Physical fitness and activity are well known to confer numerous health benefits in the prevention of certain chronic diseases (¹⁰). The recent public health message has focused on the incorporation of moderate intensity physical activity into the daily lives of all individuals (¹⁴). This strategy represents a shift from previous recommendations of vigorous exercise programs and may provide a more feasible goal for much of the population. In addition, it has been recognized that many of the benefits of physical activity may be apparent at lower intensity levels without a large effect on fitness or maximal aerobic power. Thus, current guidelines encourage people to accumulate a minimum of 30 min of “moderate intensity” activity daily (¹⁴). Moderate intensity is defined as activity requiring a rate of energy expenditure (EE) of 3–6 METs, or 3–6 times the resting EE (1 MET = 3.5 mL O₂·kg⁻1·min⁻1).

To better define the dose-response relationships between activity and health outcomes, it is necessary to accurately quantify physical activity patterns. Methodologically, this involves recording people’s daily activity habits along with accurate estimates of the energy cost of these various activities. Although short-term laboratory exercise sessions can be accurately monitored by directly measuring physiological variables, intermittent activity in a field setting is much more difficult to assess. Questionnaires are often used to classify habitual activity levels; however, these rely on self-report, which is subject to numerous sources of error (^11,12,18).

Various types of motion sensors have been developed in an attempt to monitor more objectively and accurately physical activity in the field. These include pedometers, simple and low-cost devices that record the number of steps taken with varying degrees of sensitivity. Several types of accelerometers are also commercially available. These more sophisticated devices are able to detect and record the actual magnitude of acceleration, allowing the quality or intensity of movement to be determined. The advantages of these devices include the small size, which allows subjects to wear the monitor for long periods of time without interfering with normal movement, and the ability to store data continuously over long periods of time. This information can then be analyzed to examine patterns of activity over the course of several days or weeks.

Laboratory investigations have established a linear relationship between the counts recorded using accelerometry and EE during locomotion (^4,6,9,12,17). This has led to the development of equations to predict MET level or intensity classification (i.e., light, moderate, hard, vigorous) from accelerometer recordings (⁹). However, these devices have not been validated against direct measures of energy cost in the field or during activities other than locomotion (walking or running).

Furthermore, the recent guidelines suggest that common household chores or lifestyle activities can contribute to the time spent in “moderate intensity” activity. Typical recommendations include tasks such as housecleaning, gardening, and yard work. Many recreational pursuits (not specifically intended as “exercise”) may also fall into this category. Information on the energy cost of these types of activities is often obtained from the Compendium of Physical Activities (¹). This source reports estimated MET levels for a wide variety of tasks and is used as a basis for calculating energy expenditure from commonly used physical activity questionnaires. However, the values are derived from limited data and have not been adequately verified.

The primary aim of this study was to investigate the validity of accelerometry for the assessment of EE in the field. This was conducted by comparing the relationship between accelerometer counts and metabolic cost among the selected activities (walking, golf, housecleaning, and yard work) with that obtained during locomotion. In addition, the intensity levels of these tasks were evaluated and compared to the values reported in the Compendium of PhysicalActivities.

METHODS

Subjects.

Subjects were eligible to participate if they were between 30 and 50 yr of age and were physically able to complete the protocol, including an ability to play golf. No constraints were placed on fitness level or other health habits. Volunteers were recruited from a local golf course and from the community. All subjects read and signed an informed consent document and completed a physical activity readiness questionnaire (PAR-Q) before participating in the study.

Protocol.

A habituation session was provided to familiarize the subjects with the equipment and procedures for the study. During this visit, subjects completed the consent form, PAR-Q, and a questionnaire to evaluate their habitual level of physical activity on a 7-point scale, as described by Ross and Jackson (¹⁵). Subjects then performed three test sessions, with a subset of volunteers (N = 10) repeating the testing for reliability assessment. The first session consisted of walking at four self-selected speeds on an indoor track. Subjects were instructed to walk at “leisurely” (bout 1), “comfortable” (bout 2), “moderate” (bout 3), and “brisk” (bout 4) paces for approximately 5 min each, with 5 min of rest between bouts. Pace and stride frequency were measured for each lap. The second condition consisted of two holes of golf at a local golf course. Subjects used a pull cart for their clubs, and play was continuous for the two holes. The final session involved indoor and outdoor household tasks, including 5 min each of washing windows, dusting, vacuuming, lawn mowing (using a gas-powered push mower), and planting shrubs.

