Calorimetric validation of the Caltrac accelerometer during level walking.

Calorimetric Validation of the Caltrac [R] Accelerometer During Level Walking Over the last 20 years, there has been a considerable interest in the assessment of physical activity levels and how activity levels relate to cardiovascular fitness and health. [1-3] A variety of methods have been used in measuring physical activity levels including self-reports by questionnaires and interviews, direct observation of physical activities, monitoring of heart rate by telemetry, Calorimetric measurement of oxygen uptake ([Vo.sub.2]), isotope ratio mass spectrometry, and the use of motion activity sensors. Of all of these methods, the motion sensors are reasonably nonobstructive, have the advantage of being objective, and are the most cost-effective for clinical use. [3-6]

In recent years, may different types of mechanical and electronic motion sensors have been manufactured and used in epidemiological studies to estimate physical activity levels. [3-6] The Caltrac [R] accelerometer (*1) is currently receiving increased scientific interest because of the manufacturer's claim that it is a valid predictor of caloric expenditure, or energy expenditure (Ee), for a variety of physical activities. [7-9] The Caltrac [R] device is a single-plane accelerometer that is designed to monitor the up (acceleration)-and-down (deceleration) movements of the body. The movement is converted into electrical signals and displayed as digital readout representing a level of physical activity. [7,8] The Caltrac [R] accelerometer can accumulate up to 19,999 counts, which is of sufficient capacity to last for approximately 24 hours. [8] The device uses two lithium coin cell batteries, which can last about 1,000 hours of continuous use. A major limitation of the Caltrac [R] accelerometer is that the device does not respond to isometric (static) forms of exercise and activities in which the body weight is partially supported, as in bicycling or rowing. [9]

Because of tis light weight and low cost, the Caltrac [R] accelerometer has the potential to be useful in physical therapy practice to quantify the level of activity at work or during leisure activities. [10] More specifically, it can be used to monitor patients' Ee during aerobic physical activities. It may be useful in Bariatric, cardiopulmonary rehabilitation, and pain management programs. Some skepticism, however, has been expressed concerning the validity of the Caltrac [R] accelerometer in measuring Ee. [11] This study was designed to compare the Caltrac [R] accelerometer output with measured Ee.

Review of the Literature

Previous studies have shown that the Caltrac [R] accelerometer is reliable [9, 10, 12] and sensitive to changes in walking speed. [8-10, 12] To date, however, few studies have evaluated the validity of the Caltrac [R] accelerometer. Wong and co-workers validated the Caltrac [R] accelerometer output against the calorimetric measurement of [Vo.sub.2]. [7] Fifteen subjects walked or ran on a level, motor-driven treadmill for three minutes each at 2, 3, 4, 6, and 8 mph. They reported "a roughly linear increase" in accelerometer reading with increased [Vo.sub.2], but the strength of the relationship was not documented. In another study, Servais et al found a consistent association between daily Caltrac [R] accelerometer output and the level or physical activity. [8] The Ee in the study was not measured but instead derived from energy-cost tables. These tables represent only estimates of Ee and are subject to considerable error. [9]

In 1983, Montoye and associates also validated the Caltrac [R] accelerometer readings against the criterion of [Vo.sub.2]. [9] The subjects (N = 21) performed 14 different physical activities, each lasting four minutes. The activities involved 1) walking or running on the treadmill at different speeds (2, 4, and 6 mph) and grades (0%, 6%, and 12%), 2) half-knee bends (28 and 48 [bends.min.sup.-1]), and 3) floor touches while bending at the knees (24 and 36 [touches.min.sup.-1]). They found that the Caltrac [R] accelerometer output did not reflect the increased Ee when the grade was increased during the walking or running activities. The combined (pooled) data for all the activities revealed a moderate correlation of .74, but a wide standard error of estimate (SEE) of 6.6 mL [0.sub.2.kg.sup.-1.min.sup.-1]. They derived linear regression equations to predict [Vo.sub.2] ([mL.kg.sup.-1.min.sup.-1]) from the accelerometer output. Their regression equation is of limited application in the clinical setting because patients following an exercise program are usually more interested in their Ee than the amount of oxygen they consume. Therefore, it is of clinical relevance to derive a regression equation to predict Ee [(kcal.min.sup.-1]) from the accelerometer output.

