点击显示 收起
1 From the Jean Mayer US Department of Agriculture Human Nutrition Research Center on Aging at Tufts University, Boston.
2 Supported in part by NIH grants T32AG00209 (to NPH), DK09747 (to MAM), and AG12829 and DK46124 (to SBR) and by the USDA, Agriculture Research Service, under cooperative agreement 58-1950-9-001. 3 Address reprint requests to SB Roberts, Energy Metabolism Laboratory, Jean Mayer USDA Human Nutrition Research Center on Aging at Tufts University, 711 Washington Street, Boston, MA 02111. E-mail: sroberts{at}hnrc.tufts.edu.
ABSTRACT
Background: The specific underlying causes of adult weight gain remain uncertain.
Objective: The objective was to determine the association of 3 measures of eating behavior with weight gain and body mass index (BMI; in kg/m2) in adults.
Design: Current dietary restraint, disinhibition, and hunger were assessed with the use of the Eating Inventory in 638 healthy, nonsmoking women aged 5565 y. In addition, subjects reported their current weight and height, their weight for 6 age intervals, and changes in voluntary dietary energy restriction over the past 10 y. Current weight and height were validated in 10% of subjects.
Results: Current disinhibition strongly predicted weight gain and current BMI (partial r = 0.27 and 0.34, respectively, both P < 0.001). Neither restraint nor hunger was a significant independent predictor of either variable, but the positive associations between disinhibition and both weight gain and BMI were attenuated by restraint (P = 0.016 and 0.010, respectively, after adjustment for confounding variables). In the subpopulation of women who reported a stable level of voluntary dietary energy restriction, disinhibition also strongly predicted weight gain and higher BMI, and restraint was negatively associated with weight gain (partial r = 0.17, P = 0.019).
Conclusions: Higher disinhibition is strongly associated with greater adult weight gain and higher current BMI, and dietary restraint may attenuate this association when disinhibition is high. These findings suggest that eating behavior has an important role in the prevention of adult-onset obesity and that further studies are warranted.
Key Words: Dietary restraint disinhibition body weight BMI obesity overweight postmenopausal women body mass index women
INTRODUCTION
The prevalence of obesity has increased worldwide in the past decade (13), and >55% of the adult US population is now considered overweight or obese (4, 5). National statistics show that weight gain during adulthood has increased by >50% in the past 30 y (1). Considerable uncertainty remains, however, over the specific causes (6).
Eating behavior may be important in the prevalence of obesity. The Eating Inventory (EI) of Stunkard and Messick (7) is a recognized instrument for quantifying eating behavior, specifically 3 constructs termed restraint, disinhibition, and hunger. Dietary restraint is defined as a tendency to consciously restrict food intake either to prevent weight gain or to promote weight loss by control over both energy intake and types of foods eaten (8), disinhibition is the tendency to overeat in the presence of palatable foods or other disinhibiting stimuli such as emotional distress (9), and hunger is the susceptibility to perceived body symptoms that signal the need for food (9).
Previous studies using the EI found that high disinhibition scores were consistently associated with high body mass index (BMI) (1012), whereas associations of dietary restraint scores with BMI were more contradictory. Several studies found a significant positive association of restraint scores with BMI (1315), but others found a negative association (16, 17). However, differences in BMI between individuals with high and low dietary restraint scores appear to be minimal, with perhaps an overall trend toward the association of higher scores with higher BMI values. Higher restraint scores are more clearly associated with greater weight loss during dieting (10, 1719) and better weight maintenance after weight loss (18). However, there is no published information on associations between long-term weight gain and eating behavior from studies using a valid instrument to separate the different aspects of eating behavior. Thus, the indication from current data that disinhibition promotes weight gain remains speculative. Moreover, many previous studies did not exclude smokers and individuals with chronic diseases or eating disorders (factors that are highly likely to confound the relation between eating behavior and body weight), which makes the validity of the existing data uncertain.
The present study was designed to investigate the association of eating behavior with weight change over 20 y and current body weight at 5565 y of age in a retrospective, cross-sectional investigation of a large sample of healthy women. The study was specifically designed to test the hypothesis that lower restraint and higher disinhibition and hunger are associated with greater weight gain and higher body weight.
