Whole grains (WG) are grains that contain the entire nut or seed kernel, including the endosperm, bran, and germ, from the plant from which they are produced [1
]. WG foods are those that are either 100% whole grain (e.g., WG rolled oats and WG brown rice) or foods that contain some proportion of a whole grain ingredient (e.g., whole grain bread containing whole wheat flour) [1
]. WG foods or foods containing significant quantities of WGs tend to be higher in fiber and contain more of other essential nutrients, including iron, zinc, magnesium, selenium, and B vitamins, than refined grains [1
]. Data from observational studies consistently indicate a relationship between WG intake and dietary fiber consumption. For example, comparing categories of WG intake in several cohorts in the United States and Europe shows that total dietary fiber intake is significantly associated with WG intake such that total fiber intake is generally 50–100% higher in the top versus the bottom quintile or quartile of whole WG intake [3
]. WG intake has been associated with healthful eating patterns and lower risk for several morbidities such as cardiovascular disease, diabetes, and obesity [1
]. The 2015 Dietary Guidelines for Americans (DGA) recommend that at least half of daily grain intake be from WG, and all healthy eating pattern examples, i.e., Healthy U.S.-Style Eating Pattern, Healthy Mediterranean-Style Eating Pattern, and Healthy Vegetarian Eating Pattern, in the DGA report include WG foods [1
]. Although daily intakes of total grains are close to recommended amounts, typically, Americans consume excess amounts of refined grains (e.g., white bread, grain-based desserts, white rice, etc.) and do not consume recommended amounts of WG foods (e.g., whole wheat bread, oatmeal, brown rice, etc.) [1
Findings from observational studies indicate that higher WG intakes are associated with lower risks of weight gain and incident overweight or obesity [6
]. In a review by Karl et al. (2012) on the role of WG in body weight regulation, the authors concluded that the studies completed to that point in time had not provided evidence that a hypoenergetic diet that includes 3 to 7 daily servings of WG (48–112 g/day WG) promotes greater weight loss than a control (either no intervention or foods with refined grains) hypoenergetic diet [7
]. However, results from some studies have suggested that a hypoenergetic diet including WG-containing foods may be associated with a greater reduction in body fat, particularly abdominal fat, relative to a hypoenergetic, lower WG diet [7
]. Thus, Pol et al. (2013) concluded that WG consumption does not decrease body weight compared with weight of the control group, but a small beneficial effect on body fat may be present [9
Many WG foods are good sources of dietary fiber [1
], and WG intake directly correlates with dietary fiber intake in the U.S. [4
]. However, since WG intake among average Americans is <1 serving/day WG [1
], and high quantities of fiber-poor refined grains are consumed daily [11
], refined grain-based foods are actually the primary source of dietary fiber in the U.S. [11
]. Incorporation of fiber into the diet, depending on fiber type, can favorably impact health, including attenuation of blood cholesterol and glucose levels, and improved laxation [13
]. Certain types of dietary fibers exert physiological effects that may impact weight status. Beta-glucans and resistant starch type 4, for example, have been found to increase satiety [16
], though more research is needed. In addition, WG-containing foods collectively contain other bioactive components, such as lignans and phytosterols, shown to exert metabolic effects which have potential to influence body weight and adiposity [18
]. Given that some observational studies report a link between WG intake and body weight, and several WG food components could plausibly affect body weight regulation, the aim of this review was to provide an updated quantitative analysis of data from both observational studies and RCTs examining the relationship of WG intake with body weight status and related variables.