In a recent report, Montoye et al found that the Caltrac [R] accelerometer during bench-stepping and half-knee-bending activities estimates Ee "about as well as the force plate measurements." [13] The correlation coefficient between the Caltrac [R] accelerometer readings and the [Vo.sub.2] measurements was not reported, but the scatter diagrams plotted indicated a weak relationship.

In 1985, Klesges and associates validated the accelerometer reaadings against the criterion of observed physical activity behavior. [14 Fifty adults and 30 preschool children were observed for one hour while engaged in free-play activities. The investigators recorded the subjects' physical activities using the Fargo Activity Time Sampling Survey Observational System. They reported moderate (r = .69) and low (r = .35) correlation coefficients between the Caltrac [R] accelerometer output and the observed physical activity levels for the adults and the children, respectively. They concluded that the Caltrac [R] accelerometer may not adequately assess children's physical activity level. In a follow-up study, Klesges and Klesges validated the Caltrac [R] accelerometer output against observed physical activity levels of children in their natural environment over a nine-hour duration. [15] A moderate correlation (r = .54) was obtained between the accelerometer output and the observed physical activity level. They attributed this finding to the longer asessment period used, and concluded that the Caltrac [R] accelerometer may be inappropriate for measuring the level of physical activity of children.

We found a total of six validation studies on the Caltrac [R] accelerometer in the literature; four of the studies used [Vo.sub.2] as the criterion variable, and the remaining two studies used direct observation as the criterion measure. The direct observation method is subjective and too time-consuming for clinical use. [3-6] In contrast, the calorimetric measurement of [Vo.sup.2] is highly reproducible and a valid measure of Ee. [16] New methods of measuring physical activity, therefore, are often validated against [Vo.sub.2]. [1-6,17] In laboratory and field studies, [Vo.sub.2] during physical activities is measured by the indirect calorimetric method. This method involves the collection of the expired gas using a Douglas bag or Kofranyi-Michaelis (Max Planck) respirometer. [16] At steady state, aliquot samples of the expired gas are withdrawn and analyzed for oxygen and carbon dioxide concentration, which are subsequently used in the computation of [Vo.sub.2].

Statement of the Problem

From a clinical perspective, the validity of an instrument is indeed important. The validity construct answers the question, Does the instrument measure what is purports to measure? In this case, does the Caltrac [R] accelerometer really measure Ee? Although the reproducibility of the Caltrac [R] accelerometer output is not in doubt, [9, 10, 13] the validity of the accelerometer in measuring Ee is unclear currently. [11] For the Caltrac [R] accelerometer to have a wider clinical application, further studies validating the device output against the carorimetric method are warranted.

The purpose of this study were to

1. Determine the relationship between the Caltrac [R] accelerometer output and the calorimetric measurements of [Vo.sub.2] during level walking.

2. Determine whether differences exist between the Caltrac [R] accelerometer output ([count.min.sup.-1]) and the measured Ee ([kcal.min.sup.-1]) at different work loads.

3. Derive regression equations to predict Ee from the Caltrac [R] accelerometer output, if differences were found between the Caltrac [R] readings and the measured Ee.

We expected that 1) a highly significan correlation would exist between the Caltrac [R] accelerometer readings and the [Vo.sub.2] measurements, 2) no differences would exist between the Caltrac [R] output and the measured Ee, and 3) Ee can reliably be predicted (p [is less than] .05) from the accelerometer output.

Method

Subjects

Twenty-five healthy volunteers (10 men, 15 women), between the ages of 18 and 38 yearS, from the University of Florida participated in this study. On the average, the subjects' age was 24.7 [+ or -] 5.4 years, their height was 172.7 [+ or -] 9.1 cm, and their body weight was 67.4 [+ or -] 126 kg. Most of the subjects were involved in recreational activities, but none was an elite athlete. the research protocol was approved by the Institutional Review Board of the University of Florida. The medical history of all potential subjects was reviewed, and those with cardio-respiratory and metabolic disorders were excluded from the study. Before testing, the subjects were informed of the testing procedure and risks of the study, and each subject signed an informed consent form.

