点击显示 收起
1 From the Department of Health Sciences, Boston University, Boston (LGB); the Eunice Kennedy Shriver Center, University of Massachusetts Medical School, Waltham, MA (LGB); the Department of Family Medicine and Community Health, Tufts University School of Medicine, Boston (AM, SMP, and ENN); and the Division of Physical Activity and Nutrition, Centers for Disease Control and Prevention, Atlanta (WHD)
2 The studies were conducted at the General Clinical Research Center (GCRC) at the Massachusetts Institute of Technology (MIT) and supported by MIT GCRC grants MOI-RR-01066 and MO1-RR-00088 and by NIH grants DK-HD50537 and 5P30 DK46200. 3 Reprints not available. Address correspondence to LG Bandini, Department of Health Sciences, Boston University, 635 Commonwealth Avenue, Boston, MA 02215. E-mail: lbandini{at}bu.edu.
ABSTRACT
Background: Although it is widely accepted that weight gain results when energy intake exceeds energy expenditure (EE), how reduced EE contributes to the development of obesity remains unclear.
Objective: We tested the hypothesis that reduced EE in the premenarcheal period in girls constitutes a risk factor for an increase in relative weight [body mass index (BMI) z score] and percentage of body fat (%BF) during adolescence.
Design: We measured EE at study entry in 196 premenarcheal nonobese girls. Resting metabolic rate (RMR) was measured by indirect calorimetry. Total energy expenditure (TEE) was measured by the doubly labeled water method. Activity energy expenditure (AEE) was calculated from RMR and TEE. After the baseline study, girls were followed annually until 4 y after menarche (
Conclusions: Our findings suggest that EE in the premenarcheal period is not a risk factor for increases in %BF or BMI z score in girls during adolescence.
Key Words: Adolescence • energy expenditure • energy metabolism • obesity
INTRODUCTION
Obesity results from an energy imbalance, whereby energy intake (EI) exceeds energy expenditure (EE), net of any growth. Genetic and lifestyle factors contribute to energy balance and are important in understanding the cause of obesity. Studies of EI and EE in obese children and adolescents do not suggest that obese children eat more or spend less energy than nonobese children (1-8). However, conclusions from these studies are limited by the significant underreporting of EI observed in both obese and nonobese children (1, 3, 9). Although EE in the obese does not differ from that in normal-weight individuals after adjustment for fat-free mass (FFM), reduced EE in the preobese state could be a risk factor for developing obesity. Subsequent weight gain can normalize EE (10). Ravussin et al (10) found that low resting EE appeared to be a risk factor for body-weight gain in adult Pima Indians.
In an early study, Griffiths and Payne (11) used heart-rate monitoring to measure EE in 37 children, 3–5 y old, of obese and nonobese parents. The results of their cross-sectional study suggested that the children of obese parents had lower EEs than the children of nonobese parents. However, a 12-y follow-up of their subjects showed that body fat did not differ between these 2 groups of children (12). To date, 4 studies of EE as measured by doubly labeled water were conducted in infants. In the earliest study (13), a statistically significant relation between low EE and weight gain over the first year of life was observed, but only 18 infants were studied and only 6 of those infants became overweight. Three larger studies subsequently failed to find evidence of a relation between reduced EE and weight gain (14-16).
In 2 studies, one of children aged 4–7 y followed for 3–4 y (17) and one of children aged 4.6–11 y followed for 3–5 y (18), no significant relation between reduced EE and subsequent body weight gain was observed. Neither study, however, followed the children over a critical period such as puberty.
Puberty is a period of rapid growth and development in which boys and girls increase FFM substantially, although in girls the gain is only about half that in boys (19). Because puberty in girls is associated with considerable increases in body weight and body fat mass, adolescence could be a critical period for the development of obesity (20). We hypothesized that premenarcheal girls with a reduced resting metabolic rate (RMR) or daily total energy expenditure (TEE) gain more body fat and have increased relative body weight gains during the adolescent period than girls with higher EE. In addition, we hypothesized that less energy spent on activity (AEE) during the premenarcheal period is an additional independent risk factor for increased weight gain over the adolescence period.
