Estimation of Energy Expenditure from Physical Activity Measures: Determinants of Accuracy
Diet and Human Performance Laboratory, Building 308, Room 122, Beltsville, MD 20705. E-mail: email@example.com
Objective: To describe the determinants, specifically age, body mass index, percentage of body fat, and physical activity (PA) level, associated with over- and underestimation of energy expenditure (EE) using PA records and the Stanford Seven-Day Physical Activity Recall (7DR) compared with doubly labeled water (DLW).
Research Methods and Procedures: We collected PA measures on 24 males eating a controlled diet designed to maintain body weight, and we determined EE from DLW and estimated EE from PA records and 7DR.
Results: Absolute differences in the estimation of EE between DLW and PA assessment methods were greater for the 7DR (30.6 ± 9.9%) than PA records (7.9 ± 3.2%). In PA records, overestimation of EE was greater with older age and higher body fatness; EE was overestimated by 16.7% among men 50 years and older compared with only 5.3% among men <40 years of age. For percentage of body fat, EE was overestimated by 19.7% among men with a percentage of body fat ≥30% compared with only 5.6% among men with a percentage of body fat <25%. A trend for less overestimation of EE with higher levels of PA (measured by DLW/basal metabolic rate [BMR]) also was observed in the PA records. In the 7DR, the estimates of EE varied widely and no trends were observed by age, percentage of body fat, and PA levels.
Discussion: Estimation of EE from the 7DR is considerably more variable than from PA records. Factors related to age and percentage of body fat influenced the accuracy of estimated EE in the PA record. Additional studies are needed to understand factors related to accurate reporting of PA behaviors, which are used to estimate EE in free-living adults.
Most chronic diseases, such as cardiovascular disease, type 2 diabetes, and certain cancers, are associated with lifestyle behaviors such as physical inactivity (1). Increased participation in physical activity (PA) is also associated with a reduced risk of obesity (2). There are a number of different methods for measuring PA; however, because of its complex nature, accurate measurement of PA is difficult (3). Measurement may be further complicated by several demographic and physiological factors such as age and body weight. Age and body weight may influence a person's ability to accurately assess PA (4, 5, 6, 7, 8, 9). In examining the relationship between PA and a disease or condition, such as cardiovascular disease or cancer, it is important to understand what factors are associated with improved PA assessment. Inaccurate assessment of PA can obscure important associations and limit the ability to detect significant associations between PA or energy expenditure (EE) and disease outcomes (10).
The criterion field method for estimating individual EE has been doubly labeled water (DLW) (11, 12, 13). However, the cost and availability of isotopes and the requirement for analysis by isotope ratio mass spectrometer prohibits DLW from being widely used in studies of large populations.
There are a number of different PA assessment tools available to estimate EE in studies of PA and health (14). One method to assess PA is the PA record. PA records have been used as a qualitative and quantitative direct measure of PA that provides a concurrent description of the type, perceived intensity, purpose, and duration of PA. This information can be used to estimate total EE or the energy expended from specific types of PAs (3). One of the most widely used PA questionnaires is the Stanford Seven Day Recall Questionnaire (7DR), which is a qualitative and quantitative indirect measure of PA that is widely used in epidemiological, clinical, and behavior change studies (15, 16, 17, 18). PA records and questionnaires differ from purely quantitative direct measures of EE, such as DLW, which provide a concurrent measure of the energy cost of PA but do not describe the type or purpose of PA performed.
We reported (19, 20) the comparison of PA records, 7DR, and DLW for 24 adult men. We found that PA records kept for seven consecutive days may estimate the mean EE of a population with a <8% difference between PA records and DLW. Estimation of EE from the frequency, duration, and type of activities recorded in a PA record may contribute errors, which could explain our observation of an R2 of 0.10 between PA record and DLW estimates of EE (20). The 7DR questionnaire estimated the mean EE of the population being studied with a 30% difference between the questionnaire and DLW (R2 = 0.14). Because we observed that the use of PA records to predict individual EE depends largely on the compliance of the participant being studied and the ability of the participant to correctly estimate time spent in activities of differing intensities, we conducted this analysis to ascertain the possible demographic and physiological determinants associated with over- and underestimation of EE using PA records and the 7DR compared with DLW. Thus, the purpose of this study was to ascertain the determinants associated with accurately estimated EE using PA records and the 7DR compared with DLW. Determinants of interest were age, body composition, and PA levels.