Experimental Design

A quasi-experimental protocol where each subject acted as his or her own control was used. [18] Specifically, each subject walked on a level (0% grade), motor-driven treadmill at four different speeds (54, 81, 104, and 130 [m.min.sup.-1]). Each of the four experimental trials lasted eight minutes, and all testing was completed within one hour. The subjects' [Vo.sub.2] and accelerometer output were monitored during each trial. We selected these walking speeds because they represent the range of speed adopted during most recreational activities, and we used an eight-minute exercise duration to ensure that the subjects (irrespective of their physical fitness status) attained a steady state during the exercise bouts. [19]

We did not include graded walking in the experimental design because previous studies have shown that the Caltrac [R] accelerometer tends to underestimate Ee during graded walking and running physical activities. [9, 12] The room temperature during the testing was maintained between 20[degrees] and 22[degrees]C. The relative humidity and barometric pressure were maintained between 52% and 56% and 758 and 766 mm Hg, respectively.

Procedure

On arrival in the laboratory, the subjects' weight and height were measured. These data and other personal data such as age and sex were programmed into the Caltrac [R] accelerometer. The subject were then introduced to the laboratory equipment and were allowed to practice walking on the treadmill until they became confident. Subsequently, the Caltrac [R] accelerometer was affixed to the waist belt at the lumbar region, and the subjects were progressed to walking on the treadmill. The four experimental conditions (ie, 54-, 81-, 104-, and [130-m.min.sup.-1] walking speeds) were presented in sequential order starting with the [54-m.min.sup.-1] walking speed. Each of the testing conditions lasted eight minutes, and adequate rest periods of about five minutes' duration were allowed between exercise bouts. The testing was resumed when the subjects re-attained their pre-exercise resting heart rate.

During the test, [Vo.sub.2] (in [L.min.sup.-1] and in [mL.kg.sup.-1][.min.sup.-1]) and nonprotein respiratory exchange ratio (RER) were monitored by the Beckman Horizon metabolic cart. (*2) For all the subjects, the metabolic cart's gas analyzers were calibrated against standard gas mixtures shortly before the initial exercise bout and between the second and the third exercise bouts. The [Vo.sub.2] and RER were monitored every 30 seconds, but only the values at steady state (ie, 6th-8th minutes of exercise) were used for data analysis. The Caltrac [R] accelerometer output was monitored every two minutes during the testing, but only the total readings at the end of the eighth minute were used for data analysis. All subjects completed the four experimental trials; however, two of the subject jogged instead of walked during the fastest speed (130 [m.min.sup.-1]).

Computation

The mean [Vo.sub.2] ([L.min.sup.-1] and [mL.kg.sup.-1][.min.sup.-1]) and RER from sixth through eighth minutes of exercise was computed. The Caltrac [R] accelerometer output at the end of each exercise trial was divided by 8 to convert the readings to [counts.min.sup.-1]. The [Vo.sub.2] ([L.min.sup.-1]) was converted to a caloric value ([kcal.min.sup.-1]) as described by McArdle et al [16]:

Ee ([kcal.min.sup.-1]) = [[Vo.sub.2] ([L.min.sup.-1]) X thermal equivalent of oxygen for RER]

The thermal equivalents of oxygen for the different RERs were obtained from standard nomograms. [16]

Data Analysis

We computed the Pearson product-moment correlation coefficient (r) for the combined (pooled) data to determine the relationship between the accelerometer output and the measured [Vo.sub.2] ([mL.kg.sup.-1][.min.sup.-1]) and Ee ([kcal.min.sup.-1]). A paired t test was used to determine significant differences between the Caltrac [R] accelerometer output and the Ee at different walking speeds. The paired t-test statistical procedure was two-tailed at an alpha level of .05. Both linear and polynomial regression models were applied to establish a predictive equation between the accelerometer output and the criteria variables (ie, [Vo.sub.2] and Ee). We determined the predictability of the regression equations with the analysis of variance (ANOVA) procedure. Multipe and stepwise regression models were applied to determine the contribution of age, body weight, and height to the prediction of the criteria variables. The stepwise regression procedure selects the predictor variables in the order of their relatives strengths in predicting the criteria variable. The tolerance level was set at a probability of .01 (F [is greater than or equal to] 4.00). These statistical tests were performed on a Macintosh Plus micro-computer (*3) using the StatView 512+ statistical package. (*4)