During all sessions, respiratory gas exchange was assessed using a portable metabolic measurement system (TEEM100, AeroSport, Inc., Ann Arbor, MI) carried in a pack on the lower back. The medium flow pneumotach was used and was calibrated before each test. The gas analyzers were calibrated and verified with known gases immediately before and after each test. The TEEM100 was validated for moderate exercise in previous laboratory studies (^13,19). In addition, three motion sensors were worn on elastic belts around the waist. A uniaxial CSA accelerometer (model 7164, Computer Science Applications, Inc., Shalimar, FL) was worn above the right hipbone, a triaxial Tritrac monitor (Reining International, Ltd., Madison, WI) was worn on the left hip, and a Digiwalker pedometer (SW-701, Yamax Corp., Japan) (walking only) was placed on the right hip aligned with the midline of the right thigh. Cardiorespiratory and accelerometer data were averaged over 1-min sampling intervals, whereas the pedometer reading was recorded at the end of each walking speed. The CSA and Tritrac accelerometers were initialized and downloaded according to the manufacturers’ specifications, using software provided by the respective companies. The TEEM100 was also downloaded to a computer, and all data were imported into Microsoft Excel for further processing. All instruments were synchronized to the same clock, and time was recorded at the beginning and end of each activity so that the appropriate data could be extracted.

Subjects were asked to maintain similar diet and activity patterns throughout the study. All sessions were conducted at the same time of day for each subject, and information was collected regarding the time of food and caffeine intake and exercise before testing. Body mass and environmental conditions were recorded for each session.

Analysis.

For the walking session, metabolic and accelerometer data were averaged over the final 2 min of each bout. Stride frequency was determined by measuring the time required to complete 10 full strides, with the average of two measurements used for each bout. Stride frequency was doubled to obtain step frequency, with each footstrike (either foot) counted as a step. The Digiwalker output (total number of steps) was recorded at the end of each bout and divided by the total walking time for that bout to obtain a reading in steps per min. For the other activities (golf and household tasks), metabolic and accelerometer readings were averaged over the entire bout, excluding the initial 2 min (to allow a steady state to be reached). All metabolic data were adjusted for the mass of the equipment worn (5.0 kg) by dividing the absolute V̇O₂ by the subject’s body mass plus equipment mass. For the Tritrac, count data presented are from the vector magnitude, which is a composite measure of all three axes.

Using the pooled data, we examined Pearson product moment correlation coefficients to assess the relationships between metabolic cost and the count data from the two accelerometers. These correlations were examined for the walking sessions only and for all activities combined. Regression analysis was then performed using the pooled data to develop equations predicting metabolic cost from activity counts. The equations were rearranged to derive count cutoff values corresponding to predetermined MET levels (light activity, > 1 MET to < 3.0 METs; moderate activity, ≥ 3.0 METs to < 6.0 METs; and hard activity, ≥ 6.0 METs to <9.0 METs). Cutoff values established from each equation (walking and all activities) were compared for both accelerometers. In addition, the CSA values were compared with previously reported cutoff values for these intensity levels.

The data from the four walking speeds were used to develop individual regression equations relating V̇O₂ to CSA and Tritrac counts for each subject. These equations served as individualized “calibration curves” for the accelerometers. Subsequently, the count values from the golf and household activities were substituted into the equations to predict MET values from each accelerometer. These predictions were compared with the actual measured METs using a two-factor repeated-measures ANOVA (activity × method).

The accuracy of the Digiwalker was examined using the results from the walking bouts. The pedometer reading, converted to steps per min, was compared with the investigator-determined timed step frequency and analyzed using a two-factor repeated-measures ANOVA (bout × method). For the metabolic data, mean values and 95% confidence intervals were used to characterize the intensity of each activity and to compare the observed values with MET levels reported in the Compendium of Physical Activities (¹). All statistical analyses were performed using the software package Statistica^TM version 5.1 (StatSoft, Tulsa, OK).