The energy spent on activity can be calculated from TEE and RMR. Unlike RMR, the energy spent on activity can be altered by individual behavior. In this study, we examined the changes in body mass index (BMI) z score and in body fatness as measured by bioelectrical impedance analysis (BIA) to test our hypothesis that RMR, AEE, and TEE in normal-weight premenarcheal girls are associated with increased body fat or relative weight gain during the pubertal period. The longitudinal nature of the study and the annual measures of physical activity level and dietary intake allowed us to control for changes in activity and diet over the study period.
SUBJECTS AND METHODS
Participants
Between September 1990 and June 1993, we enrolled 196 girls in the Massachusetts Institute of Technology (MIT) Growth and Development Study. The criteria for enrollment were premenarcheal status and a triceps skinfold <85th percentile for age and sex (21). Premenarcheal girls aged 8–12 y were recruited from the Cambridge and Somerville public schools in Massachusetts, the MIT summer day camp, and friends and siblings of enrolled subjects. All participants were healthy, free of disease, and not taking any medication that affected body composition or EE. The study protocol was approved by both the Committee on the Use of Humans as Experimental Subjects at MIT and the Institutional Review Board of the New England Medical Center.
Study protocol
Initial study visit
Details of the baseline visit are described elsewhere (22). Girls admitted to the Clinical Research Center at MIT for an overnight visit arrived in the late afternoon at which time a physician obtained a medical history and examined each girl to ensure that she was in good health. Girls were asked to complete a physical activity questionnaire and a food-frequency questionnaire. At 2000 each participant was given 0.25 g H218O and 0.1–0.12 g 2H2O/kg estimated total body water (TBW) for the measurement of TBW and TEE after a baseline urine sample was obtained.
The next morning, the second void of the day was collected at 0800 to measure 18O and 2H enrichment above baseline values. This same sample was used to determine TBW and the initial time point of the EE period.
Subjects wearing a hospital gown and slippers were weighed on a Seca scale (Seca, Hanover, MD) accurate to 0.1 kg. Height was measured to 0.1 cm with a wall-mounted stadiometer (Genentech Inc, San Francisco). Body composition was determined by BIA with use of a BIA 101 impedance analyzer (RJL, Clinton Township, MI) as previously described (23). RMR was measured by indirect calorimetry after an overnight fast and a 30-min rest period as previously described (22, 24). At the baseline and exit visits (4 y after menarche), and other visits when TBW was measured, body weight and BIA were measured after an overnight fast. On annual visits, body weight and BIA were measured either after an overnight fast or 2 h after a meal. Most but not all of the visits were measured in the fasting state. Weight was measured in a hospital gown on all visits.
Two weeks later the participants returned to the Clinical Research Center after an overnight fast. Collection of a urine sample (the second void of the day) marked the end of the EE period. The subjects were weighed, and their RMR was measured.
Annual visits
Girls were seen annually on the anniversary of their baseline visit. Diet and activity questionnaires, anthropometric measures, and BIA measurements were repeated. Girls completed the study when they were 4 y postmenarche. For the present analysis, girls who had at least one follow-up visit were included (n = 187).
Analyses
Total body water and energy expenditure
Mass spectroscopy analyses
Isotopic analyses for the assessment of body composition and EE were conducted at the US Department of Agriculture Human Nutrition Research Center at Tufts University, Boston, using a Hydra Gas Isotope Ratio Mass Spectrometer (PDZ Europa LTD, Northwich, United Kingdom) and a SIRA 10 (Fisons Instruments, Altrincham, United Kingdom) as previously described (22). Criteria for acceptance values were a SE for replicate measures of 0.35 for oxygen and 1.5 for deuterium.
Energy expenditure
We used Weir's equation (25) to calculate RMR from measures of indirect calorimetry and TEE from the doubly labeled water method. We determined carbon dioxide production (rCO2) from the doubly labeled water method and calculated
Body fatness was estimated for each follow-up with prediction equations developed in this cohort that used available measures of TBW by isotopic dilution of H218O as the criterion method (27). We developed separate equations for premenarcheal and postmenarcheal measurements and estimated the percentage of body fat (%BF) for each time point according to the girl's menarcheal status. A correction factor was applied to the predicted body fat measures for each girl to address the discontinuity that arises from using 2 independently generated equations. The average correction was 0.069% (interquartile range: –0.61, 0.41). The BIA prediction equations were developed from multiple measures of TBW available at different time points. A total of 422 measurements from 196 girls were used in the development of these equations. The distribution of measurement numbers were as follows: 52 participants had one measure of TBW, 84 participants had 2 measures of TBW, 42 participants had 3 measures of TBW, 14 participants had 4 measures, and 4 participants had 5 measures.