Research Methods and Procedures
Data were obtained from 24 men who were recruited from participants in an ongoing feeding study of 60 men at the United States Department of Agriculture (USDA), Agricultural Research Service, Beltsville Human Nutrition Research Center in Beltsville, Maryland. This larger study was conducted as a 6-month outpatient feeding trial of diets varying in lipid content. All the men consumed 10 meals per week under the supervision of a dietitian, and the remaining meals and snacks were provided for offsite consumption. The study reported here was conducted over 2 weeks during this controlled feeding period. No limits were placed on free-living activities by the experimental protocol. Subjects were originally recruited by advertisement at the Beltsville Agricultural Research Center in Beltsville, Maryland; at the Goddard Space Flight Center in Greenbelt, Maryland; and from the laboratory's computerized database of people known to be interested in participating in human studies. The study protocol was approved by the Institutional Review Board at the Johns Hopkins University School of Medicine and by the Human Studies Committee of USDA's Agricultural Research Service. Subjects were invited to attend an informational meeting and those interested in participating gave written informed consent. At the beginning of the clinical feeding study, each potential subject received a medical evaluation by a physician including measurement of blood pressure, height, weight, and analysis of fasting blood and urine samples to screen for absence of metabolic diseases.
Each subject was studied over a 2-week period. On day 1 the doubly labeled water (2H218O) was administered, and urine was collected over the next 14 days. Because of calorimetry scheduling, PA records were kept between days 1 and 7 or days 8 and 14 of the2H218O protocol, and the 7DR was administered on day 14 of the2H218O protocol.
Weight (WT) was determined on an electronic balance to the nearest 0.01 kg, and height (HT) was measured to the nearest 0.1 cm with a stadiometer. Body mass index (BMI) was defined as WT/HT2 and expressed in kilograms per square meter. Percentage of body fat was determined by DXA (version 1.3; DPXL Lunar Corp., Madison, WI). The subjects were asked not to consume anything 3 hours before the scan, to dress in metal-free clothes, and to remove jewelry. Depending on subject size, the scans took ∼20–37 minutes, and the subjects received from 0.03 to 0.06 mRem of radiation. The components for total body, i.e., bone mineral, soft tissue (lean body mass), and fat, were used to calculate percentage of body fat. Percentage of body fat was determined as follows; %Fat = [(fat mass)/(fat mass + lean mass + bone mineral mass)] × 100%, and body fat mass was calculated from%Fat and body weight.
Average EE was determined using the DLW (EEDLW) method (11). Baseline samples of urine were collected before the isotope dose (2H218O, 0.14 g/kg body weight;2H2O, 0.70 g/kg body weight) administration and daily for 14 days. The urine was analyzed for 2H2 by an automated infrared analysis method in our laboratory and for18O by a commercial laboratory subcontractor. Standards prepared by the investigators, but unknown to the commercial laboratory, were used to monitor the subcontractor's performance. Isotope kinetics were determined using a multipoint calculation technique (11).
Basal Metabolic Rate
Each subject entered the calorimeter at 6:30 pm and remained there for a total of 23 hours. They followed a uniform schedule that included meals, sleep, basal metabolic rate (BMR) measurement, and subsequent exercise protocols. BMR was measured on awakening at 7:00 am over a 1-hour period of time while the subjects were in the room calorimeter. Subjects were allowed to use the toilet facilities on awakening but were then confined to bed for the measurement of BMR. The 20.39 m3 indirect calorimetry system continuously measured respiratory gas exchange to within 1.5% (21). Between run intraindividual variation of this calorimeter is 4.6% (21). While the subjects were in the chamber, CO2 production and O2 consumption were measured continuously. The average BMR determined over 1 hour was used for analyses.
PA level (PAL), as defined by James et al. (22, 23), is equal to total EE divided by BMR and serves as a measure of PA. Therefore, we calculated PAL = EEDLW/BMR.
