|Home | About | Journals | Submit | Contact Us | Français|
We examined the effect of dietary energy density change on body weight in participants of a randomized trial. Intervention participants markedly increased fruit and vegetable intake while reducing energy intake from fat. Participants were 2,718 breast cancer survivors, aged 26−74 yr, with baseline mean body mass index of 27.3 kg/m2 (SD = 6.3). We assessed dietary intake by sets of four 24-h dietary recalls and validated with plasma carotenoid concentrations. Weight and height were measured at baseline, 1 yr, and 4 yr. Dietary energy density was calculated using food but excluding beverages. Intervention participants significantly reduced dietary energy density compared to controls and maintained it over 4 yr—both in cross-sectional (P < 0.0001) and longitudinal (Group × Time interaction, P < 0.0001) analyses. Total energy intake or physical activity did not vary between groups. The intervention group had a small but significant weight loss at 1 yr (Group × Time interaction, P < 0.0001), but no between-group weight difference was observed at 4 yr. Our study showed that reducing dietary energy density did not result in a reduction in total energy intake and suggests that this strategy alone is not sufficient to promote long-term weight loss in a free-living population.
Fiber, water, and fat are the 3 most important determinants of dietary energy density (1-3). Consequently, most fruit and vegetables are generally low in energy density due to their high fiber and water content (4-7). It has been observed that the volume of an individual's dietary intake remains more or less constant (8), which has led to the hypothesis that people may regulate their food intake based on volume rather than total energy. Accordingly, replacing energy-dense, high-fat foods with much less energy dense, fiber-rich foods such as vegetables and fruit should result in a reduction of energy intake and weight loss (8,9).
Various cross-sectional studies have found that individuals who eat high-energy-dense foods consume more energy and are relatively heavier than those who consume proportionately greater amounts of low-energy-dense foods (10-13). A number of feeding studies that have manipulated dietary energy density have suggested that a decrease in energy density is associated with weight loss (14,15). Although these feeding studies have addressed important questions about the association between energy density and weight loss, longer term studies of individuals eating in real-life situations are necessary to test the hypothesis.
Ad libitum randomized trials that have encouraged participants to increase their fruit and vegetable intake and/or decrease their fat intake have had mixed results in terms of the amount of validated dietary change as well as weight change (16-23). None of these trials have reported the energy density of the diets in the intervention and control groups, and thus, it is possible that those studies that did not observe a decrease in weight may not have achieved a significant change in dietary energy density.
This article investigates the relationship between change in dietary energy density and body weight as an ancillary report of the Women's Healthy Eating and Living (WHEL) Study—a large-scale randomized trial of the role of a plant-based dietary pattern in reducing breast cancer recurrence and death (24). Participants in the WHEL Study intervention group significantly increased their fruit, vegetable, and fiber intake and decreased their intake of energy from fat (25,26), a pattern characterizing a low-energy density diet, whereas the control participants consumed their usual diet. The WHEL Study has assessed dietary intake at multiple time points, and thus provides the necessary data to assess change in dietary energy density according to method reported in the literature (27). In this article, we compare dietary energy density between the intervention and the control groups at baseline and demonstrate the association between dietary energy density and body weight. Then, we investigate the relationship of change in energy density to change in weight between study groups up to 4 yr postrandomization.
In this article, we consider participants of the WHEL Study. Population characteristics, eligibility criteria, randomization procedures, and dietary intervention protocol have been described in detail elsewhere (24,26).
All women enrolled in the WHEL Study who did not have a study endpoint (death or recurrence) by 4 yr of follow-up were eligible for this study (n = 2,718). WHEL Study participants were aged 18−70 yr at cancer diagnosis; treated for primary, operable, and invasive stage I, II, or IIIA breast carcinoma; and at study entry were not receiving or scheduled for chemotherapy and had no evidence of cancer recurrence after initial treatment. Enrollment in another dietary trial, pregnancy, receiving estrogen replacement therapy, and presence of life-threatening medical conditions or diseases were key exclusion criteria.
In this study, we used WHEL baseline, 1-yr, and 4-yr follow-up data and adopted its randomized design for data analysis (control = 1,363, intervention = 1,355). Dietary data at baseline, 1 yr, and 4 yr were available for 2,713 (control = 1,360, intervention = 1,353), 2,465 (control = 1,270, intervention = 1,195), and 2,324 (control = 1,202, intervention = 1,122) women, respectively. At the same time points, 2,718 (control = 1,363, intervention = 1,355), 2,306 (control = 1,174, intervention = 1,132), and 2,146 (control = 1,116, intervention = 1,030) women had their body weight measured.
