Regular physical activity helps prevent early death and chronic diseases such as coronary heart disease, stroke, type 2 diabetes, depression, and some types of cancer (Ballard-Barbash et al., 2012; US Department of Health and Human Services, 2008b). Because of the importance of physical activity to health, national public health surveillance systems measure and track physical activity for use in planning, implementing, and evaluating public health practice (Galuska & Fulton, 2009). Physical activity levels of the U.S. population are most commonly assessed using a self-report measure, a valuable approach for monitoring and surveillance of physical activity in populations because of its feasibility, efficiency, and cost. Accelerometers, a device-based motion sensor, however, provide a precise measure of body movement (Troiano, Pettee Gabriel, Welk, Owen, & Sternfeld, 2012).

Uniaxial accelerometry, when worn on the hip, is an objective means of measuring vertical acceleration (Kozey, Staudenmayer, Troiano, & Freedson, 2010) and is one method of assessing free-living physical activity (Montoye, Kemper, Saris, & Washburn, 1996). The use of accelerometry for this purpose has become more feasible as accelerometers have become smaller, more reliable, and less expensive (Troiano et al., 2012). Although numerous studies designed to calibrate and validate accelerometers have been conducted, there is no standardized methodology to translate accelerometer output (i.e., counts per unit of time) into an estimate of physical activity (Masse et al., 2005). The most common means of doing so is to translate accelerometer output into measures of MET expenditure that reflect thresholds for specified levels of physical activity. However, as shown in Table 1, nine studies that used Actigraph accelerometer data to estimate participants’ level of aerobic activity used radically different cut points to define participation in moderate-intensity activity (cut point range:191–2743) and in vigorous-intensity activity (cut point range: 4945–7526) (Brage, Wedderkopp, Franks, Andersen, & Froberg, 2003; Brooks, Gunn, Withers, Gore, & Plummer, 2005; Freedson, Melanson, & Sirard, 1998; Heil, Higginson, Keller, & Juergens, 2003; Hendelman, Miller, Bagget, Debold, & Freedson, 2000; Leenders, Sherman, Nagaraja, & Kien, 2001; Matthews, 2005; Swartz et al., 2000; Troiano et al., 2008; Yngve, Nilsson, Sjostrom, & Ekelund, 2003).

Previous research has examined differences in physical activity estimates across multiple cut points in both youth and adults (Evenson, Buchner, & Morland, 2012; Loprinzi et al., 2012; Trost, Loprinzi, Moore, & Pfeiffer, 2011). Among youth, when compared to indirect calorimetry, only child-based cut points developed by Evenson and colleagues (Evenson, Catellier, Gill, Ondrak, & McMurray, 2008) exhibited acceptable classification accuracy for sedentary, light, moderate, and vigorous activity levels (Trost et al., 2011). In adults, the prevalence of meeting previous physical activity recommendations (Haskell et al., 2007) showed a wide discrepancy (4.5% to 97.6%) for multiple cut points (Loprinzi et al., 2012). Prevalence estimates using cut points derived with a lifestyle activity protocol (72.1%, 97.6%) were much higher than prevalence estimates using cut points derived with a walking/running activity protocol (4.5% to 36.4%) (Loprinzi et al., 2012). Cut points derived from studies using participation in ambulatory activities (walking or running) only are considerably higher than cut points derived from participation in lifestyle activities which include other activities such as household or gardening activities in addition to walking or running (Matthews, 2008). The physical activity protocol (participation in ambulatory or lifestyle activities) used to develop cut points is important because some lifestyle activities are composed of complex movement patterns that are associated with little vertical acceleration, meaning they exhibit lower counts, but expend energy through contractions of large muscle groups (Matthews, 2005; Swartz et al., 2000).

The impact of using cut points derived from purely ambulatory and lifestyle protocols to estimate physical activity levels based on the 2008 Physical Activity Guidelines for Americans (2008 Guidelines) for adults is unknown. Furthermore, there is little information regarding demographic patterns for multiple cut points derived from studies using ambulatory and lifestyle protocols. Therefore, we sought to estimate levels of physical activity as defined by different accelerometer cut points and to describe differences in levels of physical activity with cut points developed with ambulatory and lifestyle protocols. We also sought to examine patterns of physical activity prevalence, using different accelerometer cut points, across demographic and anthropometric characteristics and to determine if patterns are similar across study activity protocols.

Our findings showed that when using an accelerometer to estimate the percentage of the U.S. adult population meeting the aerobic component of the 2008 Guidelines, the prevalence varied depending on the choice of cut point. The physical activity prevalences from lifestyle protocol cut points were notably larger than estimates from ambulatory protocol cut points. For example, from 2003 to 2006, between 61% and 98% of adults would be classified as meeting 2008 Guidelines using lifestyle cut points whereas between 6% and 27% would meet 2008 Guidelines using ambulatory cut points. This results in a median difference of 65 percentage points, based on the median estimates of 77% and 12% for lifestyle and ambulatory protocol cut points, respectively. While the magnitude of the prevalence of meeting the 2008 Guidelines is different for ambulatory and lifestyle cut points, the prevalence patterns are similar. For example, both ambulatory and lifestyle cut points show more men than women meet the 2008 Guidelines.

There also appears to be a similar pattern for prevalence estimates across most demographic and anthropometric subgroups, but not necessarily all subgroups. Although sex, race/ethnicity, and age groups show similar patterns irrespective of activity protocol, the effects for body mass index and education seem to be much larger for ambulatory activities compared to lifestyle activities. The prevalence estimates for meeting the 2008 Guidelines for some subgroups are at least twice as large as other subgroups (e.g., 18% for normal weight and 5% for obese) when using cut points from ambulatory activities whereas the prevalence estimates are similar across subgroups (e.g., 80% for normal weight and 72% for obese) when using cut points from lifestyle activities.