Results

Figure 1 is a scatter diagram illustrating the relationship between the Caltrac [R] accelerometer output and the measured [Vo.sub.2] ([mL.kg.sup.-1][.min.sup.-1]). A positive correlation (r = .76, p [is less than] .001) was found between the accelerometer readings and the [Vo.sub.2]. The relationship between the Caltrac [R] accelerometer output and the measured Ee is shown in Figure 2. The scatter plot revealed a positive linear relationship (r = .91, p [is less than] .0001) between the accelerometer output and the Ee.

The results of the paired t test used to determine differences between the accelerometer output and the measured Ee are summarized in Table 1. The analysis revealed that the Caltrac [R] accelerometer output was significantly higher (p [is less than] .001) than the measured Ee at the different walking speeds. The difference between the Caltrac [R] output and the measured Ee range from 13.3% to 52.9%. Because the Caltrac [R] accelerometer overestimated Ee (Fig. 3), it was necessary to adjust for the measurement error through the use of a regression equation. The results of the regression analyses are presented below.

The correlation coefficient, coefficients of determination ([R.sup.2]), SEEs, and F ratios of the ANOVA for the different regression equations are summarized in Table 2. The linear and polynomial regression equations to predict [VO.sub.2] ([mL.kg.sup.-1.min.sup.-1]) from the accelerometer output (X) are as follows:

[VO.sub.2] ([mL.kg.sup.-1.min.sup.-1]) = 5.39 + 1.45 [X ([counts.min.sup.-1])] [VO.sub.2] ([mL.kg.sup.-1.min.sup.-1]) = 0.35 + 3.06 [X ([counts.min.sup.-1])] - 0.11 [[X.sup.2] ([counts.min.sup.-1])]

The ANOVA for both linear and polynomial regression models (Tab. 2) revealed that [VO.sub.2] ([mL.kg.sup.-1.min.sup.-1]) can reliably be predicted (p [is less than] .0001) from the accelerometer output. The quadratic equation (equation 3), however, was a better (p [is less than] .01) predictor of [VO.sub.2] than was the linear equation (equation 2). The regression curve for the quadratic equation is plotted in Figure 4.

The linear and polynomial regression equations describing the relationship between the measured Ee and the acceleromeer output (X) are as follows:

Ee ([kcal.min.sup.-1]) = 0.83 + 0.63 [X ([counts.min.sup.-1])] Ee ([kcal.min.sup.-1]) = 0.56 + 0.71 [X ([counts.min.sup.-1])] - 0.006 [[X.sup.2 ([counts.min.sup.-1])]

The results of the ANOVA for both models (Tab. 2) revealed that Ee can reliably be predicted (p [is less than] .0001) from the accelerometer output. The quadratic equation (equation 5) was not a better predictor (p [is greater than] .05) of Ee than was the linear equation (equation 4).

The addition of age, body weight, and height into an equation containing the accelerometer output (X) significantly increased (p [is less than] .05) the accuracy of prediction of [VO.sub.2] (Tab. 3). The multiple regression equation is as follows:

[VO.sub.2] ([mL.kg.sup.-1.min.sup.-1]) = 12.61 + 1.81 [X ([counts.min.sup.-1])] + 0.011 [age (yr)] - 0.243 [weight (kg)] + 0.095 [height (in) (*5)]

The results of the stepwise regression analysis for [VO.sub.2 ([mL.kg.sup.-1.min.sup.-1]) are summarized in Table 3. The analysis revealed that the best predictors of [VO.sub.2] were the accelerometer output and body weight. They accounted for 58.3% and 17.9% of the total variance (76.4%) in [VO.sub.2], respectively.

