Accurate assessment of physical activity is critically important when examining the relationship between physical activity exposure and a number of health related outcomes (i.e., cardiovascular disease, hypertension, obesity). If the exposure variable, in this case physical activity, is not precisely characterizing physical activity behavior, the strength of its relationship to a given outcome is likely reduced or eliminated ⁽³⁾. Self report of physical activity is highly susceptible to inaccuracy and/or imprecision as these instruments are dependent on subjects ability to recall or report physical activity. In fact, if the self-report measure contains a set of specific closed ended questions, then the precision of the activity measure is also dependent on whether the questions capture all physical activity.

To eliminate many problems associated with self-report measures, it is necessary to explore alternative methods that do not rely on the subject's ability to recall activity or on the quality of the self-report measure. Activity monitoring using a motion sensor certainly has the potential to eliminate these problems. Several monitors have been tested and described in the literature including the Computer Science and Applications, Inc. (CSA) Model 5032 accelerometer ^(6,7,10). The design specifications of the first generation of this instrument were described by Redmond and Hegge ⁽¹¹⁾ who used the device to study rest and activity patterns of soldiers involved in sustained military operations.

Other activity monitors such as the single plane Caltrac accelerometer or the triaxial Tritrac accelerometer have been tested and validated in laboratory ^(5,8,10,12) and field^(6,7,9) settings and allow the user to express physical activity in caloric expenditure units. This feature is particularly attractive to researchers who are attempting to address issues related to the dose-response paradigm by examining how much activity is sufficient to elicit health and/or fitness benefits. Moreover, the rate of caloric expenditure (kcal·min^-1) allows the user to define activity intensity.

The CSA monitor does not provide the user with this type of information. There are two models of this device available. The original monitor (Model 5032) has previously been validated ^(6,10). The device is accurate and sensitive, battery life is excellent (about 6 months), and it is unobtrusive to subjects (6.6 × 4.3 × 2.5 cm, 70 g), very durable, and maintains calibration extremely well over time (± 5%). Another attractive feature is the capacity to store counts over small time periods permitting evaluation of patterns of activity within a day or over several days. Janz et al. ⁽⁶⁾ established exercise intensity based count cut-points using heart rates of 75, 130, and 150 beats·min^-1 to represent boundaries for sedentary, moderate, and vigorous activity for children. These heart rates corresponded to count ranges of 25-250, 251-499, and ≥ 500 cnts·min^-1 using the CSA Model 5032.

Currently it is not possible to define the appropriate CSA count cut-points that represent meaningful intensity categories for adults. In physical activity epidemiology, researchers are often interested in minutes of time spent in selected intensity categories that are operationally defined as light, moderate, hard, and very hard intensity physical activity. Thus, the primary purpose of this study was to define CSA accelerometer activity count categories for adults that correspond to different activity intensity levels. A secondary purpose was to develop and validate an equation to estimate caloric expenditure from activity counts in adults during walking and running.

METHODS

Subjects. Twenty-five males (mean ± SD age = 24.8 ± 4.2 yr, mass = 71.8 ± 7.9 kg, height = 177.6 ± 6.7 cm) and 25 females (age = 22.9 ± 3.8 yr, mass = 63.0 ± 7.5 kg, height = 166.1 ± 6.3 cm) volunteered as participants. Subjects read and signed an informed consent document in accordance with mandated university guidelines for use of human subjects.

Procedures. Subjects performed 6 min each of the following exercise conditions on a motorized treadmill: slow walking (4.8 km·h^-1), fast walking (6.4 km·h^-1), and jogging(9.7 km·h^-1). Each condition was separated by a 5-min rest period and the order of conditions was balanced across subjects. Treadmill speed was verified with the subject on the treadmill by applying a high precision tachometer (Biddle, Inc., Plymouth Meeting, PA) to the surface of the treadmill. Subjects were asked to refrain from exercise and caffeine consumption for 4 h before testing.

