 1) citation; 2) purpose of study; 3) PAM data collection cycle(s) analyzed (ie, 2003–2004 and/or 2005–2006); 4) study sample size and age of participants in sample; 5) whether investigators reported using the NCI-supplied SAS syntax; 6) rules for defining nonwear time (ie, time that the accelerometer was not likely worn), a valid day (ie, the minimum number of wear-hours required to be considered representative of a day’s behavior), and the minimum number of valid days required for a participant to be included in the analysis; 7) accelerometer-derived variables (eg, activity counts/day, time spent in moderate-intensity activity, steps in vigorous-intensity activity); and 8) cut point definitions used for each accelerometer-derived variable (ie, values used to categorize continuous data). The second author verified the details independently. Discrepancies were discussed and consensus achieved. The results were tabulated to facilitate comparison among studies. We made no attempt to contact the articles’ authors to obtain unreported information or clarify writing; data extraction was made on face value.

7The purpose of the 54 articles varied (Table 1); they represented, for example, descriptive analyses, methodological analyses, and inferential analyses. Eighteen studies used the NHANES 2003–2004 cycle, 15 used the 2005–2006 cycle, and 21 combined data from both cycles. Sample sizes ranged from 103, representing prostate cancer survivors, to 6,329, representing participants with 1 or more days of wear in the 2003–2004 cycle. Fourteen studies focused primarily on children and/or adolescents (through age 19 y), 33 on adults (including 1 study on all participants aged ≥16 y), 2 on older adults, and 3 on all ages (ie, ≥6 y). Two reported only the mean age of cancer survivors.

8Twenty-four studies reported using the NCI-supplied SAS syntax. Eight studies cited previous work, notably the first published study, that used the SAS syntax. The remaining 22 studies did not attribute their decision rules to another source. Most studies (42 of 54) defined nonwear time as 60 minutes or more of consecutive zeros, with or without allowance for interruptions, variously defined (Table 2). One study defined 10 minutes or more of identical consecutive nonzero counts as missing data. Seven studies that focused on children or adolescents or both defined nonwear time as 20 minutes or more of consecutive zeros. Four studies did not define a valid day; however, 2 of these studies may have implemented decision rules embedded in the NCI-supplied SAS syntax, which they reported using. Regardless, 49 of 54 studies defined a valid day as 10 hours or more of wear. The minimum number of valid days required for a participant to be included in the analysis varied; 17 studies required a minimum of 1 day; 23 studies required a minimum of 4 days; and 10 required a minimum of 4 days, including 1 weekend day.

9Studies on adults typically presented multiple accelerometer-derived variables (Table 3). Definitions differed for some similarly named variables. For example, some studies defined time in sedentary behavior as less than 100 activity counts per minute; others defined it as less than 260 activity counts per minute. Time in light intensity was defined as 100 to 759 activity counts per minute, 100 to 573 activity counts per minute, 100 to 1,951 activity counts per minute, 100 to 2,019 activity counts per minute, 260 to 1,951 activity counts per minute, and 500 to 2,019 activity counts per minute. Time in moderate- to vigorous-intensity physical activity (MVPA) was defined as 500 or more activity counts per minute, 574 or more activity counts per minute, 760 or more activity counts per minute, 1,000 or more activity counts per minute, 1,500 or more activity counts per minute, 1,952 or more activity counts per minute, 2,000 or more activity counts per minute, or 2,020 or more activity counts per minute. Time in MVPA was sometimes considered as any minute above the cut point and at other times only as minutes within a bout of 10 minutes or more (which may or may not have allowed for an interruption of 1 or 2 minutes below the cut point). Step data were reported in 2 ways: 1) in a raw or uncensored format (ie, not adjusted in any way) and 2) following a process of censoring steps from any minute with less than 500 activity counts per minute. (The latter process was designed to interpret the higher values of accelerometer-based step data against lower pedometer-based scales.) Physical activity levels were categorized according to a step-defined graduated index. Additional accelerometer-derived variables included time in incremental cadence (steps per minute) bands and peak cadence indicators (defined as the highest level of physical activity, or natural best effort, measured during a given day).

10The 2 primary cut point definitions of time spent in MVPA for studies on children or adolescents or both were age-specific values, building on previous research, and 3,000 or more activity counts per minute (Table 4). Bouts were defined as any minute, 1 to 4 minutes, 5 to 9 minutes, and 10 or more minutes above threshold, again at times allowing for minimal interruptions below the cut point. Uncensored and censored steps per day were reported. Data were also presented according to a child-specific step-defined graduated index.

11An obvious advantage of NHANES accelerometer data is that they reflect objectively measured behaviors that can be examined, compared, and related to other NHANES data. NCI-supplied SAS syntax has facilitated analysis of these data. When studies in our review did not explicitly report some decision rules, they frequently reported use of SAS syntax or they cited previous methods that had used this tool, suggesting that SAS syntax was likely applied. Clearly, researchers have treated, analyzed, and reported PAM data in nonstandardized ways, which compromises the ability to make comparisons among studies. This lack of uniformity is perhaps most apparent in the multiple cut point definitions of time spent in MVPA. Inconsistent approaches will impede the ability to track behaviors over time in the United States and compare US behaviors with behaviors in other countries.