Although we found no studies whose results were directly comparable with ours, we did find three that addressed how variations in accelerometer cut points affect physical activity estimates, one among youth and adults (Loprinzi et al., 2012), one among children and youth aged 5–15 years (Trost et al., 2011), and one among older adults (Evenson et al., 2012). In the study among youth and adults, using the same NHANES data and some of the same cut points, researchers found that the prevalence of meeting the old recommendations varied significantly, depending on the cut point used (Loprinzi et al., 2012). Similar to our findings, using the same heterogeneous sample of adults (Loprinzi et al., 2012), the prevalence of meeting previous recommendations varied from 5% to 98%. In addition, the prevalence varied from 5% to 36% in studies using an ambulatory protocol and from 72% to 98% in studies using a lifestyle activity protocol. In the study among children and youth (Trost et al., 2011), researchers who compared accelerometer cut points derived from 5 youth-specific equations with indirect calorimetry, a criterion measure, found that the physical activity estimates varied widely depending on the cut points used. Estimates from two of the five cut points underestimated time spent in moderate-to-vigorous intensity physical activity by 39% to 74%. In older adults (≥ 60 years) (Evenson et al., 2012), researchers examined the impact of a set of pre-determined cut points on daily minutes of moderate to vigorous physical activity, using data from the 2003–2006 NHANES examination. Findings showed the patterns of physical activity were generally consistent across age groups, gender and race/ethnicity. For example, prevalence estimates were always lower for women, regardless of the cut point used.

The differences in cut points in the validation studies we examined may be attributed to three factors. The first is the more comprehensive activity profile used in studies that attempted to reflect energy expended in lifestyle activities (US Department of Health and Human Services, 1996), which tend to involve movement patterns with less vertical acceleration and thus produce lower accelerometer readings than ambulatory activities of the same intensity level (Bassett et al., 2000; Matthews, 2005, 2008; Swartz et al., 2000). The second factor contributing to the differences in cut points is the small, disproportionately young, and fairly homogeneous nature of the samples in the studies we examined. Future accelerometer validation studies may benefit from accelerometer cut points derived from samples more representative of the U.S. adult population. The third factor contributing to the variation is the choice of statistical methods used to determine the cut points. In general, the use of accelerometer data to estimate energy expenditure is based on the assumption of a linear relationship between vertical acceleration and energy expenditure. Although this assumption has been shown to be more valid in estimating energy expended in ambulatory activities such as walking and running, it has been shown to be less valid in estimating energy expended in lifestyle activities such as sweeping or raking (Matthews, 2005). Thus, a simple linear association between counts and physical activity intensity may not generalize to all types of physical activities (Crouter & Bassett, 2008; Crouter, Clowers, & Bassett, 2006; Crouter, Horton, & Bassett, 2011; Crouter, Kuffel, Haas, Frongillo, & Bassett, 2010).

This study has at least two limitations. The first limitation is that our findings may have been affected by selection bias inasmuch the distribution of characteristics among participants of our analytic sample (n=6547) may have differed from that among the NHANES participants excluded from analysis (n=3054). However, we found no significant differences between the two groups by sex or BMI and only small (<10%) albeit significant differences by education and race/ethnicity. The largest difference between the two groups was that 18 to 24 year olds accounted for 22% of NHANES participants excluded from our study sample but only 11% of those included in the sample. We could not determine, however, the extent to which physical activity patterns among participants in this age group may have differed from patterns among NHANES participants who were excluded from the study sample. The second limitation is that NHANES data did not include a criterion measure of physical activity energy expenditure (such as doubly labeled water) with which to compare our estimates of energy expenditure.

Although there is no universally-accepted method of measuring physical activity, estimates of physical activity based on accelerometer data offer a number of advantages over estimates based on self-reports of physical activity, including the elimination of recall bias, social desirability bias, and issues related to estimating the duration and frequency of one’s physical activities (Atienza et al., 2011). However, because accelerometers, worn on the hip, measure only vertical acceleration, they fail to account adequately for activities other than walking or running or to indicate the purpose of any physical activity. Physical activity estimates based on accelerometer data, therefore, should not be compared directly with estimates based on self-reports. Because few population-based studies of the relationship between physical activity and health outcomes have used accelerometer-derived measures of physical activity (Bowles, 2012), associations between physical activity and health (US Department of Health and Human Services, 2008b) on which national physical activity guidelines are based have been derived solely from studies that relied on subjects’ self-reports of their physical activity. The extent to which such associations would differ if physical activity estimates were based on accelerometer data is unclear. Furthermore, accelerometer data, like self-reported measures of physical activity, are subject to various sources of measurement error (Nusser et al., 2012). Currently, a study is underway to examine the error in self-reported and accelerometry-derived physical activity with known standards of energy expenditure (Bowles et al., 2012).

Our findings indicate that researchers using accelerometer data to estimate physical activity prevalence in a given population need to be aware that their choice of activity level cut points can have a marked effect on their estimates. Depending on the cut point used, physical activity prevalence may be under- or over-estimated by varying degrees. Furthermore, because of the potential bias by body mass index and education level, it may be that there is not a “one size fits all” for defining standard cut points. Generation of future accelerometer cut points may consider developing cut points for demographic subgroups using a variety of lifestyle physical activities.