SUBJECTS AND METHODS
Survey
Women aged 5565 y and living in New England were recruited by advertisement in local newspapers and by direct mail via commercially available mailing lists of names originally obtained from state motor vehicle registration records (Donnelley Marketing, Omaha) for a study described as examining nonspecified eating habits and health. After indicating their willingness to participate, subjects were mailed questionnaires and an informed consent form. Although not compensated for their completion of the survey questionnaires, subjects received a stipend if chosen to participate in the validation study component (see below). Ethical approval for the survey and the validation study was given by the New England Medical CenterTufts University Human Investigations Review Committee.
A total of 2088 women (representing a response rate of 60%) completed a medical and lifestyle history questionnaire, which included questions on current height and weight, the EI (7), and 12 supplementary questions (Appendix A). The supplementary questions concerned changes in body weight and food intake, including mean body weight at 6 age intervals (2029, 3039, 4049, 5055, 5560, and 6065 y; question 12) and change in estimated extent of self-imposed dietary energy restriction over the past 10 y (question 11).
View this table:
APPENDIX A. Supplemental questions on eating behavior
Initial scoring of the EI, which consists of 36 true or false and 15 multiple-choice questions, with 3 separate groups of questions used to calculate dietary restraint, disinhibition, and hunger scores, was completed according to published guidelines (7). For the analysis of eating behavior and weight change, a modified disinhibition score was calculated, omitting question 25 ("My weight has hardly changed at all in the last 10 y"), because this question might have undue influence on this analysis (ie, a response of "false" to this question correlates both with disinhibition and with weight change). To compensate for missing answers on some returned EI questionnaires, we calculated proportional scores ([raw score/(total number of possible answers - number of missing answers)] x total number of possible answers) for each eating behavior scale. Subjects with 4 missing answers for the restraint scale (out of 21 total questions), 3 missing answers for the disinhibition scale (out of 16 total questions), 3 missing answers for the modified disinhibition scale (out of 15 total questions), or 3 missing answers for the hunger scale (out of 14 total questions) were excluded from the data analysis. Because these criteria were used, a small number of subjects with incomplete questionnaires were included in the analysis after the calculation of proportional scores (n = 50, 41, 40, and 46 for restraint, disinhibition, modified disinhibition, and hunger scales, respectively). Although we are not aware that this scoring method has been described previously to compensate for the common problem of missing answers on EI questionnaires, its use is routine in other questionnaire studies.
A total of 653 subjects were excluded because they were current smokers or reported having a medical disorder that might influence weight or eating behavior or both (cancer, heart disease, glandular disorder, or eating disorder). Forty-two subjects were excluded because of incomplete EI questionnaires (exceeding the described limits), and 754 were excluded because of missing demographic or anthropometric information. One woman who reported an unusually large weight loss over the specified period (70 kg) was excluded as well, providing a final sample size of 638, which was 95% white, non-Hispanic; 1% black; 1% Asian; and <1% Hispanic and American Indian.
Validation study
Of the final sample, 10% (n = 67) were invited to participate in a more detailed study, described elsewhere (20), in which reported weights and heights were validated in women with a high level of restraint (EI restraint score 13) and with a low level of restraint (score 5). Fasting body weight was measured while the subject was wearing a preweighed gown, and height was measured with the use of a wall-mounted stadiometer.
The difference between reported and measured weights was not significant in the women with a low level of restraint (0.28 ± 0.47 kg) but was significant in the women with a high level of restraint (-0.88 ± 0.35 kg; P = 0.017). No significant differences were found between reported and measured heights (data not shown). There was a small but significant difference between BMI calculated with the use of reported and measured values in the group with a high level of restraint (-0.44 ± 0.17; P = 0.013) but not in the group with a low level of restraint (0.015 ± 0.22). The relatively small underestimation of body weight in eaters who exercise restraint is consistent with previous reports suggesting a generally good correlation between self-reported and measured body weights (2123) and relatively little effect of restraint on the accuracy of reporting (24).
Statistics
Statistical analyses were performed with SPSS 10.0.7 for WINDOWS and SYSTAT 9.0.1 (SPSS Inc, Chicago). Values are expressed as means ± SEMs. For the validation study analyses of reported compared with measured heights and weights, differences between groups were analyzed by using Student's independent t tests and comparisons between reported and measured variables were performed by using paired t tests. Differences were considered significant at P < 0.05, except for differences between groups, which were considered significant at P < 0.0125 after Bonferroni correction for the comparison of 4 restraint quartiles. For the survey, multiple linear regression was used to examine the relation between scores on the restraint, disinhibition, and hunger scales of the EI and both weight change from the age interval 3039 to 5560 y (calculated by subtracting the answer to question 12b from the answer to question 12e; Appendix A) and current BMI. Visual inspection of the residuals resulting from the regression models using normal probability plots failed to show any serious departures from normality.