2.1. Literature Searches
The Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines were followed for performing the meta-analyses [21
]. A comprehensive literature search was conducted using the Ovid Medline database, which covered studies published from 1946 through January 2018. The search was designed to identify publications of observational studies and RCTs that examined WG intake from WG foods (e.g., oats, quinoa, wild brown rice, etc.) or foods made with WGs (e.g., whole grain breads, whole grain ready-to-eat breakfast cereals, etc.) and not supplements or specific food additives (e.g., dietary fiber supplements). The search strategy used several terms for WG (whole grain, wholegrain, whole-wheat, wild rice, whole rye, buckwheat, oat, etc.). Full search term details are provided in Table S1
2.2. Inclusion and Exclusion Criteria Screening
Inclusion and exclusion criteria were applied through a three-level screening process. Full inclusion and exclusion criteria and details are provided in Table S2
for observational studies and Table S3
for RCTs. Final inclusion criteria included study conducted in humans, English language, intervention arm (for RCTs) or a primary exposure variable (for observational studies) where whole foods (e.g., WG bread, brown rice, etc.) are the source of WG, and the WG-containing food is independently assessable and not part of a mixed intervention such as a diet that increases fruits, vegetables and WGs simultaneously. For the observational database, a weight-based anthropometric outcome of interest (i.e., body weight, body mass index (BMI), adiposity, fat-free mass, waist circumference) had to have been examined. The final exclusion criteria included animal studies, in vitro studies, studies conducted in children (<18 years) or pregnant women, studies assessing gluten-free and/or oral rehydration interventions or associations, reviews, bibliographies, case reports, letters, and/or no WG intervention or assessment.
To identify publications, one scientist (CS) performed two separate literature searches—one aimed at capturing observational studies and another aimed at capturing intervention studies. Publications identified using the search terms underwent the first level of screening using Abstrackr (http://abstrackr.cebm.brown.edu
). Abstract screening was conducted by one scientist (KL or CS). Full texts of all publications identified as potentially eligible from the abstract screening phase were then obtained and reviewed for eligibility (level 2 screening) by one scientist (CS or KL); however, texts that were unclear with respect to eligibility were additionally reviewed by an additional scientist (NM). Studies that were excluded for not meeting eligibility criteria during level 2 screening were reviewed in duplicate prior to final exclusion. All discrepancies were resolved by a scientific team with oversight by NM. After full text reviews were completed, PICO (population, intervention, comparator, and outcome) data and results were extracted from eligible publications into one of two databases (one for intervention studies, one for observational studies) that were created with input from the research team. For the intervention study database PICO information was extracted, and then the database was searched for and restricted to studies with anthropometric outcomes of interest for the meta-analysis [22
]. The observational database was created later, with the goal of a meta-analysis already established; thus, entries were only included in the database if they had the anthropometric outcomes of interest (per inclusion criteria stated above). All data were extracted and entered into the respective database by one scientist (NM, KL, or CS) and then reviewed in full for accuracy by a second scientist.
Publications in the two databases then underwent an additional level 3 screening by two scientists (OMP and HNC) to determine eligibility for final inclusion in the meta-analyses presented here. Each scientist independently performed the level 3 screening, and disagreements in the final inclusion/exclusion criteria were discussed among the scientific team until consensus was reached. Additional inclusion criteria and exclusion criteria for level 3 screening for observational studies included screening for publications which specifically assessed a measure of weight (kg) or weight status (BMI) as an outcome measure of interest and studies providing cross-sectional data for methodological consistency. Additional inclusion criteria and exclusion criteria for level 3 screening for RCTs screened for studies which specifically assessed a measure of weight (kg) or weight status (BMI) as an outcome variable and where the intervention was at least 12 weeks in length. If additional anthropometric measurements, e.g., whole-body adiposity, waist circumference, fat-free mass, etc., were also part of a study’s outcome assessment for any of the observational studies or the RCTs, baseline data, and in the case of RCTs, end-of-treatment data, were also recorded for potential secondary analyses.
