| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
1National Cancer Institute, Bethesda, Maryland 20892; 2Centers for Disease Control and Prevention, Atlanta, Georgia 30341; 3Baylor College of Medicine, Houston, Texas 77030; 4East Carolina University, Greenville, North Carolina 27858; and 5West Texas A&M University, Canyon, Texas 79016
Submitted 25 August 2003 ; accepted in final form 4 December 2003
 |
ABSTRACT |
|---|
 TOP ABSTRACT METHODSRESULTSDISCUSSIONGRANTSACKNOWLEDGMENTSREFERENCES |
|---|
accelerometer; physical activity diary; energy expenditure; physical activity energy expenditure
Currently, no universally agreed-on standard exists for validating measures of physical activity. The doubly labeled water (DLW) methodology is believed to be the most accurate method for assessing energy expenditure in free-living conditions (22). In controlled conditions, measurement errors for the DLW measure at the group level are estimated to be within ±5% (22), whereas individual errors are estimated to range from -38 to +54% (27). Therefore, the DLW method is considered accurate for assessing group level energy expenditure and, as such, has been used as the gold standard for validating physical activity measures, including activity diaries and accelerometers. However, studies that validated accelerometers (6, 14, 16, 18, 32) and diaries (7, 11, 26) against DLW energy expenditure showed mixed results. Correlations between total energy expenditure (TEE) determined by DLW and physical activity measures ranged from -0.07 to 0.93 for accelerometers and from 0.32 to 
1.0 for diaries. Moreover, two of the diary-validation studies reported contradictory information; investigators observed a positive association with TEE determined with DLW yet failed to find an association with physical activity energy expenditure (PAEE) determined by DLW (11, 26).
To evaluate the association of physical activity measures with DLW energy expenditure, investigators often transform the data obtained from accelerometers or diaries by using measures of body mass or resting metabolic rate (RMR) (7, 10, 11, 24, 26, 28). Such transformations may falsely inflate the association of these physical activity measures with DLW energy expenditure because they reflect the strong association that exists between TEE and body mass or RMR (8). For example, to convert the metric obtained from a diary into kilocalories, investigators typically perform a mathematical transformation of the data using body mass (e.g., MET-min multiplied by body mass/60 kg, where 1 MET is the amount of energy you expend at rest) (10, 11). With this transformation, energy expenditure obtained from the diary becomes a function of body mass and is therefore expected to correlate with DLW energy expenditure because body mass explains almost half of the total variance in energy expenditure (8). Thus, when investigators use a body mass transformation, the magnitude of the association may be driven largely by body mass. To date, investigators have not examined how body mass transformations affect DLW validation coefficients, nor have they proposed alternative statistical strategies.
The main purpose of this analysis was to determine, using a sample of women aged 40 yr and older, the influence of body composition [body mass and fat-free mass (FFM)] on the associations between two measures of physical activity (diary and accelerometer) and DLW energy expenditure. Specifically, two approaches were used to determine the extent to which body composition influenced the association between physical activity and DLW energy expenditure: 1) mathematical transformation of physical activity measures and 2) statistical procedures (multiple regression, partial correlations, and allometric scaling) to determine how much of the variance in DLW energy expenditure was explained by the physical activity measures. In addition, the extent to which body mass influenced agreement between physical activity and DLW energy expenditure was assessed by using the methods-comparison analysis.
|
METHODS |
|---|
|
TOPABSTRACTMETHODSRESULTSDISCUSSIONGRANTSACKNOWLEDGMENTSREFERENCES |
|---|
Study participants were African American and Hispanic women who were recruited for the Women On The Move (WOTM) study, a 5-yr study to develop and validate questionnaires for measuring physical activity among 40- to 70-yr-old women. Women residing in the greater metropolitan area of Houston, Texas, were eligible for the study if they 1) were aged 40-70 yr, 2) self-reported their race/ethnicity as African American or Hispanic, 3) had no health limitations that would prevent them from being physically active, 4) were not pregnant, 5) were not planning to move out of the metropolitan area in the next year, 6) were literate in English or Spanish, and 7) successfully completed and returned by mail a 1-day practice physical activity diary. Of the 656 women who expressed an interest in participating in the WOTM study, 260 enrolled in the study. The sample was 50% African American and 50% Hispanic (predominantly Mexican American). Women were recruited through the media (print, television, and radio), community presentations, and posted flyers. The recruitment, screening, and retention protocol is described elsewhere (15). Data were collected from August 1997 through June 1999.