We asked individuals to record all PA for 7 days as described by Ainsworth et al. (24) (see Figure 1). Seven 24-hour PA records were completed by subjects during a 1-week period between days 7 and 14 of the DLW sample collection. Subjects recorded all activities performed throughout the day in the PA record. Each change in activity was a new entry in the PA record. For example, walking to and from the car to the house four times in a day was recorded as four entries rather than one entry. Subjects were given detailed instructions for completing the PA records and recorded the following in a book (one book per day): (1) the time of day they started each new activity, (2) body position during the activity (reclining, sitting, standing, or walking), (3) general characterization of the purpose for doing the activity (e.g., occupation, household, childcare, exercise, transportation, etc.), (4) a detailed description of the activity (e.g., typing, eating, walking for exercise, etc.), and (5) a perceived effort (light, moderate, vigorous) for the activity. Items (1) through (5) above were recorded in the PA record to help in the appropriate assignment of the correct MET (1 kcal/kg body weight per hour) level as listed in the Compendium of Physical Activities (25). One MET reflects the ratio of the associated metabolic rate for a specific activity divided by the resting metabolic rate.
Completed daily PA records were turned into study staff each morning. Each weekday evening, when the participants arrived for dinner, an investigator reviewed the PA record with the subjects to ensure the completeness of the data collection. Weekend PA records were reviewed the following Monday. The same investigator (J.M.C.) instructed the subjects in the use of these PA records and inspected the completed forms. All 24 men turned in seven complete 24-hour PA records. Two investigators (M.L.I. and B.E.A.) assigned a five-digit code to each activity as obtained from the Compendium of Physical Activities (25). Differences in assigned Compendium codes were identified by a third reviewer, consensus was reached, and corrections were made. The five-digit code links the purpose, description, and MET intensity for each activity in the PA records. PA records were scored using a statistical program written for this study. PA minutes and MET minutes (calculated as the MET intensity times the number of minutes of each activity) were summed and then averaged over the 7-day period. EERECORD (in kilocalories) was computed as follows: MET − minutes × (body weight in kilograms/60).
The 1985 version of the 7DR presented by Blair et al. (15) was used in this study. The interview-administered questionnaire takes ∼10 minutes to complete (15). The questionnaire has 14 items that assess the number of hours and minutes spent sleeping (1.0 MET) and in moderate (4.0 METs), hard (6.0 METs), and very hard (10.0 METs) intensity physical activities for weekend and weekdays separately. 7DR validation studies demonstrate acceptable validity and reliability when compared with various direct and indirect measures of PA and EE (14). Time spent in light intensity activities was determined by subtracting the time sleeping and in the time spent in moderate, hard, and very hard activities from 24 h/d. The 7DR was administered twice for the purposes of determining physical activity before and during the DLW period; at the beginning of the 14-day DLW measurement period and again at the end of the 14-day period. We used the scores from the second administration for data analysis because this 7-day recall coinsided with DLW data collection. Surveys were edited for clarity and accuracy by an investigator (J.M.C.) in the presence of the subject. PA data were computed as min/d and MET min/d (calculated as the MET intensity times the minutes reported for each type of activity). EESDR (in kilocalories) was computed as follows: MET minutes × (body weight in kilograms/60).
Statistical analyses were performed with PC-SAS (SAS Institute Inc., Cary, NC). Means were calculated as the measure of central tendency and SEs of the mean were computed as a determination of variability. Differences in EE (kilocalories per day) between DLW and the PA assessment methodologies were computed as follows: EEΔSDR = EESDR − EEDLW and EEΔRECORD = EERECORD − EEDLW. Pearson product-moment correlations were used to determine the association between specific determinants of EE (e.g., age, BMI, percentage of body fat, and PAL) and EEΔSDR and EEΔRECORD. General linear models were used to determine least-square means after removing the covariation of age, BMI, and percentage of body fat on the prediction of EEDLW by the PA assessment methodologies. Forward stepwise regression analysis further was used to determine the proportion of variance in EEΔSDR and EEΔRECORD explained by age, BMI, percentage of body fat, and PAL.
The study population (Table 1) ranged in age from 27 to 65 years and also had a broad range of body weight, BMI, and percentage of body fat. Twenty men (83%) were non-Hispanic white and 23 (96%) were employed full-time.