2.3. Meta-Regression Analysis of WG Intake: Cross-Sectional Studies
A meta-regression analysis was performed on results of the observational studies to evaluate cross-sectional associations, which applied several assumptions. When mean or median WG intake for each category of WG intake was reported, this value was used in the analysis. When intake was not reported (n = 4 studies from 3 publications) [3
], the midpoint of the range of values reported within a category was employed. For the study reported by Albertson et al. (2016), WG intake was presented in categories consisting of 0 servings/day, >0 to <1 servings/day, or ≥1 serving/day [23
]. The >0 to <1 servings/day was estimated to be equivalent to 0.1 g/day to 15.9 g/day and the ≥1 serving/day was estimated to be equivalent to ≥16 g/day. WG intake in the highest category of ≥1 serving/day (≥16 g/day) was further estimated, for this study only, by assuming that approximately 70% of the incremental dietary fiber between the middle and highest WG intake groups was attributable to WG intake, with 3.58 g of fiber per 16 g of WG. Sensitivity analyses were completed to assess the degree to which different assumptions for the Albertson et al. (2016) study impacted the overall results [23
]. Varying the estimate for the highest WG intake group from 16 to 40 g/day did not materially alter parameter estimates for the study.
Because of differences in analytical (e.g., statistical, dietary assessment, etc.) methods employed and reporting of multiple analyses over different follow-up periods, some within the same cohorts, it was not possible to conduct a meaningful pooled analysis of data from prospective cohort studies. Therefore, a qualitative assessment of the results was undertaken to evaluate strength, consistency, and dose–response for the associations between baseline WG intake and change in WG intake and change in measures of body weight.
2.4. Meta-Regression Analysis of WG Intake: RCTs
The primary outcome of the RCT data meta-analysis was change in body weight (kg), expressed as the standardized mean difference between the exposed group with the highest WG intake reported and the control group. Secondary and sensitivity analyses were conducted to assess the relationship of higher WG versus a control on (1) change in waist circumference (cm), (2) change in body fat percentage, (3) weight change (kg) in a subset of studies that included subjects of both sexes, and (4) weight change (kg) in hypocaloric intervention studies.
Cochrane risk of bias for clinical trials was assessed [25
] with nutrition-specific items from a critical appraisal of systematic reviews in the field of nutrition [26
]. The methodologic quality of each study was evaluated based on predefined criteria, in accordance with the Agency for Healthcare Research and Quality recommendations for systematic reviews [27
]. Study quality for individual domains was determined in duplicate (OMP and HNC), and discrepancies were resolved by consensus in group conference. Preliminary study quality screening included ensuring studies met all the predetermined inclusion and none of the exclusion criteria as well as adequate study length and statistical power for RCTs and the inclusion of relevant cofounding analyses for observational studies. Study quality was assessed in duplicate by two scientists (KCM and OMP) using the Heyland Methodologic Quality Score (MQS)—a tool which rates study methodologic quality on the basis of nine criteria: random assignment, analysis, blinding, patient selection, baseline group comparability, extent of follow-up, treatment protocol, co-interventions, and outcomes [28
]. Studies are rated between 0 (lowest quality) and 14 (highest quality), and studies with a rating of ≥8 are considered high-quality trials (Table S4
2.5. Statistical Analyses
Descriptive statistics and both unweighted and weighted meta-regression analyses were completed using SPSS Statistics, version 25.0 (IBM, Armonk, NY, USA). Since insufficient data were available for inverse variance weighting for all studies, the weighting scheme used the number of subjects in each group as the weighting factor. Unless otherwise specified, an alpha level of 0.05 was used to define statistical significance.
Pooled analyses were completed using the Meta-analysis with Interactive eXplanations (MIX, version 2.0) program [30
]. Within-group changes and standard error (SE) for within-group change were based on reported values obtained from the publication; when these values were not reported, they were calculated from the reported group mean and SE or standard deviation (SD) for the baseline and final values within each group:
SEChange = (SE2Final − SE2Baseline − 2 × r × SEFinal × SEBaseline)0.5,
where r is the correlation coefficient between baseline and final values (within-group). A value of 0.59 was used for r, as suggested in an empirical evaluation of within-group correlations [31
]. The SE of the difference in response between groups was calculated as follows
SEDifference = (SE2ChangeWG + SE2ChangeControl)0.5.