12The intent of this review was neither to judge researchers’ decisions about examining PAM data nor to make pronouncements on the most appropriate strategies. Unique research questions may require equally unique analytical approaches; some degree of inconsistency must be tolerated if scientific discovery is to be encouraged. That being said, consumers (including the research community) of these data must be informed about inconsistencies, especially when different cut point definitions are used for similarly named variables.

13One of the primary challenges to implementing a measure of accelerometer-based physical activity in a study is ensuring compliance with monitoring protocols. The 2003–2004 and 2005–2006 NHANES protocols asked participants to remove the accelerometer only during sleep and water-based activities (eg, swimming, showering, bathing). Conclusions about accelerometer-based behavior are affected by definitions of nonwear time and a valid day. Most studies included in our review defined nonwear time as 60 minutes or more of consecutive zeros. Differences between allowances for interruptions may simply represent reporting discrepancies, especially because many of the studies catalogued in Table 2 also reported using SAS syntax (or cited methods of previous work that did). The NCI decision to use 60 minutes of consecutive zeros to identify nonwear time was based on research by Mâsse et al. These researchers demonstrated that sample sizes were optimized when nonwear time was defined as 60 minutes, rather than 20 minutes, of continuous zeros. A recent study indicated that 90 minutes of continuous zeros may provide more accurate estimates of time in sedentary and active behaviors.

14Researchers were also almost perfectly consistent in defining a valid day as 10 or more hours of wear time, which is also the definition provided by SAS syntax. (In 2 of the 4 studies that did not report nonwear criteria, we assumed that they used this definition because they reported using SAS syntax.) Mâsse et al compared results of studies that used different definitions of a valid day, and although they did not recommend a specific number of hours of wear time to define a valid day, they noted that the strictest requirement (≥12 h/d) negatively affected sample size. They also speculated that stricter requirements might unduly limit inclusion of inactive people, thereby affecting overall data distribution. The popularity of using 10 or more hours of wear time to define a valid day is likely due to numerous factors: 1) it was a component of one of the decision-rule algorithms evaluated by Mâsse et al; 2) it was used in the seminal NHANES PAM data publication; and 3) it was built into the NCI-provided SAS syntax to accompany the NHANES accelerometer data. Using 2005–2006 NHANES data, Tudor-Locke et al showed that, as population estimates of nonwear increase, all other time in intensity (eg, MVPA) and volume (ie, activity counts/d, steps/d) indicators decrease to some degree, but the negative effect is most pronounced on estimates of time spent in sedentary behavior. Nonwear time is more likely to reflect time spent in sedentary behaviors than in active behaviors. Mâsse et al reached a similar conclusion: varying minimal wear-time requirements primarily affected minutes of inactivity (their preferred term). Others concluded the same. The effect of reduced wear time on estimates of sedentary behavior should not differ by age. Although there is apparent consensus that 10 or more hours of wear time is adequate to define a valid day, a 24-hour wear-time protocol would remove much ambiguity from analysis.

15Regardless of how scientists have analyzed NHANES accelerometer data, however, it remains clear that the US citizenry is not very active. Troiano et al reported that less than 5% of adults achieve public health guidelines, although this low estimate may be an artifact of the minimal bout criterion and a cut point definition that was based primarily on locomotor activities. Matthews et al reported that more than 50% of monitored time is spent in sedentary behaviors. Tudor-Locke et al reported that NHANES adults took an average of approximately 6,500 steps/day (considered “low active” on a pedometer-based scale). Using identical accelerometer models and analytic methods to directly compare Swedish data with NHANES data, Hagströmer et al showed that the amount of time spent in MVPA was not uniformly greater in Sweden than in the United States, even though Sweden has a population that is generally considered to be quite active.

16Although we limited our online search to English-language articles indexed in PubMed, we are confident that this search engine was the best one for identifying articles on NHANES accelerometer data. This free resource is maintained by the National Center for Biotechnology Information at the US National Library of Medicine, which is located at the National Institutes of Health. We included only 1 article that was not indexed in PubMed. Our search spanned 2007 (the year that these data were released) through 2011; however, we acknowledge there may be additional forthcoming articles that we did not identify. This review necessarily represents a limited time frame.

17NHANES accelerometer data represent an important public use resource for researchers and practitioners engaged in designing and directing health programs and services and developing public health policy. This review was undertaken to summarize existing research that has used these data. The studies we identified bear evidence of the multiple and diverse uses of these data, and we can anticipate that they will continue to be used in epidemiologic and health sciences research. We hope that the resulting catalog of accelerometer decision rules, derived variables, and cut point definitions used to analyze these NHANES data serves as a useful starting point for future researchers to consider as they plan and report their own analyses.