Objective: To examine the reliability and validity of the SenseWear Pro 2 Armband (SWA; Body Media, Pittsburgh, PA) during rest and exercise compared with indirect calorimetry (IC) in obese individuals.
Research Methods and Procedures: Energy expenditure was assessed during rest with the SWA and IC in 142 obese adults (37 men and 105 women, BMI = 42.3 ± 7.0) and in 25 lean and overweight adults (BMI = 25.3 ± 3.2) who were used as a comparison group. Twenty-nine of the obese adults also participated in three separate short exercise sessions including cycle ergometry, stair stepping, and treadmill walking.
Results: The repeatability of SWA estimates in obese subjects was high (r = 0.88, p < 0.001). The SWA generally underestimated the resting energy expenditure (REE) (1811 ± 346 vs. 1880 ± 382 kcal/d) and highly overestimated the energy expenditure during the exercise sessions in obese individuals. REE estimations by SWA were significantly correlated with fat-free mass (r = 0.88, p < 0.001). Bland-Altman plots based statistical analysis for the estimated REE, and measured IC showed a low agreement (Total Error > 20% but Systematic Error < 5%) between the two methods in obese subjects, although they showed a high correlation and a very good agreement in lean and overweight patients.
Discussion: The SWA is an easy to handle, practical, new portable device for measuring energy expenditure. The accuracy of the SWA appeared to be poor in the obese subjects we examined, especially those with high REE both in rest and exercise. We believe that it is necessary to incorporate new, obesity-specific algorithms in the relative software.
The statistics on obesity are appalling, with nearly one-half billion of the world's population now considered to be overweight or obese (1). The development of obesity occurs when caloric intake is disproportionate to energy expended. Three metabolic factors have been reported to be predictive of weight gain: a low adjusted sedentary energy expenditure, a high respiratory quotient, and a low level of spontaneous physical activity (2).
Daily energy expenditure can be measured in a respiratory chamber or in free-living people by the doubly labeled water (DLW)1 method. DLW and room calorimetry are considered to be the gold standards for energy expenditure assessment (3). Metabolic carts have also been shown to be reliable and accurate instruments for measuring resting energy expenditure (REE) when compared with direct calorimetry. The use of metabolic carts is the standard in research by which REE is measured (4). However, indirect calorimetry (IC) cannot easily assess free-living subjects, whereas DLW does not provide information on the pattern or intensity of physical activity. Furthermore, the expense of the equipment and supplies, time needed in the laboratory for subsequent analysis, burden for subjects, and significant amount of technical expertise required limit the use of these techniques (5). Prediction equations are often used as alternatives to measurements of energy expenditure; however, the accuracy of these equations has often been criticized (6, 7).
Historically, investigators have relied on self-report when attempting to quantify physical activity (8). However, intrinsic validity and reliability issues with self-report support the use of objective measures of physical activity (9).
To overcome the disadvantages of metabolic carts and to improve on the subjective nature of physical activity records and questionnaires, smaller, more manageable devices to measure energy expenditure have been developed that are more affordable and, hence, have more widespread use and applicability (9, 10). The SenseWear Pro 2 Armband (SWA; Body Media, Inc., Pittsburgh, PA) is a newly developed commercially available device to assess energy expenditure. This device is worn on the right upper arm over the triceps muscle and monitors various physiological and movement parameters. Data from a variety of parameters including heat flux, accelerometer, galvanic skin response, skin temperature, near-body temperature, and demographic characteristics including gender, age, height, and weight are used to estimate energy expenditure using proprietary equations developed by the manufacturer (10, 11, 12).
Previous studies have reported that the SWA has shown to be highly reliable at estimating the energy expenditure of rest, but has provided less accurate estimates of energy expenditure when compared with IC during various exercise protocols (10, 11, 12). Because the previous studies evaluated the SWA in relatively young, normal-weight adults, it is unclear whether similar results would be observed in individuals of different ages and body weights. The primary purpose of this investigation was to examine the reliability and validity of the SWA energy expenditure estimation at rest and during three different modes of exercise in obese individuals in a laboratory setting compared with simultaneous IC measurements.