Informed written consent from study participants was collected in the WHEL Study. The Human Subjects Committee of the University of California, San Diego, and all participating institutions approved the study procedures.
Participants in the intervention group were encouraged to maintain a dietary pattern that included a daily consumption of at least 5 vegetable servings, 16 ounces of vegetable juice (or equivalent vegetable servings), 3 fruit servings, 30 g of fiber (18 g/1,000 kcal), and 15−20% energy from fat (24,26). Telephone counseling, monthly cooking classes, and newsletters were the principal methods to promote dietary change in the intervention participants. Control group participants received print materials that included dietary guidelines from the U.S. Department of Agriculture (28) and the National Cancer Institute (29) and a bimonthly cohort maintenance newsletter with general health and nutrition information unrelated to the intervention group's dietary goals.
Dietary intake was assessed through a set of four 24-h dietary recalls at baseline, 1 yr, and 4 yr. Trained dietary assessors conducted these recalls by telephone on randomly selected days, stratified for weekend vs. weekdays, over a 3-wk period. The Nutrition Data System for Research (NDS-R) software was used to collect and estimate dietary intakes (NDS-R version 6.0, 2006, University of Minnesota, Minneapolis, MN). NDS-R included more than 18,000 food codes, including many ethnic foods, and over 8,000 brand-name products.
A number of strategies were used to maximize the accuracy of dietary recall data (30). Dietary assessors completed a training program that included standardized data collection, proper interview technique, and efficient use of dietary analysis software. Participants were trained, before study enrollment, to estimate serving sizes with food models, measuring cups, and spoons, and were provided with 2-dimensional food models for reference during recalls. In addition, assessors used a multipass method that improved recall accuracy by prompting to obtain detailed data about type, amount, and preparation method of foods eaten.
We determined a participant's dietary energy density (kcal/g; 1 kcal = 4.18 kJ) for a dietary recall day by estimating total energy intake (kcal) for that day and dividing it by the total amount (g) of food reported being consumed on that day. Energy density values of the set of 4 days were averaged to derive a mean dietary energy density value for each participant. In our calculations, we excluded all beverages.
Physical activity was determined from the Personal Habits questionnaire developed for Women's Health Initiative (WHI) (31), expressed as metabolic equivalents per week (Metmin/wk) (32), and completed at baseline, 1 yr, and 4 yr. For the WHEL Study, this questionnaire was calibrated with the standard 7-Day Physical Activity Recall (PAR) (33) and validated with an accelerometer reading (34). The accelerometer measured an average of 165 total min of physical activity per week, which was not statistically different from the 187 min reported for the PAR or the 171 min reported with the WHI 9-item questionnaire.
Weight and height were measured—with the participants wearing light clothing and no shoes—during clinic visits (baseline, Yr 1, and Yr 4) scheduled in the WHEL Study. Body mass index (BMI) was calculated as weight (kg)/height (m2).
Information on cancer stage (I, II, IIIA) and demography was ascertained through medical records and questionnaire, respectively. Age at study entry was categorized into 10-yr age groups (<44, 45−54, 55−64, and ≥65 yr), and race was categorized as non-Hispanic White, African American, Hispanic, Asian American, and others. Other variables included were education (college graduate vs. nongraduate), employment status (yes, no), marital status (married vs. not married), and smoking (current, past, and never). We calculated summary variables such as total fruit and vegetable intake (servings/day) and percent energy intake from fat/day from 24-h dietary recalls.
Plasma carotenoids are well-known biomarkers of fruit and vegetable intake (35). The WHEL Study measured plasma carotenoid concentrations on a 28% random sample of subjects identified at baseline and has published plasma carotenoid measurement procedures and baseline to 1-yr results (25,36). In this analysis, we report total plasma carotenoid concentrations on the available population (n = 881) at baseline, 1 yr, and 4 yr. Total plasma carotenoids are=the sum of the individual carotenoids separated and quantified (α-carotene, β-carotene, β-cryptoxanthin, lycopene, and lutein plus zeaxanthin) using high-performance liquid chromatography methodology (25). The mean laboratory day-to-day coefficient of variation for total plasma carotenoids was less than 7%.
We compared baseline characteristics of the control and the intervention groups; demographic, behavioral, and cancer related variables, thought to be potential confounders of the relationship between dietary intake and weight, were examined in this respect.