Study Protocol
Approval for the study was obtained from the Institutional Review Board for the protection of human subjects at the University of Texas-Houston, Health Science Center, and the Centers for Disease Control and Prevention. Data reported in this study were collected during the first 2-wk period of the validation protocol.
Study participants were screened initially by telephone and at an in-person screening, during which demographic information was collected. To estimate energy expenditure, participants were given a dose of DLW on the first day of the protocol, and seven urine samples were collected from each participant during the next 14 days. To monitor physical activity, participants were asked to wear an accelerometer and to keep a physical activity diary for 7 days during week 2 of the protocol (21). RMR data were collected after the 14-day DLW protocol was completed. Participants were compensated for their participation in the study.
DLW Energy Expenditure
TEE was estimated over the 14-day study period in free-living conditions by use of DLW (19). After providing a baseline urine sample, each participant ingested a known quantity of 18O and 2H (18O ingested as H218O at 100 mg/kg of body mass and 2H ingested as 2H2O at 100 mg/kg of body mass; Isotec, Miamisburg, OH). The 2H218O container was rinsed three times with 20 ml of drinking water, and the participant ingested all three rinses. The baseline urine sample and seven subsequent urine samples were collected on days 1, 3, 5, 7, 10, 12, and 14 of the protocol. Participants were instructed not to collect the first daily void of urine. Urine samples at baseline and on days 7 and 14 were collected in the presence of the research staff who recorded date and time of the sample. Other samples were self-collected at the participant's home, and participants were given a small cooler in which to transport the samples. All urine samples were transferred into aliquot tubes and frozen.
Baseline and postdose urine samples were analyzed for deuterium and 18O isotopic enrichment by gas-isotope-ratio mass spectrometry (34). For hydrogen isotope ratio measurements, 10 µl of urine without further treatment were reduced to hydrogen gas with 200 mg of zinc reagent at 500°C for 30 min (34). The 2H-to-1H isotope ratios of the hydrogen gas were measured with a Finnigan Delta-E gas-isotope-ratio mass spectrometer (Finnigan MAT, San Jose, CA). For O2 isotope ratio measurements, 100 µl of urine were allowed to equilibrate with 300 mbar of CO2 of known 18O content at 25°C for 10 h using a VG ISOPREP-18 water-CO2 equilibration system (VG Isogas, Cheshire, England). At the end of the equilibration, the 18O-to-16O isotope ratios of the CO2 were measured with a VG SIRA-12 gas-isotope-ratio mass spectrometer (VG Isogas, Limited, Cheshire, England). Fractional turnover rates for each isotope and their dilution spaces were computed and used to compute CO2 expiration rate (rCO2). Finally, the rCO2 was converted to TEE by use of the Weir equation (12), as follows
 |
where rO2 (O2 consumption rate) was calculated from the food quotient (FQ) (3) calculated from food intakes using rO2 = rCO2/FQ and UN as the 24-h urinary nitrogen excretion (in g).
The multipoint calculation method was employed to estimate isotope decay over 7 days and 14 days. TEE for week 2 (used for the analyses in this study) was obtained by subtracting the 14-day estimate from the 7-day estimate. PAEE was estimated with the equation TEE - thermic effect of food (TEF) - RMR, where TEF was estimated to be 10% of TEE.
RMR
On the morning of RMR testing, participants arrived at the laboratory between 6 AM and 8 AM after having fasted for 
12 h. Participants had been asked to minimize their activities for the 40 h before testing. Participants were familiarized with the equipment and the laboratory environment. Each participant (wearing a hospital gown) was weighed on a physician's balance scale (Detecto, Webb City, MO). After the participants had taken a 20-min rest in a supine position, expiratory gas exchange was collected by indirect calorimetry using a SensorMedics Vmax 229 ventilated open-hood system (SensorMedics, Yorba Linda, CA) for 40 min. O2 variation of ±25 ml/min was used to determine whether the collection was successful (29). Each participant was monitored periodically to ensure that she remained awake. Data collection took place in a thermal-regulated environment with minimal light and noise.