The multiple regression equation for the measured Ee is as follows:

Ee ([kcal.min.sup.-1]) = -1.32 + 064 [X (counts*min.sub.-1)] + 0.008 [age (yr)] - 0.019 [weight (kg) + 0.046 [height (in)]

The stepwise regression analysis of the measured Ee data revealed that the only viable predictor of Ee was the accelerometer output. The contribution of age, body weight, and height toward the prediction of Ee was minimal. The accelerometer output accounted for 83.8% of the total variance (84.2%) in Ee (Tab. 2).

Discussion

This study was designed to compared the Caltrac [R] accelerometer output with measured Ee. We obtained a moderate correlation of .76 (p [is less than] .001) between the accelerometer readings and the measured [VO.sub.2] ([mL.kg.sup.-1.min.sup.-1]) and a high correlation of .92 (p [is less than] .0001) between the accelerometer output and the calculated Ee. These findings supported our research hypothesis. The correlation coefficient that we obtained between the [VO.sub.2] and the accelerometer output (r = .76) is consistent with the value (r = .74) reported by Montoye et al under similar laboratory testing conditions. [9] The SEE of 3.8 [mL.kg.sup.-1.min.sup.-1] that we obtained for the linear regression equation (equation 3), however, is much lower than the 6.6 [mL.kg.sup.-1.min.sup.-1] reported by Montoye and associates. [9] The higher SEE obtained in the study of Montoye et al may partially be attributed to the variety of physical activities included in their design. Some of the activities (ie, half-knee bending and floor touches while bending at the knees) are predominantly isometric forms of exercise, which single-plane accelerometers cannot measure. [10,11] Because previous validation studies [7-9,13-15] did not derive regression equations to predict Ee, further comparisons cannot be made.

The Caltrac [R] operational manual indicates that the accelerometer "has been customized to measure the calories your body needs based upon your weight, height, age, sex, caloric intake and activity level." [20] Based on this information, we did not expect to find a significant difference between the accelerometer output and the measured Ee. Contrary to our expectation, we found that the Caltrac [R] accelerometer significantly overestimated Ee (p [is less than] .01) during level walking (Tab. 1). We used the Student's t test statistical procedure to determine whether there was a significant difference in the accelerometer output and the measured [VO.sub.2] between the male and female subjects. The results of the analysis revealed that the accelerometer output (Tab. 4) and the energy expended (Tab. 5) by the male subjects were significantly higher (p [is less than] .001) than the female subjects at the different walking speeds. An analysis of the subjects' physical characteristics data revealed that the male subjects were heavier (t = 5.6, p [is less than] .0001), taller (t = 3.4, p [is less than] .01), and older (t = 3.5, p [is less than] .01) than the female subjects. The higher accelerometer and caloric outputs by the male subjects, therefore, may be attributed to their greater body mass. The results of the Student's t test for [VO.sub.2, expressed as [mL.kg.sup.-1.min.sup.-1] to eliminate the effects of body weight, showed no significant difference (p [is greater than] .05) between the male f and female subjects (Tab. 6), thus supporting our speculation.

We used different regression models in an attempt to obtain the most accurate regression formula for the [VO.sub.2] and Ee criteria variables. When several models are significant, there are no universally accepted rules for selecting the best regression equation. [21] Apparently, clinicians must decide which equation best meets their needs taking into consideration the time and cost of measuring the predictor variables. For example, the addition of age, body weight, and height to the [VO.sub.2] linear regression equation containing the accelerometer output increased the coefficient of determination from .583 to .764 (31%) and decreased the SEE from 3.78 to 2.88 (24%) (Tab. 2). The ease with which personal data such as age. body weight, and height could be measured and the significant increase (p [is less than] .05) in the predictive accuracy of the regression support the use of the multiple regression equation when an accurate estimate of [VO.sub.2] is sought from the accelerometer output. We found no significant difference in the predictive accuracy between linear, polynomial, and multiple regression formulas for Ee. Because of its simplicity, it will be more appropriate to use the linear regression formula to predict Ee from the accelerometer output.

The results of the stepwise regression analyses revealed that the best predictors of ([VO.sub.2] were the accelerometer output and body weight. The accelerometer output was the only viable predictor of Ee (kcal*[min.sup.-1]). These findings suggest that the Caltrac [R] accelerometer output is a valid predictor of Ee during level walking when the appropriate regression equation is used.