Analyses were performed on 3 data sets. The primary survey analyses presented are on the healthy individuals who had complete information on anthropometric measurements and eating behavior and who did not smoke or report medical problems or eating disorders (n = 638). Although nearly 70% of the initial sample was excluded in this data set, this was only for the purpose of obtaining a clean sample without potential confounding factors such as missing data and inclusion of smokers and individuals with health or eating disorders. As described below, the entire valid data set was also analyzed (n = 1959 for BMI and 1458 for weight change), and the results were essentially identical to those from the main analyses. The third data set consisted of the subset of individuals who reported no or only a slight change in level of energy restriction during the past 10 y (response 2, 3, or 4 to question 11) and for whom weight change data were available for the same period (n = 199). There were 428 subjects reporting no or only a slight change in the level of energy restriction during the past 10 y, but only the 199 women aged 6065 y could be used in this component of the analysis because the younger women fell into an age group for which we did not have reported body weight for the same 10-y interval. This subset was analyzed to determine whether similar results would be obtained in individuals reporting relatively stable energy restriction. Differences in basic demographic characteristics between the 3 data sets were examined with the use of one-way analysis of variance, and Tukey's honestly significant difference test was used for post hoc comparisons of significant mean group differences.
Both main effects and interactions between independent variables were examined. Analyses were performed both with and without covariates of current age (y), years since menopause, parity, current hormone replacement therapy (no, yes), education level (low: postsecondary education = none, vocational school, or 2 y of college; high: postsecondary education = 4 y of college, graduate school, or professional school), initial BMI (BMI at age 3039 or 5055 y, depending on sample subset analyzed), and past smoking status (never, ever) to adjust for the potential influences of these variables.
RESULTS
Survey: comparison of 3 data sets
Characteristics of the 3 data sets of the women in the survey population are shown in Table 1. There were no significant differences in BMI between the total sample of subjects (n = 1959) and the healthy women whose responses made up the primary data set (n = 638). The BMI of the subsample who reported no change or a minimal change in dietary energy restriction over the past 10 y and who also reported weight change during the same period (n = 199) was significantly lower than the total sample, but only by a small amount. All 3 groups had mean BMI values that closely approximated the mean BMI of older women in a national survey (25). In addition, the mean reported weight changes from 3039 to 5560 y were consistent with previous reports (26).
View this table:
TABLE 1 . Characteristics of the 3 data sets of the survey population1
The primary data set and the entire study population
In the group of healthy subjects used for the primary study analyses, 87% gained weight, 8% lost weight, and 5% maintained a constant weight over time. Disinhibition and hunger scores ranged from the minimum to the maximum possible score (016 and 014, respectively), with the range of restraint scores nearly as wide (020 out of 021 possible). Higher scores reflect a greater tendency to exhibit that particular eating behavior characteristic.
Results of the multiple linear regression to evaluate current dietary restraint, disinhibition, and hunger scores as correlates of weight change from the age interval 3039 to 5560 y are shown in Table 2. Disinhibition (calculated with the omission of question 25 from the EI, as explained in Methods, although use of standard disinhibition scores in this model did not change the basic relation) was the only significant independent predictor of weight change, and there was a significant interaction between restraint and disinhibition. Also shown in Table 2 is an adjusted model predicting weight change, in which current age, parity, and education level were all significant. However, the inclusion of these variables did not alter the basic relation between eating behavior and weight change. In this analysis, current age probably served as a surrogate variable for demographic changes in weight gain that occurred over the 10-y age interval of the study population [ie, younger women may have gained more weight between the age intervals 3039 and 5560 y than did older women because their time interval occurred more recently (1)]. Past smoking status and BMI at the age interval 3039 y were also examined as potential confounders but were not significant; their inclusion in the model did not significantly alter the restraint, disinhibition, or interaction coefficients.
View this table:
TABLE 2 . Multiple linear regression analyses of factors associated with reported change in weight from the age intervals 3039 to 5560 y in nonsmoking, healthy women1
The adjusted interaction between restraint and disinhibition in predicting weight change is illustrated in Figure 1. Higher disinhibition was associated with greater weight gain, but higher restraint attenuated weight gain at high levels of disinhibition. Overall, disinhibition and restraint accounted for 19% of the variance in weight change in the unadjusted model and together with the confounding variables accounted for 23% in the adjusted model.