Relative weight
To provide a measure of relative weight, a BMI z score was calculated for each BMI measure with the reference to age- and sex-specific limits provided by the Centers for Disease Control and Prevention growth reference standards (28). Three limits were estimated: the median (M), the SD (S), and the power in the Box-Cox transformation (L). The equation for the LMS is the following: Centile = M (1 + LSZ)1/L, where Z is the z score. The final set of percentile curves available from the Centers for Disease Control and Prevention was produced by using the modified LMS estimation procedure, as described in detail in the documentation that accompanied the release of these growth charts (29). These limits are based on BMIs calculated from heights and weights from national surveys conducted in the United States between 1963 and 1980 (30).
Physical activity and inactivity measures
Participants completed a questionnaire that identified usual patterns of physical activity. Participants were presented with two 24-h timetables (school day and weekend day) and asked to recall, on an hour-by-hour basis, their participation in 5 types of activities during each time block: sleeping or lying down, sitting, standing, walking, and vigorous activity (exercising, playing, or being involved in sports). In addition, participants completed a similar grid on which they reported television-viewing time on an hourly basis (including time spent watching videos or playing video games).
The average time spent daily in each activity was computed as a weighted average of the school day (5/7) and weekend day (2/7) reports. Information on the reliability of this physical activity assessment protocol was published elsewhere (31).
To develop an index to assess the time spent and level of physical activity, we examined the relation between AEE adjusted for body weight and time spent sleeping, sitting, standing, walking, in vigorous activity, and television viewing. We then created an activity variable to reflect time spent and intensity of vigorous activity and walking and an inactivity variable to reflect time spent sitting, standing, and sleeping. The average daily hours spent walking and in vigorous activity were combined and weighted by their intensity (using a metabolic equivalent value) to create an activity index. Average hours spent daily in sleeping or lying down, sitting, and standing were summed to create an inactivity index. The activity index was the best correlate of the activity indexes with AEE adjusted for body weight (r = 0.29, P < 0.0001). The inactivity variable was inversely correlated with adjusted AEE (r = –0.28, P < 0.001) (L Bandini, A Must, unpublished observations, 2003).
Dietary assessment
All participants completed a semiquantitative food-frequency questionnaire designed for children aged 9–18 y and based on one validated for adults (32). Similar to the adult version, this questionnaire was designed to be self-administered, but participants were given oral or written instructions or both on how to complete the forms properly. Specific details about the questionnaire are described elsewhere (33). With data from questionnaires, we calculated daily servings of fruit and vegetables, percentage of daily calories from sugar-sweetened soda, and percentage of daily calories from macronutrients.
Other variables
Early in the study, we measured the height and weight (dressed but without shoes) of the participant's biological parents and calculated BMI (in kg/m2). Parental overweight was defined as a BMI 25 in at least 1 biological parent. Participants were classified as having lean parents only if neither biological parent were overweight. The cohort includes 19 sister pairs and 2 sets of 3 sisters, which represent 22.4% (44 of 196) of the cohort. One of the sister sets included a set of identical twins (one of whom we dropped from the analysis). We were concerned that including nonidentical twin sisters in the analysis would present a potential problem in separating endogenous (genetic) and exogenous (environmental) components of the identified factors that affect individual growth trajectories. To test whether this concern was founded for our study population, we conducted a series of analyses to assess similarities in the individual BMI z score and %BF trajectories observed over the study relative to the time of follow-up. First, we assessed similarities in the BMI z score and %BF trajectories within each sister pair by using Pearson correlation. The range of correlation coefficients for sister pairs was –0.41 to 0.91 for BMI z score and –0.62 to 0.92 for %BF. Then, we assessed similarities in the BMI z score and %BF trajectories for each sister and the rest of the cohort. With only one exception, for every sister pair, the cohort included 3 nonsisters whose BMI z score and body fat trajectory was more similar to her own than to her sister's. Therefore, we retained all sisters in the analyses herein, except for one sister from the pair of identical twins who we dropped entirely.