Accelerometer
The CSA accelerometer (model 7164 WAM, Computer Science and Applications, Shalimar, FL) was used to estimate physical activity. This small, uniaxial device (dimensions: 5.1 x 4.9 x 1.6 cm; wt: 39.8 g) is specifically designed to assess human motion as it registers acceleration ranging from 0.05 to 2.0 G (where G, the constant of gravitation, is equal to 9.825 m/s2). It suppresses high accelerations in the 0.25- to 2.5-Hz range to eliminate nonhuman motion, such as automobile acceleration (Computer Science and Applications, 1995). Activity counts are the measurement units used by the accelerometer, which was programmed to register activity counts in 1-min epochs. The criterion for moderate-intensity physical activity was 3 METs, estimated as 1,952 counts/min (17). One MET equals the value of resting O2 uptake relative to total body mass and generally approximates the value of 3.5 ml of O2 per kilogram of body mass per min or roughly 1 kcal/min for a person who weighs 60 kg. During week 2 of the DLW protocol, participants wore the accelerometer on the right hip under their clothes to minimize superfluous movement. Participants were instructed to wear the accelerometer for 7 consecutive days, except when sleeping or in contact with water.
Physical Activity Diary
Each participant completed a physical activity diary during week 2 of the study for 7 consecutive days and recorded activities that lasted 10 min. Participants were instructed to record five characteristics about each activity: 1) type (occupation, sport and exercise, walking, house/yard, inactivity, personal care, free time/entertainment, transportation, or miscellaneous); 2) brief description; 3) estimated intensity (low, medium, or high); 4) position (lying/resting, sitting, standing, or moving around); and 5) time spent (min). If the activity involved walking, the participant recorded the walking pace (moving about, slow pace, medium pace, brisk walking, or stair climbing). If the activity involved carrying or lifting a load, the estimated weight carried was recorded. In addition, for each activity, the participants indicated whether they were wearing the accelerometer. Instructions for completing the diary and a sample entry were included in each diary. The physical activity diary format and instructions are described elsewhere (13). After each participant had returned the diary, an interviewer reviewed the diary with her to clarify entries. Diary entries were coded using a standard physical activity reference guide (1) to determine the MET values of physical activities reported in the diary. Diaries were coded by two observers and reconciled by a third observer; the coding protocol is described elsewhere (21). The unit of measure for the diary was metabolic equivalent (MET)-min and was computed by multiplying activity duration by activity MET.
Anthropometric Indexes
Anthropometric measurements of participants without their shoes were taken at the initial screening assessment visit. Two measurements of body mass and height were taken without shoes using a Seca alpha-digital scale (QuickMedical, Snoqualmie, WA) and a portable Accustat stadiometer (Genentech, San Francisco, CA) that were calibrated monthly. Body mass index (BMI) was calculated as body mass (kg) divided by height squared (m2). All measurements were taken by trained observers using a standardized anthropometric protocol (20). Means of the two measurement values were used in the analyses. In addition, FFM was calculated from the isotope dilution space of 18O by using the appropriate hydration constant for FFM, which was 0.725 (31). Body fat was calculated as the difference between body mass and FFM.
Statistical Analyses
All data were analyzed using the SPSS statistical software package (release 10.0 for Windows, SPSS, Chicago, IL). Means and standard deviations were computed for continuous data, and frequency distributions were computed for categorical data. Two approaches (mathematical transformations and statistical procedures) were used to determine the extent to which body composition (body mass and FFM) influenced the association between the physical activity measures (accelerometer and diary) and DLW energy expenditure.
Mathematical transformations. Activity counts from the accelerometer and MET-min from the diary were adjusted for body mass using a multiplicative adjustment factor: body mass (kg) divided by standard weight (60 kg) (10, 11). Also, multiplicative transformations were performed to adjust accelerometer counts and diary MET-min for FFM [FFM (kg) divided by standard FFM; 43.2 kg was estimated from the published literature to be the standard FFM for a 60-kg woman] (2). The body mass transformation changed the physical activity diary MET-min into kilocalories per day, which is equivalent to the measurement unit for DLW. Pearson product moment correlations were used to assess the association between the physical activity measures and energy expenditure by DLW (TEE and PAEE).