Oxygen consumption was measured minute-by-minute using open circuit spirometry with a computer-based data acquisition system. Subjects breathed through a Hans Rudolph (Kansas City, MO) high velocity 2-way nonrebreathing valve (Model 2700, dead space = 95 mL). Inspired air volume was measured with a calibrated dry gas meter (Rayfield Equipment, Waitsfield, VT). Expired air was directed through a 5-L mixing chamber and a dried gas sample was continuously analyzed for O₂ and CO₂ concentration using Ametek(Pittsburgh, PA) oxygen (Model S-3A1) and carbon dioxide (Model CD-3A) analyzers. The analyzers were calibrated before each test using verified gases of known concentration. Analog data from the analyzers and dry gas meter were converted to digital signals and transmitted to a Leading Edge (Model D2/LPS) personal computer equipped with a respiratory gas exchange software package(˙VO₂PLUS, Exeter Research, Exeter, NH). Steady-state parameters were calculated by averaging the final 3 min (minutes 4-6) of exercise for each condition. METs were calculated by dividing the steady-state˙VO₂ by 3.5 mL·kg^-1·min^-1.

The Computer Science and Applications Model 7164 is a smaller version (5.1× 4.1 × 1.5 cm, 43 g) of the Model 5032. It is a uniaxial accelerometer that assesses accelerations ranging from 0.05-2.0 G and is band limited with a frequency response from 0.25-2.5 Hz ⁽⁴⁾. These parameters detect normal body motion and filter out high frequency movement such as vibrations ⁽⁴⁾. The acceleration signal is filtered by an analog bandpass filter and digitized by an 8 bit A/D converter at a sampling rate of 10 samples per second⁽⁴⁾. Each digitized signal is summed over a user specified time interval (epoch), and at the end of each epoch the activity count is stored internally and the accumulator is reset to zero. In the current study, a 60-s epoch was used and activity counts were expressed as the average counts per minute over the 6 min of exercise.

The CSA accelerometer was firmly secured to a belt and positioned on the right hip. The accelerometer was initialized before each exercise session according to manufacturer specifications. An external timepiece was used to synchronize the accelerometer internal clock with the ˙VO₂PLUS software program.

Statistical analysis. A one-way repeated measures ANOVA was used to examine the condition effect (exercise intensity) for all dependent variables. Linear regression was used to establish the relationship between metabolic cost and counts. To estimate caloric expenditure, a multiple linear regression approach was used and cross validated on a holdout sample. A probability level of P < 0.05 was used to establish statistical significance.

RESULTS

Figures 1 and 2 illustrate the metabolic and CSA count data for males and females separately. Since no gender differences were observed, subsequent analyses were performed with the total sample.Table 1 presents the pooled metabolic and activity count data. A one-way repeated measures ANOVA revealed a significant condition effect for all dependent variables. Between 4.8 and 6.4 km·h^-1, the average count difference was 2191 cnts·min^-1 and between 6.4 and 9.7 km·h^-1, the average count difference was 4554 cnts·min^-1. The range of METs represented by the three walking and jogging conditions was from 3.7-9.7 METs.

Figure 3 illustrates the relationship between cnts·min^-1 and METs. The relationship is linear (r = 0.88) with an increase in the variability about the regression line evident at 7 METs and above in comparison to the count responses observed from 3 to 6 METs. No data points are available for the 6.5-8 MET range since these intensities represent the walk-run transition interval.

The following regression equation for estimating METs from cnts·min^-1 was used to establish count ranges corresponding to MET level categories typically used in the literature to define light (≤ 2.99 METs), moderate (3.0-5.99 METs), hard (6.0-8.99 METs), and very hard activity (≥ 9.0 METs) (Table 2): METs = 1.439008 +(0.000795 * cnts·min^-1) (r² = 0.82; SEE = ± 1.12 METs). The count range corresponding to each intensity was determined by solving the rearranged regression equation for counts and inserting the lower and upper limits for METs. For example, to set the upper boundary for moderate activity (5.99 METs) the following equation was solved: These data provide a simple template to convert counts-perminute ranges to activity intensity categories.

In certain cases it may be desirable to obtain an estimate of energy expenditure from counts. We developed a regression equation on a sample of 35 subjects randomly selected from the original sample. The equation was: kcal·min^-1 = (0.00094 * cnts·min^-1) + (0.1346 * mass in kg) - 7.37418 (r² = 0.82, SEE = ± 1.40 kcal·min^-1). The equation was subsequently cross validated on the holdout sample of 15 subjects. The cross-validation results are presented in Table 3. Mean differences between actual and predicted kcal min were: -0.19, -0.46, and 0.12 kcal·min^-1 for 4.8, 6.4, and 9.7 km·h^-1, respectively (P > 0.05).Figure 4 illustrates the individual data for the cross validation. Across all speeds, the correlation between actual and predicted energy expenditure was r = 0.93 (SEE = ±0.93 kcal·min^-1).