Research Methods and Procedures
In all, 142 obese individuals (BMI >30) participated in the study. Twenty-nine of them also participated in the exercise protocols. All participants had been admitted to the Department of Internal Medicine of San Giuseppe Hospital (Verbania, Italy) for diagnostic and therapeutic problems related to obesity. Subjects with a medical condition that could prevent safe participation in moderate intensity exercise or with a medical condition that would require medical clearance before participation were excluded from this study. Diabetics and patients on medication that might have affected the SWA or the thermoregulatory process (e.g., sibutramine, anti-cholinergics, psychotrops) were not included in this study to minimize the number of confounding factors. In addition, 25 healthy lean and overweight (18.5 < BMI < 30) volunteers who were recruited from the hospital personnel participated in the resting protocol. The demographic characteristics of all participants are reported in Table 1. The individuals who participated in the exercise protocols were 14 men and 15 women with a mean age of 31.2 ± 3.2 years and a mean BMI of 43.2 ± 5.3. The aim and design of the study were approved by the Hospital Ethics Committee, and informed consent was obtained from all participants.
Table 1. Demographic characteristics of the participants
Healthy lean and overweight subjects
43.1 ± 14.4
48.3 ± 13.9
46.9 ± 14.2
39.2 ± 14.2
35.8 ± 15.2
36.5 ± 14.7
40.9 ± 6.5
42.8 ± 7.1
42.3 ± 7.0
26.8 ± 2.5
24.9 ± 3.3
25.3 ± 3.2
Resting and Exercise Protocols
Measurements of REE were conducted between 9 am and 11 am after an overnight fast (at least 12 hours) by means of the gas dilution method using the SensorMedics Vmax29 metabolic cart (SensorMedics Corporation, Yorba Linda, CA). Before commencement of the measurements, subjects rested for 30 minutes, during which time the traditional indirect calorimeter was calibrated. Subjects were asked to remain awake and motionless for the duration of the simultaneous measurements. The SensorMedics Vmax29 measurements were obtained for 20 to 25 minutes. Because there is no a reliable approach to assess interday variability of the REE assessment, we consider a measurement valid when 15 minutes of steady state, determined as a coefficient of variation <5% in minute respiratory quotient and minute oxygen consumption, is obtained.
The SWA measurements were obtained for 20 to 30 minutes. The SWA was worn on the right arm over the triceps muscle at the midpoint between the acromion and olecranon processes. The armband was placed on the subject's arm for a period of 5 to 10 minutes before data collection to allow for acclimation to skin temperature. Energy expenditure during exercise was computed at 1-minute intervals. The exact time (start and stop) of each IC measurement was recorded to synchronize with the SWA estimate. For the analysis of the SWA data, the appropriate software was used (InnerView Research Software, version 4.0, build 0-610; BodyMedia, Inc., Pittsburgh, PA), which, to our knowledge, is the most recent version. The subject's gender, age, height, and weight were programmed into the SWA before each trial. To explore the repeatability of the portable device, the SWA was worn by the same subject the next day under the same conditions. A total of three identical SWA instruments were used indiscriminately.
For each subject, REE (in kilocalories per day) was also predicted using the Harris-Benedict (H-B) equations (13):
where H is the height in centimeters, W is the body weight in kilograms, and A is the age in years.
Twenty-nine of the obese subjects also participated in three separate, easy-to-perform modes of exercise on different occasions that included cycle ergometry, stair stepping, and treadmill walking. These exercise protocols were chosen to test recording of the activities in the horizontal and vertical plane plus a static exercise. The order in which the different exercise protocols were performed was randomly assigned. All exercises began after a brief warm-up. The cycle ergometer protocol involved 5 minutes of pedaling at 60 rpm at a fixed load of 60 watts (Cornival 400; Lobe B.V., Groningen, The Netherlands). Stair stepping was performed on a 16-cm bench for 5 minutes. Walking was performed on a motorized treadmill (Marquette, series 2000 treadmill; GE Healthcare, Little Chalfont, Buckinghamshire, UK) at a speed of 3 km/h for 5 minutes. The time frame (5 minutes) was considered as relatively manageable for an obese person to complete at the fixed intensity of the exercises. During each exercise protocol, energy expenditure was measured simultaneously using the open-circuit calorimetry and the SWA.
Fat-free mass (FFM), total body water, and extracellular water (ECW) were determined by bioelectrical impedance analysis (BIA 101/S; Akern, S.r.I., Firenze, Italy) in the fasting state. Analysis was performed using Bodygram software version 1.2 (Akern, S.r.I.).
The intraclass correlation coefficient was calculated for testing the reliability of the SWA REE estimations. Univariate and multivariate linear regression analyses were employed to assess the relation between various anthropometric and metabolic indices and the estimated energy expenditure. Skewed variables were logarithm-transformed before statistical testing.