Table 1. Demographic characteristics of the study population (n = 24)
|Age (years)||41.2 ± 9.6||27–65|
|Weight (kg)||79.5 ± 9.0||63.9–94.5|
|Height (m)||1.78 ± 0.07||1.66–1.95|
|BMI (kg/m2)||25.1 ± 2.7||20.9–31.5|
|Percentage of body fat*||21.1 ± 7.1||9.8–43.3|
|Ethnicity (% NHW)||83%|| |
|Employment (% FTE)||96%|| |
The EE measured by DLW, PA records, and 7DR for each participant and for all participants combined is shown in Table 2. The mean EE determined from 7DR (4157 kcal/d) was significantly higher than the mean EE determined from DLW (3169 kcal/d) (p < 0.05). There was no statistically significant difference in mean EE as determined from PA records (3385 kcal/d) compared with the mean DLW (3169 kcal/d).
Table 2. Mean (range) EE*,† measured by DLW, PA records, and 7DR for each subject
|Mean± SEM||3169 ± 84||3385 ± 38||224 ± 100||4157 ± 347||987 ± 324|
|Percent difference‡|| || ||7.9 ± 3.2|| ||30.6 ± 9.9|
Table 3 compares the absolute kilocalories per day and relative percentage values for EEΔRECORD and EEΔSDR adjusted by age, BMI, percentage of body fat, and PAL groups. Increasing age, BMI, and percentage of body fat were associated with greater overestimation of EE in both the PA records and 7DR; however, these associations were not statistically significant (p > 0.05). In general, there was better agreement between EEDLW and EERECORD as PAL increased (p = 0.03).
Table 3. Least-square means* for the mean and percentage difference between EE determined from DLW and PA records or 7DR by age, BMI, percentage of body fat, and PA level groups
|Age (years)|| || || || || || |
|<40||12||114 ± 130||5.3%||12||561 ± 454||16.1%|
|40–49||8||242 ± 158||10.5%||8||1824 ± 557||57.6%|
|50+||4||472 ± 223||16.7%||4||596 ± 787||20.3%|
|BMI (kg/m2)|| || || || || || |
|<25||14||119 ± 129||6.0%||14||949 ± 454||28.6%|
|25–30||9||317 ± 161||11.7%||9||1035 ± 567||33.4%|
|≥30||1||670 ± 488||21.2%||1||1109 ± 1716||35.3%|
|Percentage of body fat|| || || || || || |
|<25||18||108 ± 111||5.6%||18||1063 ± 405||32.4%|
|25–30||4||511 ± 245||17.4%||4||717 ± 895||24.3%|
|≥30||2||603 ± 329||19.7%||2||851 ± 1199||27.4%|
|PA level‡|| || || || || || |
|Least active||8||431 ± 143||19.3%||8||827 ± 619||29.0%|
|Moderately active||8||332 ± 157¶||10.6%||8||591 ± 681||18.6%|
|Most active||8||−113 ± 165§||2.0%||8||1546 ± 714||44.4%|
Correlations between age, BMI, percentage of body fat, and PAL with both EEΔSDR and EEΔRECORD are shown in Table 4. For PA records, higher percentage of body fat was assoiated with an overestimate of EE (r = 0.53, p < 0.05). Conversely, with higher PAL, mean differences between EEDLW and EERECORD decreased (r = −0.53, p < 0.05). In a regression model, percentage of body fat and PAL explained 45% of the variance in the EEΔRECORD (p < 0.05) (Table 5).
Table 4. Correlation between age, BMI, percentage of body fat, and PA level and difference between EEDLW* and the PA assessment methods (n = 24)
|Percentage of body fat||0.53¶||−0.26|
Table 5. Determinants associated with inaccurately estimating EE* from PA measures in adult men (n = 24)
|Percentage of body fat||0.29||0.007||0.29|| |
|PA level||0.16||0.007||0.45|| |
| ||Seven day recall|
|Determinant||Partial R2||Partial p value||Model R2||Model p value|
|Percentage of body fat||0.07||0.21||0.07|| |
|PA level||0.01||0.60||0.19|| |
DLW is the gold standard for measuring EE in a free-living population; however, the isotope availability fluctuates and the cost per subject can be as much as $1800, which is not feasible for studies of large populations. Thus, other less expensive and easily obtainable measures of EE are needed for studies of EE and health (3). Measures of PA, such as PA records and recall questionnaires, have been used in epidemiological and clinical studies to estimate EE. This is despite the lack of adequate testing of the validity of EE estimates using these measures. In this study we compared EE estimated by PA measures with DLW in a field-validity study. The mean difference in EE measured by PA records and DLW was 8% compared with 30% between EE from 7DR and DLW.