FIGURE 1. . Mean (±SEM) weight change between the age intervals 3039 and 5560 y in relation to tertiles of dietary disinhibition and restraint (7) in the primary survey population of healthy women (n = 638). Values are adjusted for current age, parity, and education level.
Unadjusted and adjusted eating behavior models predicting current BMI are shown in Table 3. The models predicting BMI were very similar to the models predicting weight change. The unadjusted model showed that disinhibition was the major independent predictor of BMI, with greater disinhibition predicting higher BMI. Hunger and restraint were not independently significant, but there was a significant interaction between restraint and disinhibition. An adjusted model predicting BMIin which current age, current hormone replacement therapy, parity, and education level were all significantis also shown in Table 3, but the inclusion of these confounding variables did not alter the basic relation between eating behavior and BMI that was observed in the unadjusted model. Past smoking status and number of years since menopause were also initially included in the model, but were not significant; their inclusion in the model did not alter the restraint, disinhibition, or interaction coefficients.
View this table:
TABLE 3 . Multiple linear regression analyses of factors associated with BMI at age 5565 y in nonsmoking, healthy women1
The adjusted relation between current BMI and tertiles of dietary disinhibition and restraint are presented in Figure 2. Women with a high level of disinhibition and high restraint scores had a lower BMI than did women with a high level of disinhibition and low restraint scores. Restraint influenced BMI in women with a high and medium level of disinhibition, but had little influence on BMI in women with a low level of disinhibition. Overall, disinhibition and restraint accounted for 29% of the variance in BMI in the unadjusted model and together with the confounding variables accounted for 32% in the adjusted model.
FIGURE 2. . Mean (±SEM) current BMI in relation to tertiles of dietary disinhibition and restraint (7) in the primary survey population of healthy women (n = 638). Values are adjusted for current age, current hormone replacement therapy, parity, and education level.
For the complete study populationincluding those women excluded on the basis of chronic diseases, smoking, and reported eating disordersthe results were essentially identical to those of the primary study. In particular, regression models predicting weight change and BMI using the larger samples gave essentially the same results as those reported here, and only very small changes in the unstandardized ß coefficients were observed (data not shown).
Subjects with a stable level of energy restriction over 10 y
In the 199 subjects who reported no or only a slight change in dietary energy restriction over the past 10 y and for whom data on weight change for the same period were available, disinhibition was again a significant independent predictor of weight change (Table 4 and Figure 3; P < 0.001). In this analysis, there was no significant effect of the potential confounders used in the primary analysis, and restraint was a weakly independent predictor of weight change (P = 0.019) in contrast with its significant interaction with disinhibition in the primary analysis. Note that the reason that weight gain appears to be lower in this analysis than in the analysis of the primary data set is that the time between measurements is less (10 y compared with 20 y). Disinhibition, but not restraint, also significantly predicted current BMI in this subpopulation (partial r = 0.61, P < 0.001).
View this table:
TABLE 4 . Multiple linear regression analysis of factors associated with reported change in weight from the age intervals 5055 to 6065 y in nonsmoking, healthy women with no or only a slight change in the reported level of dietary energy restriction over the past 10 y1
FIGURE 3. . Mean (±SEM) weight change between the age intervals 5055 and 6065 y in relation to tertiles of dietary disinhibition and restraint (7) in the subset of women with a stable level of energy restriction over 10 y (n = 199).
DISCUSSION
The results of this study are consistent with the suggestion that eating behavior may be an important factor determining excess weight gain during adulthood. In particular, we found a strong positive association between current scores on the EI disinhibition scale and reported adult weight gain. To our knowledge, this is the first report of an association between adult weight gain and eating behavior when the 3 constructs of eating behavior (restraint, disinhibition, and hunger) (7) were assessed separately. We also found a strong positive association between disinhibition and current BMI, with a substantial 15.2-kg weight difference (normalized for height) between groups of subjects with high (8) and low (3) scores for disinhibition. An association between current BMI and disinhibition was observed by Williamson et al (12) in a smaller study of women aged 1778 y and by Westenhoefer et al (10) in a study of readers of a women's magazine. In those studies, other factors influencing weight status were not taken into account or used to exclude unsuitable subjects; thus, the findings were uncertain. However, we obtained very similar results after excluding individuals reporting known factors that might confound the relation between eating behavior and body weight, including current smoking, eating disorders, and major diseases. The combination of our new findings and those of previous research suggests that disinhibition promotes adult weight gain and that strategies aimed at reducing disinhibition may be broadly effective in curbing the current epidemic of obesity.