Values are reported as standardized mean differences between treatment groups with corresponding 95% confidence intervals (CIs). Pooled estimates with 95% CIs and p
-values were calculated from random effects meta-analysis models. For the RCT analyses, between-study heterogeneity was assessed with the Q and the I2
The results of the meta-regression analysis of the cross-sectional evidence show a significant inverse relationship between WG intake and BMI. Findings from prospective cohort studies support this relationship, with baseline WG intake and change in WG intake generally showing inverse associations with weight change during follow-up periods of four to 20 years, particularly in the studies with larger numbers of subjects. These relationships remained statistically significant in most cases after adjustment for a variety of covariates and potential confounders. These results are consistent with those from a recent meta-analysis of prospective studies assessing the association of food group intake and risk for overweight/obesity and weight gain [52
]. In that analysis, five studies were included in the meta-analysis on incident overweight and/or obesity, which yielded a summary relative risk of 0.85 (95% CI: 0.79–0.91) for high vs. low WG product intake (I2
= 0%). Three studies were included in that paper’s final dose–response meta-analysis for the WG food group [37
], where an inverse relationship (relative risk for overweight/obesity 0.93, 95% CI 0.89–0.96) per 30 g/day higher intake of WG products was observed. Three studies were included in the analysis of WG product consumption and the risk for weight gain, with weight gain defined as >2 kg during a mean period of 4 years, ≥10 kg during 13 years, or ≥25 kg during an average period of 12 years. The summary relative risk (95% CI) for weight gain was 0.83 (0.70 to 0.97), with I2
= 16% in the high compared with low intake analysis, and 0.91 (0.82 to 1.02), I2
= 69%, for each increase of 30 g of whole grain products/day.
Results from the primary and secondary meta-analyses of RCT evidence in the present investigation failed to show a significant effect of higher WG intake on body weight, consistent with findings from some prior reviews [6
], but not with those from another recent meta-analysis [54
]. Reynolds et al. (2019) reported on the results of meta-analysis of 11 RCTs (919 adult participants) assessing the effect of WG on body weight and concluded that WG intake has a significant (mean difference: −0.62 kg, (95% CI −1.19 to −0.05)) association with change in body weight [54
]. Some differences exist between the present analyses compared to that of Reynolds et al. (2019). Of the nine studies included in the present meta-analysis and the 11 included in the Reynolds et al. (2019) analysis, only Brownlee et al. (2010) and Chang et al. (2013) were included in both [49
]. The remaining nine RCTs included by Reynolds et al. (2019) did not meet our inclusion/exclusion criterion of a 12-wk minimum intervention period [55
], did not provide sufficient information to quantify the amount of WG in the intervention [59
], did not provide WG in food (e.g., the intervention was provided as a high-fiber fraction of WG) [61
], or instructed subjects to maintain stable body weight [59
]. Reynolds et al. (2019) excluded trials that employed a hypocaloric (weight loss) diet as part of the intervention, whereas they were included in the present analysis.