Age and percentage of body fat have received little study in relation to the accuracy of estimating EE from direct (i.e., DLW and PA records) and indirect measures (i.e., PA questionnaires such as the 7DR) of PA. In this study, age, BMI, percentage of body fat, and PAL were determinants of reporting accuracy of EE using PA records compared with measuring EE from DLW. Men who overestimated EE on the PA records had a significantly higher BMI and percentage of body fat compared with men who accurately estimated their EE (Tables 3 and 4). Accuracy of estimated EE increased to within 2% of EEDLW as PAL increased. A possible explanation for the increased accuracy between EEDLW and EERECORD may be that active men planned their daily activities and thus were more aware of their PA behaviors.
The social desirability theory describes the tendency of individuals to present themselves in an overly positive or culturally appropriate light for self-esteem enhancement (4). This theory provides an explanation why the overweight participants may have overestimated their EE from PA records. Food-intake studies show that adults with a BMI > 25 kg/m2 tend to underreport food intake (26). In this study, men with a BMI > 25 kg/m2 overestimated EE from PA records and 7DR. The observations are comparable in relation to possible social desirability bias. Klesges et al. (6) showed that a higher BMI is associated with underreporting of time spent in aerobic activities. Others (7, 8, 9) have shown that obese individuals overestimate their duration of moderate-to-higher intensity PA (7), the total minutes of exercise performed (8), and total EE (9). Our results are consistent with studies showing that adults with BMI > 25 kg/m2 tend to overestimate their self-reported PA.
Wilcox and King (4) found that older age was associated with overestimating activity level in women relative to peers, but Blair et al. (5) found that age was unrelated to the accuracy of retrospective recalls of PA. In the present study, we observed a trend toward greater overreporting with older age; however, this trend was not statistically significant (p = 0.07).
There may be other reasons, besides social desirability, for the apparent overestimation of EE on the PA records among older or overweight men in this study. For the PA records, study investigators assigned a five-digit compendium code to each activity, with the activity linked to a MET intensity level. The MET levels used are average values obtained from various studies (25). In the use of the Compendium of PA, Schmitz (27) noted that the measured energy cost of PA was higher among obese women for selected activities than presented in the Compendium of PA, resulting in a higher measured EE than obtained by the EE computed from the PA records (25).
Another plausible explanation for the higher estimated EE from the PA records may be the variability of the energy cost of movement among individuals in the study. Recent studies by Hendelman et al. (28) and Bassett et al. (29) quantified the energy cost of various PAs using a portable metabolic measurement system. The values measured in both studies were significantly different from the MET levels cited in the Compendium of PA. For example, the studies showed the energy cost of golf and walking slowly to be lower than those in the Compendium of PA, whereas yard work and fast walking were higher. Further, individual variation, reflected by the SDs in the energy cost of each activity measured by Hendelman et al. (28) and Bassett et al. (29) were on the order of 0.5 to 1.5 METs. This shows the individual variation of movement that the Compendium of PA does not reflect because it provides absolute MET levels. Therefore, factors of age and body weight could cause significant variability in the true energy cost of PA and estimated EE.
Also, the occupations of the men who participated in this study were very diverse and included administrators, scientists, technicians, carpenters, construction workers, farm laborers, electricians, and an automobile detailer. Some men worked two jobs (e.g., a recreational therapist who had a second job evenings and weekends as a disc jockey and an engineer who had a nighttime and weekend office-cleaning service). The largest overestimations of the time spent in hard and very hard PA were made by subject 4 (7DR was 204% of DLW), whose occupation was to detail cars for an automobile wholesaler using a 40-lb machine to buff the cars and subject 11 (7DR 116% of DLW) who was a carpenter. Other overestimations were made by men with more than one job or by men with second jobs that involved intermittent heavy PA. Therefore, some of the overestimations could easily be explained by the subjects’ perceptions of the exertion of their PA, the complexity of their activities, the length of time during the 7 days when they were active, or imprecise MET estimates of the energy costs of selected occupational activities in the Compendium of PA.