Bland-Altman bias plots were created to assess the agreement between the IC measurements and SWA estimations of energy expenditure, as well as the reliability of the SWA estimate across the entire range of measurements (14, 15). Limits of agreement involved the percentage mean difference between the two measurement tools ± 1.96 standard deviation of the percentage mean difference. The upper and lower 95% confidence limits of 1.96 standard deviation of the differences between the methods were calculated to examine whether they were equal to or smaller than a pre-defined limit of 15% for total error (TE). We additionally investigated whether the 95% confidence limits of the mean between the methods were equal to or smaller than a pre-defined limit of 5% for systematic error (16). p < 0.05 was regarded as statistically significant. Unless otherwise stated, data are presented as mean ± standard deviation. Statistical analysis was performed using the SPSS/PC statistical program (version 11.5 for Windows; SPSS, Inc., Chicago, IL).
The two SWA estimates from two different rest trials in obese individuals were compared to assess the reliability of this device. This produced highly reliable estimates of energy expenditure at rest with an intraclass correlation coefficient of r = 0.88 [95% confidence interval (CI), 0.84 to 0.91; p < 0.001]. The SWA estimates for REE of obese subjects were, in general, lower than those measured with IC (1811 ± 346 vs. 1880 ± 382 kcal/d), whereas the H-B equation-based predictions overestimated the REE (1908 ± 385 kcal/d) (Figure 1). We found significant linear correlations between REE (SWA) and REE (IC) (r = 0.88, p < 0.001) (Figure 2A) and between REE (SWA) and REE (H-B) (r = 0.96, p < 0.001) (Figure 2B).
In univariate regression analyses, no relationship was found between the biases and age, BMI, FFM (kg), total body water (percentage), and ECW (percentage) of subjects. However, there was a significant positive association of the biases and REE (IC) (r = 0.42, p < 0.001) and a borderline, but statistically significant, negative association between the biases and the subjects’ weight (r = −0.173, p = 0.04). The former but not the latter remained significant (p < 0.001) in a step-wise multivariable regression model. REEs measured by SWA and SensorMedics were both significantly correlated with FFM (r = 0.88, p < 0.001 and r = 0.84, p < 0.001, respectively). There was no statistically significant difference in the biases between men and women (p = 0.87, Student's t test).
Bland-Altman plots (14) were used to show the mean bias, limits of agreement, and spread of the bias over the range of values between REE measured by the SWA and IC (Figure 3). The x-axis shows the mean of the results of the two methods, whereas the y-axis represents the differences of the two methods, expressed in percentage of the values of the average as is recommended (16). The 95% CI of the upper limit was higher than the pre-defined limit of 15% for TE, but the 95% confidence limit of the mean was smaller than the pre-defined limit of 5% for systematic error (5%). Figure 3 also shows a tendency for a fanning effect with a wider spread of biases at higher REE measurements. The percentage biases for four subjects with an REE > 2000 kcal/d were outliers, whereas bias of only one subject with an REE < 2000 was an outlier. The confidence limits were not less than the pre-defined limits in the subgroups of patients with 30 < BMI < 40 (n = 60) and BMI > 40 (n = 82) (plots not shown). In contrast, the Bland-Altman plot for the IC measures and SWA estimates in lean and overweight individuals show a high correlation (r = 0.96, p < 0.001) and a very good agreement with a TE < 15% and a systematic error < 3% (plot not shown).
Of the 29 obese subjects who participated in the exercise study, complete data were available for 25 subjects for cycle exercise, 26 subjects for stepping exercise, and 20 subjects for treadmill walking exercise. The missing data resulted from either failure of the SWA to provide complete data (cycle ergometer, four; stepping, one; walking, one) or the unwillingness or incapacity of the subject to complete the exercise session. The energy expenditure estimates for the three modes of exercise were higher than those measured with IC and with very low intraclass correlations (Table 2).
Table 2. Comparison of measured and predicted energy expenditure and intra-class correlations of the exercise protocols
Bland-Altman plots did not show any agreement between the two methods (Figure 4 AB to C, for the three modes of exercise). Interestingly, there was a statistically significant positive correlation between the biases of the exercise measurements and the REEs of the subjects for the stepping (r = 0.39, p = 0.49) and the walking (r = 0.77, p < 0.001) modes of exercise.
The advent of portable IC has allowed for measurement of more realistic simulations of everyday activities. To our knowledge, this study is the first to examine the validity and reliability of one of those new hand-held devices for estimating energy expenditure in comparison with simultaneous IC measurements in obese adults. The SWA device was user-friendly for the subjects in terms of easy attachment/detachment, minimal discomfort, and little or no interference in activity. The present study showed that REE measurement using the SWA in obese subjects was underestimated by a mean value of 8.8%. The magnitude of the difference in REE between the SWA and IC appeared to increase as the REE increased across individual participants in this study, a finding that is consistent with that of Jakicic et al. (12). However, a highly significant correlation was demonstrated between the REE (SWA) estimates and the measured values by IC. In addition, high correlation and very good agreement were found in the measurements of the REE between the two methods in a sample of lean and overweight individuals.