Our results also suggest a role for dietary restraint in adult weight regulation. In the primary analysis of survey data from the healthy women, dietary restraint was not itself an independent predictor of current weight and adult weight change but instead moderated the association of disinhibition with obesity and weight gain. In the women who reported a stable level of self-imposed dietary energy restriction over the past 10 y, restraint was a weak independent predictor of weight change, with upper and lower tertiles for restraint predicting average weight gains of 3.0 and 5.3 kg, respectively. Viewed from these perspectives, dietary restraint is either beneficial in preventing weight gain in all individuals (ie, the subpopulation analysis) or beneficial only in those with high levels of disinhibition (ie, the primary survey analysis). Dietary restraint has been suggested to be an undesirable trait because of its reported association with increased body weight in some studies (1315), neuroticism (27), subclinical menstrual disturbances (28), lower bone mineral content (29), and higher cortisol excretion (30). The results of the present study alternatively suggest that restraint may be a critical moderator of adult weight gain; thus, they support and extend previous consistent findings (10, 12) made in studies that did not exclude individuals with medical disorders or smokers.
There are some strengths of our study that deserve mention. In particular, to our knowledge this is the first study to examine eating behavior predictors of adult weight gain using the most current instrument for assessing eating behavior that is not confounded by concern for dieting and weight fluctuation. In addition, we excluded smokers and subjects with eating disorders or chronic health problems, who may have confounded associations between eating behavior and body weight in previous studies.
There are also several important limitations to our findings. First, our approach of combining cross-sectional and retrospective study components cannot distinguish between eating behavior constructs as consequences or causes of weight gain, and further studies are needed in which subjects are followed over time. Second, the possibility exists that the population studied may not represent the general population of healthy women aged 5565 y. Finally, we were not able to validate the self-reports of past body weight. Concerning the question of whether our study population was representative of the general population, the recruiting information given to subjects was purposely vague to minimize selection bias. In addition, the mean reported weight change and BMI of our sample were very similar to those of other study population medians (25, 26). The further concern that subjects did not accurately report past body weight cannot be avoided. However, several previous studies suggested that underestimation of past weight is relatively minor (03 kg between reported and actual weight) (3134), indicating that the results of the present study are likely to be accurate.
Because of the observational nature of this study, no firm conclusions can be drawn regarding the direction of the association between eating behavior variables and weight gain. However, it is possible that disinhibition is both a cause and a consequence of excess weight, analogous to the interrelation between physical activity and body weight (35). Evidence suggesting that disinhibition promotes weight gain comes from studies showing that disinhibition is associated with both increased energy intake (36) and increased frequency of consumption of such high-energy foods as sweets, pastries, and butter or margarine (37). Moreover, because energy intake is known to be higher in overweight than in nonoverweight individuals (38), factors such as disinhibition that promote increased energy intake may also contribute to maintenance of excess weight. To address the issue of causality, we also analyzed a subpopulation of the total sample in whom reported level of self-imposed dietary energy restriction was stable over the past 10 y, anticipating that this might identify a group in whom dietary restraint had also been relatively stable, although not all studies have found a significant negative correlation between dietary restraint and energy intake (11, 39). In this analysis, both disinhibition and restraint were independent predictors of weight gain (with disinhibition again being strongly predictive), suggesting again that high levels of disinhibition may contribute to weight gain over time and further that high levels of restraint may help prevent weight gain. Further studies are needed to directly test the hypothesis that disinhibition leads to weight gain.
There are many possible reasons for the reported association of overeating with disinhibited eating (36), including an individual's attitude toward overeating (40), which may partly be mediated by the eating environment during childhood (41) and cultural norms. Another possibility is that the overeating associated with disinhibition is partly an inadvertent consequence of unhealthy dietary patterns, a high dietary variety from highly energy-dense foods, or both. In studies unrelated to disinhibition, high dietary fat consumption consistently increases energy intake (42), and high dietary variety is one of the strongest predictors of greater food consumption (4346) and body fatness (47) in humans. Studies of the relation between dietary disinhibition, dietary composition, and dietary variety are thus warranted.
In conclusion, the rising prevalence of obesity in the United States (4, 5) underscores the need for a better understanding of the causes of excess weight gain. Our results, combined with previous work in the field, suggest that high levels of dietary disinhibition and low levels of dietary restraint may be important contributors to the current high levels of adult weight gain and maintenance of that excess weight. Further studies in this area are warrantedespecially longitudinal and intervention studiesto provide more conclusive evidence on the direction of association between eating behavior characteristics and weight change.
REFERENCES