Energy density was calculated using “food only” values. We used baseline values to assess univariable associations of energy density with categories of age, race, and BMI; one-way analysis of variance compared category means against a referent category. We also grouped participants into tertiles of baseline dietary energy density, calculated mean values of total energy intake, physical activity, and body weight for each tertile and compared tertiles using the lowest tertile as referent. We then compared baseline dietary energy density between the control and the intervention group and graphed energy density in each study group at each time period.
We also computed and compared total energy intake, physical activity, and body weight values in each study group at baseline, 1 yr, and 4 yr, testing for group differences with t-tests.
Finally, we used mixed effect models to assess change in energy density, total plasma carotenoids, total energy intake, physical activity, and body weight over the study follow-up period. We chose mixed models, as they are the best option available for correlated data and for data with random missing values. “Unstructured” covariance provided the smallest Akaike's information criterion value and was used in the mixed models.
All calculations were performed using SAS version 9.1 (SAS Institute, Cary, NC). All statistical tests were two-tailed with an alpha level of 0.05.
Baseline characteristics did not differ significantly between the randomly assigned control and intervention groups (Table 1). Women were 26−74 yr of age (mean age = 53.4, SD = 8.8). The mean BMI was 27.3 (SD = 6.3), and 57% were overweight or obese. Although predominantly non-Hispanic White (85%), the cohort also included a small but varied group of minority women (4% African American, 3% Asian American, 5%, Hispanic, and 3% other ethnicities). Well-educated [college graduate (54%)] and predominantly employed (72%), 70% of the WHEL women were also married. Only a small percentage (<5%) was diagnosed with either stage IIIA cancer or was currently smoking. The mean energy intake and physical activity were 1,717 kcal/day (SD = 407) [7,184(1,703) kJ/day] and 868 metabolic equivalent task (MET)-min/wk (SD = 879), respectively (data not shown).
At baseline, energy density was inversely associated with age (P for trend < 0.0001) and directly associated with BMI (P for trend < 0.0001). Asian-American participants reported the highest intake of fruit and vegetables and the lowest energy intake from fat (data not shown), making the energy density of their diets significantly lower than any other racial/ethnic group (Table 2). We observed strong linear trends (P < 0.0001) across tertiles of energy density, with energy intake and body weight having strong positive associations and physical activity having a strong negative association. Participants in the highest tertile of energy density consumed, on average, approximately 300 kcal/day (1 kcal = 4.18 kJ) more and performed 450 MET-min/wk less physical activity than participants in the lowest tertile; mean body weight differed by 6.8 kg between these 2 tertiles (Table 3).
Mean dietary energy density did not differ between the intervention and the control subjects at baseline, although we observed a significant difference in dietary energy density between groups at 1 yr and 4 yr (P values < 0.0001). At 1 yr, the intervention group reported consuming a diet that was 25% less energy dense than their baseline diet. At 4 yr, this difference was still highly significant but had declined to 15% (Fig. 1). The multivariate analysis (Table 4) shows that these group differences in energy density were statistically significant at both1 yr and 4 yr (P values for group by time interaction <0.0001).
Total plasma carotenoid concentrations corroborated the between-group differences in fruit and vegetable intake as assessed by 24-h recall. In the validation sample (36), no significant differences were observed between groups at baseline, and carotenoid values in the control group were relatively unchanged at 1 yr and 4 yr. In contrast, total plasma carotenoid concentrations in the intervention group increased substantially, resulting in a 66% difference between groups at 1 yr and a 41% difference at 4 yr (data not shown).
Data for energy intake, physical activity, and body weight are presented in Fig. 1 and Table 4. At baseline, mean weight in the intervention group was slightly higher than the control group (+0.2%). At 1 yr, weight in the control group increased by 0.71 kg, whereas weight decreased by 0.05 kg in the intervention group, resulting in a mean weight in the intervention group that was 0.7% lower than that of the control group. The multivariate analysis identified this difference as statistically significant (Group × Time interaction: P < 0.0001). At 4 yr, both groups had gained weight, and the mean weight for the intervention group was 0.7% higher than that of the control group. The longitudinal analysis did not identify this as statistically significant (Group × Time interaction: P = 0.23).
Reported energy intake was essentially the same at baseline and 1 yr, and there was a nonsignificant 1.4% difference between groups at 4 yr. At baseline, the intervention group performed 5% less physical activity than the control group. Although both groups reported increasing their physical activity, the intervention group performed 3.6% less physical activity than the control group at 1 yr and 0.3% less at 4 yr. This change in physical activity was borderline significant at 4 yr (Group × Time interaction: P = 0.04).