On baseline questionnaires, the participants were asked to indicate their ethnicity (white, black, Hispanic, Asian, or other). At each visit, any prescription medication reported by the subject was noted. Medications considered as possibly influencing EE or weight gain included oral contraceptives, antidepressants, antipsychotics, and steroids.
Statistical analysis
Descriptive analyses for the subjects at baseline (study entry) and 4 y after menarche (study completion) were undertaken to provide a picture of the cohort at these 2 points in time. For those girls (n = 156) remaining in the study at 4 y after menarche, differences at the 2 time points were formally tested with a paired t test.
Linear mixed effects modeling was used to evaluate the longitudinal relation between repeated measures of BMI z score and percentage of body fat (outcome) and measures of baseline EE adjusted for body composition (predictors). To adjust the EE variable for these variables established as related at baseline (22), we extracted residuals from a regression with the EE component as the dependent variable and with body composition variables or age as independent variables. We then added the mean EE value to each subject's residual to return the variable to its original scale and used the resulting variables as predictors in the mixed effects models. For RMR and AEE, we analyzed the residuals after adjustment for fat, FFM, and age. For TEE, we used the residuals after adjustment for FFM and age.
The mixed effects model that we applied consists of 2 parts: fixed and random effects. Fixed effects describe a population intercept (bo) and population slopes (b1) for a set of considered covariates, which include predictors as well as confounders. Random effects describe individual variability in the outcome (ßo)and changes over time(ß1). By considering individual random slopes and intercepts, this model allowed us to examine the influence of predictors on the change in outcome over time.
The linear mixed effects model also accounts for the correlation between repeated measurements on the same subject and the different numbers of measurements per subject. Because we had 2 outcomes (%BF and BMI z score) and 3 exposure variables (RMR, AEE, and TEE), 6 separate LME models were evaluated. To control for possible confounders in the relation between either %BF or BMI z score and the measures of EE, the following strategy was used. Simple linear or longitudinal models were evaluated to determine which potential covariates were significant predictors of any of the EE variables and either %BF or BMI z score. The following covariates were considered: physical activity index, inactivity index, parental overweight, race or ethnicity (coded as 2 dummy variables for black and "other" with white as the reference category), daily servings of fruit and vegetables, percentage of daily calories from sugar-sweetened soda, and percentage of daily calories from fat. For models with BMI z score as the outcome variable, age was expressed as chronologic age, and age at menarche was included as a covariate in all models. For models with %BF as the outcome, age was expressed relative to age at menarche (34). To examine directly, the effect of parental obesity on these longitudinal relations, we also applied the LME models in the 2 parental obesity groups. Finally, to assess the possibility that medication use influenced our analyses, the models were run excluding any visit at which a subject reported a medication that might influence EE (n = 84). To establish our statistical power for the size of our sample (given the additional time points) with consideration of the repeated measurements provided by a longitudinal analysis, we performed a recalculation of sample size by using established equations (27). Given our sample of at least 132 subjects with 7.5 measurements/subject on average, our estimated minimum detectable effect that exceeds age-related change was 0.06 BMI z score units/y and 0.23% body fat/y at 80% power (1-ß) and a type I error rate = 0.05.
Data were analyzed with SAS (Version 8.0; SAS Institute, Cary, NC) and S-PLUS software (version 4.5; MathSoft Inc, Seattle). Alpha was set at 0.05 for all analyses.
RESULTS
Body composition and EE of the participants at baseline and study completion are presented in Table 1. BMI z score at baseline (
View this table:
TABLE 1. Characteristics of the cohort at study entry (baseline) and study exit (4 y postmenarche)1
The relation of BMI z score and %BF with baseline EE over the pubertal period was explored with use of longitudinal models (Tables 2, 3, and 4). Models are presented with (model 1) and without (model 2) parental overweight as covariates.