Limitations of Study

The regression equation derived in this study can be used by adults between 18 and 38 years of age to predict their Ee while walking on a level terrain with the accelerometer device affixed to the lumbar region. The regression equation is not valid while walking on a slope during running activities. The equation may not be applicable when the Caltrac [R] device is affixed to other body locations. In a recent study, Balogun et al obtained significantly higher (p [is less than] .05) accelerometer output when the Caltrac [R] device was worn at the waistline as compared with when it was affixed to the chest line. [12] It is currently unclear whether the accelerometer output will be affected significantly if the device is affixed to a different location on the waistline. This problem warrants investigation because some clients prefer to wear the Catrac [R] device at the back of the waistline, whereas other clients prefer to wear it at the sides or at the abdominal region of the waistline.

We obtained a high correlation (r = .87) for the [VO.sub.2 multiple regression equation. The SEE for the regression equation, however, was 2.88 [mL*kg.sup.-1]. The SEE indicates that considerable variability remains, despite the srong predictive power of the equation. Even though the SEEs obtained in this investigation for the regression equations (Tab. 2) are much lower than the 6.6 [mL*kg.sup.-1]*[min.sup.-1] reported by the manufacturers of the Caltrac [R] device, [9] the high SEE must be appreciated by physical therapists using the accelerometer to predict [VO.sub.2] during level walking.

Clinical Implications

The Caltrac [R] accelerometer output can be used by the physical therapist to objectively monitor the patients' relative level of activity during aerobic exercise level-walking programs. Because the Caltrac [R] device tends to overestimate Ee (Fig. 3), the accelerometer output may create a false impression on the amount of calories expended during physical activities. The Caltrac [R] accelerometer output, therefore, should be applied with caution in making clinical decisions on patients' Ee. Patients using the Caltrac [R] device should be informed about the predictive accuracy of the accelerometer output. The physical therapist may educate the patients on the possible use of the regression equation when an accurate estimate of the Ee is needed. It is necessary to apply the regression equation in weight-loss programs because the feedback provided by the accelerometer serves as an impetus for a positive change in behavioral modification of the patients.

Conclusion

Our study revealed a moderate correlation of .76 between the accelerometer output and the measured [VO.sub.2] ([ml*kg.sup.-1*min.sup.-1]) and a high correlation of .92 between the measured Ee and the accelerometer readings. The Caltrac [R] accelerometer output was significantly higher (p [is less than] .01) than the measured Ee at the different walking speeds. We found that the best predictors of [VO.sub.2] were the accelerometer output and body weight, although the accelerometer output was the only viable predictor of Ee. Our findings revealed that the Caltrac [R] accelerometer output is a valid predictor of Ee during level walking when the appropriate regression equation is used. Because the accelerometer device tends to overestimate Ee, the raw accelerometer readings should be used with caution.

Acknowledgments

We thank Richard R. Gutekunst, PhD, Dean College of Health Related Professions, University of Florida, for his assistance in processing the research grant. Also, we would like to acknowledge the cooperation of the physical therapy students at the University of Florida.

(*1) Hemokinetics, Inc, 2923 Osmundsen Rd, Madison, WI 53711.

(*2) SensorMedics Corp, 1630 S State College Blvd, Anaheim, CA 92806.

(*3) apple Computer, Inc, 20525 Mariani Ave, Cupertino, CA 95014.

(*4) Brain Power, Inc, 24009 Ventura Blvd, Suite 250, Calabasas, CA 91302.

(*5) 1 in = 2.54 cm.