RESULTS

Twenty-five subjects (10 male, 15 female) completed the study. Average (mean ± SD) age was 40.8 ± 7.2 yr, height was 171.3 ± 8.5 cm, body mass was 71.5 ± 13.7 kg, and physical activity score was 4.9 ± 1.4 units on a 7-point rating scale (¹⁴). The male subjects were significantly taller and heavier than the female subjects (P < 0.01); however, there were no differences in self-reported ratings of habitual physical activity. All subsequent analyses are based on the entire sample. Results from the track walking session are shown in Table 1, including the self-selected walking speed, metabolic cost in METs (adjusted for the equipment mass), and the mean count values from the CSA and Tritrac accelerometers. Table 2 displays the metabolic and accelerometer data from the golf and household activities. For the 10 subjects who performed repeat testing to assess the reliability of the measures, paired t-test results revealed no differences in energy cost or accelerometer counts for any of the activities, with the exception of planting shrubs. For this task, the intensity was reduced in the second trial and was reflected in the Tritrac counts, which were also significantly lower.

Correlations between metabolic cost and accelerometer output were examined in the pooled data set. For the walking trials only, the correlation between CSA counts and METs was r = 0.77, whereas the Tritrac yielded a higher correlation at r = 0.89. When all activities were combined, these correlations were reduced to r = 0.59 (CSA) and r = 0.62 (Tritrac). In both situations, correlations between the two accelerometers were high (r = 0.87 for walking only and r = 0.93 for all activities). When the individual axes of the Tritrac were examined, the anterior-posterior plane showed very little association with overall metabolic cost for walking (r = 0.17) or for all activities combined (r = 0.24). Figures 1 and 2 illustrate the scatterplots for each accelerometer based on walking only and for all activities pooled, with the regression equations presented in Table 3. Good reproducibility was seen in the count versus MET relationships, with similar regression equations and correlations observed for both trials for the subjects who performed reliability testing. For the CSA monitor, these correlations were r = 0.78 (walking only) and r = 0.58 (all activities) for trial 1 and r = 0.78 and 0.65 for trial 2. For the Tritrac, these values were r = 0.88 and 0.60 (trial 1) and r = 0.88 and 0.68 (trial 2).

The regression equations developed on the pooled data set were also used to determine the count values corresponding to the MET ranges defining light (1.0 MET to < 3.0 METs), moderate (≥3 METs to < 6 METs), and hard (≥6 METs to <9 METs) activity categories. To derive the count cutoff values shown in Table 4, the equations from Table 3 were rearranged, and the MET levels of 3, 6, and 9 were substituted into each equation.

Individual regression equations were developed for each subject based on the metabolic and accelerometer data across the four walking speeds. These equations were then applied to the count data recorded during the recreational and household activities to predict METs, providing a within-subject comparison of the ability of the accelerometers to capture activity intensities during varied field tasks. The actual MET values along with the mean of the predicted values (from the individual calibration equations) from each accelerometer are presented in Table 5 and Figure 3. Both the CSA and Tritrac substantially underestimated the intensity of each of the activities (30.5–56.8% underestimation). Statistical analysis revealed significant effects of activity (F = 164.6, P < 0.001) and method of measurement (actual vs CSA prediction vs Tritrac prediction) for V̇O₂ (F =168.7, P < 0.001) as well as a significant interaction effect (F = 20.3, P < 0.001). The interaction was due to a difference in the magnitude of the underprediction among the different activities, with the discrepancy being lower for golfing and lawn mowing (closer to 30%) than for the housecleaning and planting activities (generally above 50%). However, despite this interaction effect, it is clear that all activities were substantially and significantly underestimated when the prediction equations generated from the walking protocol were applied to the observed accelerometer count data.

The accuracy of the Digiwalker was examined by comparing the pedometer output with the measured step frequency across the bouts. ANOVA results revealed significant main effects (bout:F = 82.8, P < 0.001; method:F = 26.7, P < 0.001) as well as a significant interaction effect (F =13.2, P < 0.001). As shown in Figure 4, the Digiwalker underestimated the number of steps per minute at lower speeds (bouts 1 and 2), although the values became more similar for the faster walking bouts. At the fastest speed, the difference between the methods was no longer significant (P = 0.089). This was supported by a correlation of r = 0.84 between the two measures for all bouts combined. Digiwalker counts per minute were also correlated with walking speed (r = 0.86) and V̇O₂ (r = 0.75) in the pooled data.