View this table:
TABLE 2. Longitudinal relation of percentage body fat and BMI z score with adjusted resting metabolic rate (RMR) at baseline
View this table:
TABLE 3. Longitudinal relation of percentage body fat and BMI z score with adjusted total energy expenditure (TEE) at baseline
View this table:
TABLE 4. Longitudinal relation of percentage body fat and BMI z score with adjusted activity energy expenditure (AEE) at baseline
Adjusted RMR at baseline was not related to change in %BF longitudinally but was positively related to BMI z score (Table 2). Because parental overweight was associated with both body fat and BMI z score, we controlled for parental overweight in the model. With parental overweight in the model, adjusted RMR was no longer significantly related to either %BF or BMI z score. We also stratified our analysis by parental overweight to explore the relation between adjusted RMR in girls with an overweight parent and in girls with no overweight parent. We found no relation of %BF or BMI z score with adjusted RMR in either group (Table 5 and Table 6).
View this table:
TABLE 5. Longitudinal relation of percentage body fat with measures of energy expenditure at baseline by category of parental overweight1
View this table:
TABLE 6. Longitudinal relation of BMI z score with measures of energy expenditure at baseline by category of parental overweight1
Adjusted TEE (Table 3) was not related to change in %BF, but it was positively related to the change in BMI z score. After controlling for parental overweight, the relation between adjusted TEE and BMI z score remained significant. When we stratified by parental overweight, there was no statistically significant relation between adjusted TEE and change in body fat in either group (Table 5), but a statistically significant relation was observed between BMI z score and adjusted TEE only in the group with at least 1 overweight parent (P < 0.05; Table 6). The estimates were of similar magnitude, however.
Adjusted AEE (Table 4) was not related to change in %BF, but it was positively related to the change in BMI z score. After controlling for parental overweight, the relation between adjusted AEE and BMI z score remained significant. When we stratified by parental overweight, there was no relation between adjusted AEE and change in body fat (Table 5), but we did observe a significant relation with BMI z score and adjusted AEE in the group with one overweight parent (P < 0.05) (Table 6). Again, the estimates of the longitudinal relation between BMI z score and adjusted AEE were similar in both parental weight groups.
At study entry none of the subjects reported taking any medications. Over the course of the study, however, some subjects did report the use of various medications at the time of their annual visit. Because of the uncertain effects of some medications on EI and EE, we repeated all of the analyses reported, eliminating the 84 study visits (which involved 50 individual subjects) at which subjects reported taking oral contraceptives, antidepressants, antipsychotics, or steroids. The results were essentially unchanged.
DISCUSSION
Our findings provide no support for the hypothesis that reduced RMR, or TEE, in the premenarcheal period is associated with increased relative weight gain or body fatness. In fact, an unexpected small positive association was observed between AEE and TEE and BMI z score (P < 0.03; estimate 0.00095). This finding suggests that a slightly higher AEE and TEE is associated with weight gain. The absence of a significant association of AEE or TEE with %BF, however, suggests that %BF and the BMI z score might not be measuring the same thing, or they could reflect differences in the range of the 2 outcome variables. It is possible that the discrepancy between BMI z score and %BF reflects the influence of body size (body mass, ie, the greater energy cost to move a larger mass) on AEE.
When we stratified our analyses by parental overweight, the relation between AEE and TEE and %BF remained nonsignificant in both parental overweight groups. When BMI z score was the outcome variable, AEE and TEE were not significant predictors in children of nonoverweight parents, but AEE and TEE were positively related in the children of overweight parents (AEE: P < 0.05, estimate 0.00115; TEE: P < 0.05, estimate 0.00086). Our findings are similar to those of Johnson et al (18) who observed a positive relation between AEE and adiposity in their cohort. They suggested that this association might be due to the increased energy cost associated with moving a greater fat mass. Note that on the basis of our selection criteria, our cohort of children who had overweight parents were not overweight.