In this group of breast cancer survivors participating in a long-term dietary trial, we observed that an increase in fruit and vegetable intake and decrease in percent energy from fat resulted in a substantial decrease in dietary energy density that was not accompanied by weight loss. Specifically, intervention participants significantly increased their intake of fruit and vegetables (2.7 and 2.3 servings/day, respectively, at 1 and 4 yr; data not shown) and decreased their percent energy intake from fat (5.7% and 4.3%, respectively, at 1 and 4 yr; data not shown). These dietary changes resulted in a large decrease in dietary energy density compared to the control group whose diets and energy density remained relatively unchanged.
At 1 yr, we observed a 25% between-group difference in dietary energy density, which was associated with small (0.7%) difference in weight in the hypothesized direction; although significant, this weight loss was much less than meets general guidelines for successful weight change (37-39). However, the intervention group sustained their reduction of dietary energy density through 4 yr, and this reduction was not associated with a maintained lower weight. Accordingly, the results of this study do not support the hypothesis that a major reduction in dietary energy density will independently result in weight loss.
A key component of the energy density–weight loss hypothesis is the assumption that people who adopt a low energy density dietary pattern will regulate their food intake by volume rather than by total energy. We did not observe this phenomenon in our study population. Despite a substantial increase in fruit and vegetable intake in the intervention group, their total energy intake did not change at either follow-up point. Likewise, we observed no meaningful difference in change of physical activity, a surrogate marker of energy expenditure, between the study groups. Thus, physical activity does not explain the finding of no difference in weight change between groups.
This study is one of the few to examine a longitudinal association between a change in dietary energy density and body weight. Our findings differ from the results of the 2 other trials in the literature that have examined this association (40,41). In both trials, weight loss was significantly correlated with decrease in dietary energy density. However, differences in the study population, intervention, and duration of follow-up between those 2 trials (40,41) and this one are substantial. The intervention in PREMIER trial (41) involves many more components than the dietary intervention in our study. In addition to promoting a high-fiber and low-fat diet, it also promoted weight loss and physical activity and restricted alcohol and sodium in-take. Unlike our study, both trials (40,41) focused on overweight or obese participants, setting up the possibility of a regression to the mean effect on weight. Further, subjects in our study maintained their dietary pattern across 4 yr, allowing us to investigate the long-term influence of such a dietary pattern.
All dietary studies need to address measurement error, and low-energy reporting is a concern, as several studies have observed higher frequency of low-energy reporting in their intervention groups (42-45). A related issue is whether intervention subjects were more prone to bias and reported eating more “socially desirable” foods such as vegetables and fruits or less fat than actually consumed, which would directly influence dietary energy density. Although differential underre-porting and social desirability bias among intervention subjects is possible, that could not explain the dietary difference observed between our study groups. Total plasma carotenoids—a biomarker of fruit and vegetable consumption—increased significantly among intervention subjects throughout the follow-up period but remained unchanged in the control group (Table 4).
This study has a number of strengths; primarily, its randomized trial design whereby randomization theoretically distributes all attributes of the study subjects, both measured and unmeasured, evenly between the groups. Neither reported caloric intake nor physical activity expenditure were different between study groups at any time point. The huge difference achieved in dietary energy density was confirmed with the accepted biomarker of vegetables and fruit. Further, in this study, we measured body weight and height, unlike many other studies that have used self-reported weight and height (46-48). Hence, the accuracy of outcome measures was higher. Finally, the cross-sectional associations of dietary energy density we described in this article are consistent with findings from previous studies (10,11,27,49).
However, this study was not a random sample of the population. WHEL participants were breast cancer survivors, generally White, highly educated, and predominantly employed; therefore, these results may not be generalizable to the population at large. Follow-up measured weight data were not available for 10% of subjects who did not attend clinic visits; however, this missing data did not differ between study groups (control = 9.9%, intervention = 11.2%). Finally, this study could not address the hypothesis of whether low energy density in conjunction with caloric restriction leads to long-term weight loss.
In summary, the intervention in this randomized trial significantly reduced dietary energy density and maintained this change over 4 yr. This change in dietary pattern was not associated with a change in energy balance (total energy intake vs. expenditure), and it did not result in a meaningful change in weight in free-living individuals. As a strategy to specifically reduce total energy intake, reducing dietary energy density may be a useful component of weight management. However, changing this characteristic of the diet without a targeted reduction in energy intake does not appear to result in either reduced energy intake or weight loss.
The authors thank Christine Hayes for her editorial support. This study was initiated with the support of the Walton Family Foundation and continued with funding from National Cancer Institute (NIH) Grant CA 69375. Some of the data were collected from General Clinical Research Centers, NIH grants M01–RR00070, M01–RR00079, and M01–RR00827.