Statistical procedures. Multiple regression, partial correlation, and allometric scaling procedures were used to assess the influence of body composition (body mass and FFM) on the associations between the physical activity measures and DLW energy expenditure. Multiple-regression procedures were used to determine the amount of variance that was accounted for by the physical activity measures after the effect of body mass or FFM was removed. Because body mass and FFM may be correlated with both the physical activity measures and DLW energy expenditure, partial correlations were computed to determine the association between the physical activity measures and DLW. Finally, because the effect of body mass and FFM may not be linear (i.e., the effect may increase as people perform more physical activity) (25), allometric scaling was used to remove the influence of body composition. Once the effect of body mass or FFM was removed, Pearson's product-moment correlations were used to assess the association between the two physical activity measures and DLW energy expenditure. A simple linear regression was used on the log-transformed variables to derive the allometric scaling exponents (33).
The methods-comparison analysis (4) was used to explore the extent to which body mass influenced agreement with the DLW estimates. Agreement was assessed only with the diary method adjusted by body mass, because this method and transformation changed the data into a metric equivalent to the DLW methodology (a requirement for this procedure). Bland-Altman plots were generated to assess the agreement between the diary method adjusted for body mass and TEE, and the limits of agreement were plotted at 2 standard deviations below and above zero. To isolate the portions of agreement attributable to diary and body mass, the following procedures were performed: 1) a random normal deviate with its mean and standard deviation equal to the activity diary was generated, 2) the random normal deviate was multiplied by body mass, 3) the difference between the random normal deviate adjusted by body mass and TEE was assessed, and 4) the agreement between the diary method adjusted by body mass and TEE was compared with the agreement between the random normal deviate adjusted by body mass and TEE. The randomly generated data provided the methodology for isolating the effect of body mass on agreement as well as a reference point for interpreting the Bland-Altman plots.
Given that the correlation coefficient is highly affected by sample size (9), partial and multiple correlations of >0.30 were considered meaningful. In addition, for the multiple-correlation analyses, a 5% incremental improvement in r2 was considered meaningful.
|
RESULTS |
|---|
|
TOPABSTRACTMETHODSRESULTSDISCUSSIONGRANTSACKNOWLEDGMENTSREFERENCES |
|---|
|
RMR, energy expenditures from DLW and diary data, and activity counts from the accelerometer are presented in Table 2. On average, women expended 2,306 kcal/day in TEE, of which 1,371 kcal/day (59.4%) were attributable to RMR and 705 kcal/day (30.6%) were attributable to PAEE. Mean estimates of PAEE from the diary were 150 kcal/day higher than mean estimates derived from DLW. The PAEE ratio (TEE/RMR) indicated that, on average, women performed some physical activity (mean PAEE ratio = 1.7). The PAEE ratios, which ranged from 1.2 to 2.5, indicated that women had varied levels of physical activity (1.4-1.5 suggests moderate physical activity and 2.0-2.4 suggests vigorous activities) (2).
|
Mathematical Transformations
The effects of mathematically transforming the physical activity measures on the associations of these measures with DLW energy expenditure are presented in Table 3. For both the diary and accelerometer data, nontransformed associations with DLW energy expenditure were <0.30 (range 0.049 to 0.208). These associations were lower than the association between the diary and accelerometer (r = 0.43). When the data were mathematically transformed to account for body mass or FFM, associations with TEE increased (range 0.474 to 0.506). The associations with PAEE also increased, but the increase was smaller than that with TEE (range 0.225 to 0.330). Although the diary associations with PAEE increased, they were both <0.30.
|
Statistical Procedures
Associations that used multiple correlations, partial correlations, or allometric scaling to adjust for the effect of body mass or FFM are presented in Tables 4 and 5. In general, the multiple correlations with TEE were higher than those with PAEE (see Table 4). The multiple-correlation models that included body mass or FFM with either physical activity measure explained between 36 and 44% of the variance in TEE and explained between 6 and 13% of the variance in PAEE. The results indicated that the multiple correlations were higher with FFM than they were with body mass. For both instruments, the incremental increases in r2 all were smaller than 5%. The incremental r2 for the accelerometer adjusted by body mass was near 5%, which suggests that a meaningful portion of the variance in TEE and PAEE was explained by the accelerometer data.