The overestimation of EE by the PA records among overweight and older individuals in this study may relate to cognitive memory processes and the time available to complete the PA record. Also, it is possible that the overweight and obese subjects overestimated their duration of higher-intensity PAs. The current DLW data preclude comment on possible overreporting of PA behaviors in the PA records and 7DR, because DLW does not measure the types of PA performed but only the EE associated with PA. Last, studies of cognition and recall of PA are limited with many questions remaining about optimal strategies to enhance the recall of PA using questionnaires (30). In particular, the role of age as a factor in the recall of PA needs more study.
Overall, when comparing the EEDLW with EE from the PA assessment methods, we observed a significant correlation for EE7DR (r = 0.51, p < 0.01) but not for EERECORD (r = 0.28, p = 0.18) (19). Evaluation of the source of variation between EEDLW and EERECORD revealed a statistically significant correlation between EEΔRECORD and percentage of body fat (r = 0.53) and a significant inverse association between EEΔRECORD and PAL (r = −0.53). Together, percentage of body fat and PAL explained 45% of the variance of EEΔRECORD (Table 5). No associations were observed for the 7DR. This suggests that the amount of detail applied to completing the PA records and the external validity of the methods used to compute EE estimates from the PA record could be population specific and could depend on more than one population descriptive characteristic such as age, percentage of body fat, and physical activity level.
PA records, 7DR, and DLW have strengths and limitations that should be mentioned. PA records provide a detailed recording of all PAs performed during a discreet period of time. Very active individuals may list over 90 discreet activities performed in a 24-hour period, whereas less active individuals may record as few as 10 discreet activities. While time-consuming to complete and process, PA records provide optimal information about the quantity and quality of PA. This is contrasted to PA-recall questionnaires that take less time and effort to complete and process; however, they may provide only limited data about the type, frequency, and duration of PA. DLW is considered the gold-standard measure for EE in a field setting. However, DLW does not provide information about PA behaviors in relation to the type, frequency, duration, and intensity. Thus, little information is obtained from DLW to help clinicians and PA specialists identify appropriate types of intervention strategies to modify or increase PA behaviors.
There are several limitations to this study that may impact the generalization of the study findings. First, the sample size is small and may limit the applicability of our findings to larger and different populations of men. Second, because the study was performed in men only, it is unknown if similar findings related to the overestimation of PA exist in women. Therefore, similar studies should be conducted in a population of women. Third, because PA is a behavior and EE is an outcome of PA behaviors, the terms are not synonymous and should not be considered to be measuring the same thing (3). Equations have been developed to estimate EE from PA; however, the equations are often population-specific and are associated with random and systematic error (3, 31). Additionally, whereas DLW is considered a direct field measure of EE, it still uses indirect measures of metabolism (calculated oxygen consumption, carbon dioxide production, and food quotient) to estimate EE (32). These limitations should be considered when using field measures to quantify PA and EE.
Regular participation in PA is associated with a host of physical and psychological health benefits (1). Yet, national surveys indicate that the majority of U.S. adults are not active at levels needed to attain these benefits. This study highlights factors related to the over- and underestimation of PA and EE among adult men; older and heavier men tended to overestimate PA and resultant EE compared with younger and leaner men in the PA records. Further, errors in the estimation of EEDLW obtained from 7DR were nearly four times larger than those observed with PA records. The implications of these findings is that the actual proportion of the older and overweight men in the U.S. meeting the national PA recommendations may vary considerably from the current national estimates. Additional studies in larger populations are needed to understand the PA reporting patterns among men, women, children, and minorities, using a range of body fat and physical activity levels to obtain more accurate estimates of PA for surveillance purposes and in studies of PA and health.
This research was funded by the United States Department of Agriculture, Agricultural Research Service, National Program in Human Nutrition. The authors would like to thank Drs. David R. Jacobs, Jr. and James L. Seale for their contributions, Robert Staples and Demetria Fletcher for technical assistance, and their subjects for their dedication.