To our knowledge, this study is the first to examine the relation between weight gain and reduced EE over the pubertal period. The strengths of the study include the adjustment for covariates known to be associated with RMR and TEE, the length of follow-up (an average of 7.1 ± 2.6 y), the large sample size at baseline (n = 196) with a retention rate of 79.5%, and the inclusion of multiple longitudinal data points. Because our cross-sectional analyses of the cohort at study entry (22) demonstrated that RMR adjusted for FFM, fat, and age was significantly related to parental weight status, all of our longitudinal analyses were adjusted for parental obesity. Because the additional time points in a longitudinal analysis provide greater precision, we repeated the power calculations to account for the additional measurements on each subject and demonstrated that our statistical approach was powerful. Our statistical approach is powerful: given our sample of at least 132 subjects with 7.5 measurements per subject on average, our estimated minimum detectable effect that exceeds age-related change was 0.06 BMI z score units/y and 0.23% body fat/y at 80% power (1-ß) and a type I error rate = 0.05
Only a few studies examined this relation, and many of them have serious limitations. Two previous studies of children (17, 18) also observed no significant relation between reduced EE and weight gain. However, one of those studies failed to measure RMR under standardized conditions (17); another found unusually low AEE in some subjects (18) and did not consider sex differences in the analyses (18). A lack of consideration of sex might have distorted their reported results because pubertal changes in body weight and body composition differ considerably among boys and girls. Furthermore, in the study by Goran et al (17), some subjects were overweight at the time of study entry, which could have eliminated any prior differences in EE. In the study by Johnson et al (18), the weight status of the subjects was not presented, so it is not clear whether any of the subjects were overweight before the study began. In another study of children in which TEE and sleeping EE were measured in a respiration chamber, no relation between sleeping metabolic rate or TEE and weight gain was observed over a 2–3-y period (35). A summary of these studies is presented in Table 7.
View this table:
TABLE 7. Studies of energy expenditure and its relation to body fatness in infants, children, and adolescents
The lack of an observed association between RMR and weight gain during the pubertal period suggests that genetic factors associated with EE are not risk factors for increased body fatness over the pubertal period. However, the lack of association between AEE and weight gain during this period does not mean that activity is unimportant.
Our study had several notable limitations. One was that EE was measured only once in the premenarcheal period. Thus, we assumed that the 2-wk period in which we measured AEE and TEE represents activity levels throughout the preadolescent period. In adults, the CV among repeated measurements of EE appeared to range from 8.2% to 15.4% (36) and may have been related to both the precision associated with the mass spectrometry and the variability among daily EE. Furthermore, a small energy imbalance, if prolonged, can lead to significant increases in body fat stores. The doubly labeled water technique is not sensitive enough to measure the small reductions in EE that can contribute to a positive energy balance, and that could, if maintained for a long time, result in significant increases in body fat. Our outcome measures were based on %BF by BIA and BMI z score, neither of which is considered a criterion measure. However, we also examined the relation between reduced EE to change in body fat measured by TBW at baseline and study completion at 4 y after menarche. The analyses with TBW involved only 2 points. We did not observe a significant relation between RMR, AEE, or TEE and %BF assessed by TBW. For analyses using %BF as the outcome, we adjusted AEE for FFM, fat, and age at menarche; for analyses using BMI z score as the outcome, we adjusted AEE for FFM, fat, age, and age at menarche. Our decision to adjust AEE for the variables outlined earlier was based on preliminary analyses that showed a significant relation between AEE and FFM, fat, and age. In summary, our study suggests that lower EE, because of either metabolic or behavioral factors, in the premenarcheal period is not a risk factor for increases in relative weight gain or body fatness in girls during the adolescent period.
ACKNOWLEDGMENTS
We thank Pamela Ching, Julie Morelli, Katherine Getzwitch, Helene Cyr, Zoom Vu, Jennifer Spadano, the nursing staff at the Clinical Research Center for their assistance with the study, and the girls who enrolled for their commitment to the study.
LGB was principal investigator of the study and contributed to the design and data collection and drafted the manuscript. AM was a coinvestigator on the study and contributed to the analytic plan and manuscript preparation. SMP contributed to the data analysis. ENN provided assistance with the analytic approach and interpretation of models. WHD was the original principal investigator of the longitudinal study and contributed to the study design and data collection and contributed to the preparation of the manuscript. There was no conflict of interest for any of the authors.
REFERENCES