References

[1] LaPorte RE, Kuller LH, Kupfer DJ, et al: An objective measure of physical activity for epidemiological research. Am J Epidemiol 109: 154-167, 1979

[2] Montoye HJ, Taylor HL: Measurement of physical activity in population: A review. Human Biol 56:195-216, 1984

[3] Saris WHM: Habitual physical activity in children: Methodology and findings in health and disease. Med Sci Sports Exerc 18:523-263, 1986

[4] LaPorte RE, Black-Sander R, Caauley JA, et al: The epidemiology of physical activity in children, college students, middle-aged men, menopausal females and monkeys. J Chronic Dis 75:787-795, 1982

[LaPorte RE, Adams LL, Savage DD, et al: The spectrum of physical activity, cardiovascular disease and health: An epidemiologic perspective. Am J Epidemiol 120:507-517, 1984

[6] LaPorte RE, Montoye HJ, Casperson CJ: Assessment of physical activity in epidemiologic research: Problems and prospects. Public Health Rep 100:131-146, 1985.

[7] Wong TC, Webster JG, Monteye HJ, et al: Portable accelerometer device for measuring human energy expenditure. IEEE Trans Biomed Eng 6:467-471, 1981

[8] Servais SB, Webster JG, Montoye HJ: Estimating human energy expenditure using an accelerometer device. IEEE Frontiers of Engineering in Health Care 8:309-312, 1982

[9] Montoye HJ, Washburn R, Servais SB, et al: Estimationof energy expenditure by a portable accelerometer. Med Sci Sports Exerc 15: 403-407, 1983

[10] Bologun JA, Farina NT, Fay E, et al: Energy cost determination using a portable accelerometer. Phys Ther 66:1102-1107, 1986

[11] Smith LK: Commentary. Phys Ther 66: 1107-1108, 1986

[12] Balogun JA, Amusa LO, Onyewadume IU: Factors affecting Caltrac [R] and Calcount [R] accelerometer output Phys Ther 68:1500-1504, 1988

[13] Montoye HJ, Servais SB, Webster JG: Estimation of energy expenditure from a force platform and an accelerometer. In: Proceedings of the Eighth Commonwealth and International Conference on Sports, Physical Education, Dance, Recreation and Health. London, England, 1986, pp 375-380.

[14] Klesges RC, Kiesges LM, Swenson AM, et al: A validation of two motion sensors in the prediction of child and adult physical activity levels. Am J Epidemiol 122:400-410, 1985

[15] Klesges LM, Klesges RC: The assessment of children's physical activity: A comparison of methods. Med Sci Sports Exerc 19:511-517, 1987

[16] McArdle WD, Katch FI, Katch VL: Exercise Physiology: Energy, Nutrition and Human Performance. Philadelphia, PA, Lea & Febigerr, 1981

[17] Schoeller DA, Ravussin E, Schutz Y, et al: Energy expenditure by doubly labeled water: Validation in humans and proposed calculation. Am J Physiol 250:823-830, 1986

[18] Campbell DT, Stanley JC: Experimental and Quasi-Experimental Designs for Research. Boston, MA, Houghton Mifflin Co, 1966

[19] Bubb WJ, Martin AD, Howley ET: Predicting oxygen uptake during level walking at speeds of 80-130 m/min. Journal of Cardiopulmonary Rehabilitation 5:462-465, 1985

[20] Caltrac [R]: Personal Activity Computer, Instruction Manual. Madison, WI, Caltronics Div of Hemokinetics, Inc, 1984

[21] Martin D, Acker JE: Predicting aerobic capacity during the modified Naughton treadmill protocol in patients with coronary artery disease. Journal of Cardiopulmonary Rehabilitation 8:297-302, 1988

J Balogun, Ph D, is Senior Lecturer, DEpartment of Medical Rehabilitation, FAculty of Health Sciences, Obafemi Awolowo University , Ile-Ife, Oyo State, Nigeria, West Africa. He was Visiting Research Scholar, Department of Physical Therapy, University of Florida, when this study was completed.

D Martin, PhD, is Assistant Professor, Department of Physical Therapy, Health Science Center, PO Box J-154, University of Florida, Gainesville, Fl 32610 (USA). Address correspondence to Dr. Martin

M Clendenin, PhD, PT, is Professor and Chairman, Department of Physical Therapy, Health Science Center, University of Florida.

This research was supported by a grant from the Division of Sponsored Research, University of Florida.

This article was submitted September 20, 1988; was with the authors for revision for 10 weeks; and was accepted February 23, 1989.

Physical Therapy/Volume 69, Number 6/June 1989