Comparing the REE estimations by the SWA with those predicted from H-B equations, we found a high correlation between the two variables. It is possible that the SWA relies more on subject characteristics such as age, gender, and weight to calculate REE and maybe is not very sensitive to small changes in energy expenditure from day to day. The prediction model used to make the SWA REE estimate is proprietary, but the model uses both sensor data and characteristics of the wearer (age, height, weight). More research is needed to determine whether the incorporation of sensor data are more or less accurate in estimating REE than established prediction equations for obese individuals.
Regarding the energy expenditure during three simple modes of 5 minutes of exercise, the SWA has generally overestimated them, although the three activities were mainly of the lower body without, many times, corresponding arm movement. It has been suggested (12) that accelerometry-based physical activity monitors may overestimate energy expenditure of obese subjects due to excessive body motion (greater body movement associated with reduced mechanical efficiency). Interestingly, there was also a statistically significant positive correlation between the biases of the stepping and treadmill walking measurements and the REE of the subjects.
Previously, comparisons were based on the product moment correlation coefficient. However, a high degree of correlation does not imply agreement between the two methods. It is widely agreed that the standard statistical method for the comparison of a new and established measurement technique is that described by Bland and Altman. Using the Bland and Altman technique in this comparison, we demonstrated that the SWA device, when compared with the IC, has clinically significant limits of agreement. This implies that the two measurement methods may not be used interchangeably in obese subjects.
This study did not investigate the reproducibility of the IC, but reproducibility of traditional IC methods is often reported to be high, with measurement error in the order of <5% (17, 18, 19). A limitation of our study was the use of bioelectrical impedance analysis for the assessment of body composition because it is well known that it tends to overestimate FFM in obese individuals, especially in the severely obese. H-B equations, which were used in our study, remain one of the commonest methods for calculating REE for clinical and research purposes and have provided valid estimations of REE in both normal-weight and overweight individuals of the same ethnicity (20).
The SWA looks promising because, theoretically, it does not have the disadvantage of being subject to the detection of false motion and enables accurate detection of non-ambulatory physical activity. The multiple-sensor array was designed to overcome the limitations of other objective energy expenditure assessment tools and, theoretically, make it a very useful tool for answering questions regarding patterns of physical activity, which cannot be determined by other measures of EE, such as doubly labeled water or extended duration oxygen consumption. Unfortunately, given the expense of the SWA, the use of this device for large-scale epidemiological trials is unlikely.
It seems that skin qualities like hydration, roughness, and laxity at the attachment position do not play a major role in the accuracy of the measurements because we did not find any significant correlation between age, total body, or ECW and the biases. The SWA was shown to be highly reliable at estimating the energy expenditure of rest in lean and overweight subjects but not in obese subjects. The results of this study suggest that specific resting and exercise specific algorithms have to be used for obese (more strictly speaking according to our findings for individuals with high REE) to provide more accurate estimates of energy expenditure. Alternatively, an enhancement of the accuracy of the generalized algorithm is needed to provide an estimate of energy expenditure across a wide range of REEs, modes, and intensities of activities. Improvements could come in the form of field-based equations for specific activities, used for calculating energy expenditure from raw data or, perhaps, an option to enter subject-specific REE. It is necessary to test the reliability of the SWA in estimating the energy expenditure of various activities in various populations before conclusions can be made about the overall reliability of this device. However, it may have the potential to provide a feasible assessment of free-living energy expenditure. Future studies examining energy expenditure during free-living conditions along with DLW because the criterion measure of energy expenditure will also be of great value. Furthermore, it would be important to investigate whether the SWA could detect and accurately estimate the slight elevation of energy expenditure over resting levels that occur with very light activities because the majority of adults spend much of their time in primarily sedentary pursuits. Although SWA seems comfortable for short assessment periods, it is debatable whether all subjects accept carrying a portable device for long-term use or assessment.
The ability to accurately and reliably monitor physical activity and energy expenditure has emerged as a critical component to weight management and the prevention of lifestyle-related health problems. If the SWA proves to be a useful measure of free-living physical activity, it would allow for collection of objective data in a way that is cost effective, easy to administer, and without interference with normal activities.
There was no funding/outside support for this study.
The costs of publication of this article were defrayed, in part, by the payment of page charges. This article must, therefore, be hereby marked “advertisement” in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.