|
|
The partial correlations between the physical activity measures and TEE and PAEE are shown in Table 5. All partial correlations were small (<0.30) and showed that, once the effect of body mass or FFM was taken into account, the accelerometer and diary data were not meaningfully correlated with TEE or PAEE. Finally, the correlations adjusted by allometric scaling all were <0.30. The largest correlation was between TEE (adjusted for body mass) and the accelerometer (0.27). The allometric scaling correlations were similar to the partial correlations.
Results of the methods-comparison analysis show the agreement between the diary and DLW and between the random normal deviate and DLW for nontransformed and body-mass-transformed data (Table 6). Because no association was observed with PAEE, agreement is shown only for TEE. The diary overestimated TEE by 10% without body mass adjustment and by 19% with body mass adjustment. Variance accounted for in TEE was higher when the data were mathematically transformed with body mass (25 vs. 0.2%). The use of a random normal deviate showed that the agreement and the percentage of variance accounted for in TEE were similar to findings from the diary for both the nontransformed and body-mass-transformed data. The diary data, mathematically transformed by body mass, accounted for 25% of the total variance in TEE; and the random normal deviate, mathematically transformed by body mass, accounted for 21% of the variance in TEE.
|
Bland-Altman plots for the mathematically transformed data (Fig. 1) show the difference between the diary data (transformed by body mass) and TEE and between the random normal deviate (transformed by body mass) and TEE. The figure shows that the patterns of agreement with TEE for both the random normal deviate and the physical activity diary data were the same (i.e., the plots were the same).
|
|
DISCUSSION |
|---|
|
TOPABSTRACTMETHODSRESULTSDISCUSSIONGRANTSACKNOWLEDGMENTSREFERENCES |
|---|
In this study, mathematically transforming the physical activity measures with body mass or FFM was found to have a direct effect on their associations with DLW energy expenditure. Specifically, no meaningful associations were observed between the diary and accelerometer with DLW energy expenditure when these measures were not transformed. Such effects were somewhat expected, because energy expenditures for weight-bearing activities are affected by body mass or FFM (25). Finding higher associations between the mathematically transformed measures and TEE than between these measures and PAEE, however, suggested that the associations may be driven by the association that exists between body mass and FFM or RMR (2). In this sample, the correlation with RMR was 0.74 for body mass and 0.73 for FFM. Thus perhaps body mass and FFM or RMR can account for the higher associations observed between TEE and the mathematically transformed measures compared with the associations observed with PAEE.
In addition to assessing the effect of mathematical transformations on associations, this study assessed how these transformations affected agreement. The methods-comparison analysis requires a common metric; therefore, agreement was assessed only with the diary. To provide a reference point for assessing the agreement between the diary and TEE, a random normal deviate variable was generated. The results indicated that the agreement between the diary and TEE was similar to the agreement observed between the random normal deviate and TEE. Interestingly, mathematically transforming the diary and random normal deviate data with body mass had similar effects on agreement, change in r2, and percentage of variance accounted for. These results suggest that the diary did not improve agreement, change in r2, and percentage of variance accounted for in TEE beyond what can be accounted for by body mass. Thus these results showed that reporting only mathematically transformed associations or agreements may be misleading. Furthermore, when mathematically transformed associations and agreement are assessed, it is suggested that a reference point be used to judge the effect of these transformations on the interpretation of the results. In this study, a random normal deviate was used to anchor these comparisons and provided a better understanding of the effect of mathematically transforming the data.