Measured values and 95% confidence intervals for the metabolic cost of each activity are shown in Table 6 along with corresponding values from the Compendium of Physical Activities (¹). Compared with the present data, the Compendium appears to slightly but significantly overestimate the EE of golf while underestimating the intensity of light cleaning (including dusting and vacuuming) and outdoor garden chores. In addition, the Compendium substantially overestimates the metabolic cost of window washing, which was comparable to light cleaning in the current study. Because the Compendium lists MET levels for walking at specific speeds only, a regression line was fit to the current data (METs = −1.09 + 1.55 × speed [mph]; R² = 80.8%, SEE [standard error of the estimate] = 0.59) to determine metabolic cost at those speeds for comparison. MET values (and corresponding confidence intervals) at 2.0, 3.0, and 4.0 mph (3.2, 4.8, and 6.4 km·h⁻1) were interpolated using this equation and are presented in Table 6. The results for walking at 3 (4.8 km·h⁻1) were comparable to the Compendium figures; however, the Compendium appeared to overestimate the energy cost of slow walking (2 mph; 3.2 km·h⁻1) while underestimating that of brisk walking (4 mph; 6.4 km·h⁻1).

DISCUSSION

Current guidelines for physical activity from the Surgeon General and the American College of Sports Medicine encourage the accumulation of 30 min of moderate-intensity activity throughout each day (¹⁴). “Moderate intensity” is generally defined as activity requiring a rate of EE of 3–6 METs, and common recommendations include housework, walking, gardening, and various recreational pursuits. However, estimates of the actual energy cost of these types of activities are based on limited data, because accurate metabolic measurements in the field have been difficult to obtain. In the current study, the energy requirements of various tasks in the field were quantified using a portable metabolic measurement system. The results confirmed that the selected activities (self-paced walking from slow to brisk, golf, housecleaning, and yard work) did fall within the moderate intensity range (3–6 METs). However, the indoor household tasks of window washing, dusting, and vacuuming as well as slow-paced walking were at the lower limit of this range.

The Compendium of Physical Activities (¹) is the most common source of information regarding the intensities of various activities and is often used as the basis for quantifying the data from physical activity assessment questionnaires. However, the values measured in the current study were significantly different from the Compendium listings. The energy cost of golf was overestimated by the Compendium, whereas light cleaning and yard work (lawn mowing and planting shrubs) were underestimated. The most substantial difference was seen in window washing, which was comparable to the other cleaning activities in the current results (approximately 3 METs), whereas being listed as much more intense (4.5 versus 2.5 METs for light cleaning) in the Compendium. However, it may be expected that the EE of these types of activities would be highly variable depending on environmental conditions, terrain (for outdoor tasks), and individual differences in performing the tasks.

For walking, the energy costs at the speeds listed in the Compendium were interpolated from the speed versus MET relationship seen for the overground walking in the current study. Although results were similar for the moderate speed (3 mph; 4.8 km·h⁻1), the Compendium appeared to overestimate the intensity of slow walking and underestimate that of faster speeds. Future work might revise the currently used prediction equations for V̇O₂ for locomotion and also more accurately determine the energy cost of other common activities using newer, portable metabolic data acquisition systems.

The primary goal of this study was to examine the ability of accelerometry to assess intensity during various activities in the field. Traditionally, accelerometers have been validated during laboratory tasks such as treadmill walking and running, with the observed relationships between motion and metabolic cost then applied to the assessment of EE in free-living situations. During level locomotion at a range of speeds, the relationship between accelerometry (counts·min⁻1) and energy cost has been shown to be linear and highly correlated for both uniaxial and triaxial accelerometers (^4,6,9,12,17). This relationship was supported by the current results, although the correlations were lower than demonstrated in previous studies. This may be due to the lower range of speeds (walking only) in the current study and the less constrained task of overground, self-paced walking. For example, the observed correlation between CSA counts and V̇O₂ was r = 0.76 for the walking trials, which encompassed a range of approximately 3–6 METs. In comparison, Freedson et al. (⁹) reported a correlation of r = 0.91 for walking and running at 3.7–9.7 METs, Melanson and Freedson (¹²) found a correlation of r = 0.82 using walking and running (3.6–8.6 METs), and Trost et al. (¹⁷) reported a correlation of r = 0.86 in children walking and jogging at 4.3–7.7 METs. The current study also revealed a higher correlation for the Tritrac, a triaxial accelerometer, than the uniaxial CSA monitor for the walking trials. However, the anterior-posterior plane measurement did not appear to contribute to the improved result, since it was not significantly correlated with the vector magnitude value or with V̇O₂.