One possible explanation for the lack of apparent effect of WG intake on body weight in RCTs in the present analysis may be that WG, per se, may not be causally related to body weight or related anthropometric variables. WG intake could be a marker for lifestyle or habits conducive to lower body weight, and the relationships in observational studies could be attributable to residual confounding. For example, WG intake may correlate with healthful lifestyle factors such as healthy dietary patterns, mindful eating behaviors, greater physical activity levels, and/or longer sleep duration [63
]. This phenomenon has been observed in children where oatmeal intake at the breakfast meal was a marker for better overall diet quality and nutrient intake versus other typical breakfast foods (e.g., eggs; ready-to-eat, high-sugar cereals; and pancakes/waffles) [67
]. In U.S. adults, WG intake is positively associated with higher diet quality and higher intakes of most micronutrients, dietary fiber, polyunsaturated fatty acids, and total energy, and inversely associated with intake of total and added sugars, monounsaturated fatty acids, saturated fatty acids, and cholesterol [68
]. In the United Kingdom, WG intake is associated with higher intakes of magnesium, fiber, and iron, and lower intake of sodium [69
]. Therefore, it is possible that the association of WG intake with lower weight status is due to residual confounding and is noncausal [70
]. The potential for bias and other types of confounding is an inherent limitation of observational studies, and, ideally, such associations should be confirmed with evidence from well-controlled RCTs [70
Another possible explanation for the differing relationships in the observational and RCT analyses is that the RCTs may not have been adequate to assess longer-term effects of WG intake on body weight and composition. There are several biologically plausible mechanisms through which higher WG intake could affect energy balance and body composition, including effects on appetite and energy expenditure [7
]. For example, in a 3-week crossover, blind intervention study assessing the effect of daily breakfast intake of WG rye porridge versus refined flour wheat bread, increases in postprandial subjective ratings of satiety were observed with the rye porridge in healthy adults [71
]. In addition, there are some potentially relevant mechanisms that may be mediated by effects of WGs and components, such as fermentable fibers, on gut microbiota [14
]. For example, consuming an evening meal containing WG rye flour bread, versus the refined flour, wheat-based bread meal, reduced circulating free fatty acids and increased breath hydrogen, two indicators of increased gut fermentation, in healthy adults [74
]. These influences may be too small to have a meaningful impact on the short-term and may require longer periods to manifest. The longest follow-up period in the RCTs assessed was 16 weeks and only two of the nine RCT comparisons were from interventions longer than 12 weeks [51
An additional consideration is variation in the definitions of WG foods used in both RCTs and observational studies [22
]. In 2006, the United States Food and Drug Administration adopted a WG definition that includes intact, ground, cracked, or flaked fruit of grains whose principal components (the starchy endosperm, germ, and bran) are present in the same relative proportions as in the intact grain [22
]. Prior to that, some studies included bran and other high dietary fiber foods in their definitions of WG [75
]. Since five of the 12 studies included in the observational data for this study are from 2006 or earlier, reported levels of WG intake in those studies potentially include foods/ingredients that are no longer defined as WG. In addition, an analysis of WG intervention study designs found that 73% of WG intervention studies did not specify a definition for the WG product or food, and only 55% of longer-term WG intervention trials reported the amount (as grams or servings) of WG used [22
]. With respect to variation in WG exposure within these RCTs, WG-containing foods vary in the quantity of WG within a food or product. A specific threshold of WG intake, or one or more components, may be needed to achieve some physiological effects; for example, thresholds of intake and quality are required to yield adequate viscosity in the stomach with oat intake to affect postprandial glycemia [76
]. Furthermore, variations may exist with respect to the type of dietary fiber and other potentially bioactive compounds. For instance, both oats and barley are rich in β-glucans [77
], but the major phenolic antioxidants in oats are p
-hydroxybenzoic acid and vanillic acid, while barley has higher levels of ferulic acid, p
-coumaric acid, and sinapic acid [77
Lastly, although no significant effects of WGs were observed in RCTs on body composition, the analysis included relatively few studies within which there was marked heterogeneity of results, and results from some trials suggest that body composition and/or body fat distribution may be influenced by WG intake [7
]. Thus, research with longer intervention periods is needed to assess parameters such as adiposity and waist circumference. Additional RCTs are also needed to assess possible influences of WG intake on the determinants of energy balance (appetite and energy expenditure). Since different WG types likely exert varying physiological effects, RCTs assessing the influences of specific WG types on weight status and related anthropometrics are needed. Additionally, well-controlled RCTs with a clear and standardized definition as to what constitutes a WG food (e.g., ≥51% WG ingredient by weight, only 100% WG food, etc.) are needed to reduce heterogeneity, and exploration of dose–response also warrant further investigation.