With the exception of the accelerometer multiple-correlation results, using different statistical procedures (multiple regression, partial correlation, and allometric scaling) to account for body mass and FFM yielded results similar to those of the mathematical transformations. The accelerometer explained a small but meaningful portion of the variance in TEE and PAEE. Almost 5% of the variance was attributable to the accelerometer after body mass was accounted for statistically in the model; however, when FFM was accounted for, the accelerometer did not explain any meaningful portion of the total variance in DLW energy expenditure. As with the mathematically transformed results, investigators should report incremental multiple-correlation models, because the aggregated statistical model can be misleading. For example, the multiple-correlation coefficient between TEE and a model that includes FFM and the diary was equal to 0.66. Reporting this value only may lead investigators to conclude that the diary is associated with TEE. In this study, however, the association was driven by FFM; only 1.5% of the variance was attributed to the diary. Finally, the statistical procedures revealed that associations with DLW were different for the diary and the accelerometer. The small (almost 5%) but meaningful proportion of the total variance explained by the accelerometer after body mass was accounted for was observed for both TEE and PAEE. Therefore, in contrast to the diary, the accelerometer seemed to relate to DLW energy expenditure beyond what had already been accounted for by body mass. When FFM was accounted for, however, the variance explained by the accelerometer was <5%.
Comparison of our findings with published studies showed that three of five accelerometer validation studies (6, 14, 16, 18, 32) and two of three diary validation studies (7, 11, 26) had findings similar to ours. The present study was not unique in finding a poor association or no association with PAEE, although other studies have shown high associations between accelerometer output and PAEE (6, 14, 32). The mixed findings may be related in part to the characteristics of the sample (i.e., level of physical activity and demographic characteristics). It is well known that the correlation coefficient is attenuated by restriction of range, where low levels of physical activity can attenuate the correlation with PAEE. The fact that the women in our study were found to be moderately active and that few of them did vigorous physical activity may explain the poor association between the physical activity measures and PAEE. Finally, in our sample, a large proportion of the women were overweight and obese. In such samples, it is critical to determine the best approach (i.e., mathematical transformation or statistical procedures) to account for the effects of body mass and FFM. In contrast, studies that have used adolescent male athletes (14) may be less likely to have associations affected by body mass or FFM because there is less variation in body mass and FFM in this sample. One such study by Ekelund et al. (14) observed the highest correlations between the accelerometer and DLW TEE (0.93) and DLW PAEE (0.96).
It is difficult to account for the effect of body mass or FFM because physical inactivity may be a cause of obesity or a consequence of the condition (25). For the same reason, removing the effect of body mass or FFM may underestimate the association that exists between the physical activity measures and DLW energy expenditure. Body mass and FFM may be biomarkers of physical activity behavior. To determine whether removing the effect of body mass or FFM may have underestimated the associations between the two physical activity measures, we examined the correlation between body mass and FFM and the two measures. Associations were <0.10 in absolute value. Thus removing the effect of body mass or FFM did not appear to underestimate the associations between the physical activity measures and DLW.
This study highlights the difficulty of using DLW as a validation standard for measuring physical activity. We also demonstrated the effect of body mass and FFM on the associations and agreement between the physical activity measures and DLW. Clearly, mathematically transforming the data with body mass or FFM can be misleading and can inflate associations with DLW. This study showed the usefulness of using a random normal deviate as a reference point to isolate the effect of body mass or FFM on associations and agreement with DLW. Although this study demonstrated these effects in a population of minority women, the effects are expected to be observed in other populations (i.e., Caucasians and men), given that the relationship between body mass or FFM with RMR is similar across populations (2). Finally, the impact of body mass and FFM on the association and agreement between the physical activity measures and DLW would lessen in a normal-weight population (due to restriction of range); however, it remains important to identify how much impact body mass or FFM has on the associations and agreement. Until such approaches are used, researchers are cautioned to evaluate more critically physical activity validation studies that use DLW as the validation standard.
|
GRANTS |
|---|
|
TOPABSTRACTMETHODSRESULTSDISCUSSIONGRANTSACKNOWLEDGMENTSREFERENCES |
|---|
|
ACKNOWLEDGMENTS |
|---|
|
TOPABSTRACTMETHODSRESULTSDISCUSSIONGRANTSACKNOWLEDGMENTSREFERENCES |
|---|
|
FOOTNOTES |
|---|
The costs of publication of this article were defrayed in part by the payment of page charges. The article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
|
REFERENCES |
|---|
|
TOPABSTRACTMETHODSRESULTSDISCUSSIONGRANTSACKNOWLEDGMENTSREFERENCES |
|---|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| Visit Other APS Journals Online |