The Yamax Digiwalker, a low-cost, commercially available pedometer, was analyzed for the walking session only in this study. It was found to detect the number of steps quite well, especially at higher speeds. This agrees with the findings of Bassett et al. (²), who demonstrated high accuracy (within 1% of actual steps) using the Yamax pedometer during overground and treadmill walking, except at low treadmill speeds. Very slow walking may not generate forces of enough impact to be captured by the pedometer. However, the first walking bout in the current study was much slower than the subjects’ normal walking speeds and would not likely be used in daily living; therefore, this would not be expected to be an important source of error in field studies. Furthermore, the difference between measured and actual step frequency at moderate and high speeds in the current study, although statistically significant, was small (2–4 steps·min⁻1) and would be considered acceptable for a field measure. It should also be noted that the criterion measure of stride frequency was obtained by measuring the time taken for 10 strides for each lap of the track, so that slight inconsistencies in pace or stride frequency over the course of each bout may have confounded this measurement. However, the pedometer may be limited in its ability to assess distance walked or energy expended, since stride length tends to increase with walking speed (²). In addition, similar to other accelerometers, the pedometer would not be expected to detect increased metabolic cost because of graded walking (^7,12) or load carriage.

Despite the fairly strong relationship between counts and V̇O₂ during walking for the CSA and Tritrac monitors, when these equations were applied to the other (nonlocomotion) activities, the accelerometers were unable to accurately assess energy cost. The regression equations developed from the four walking speeds for each subject provided an individually calibrated, within-subject comparison of the counts–V̇O₂ relationship for walking versus each of the other tasks. The metabolic costs of all other activities were significantly underestimated from the equations using either accelerometer, implying that at similar intensity levels, the activity counts were much lower than for walking. Thus, it appears that the count-V̇O₂ relationship is specific to a given type of activity or movement.

The inability of accelerometers to detect increased EE because of graded locomotion was demonstrated for the CSA monitor (¹²) and the Caltrac and Tritrac (⁷). Both of these investigations found no change in counts (and therefore in estimated EE) with increasing grade (and thus increasing energy cost) at a constant treadmill speed. Discrepancies in the count–V̇O₂ relationship have also been observed among different modes of activity. Examination of the data presented by Eston et al. (⁶) revealed differences in accelerometer counts for two of the conditions eliciting similar metabolic costs. Similarly, Fehling et al. (⁷) demonstrated much lower Tritrac counts for bench stepping compared with level walking at a similar V̇O₂, resulting in a significant underestimation of energy expenditure for bench stepping.

It thus appears that if the energy cost of an activity is related to muscular loading owing to isometric contractions, upper body movement, added weight bearing (carrying, lifting, pushing, etc.), or graded or soft surfaces, it will not be reflected by an increase in counts detected by an accelerometer. This limitation was proposed by Bouten et al. (⁴) in one of the first investigations of triaxial accelerometry. Interestingly, in the current results, the degree of underprediction of V̇O₂ during golf and lawn mowing, which involve continuous locomotion, was lower than the magnitude of underprediction during the other tasks. However, both golf and lawn mowing were still substantially underestimated, most likely because of the differences in terrain (grass surface and slight undulations in terrain) and load carriage (pulling clubs/pushing mower) compared with track walking. Furthermore, both monitors showed a similar degree of error, suggesting that a triaxial measurement does not improve the ability to capture EE based on motion at the hip. This is in agreement with Welk and Corbin (¹⁸), who found that uniaxial (Caltrac) and triaxial (Tritrac) monitors were highly correlated (r = 0.88) and showed similar associations with heart rate monitoring during 3 d of free-living monitoring in children.

The differences in the count–V̇O₂ relationship between different types of activities is further supported by the weakening of the regression equations when all activities were combined as opposed to those based on walking only (Table 5). Both the CSA and the Tritrac monitors were similarly affected, with correlations between counts and energy cost reduced to r = 0.59 and 0.62, respectively. Using a variety of different activities, including treadmill and “unregulated play,” Eston et al. (⁶) found stronger associations between EE and accelerometry (Tritrac, r = 0.91; CSA, r = 0.78); however, the range of energy cost was much larger (2.5–10.8 METs) compared with the range found in the current study (3.0–6.0 METs). Discrepancies in the prediction equations for different types of activities were not examined in that study (⁶).

Although validation studies tend to examine the accuracy of accelerometers in assessing energy cost for specific activities over brief time periods, the intended use of these monitors is for measuring physical activity and EE over extended time frames in free-living situations. This information is intended to refine our knowledge about the amount, type, and pattern of activity that are conducive to optimal health and disease prevention. Accelerometers are thought to be able to improve the assessment of physical activity because of their objectivity (compared with self-report methods, such as questionnaires or diaries) and their ease of use in the field. Because individuals may participate in a vast array of different types of movement and activity throughout the day, equations derived from a specific set of laboratory activities cannot necessarily be applied to free-living situations. The current study, along with some previous findings, suggest that the count versus EE relationship is specific to the activity being performed. This would be a limitation for daily free-living use and may lead to substantial misclassification of activity levels.