Three days of CSA accelerometer data were collected on one subject to demonstrate the application of this calibration procedure in a field setting. The monitor was worn on the right hip during nonsleep time for three consecutive days. The subject maintained an activity log similar to the instrument described by Bouchard et al. ⁽²⁾. Each hour of the day, time spent in inactivity, and various levels of activity was recorded to the nearest 0.25 h. The activity log data were then transposed into MET equivalents and expressed as the number of minutes spent in inactive/light(<3 METs), moderate (3-5.99 METs), hard (6.0-8.99 METs), and very hard(≥9 METs) activities. These data were then compared to minutes spent in comparable activity categories using the CSA accelerometer calibration data applied to the CSA field data averaged over 15-min time blocks. The results are presented in Table 4. Based on the activity log, 83-97% of each day was spent in light activity (<3 METs) and moderate and above intensity activity occurred for 9-17% of the day (30-150 min). According to the CSA monitor, 84-96% of each day was spent in light activities, and 4-16% of the activity was in at least the moderate intensity category (45-135 min). Although the absolute amount of time spent in activity categories differed between methods, both assessment techniques ranked the days similarly. Most moderate and above intensity activity occurred on Day 3 and the least amount of minutes of moderate and above intensity activity occurred on Day 2.

DISCUSSION

The primary aim of this study was to develop CSA accelerometer count cut-points that corresponded to different intensities of exercise using commonly employed MET categories. There appears to be adequate discrimination between count ranges to discern different intensities of exercise. These data suggest that 1258 counts corresponds to a 1 MET change between 3 and 9 METs using 1951 counts as the “baseline” to define the 3 MET level.

This information provides the user with physiologically valid count ranges that correspond to standardized activity intensity categories. These data can be used to establish time periods that a subject is engaged in different intensities of activity within the course of a day or over several days. Health outcomes positively associated with physical activity exposure generally must fall into at least the moderate intensity level. The CSA accelerometer may be used to determine both the quality and quantity of activity using the 1951 cnts·min^-1 lower boundary to define the minimal intensity level necessary for health benefits.

The equation to predict energy expenditure from counts and body mass functions reasonably well during treadmill walking and jogging. Caution should be exercised when using this equation as the point estimates of caloric expenditure because it was developed only using treadmill walking and running as the modes of exercise. It is recommended that this equation be tested during overground locomotion where the relationship between energy expenditure and counts may be different.

These results are based on a laboratory investigation using treadmill walking and running as the modes of exercise. Although it is likely that a large proportion of an adult's daily activity is spent in locomotion⁽¹⁾, it is not known how well the results from the present investigation apply to a field situation. Studies employing doubly labeled water, activity logs, or direct observation as criterion measures to estimate total daily energy expenditure are needed to further assess the validity of the CSA accelerometer. A recent study by Welk and Corbin⁽¹³⁾ that used direct observation as the criterion measure of activity, reported that the Tritrac accelerometer provides an accurate assessment of time spent in both sedentary and active activity in 10- to 12-yr-old boys. Although a similar investigation has not been conducted with the CSA accelerometer, the authors suggest that similar results would likely be seen with other types of activity monitors⁽¹³⁾. Janz et al. ⁽⁶⁾ reported that activity counts increased with an increase in heart rate in children, suggesting that motion as measured with the CSA accelerometer (Model 5032) paralleled a physiologically based marker of exercise intensity.

Until now it was not possible to define the appropriate count cut-points that represent exercise intensity categories for adults using the CSA monitor. In the field of physical activity epidemiology, researchers are often concerned about the number of minutes individuals spend in selected intensity categories that are operationally defined as light, moderate, hard, and very hard activity. Also of interest is the sustained minutes in moderate activity(i.e., 20 min or more) to determine whether recommended physical activity objectives are satisfied. Results from this study suggest that the CSA monitor may be used to examine patterns of physical activity to evaluate both the quantity and quality of physical activity. The device is capable of collecting information over several days and may provide a more precise profile of the quality of physical activity. This added feature of enhanced precision in characterizing intensity level with a motion sensor is likely to improve the accuracy of the measure of activity exposure. As discussed by Casperson⁽³⁾, the result may be a stronger association between risk factors and/or health outcomes and physical activity.