Several studies demonstrated relatively high correlations between different methods of assessment, including accelerometry, activity questionnaires, calorimetry, and doubly labeled water (^3,5,11). However, these studies generally compared estimates of daily EE as opposed to physical activity itself. Monitors such as the Caltrac and Tritrac provide an output of total EE that incorporates an estimate of resting EE. Because a large portion of the day is spent sleeping and in sedentary or light activity, these comparisons do not actually test the validity of the monitors in detecting physical activity but, rather, verify the accuracy of estimating basal metabolism. These correlations can also be increased by large interindividual variability in total EE (largely attributable to characteristics such as body size and age) more than by physical activity habits. For example, Fogelholm et al. (⁸) found low total error in estimating daily EE using a Tritrac in a group of overweight women; however, the accelerometer was not correlated to energy expended in physical activity (i.e., adjusted for resting EE). Sherman et al. (¹⁶) compared Tritrac measures with EE during rest and treadmill exercise, finding that although EE estimated by the Tritrac was highly related to actual EE (r = 0.96), the correlation with activity counts was much lower (r = 0.31). Similarly, high correlations were observed for EE estimated by the Caltrac monitor versus 24-h calorimetry in adolescent females, although the corresponding relationships between counts and actual EE were negligible (⁵). Furthermore, the equations used to convert activity counts to EE for the Tritrac are not made available by the company, and thus the derivation and validity of these equations have not been investigated.

Another confounder that may spuriously elevate correlations is the variability in activity levels within the monitoring period. This was suggested by Welk and Corbin (¹⁸), who found strong relationships between accelerometry and heart rate recording in a group of children during unstructured time periods involving a wide range of activities (such as during recess and after school). When activity was limited and thus more homogeneous (in the classroom), or when the entire day was considered, with a large proportion of the time spent in sedentary or low-level activity, correlations were lower.

Acknowledging the limitations of point estimates of EE based on accelerometry, a recent study proposed the use of “count cutoffs” to classify daily activity into intensity ranges (⁹). This method could then be used to determine the amount of time spent in the various categories in assessing multiple days of continuous activity recording by accelerometry. However, the current results implied that the protocol and type of activities used to determine the cutoffs corresponding to specific intensity levels may greatly affect the values, owing to the differences in the count–V̇O₂ relationship for different activities. Because the metabolic cost of the activities in this study ranged only between 3 and 6 METs, it is inappropriate to extrapolate the line to a cutoff for 9 METs, but even when only the “moderate” level is considered, discrepancies of several thousand counts were observed depending on which equation was used. This could potentially lead to substantial misclassification of activity levels precisely in the range of interest to the public health message concerning physical activity.

In conclusion, the results of this study suggest that the relationship of accelerometry to energy cost is highly dependent on the type of activity being performed. It may therefore be inappropriate to apply equations based on laboratory tasks or locomotion to free-living situations in attempts to determine physical activity EE or to classify activity levels. Accelerometers placed at the hip, either uniaxial or triaxial placement, appear unable to detect increased energy cost of upper body movement, load carriage, or changes in surface or terrain. These discrepancies in counts generated for a given intensity level may lead to substantial misclassification of activity habits. Therefore, information about the type of activities being performed is necessary to accurately assess EE. Results from several studies to date support the use of accelerometry to assess energy expended during locomotion; this may be applicable to longitudinal studies to monitor the duration or intensity of walking or running exercise. Future research may be able to determine the proportion of time spent in different modes of activity for a typical adult to assess the amount of error introduced by applying standard cut points to classify intensity levels. In addition, the effects of vehicular transportation should be examined as another source of confounding when accelerometers are used to record daily activity in free-living situations. It may be useful to combine accelerometry with questionnaire-based assessments to capture the various forms of activity that individuals may perform in daily life.

This research was funded by a grant from the International Life Sciences Institute Center for Health Promotion (ILSI CHP). The use of trade names and commercial sources in this document is for purposes of identification only and does not imply endorsement. In addition, the views expressed herein are those of the individual authors and/or their organizations and do not necessarily reflect those of ILSI CHP.