Whole grain intake, compared to refined grain, improves postprandial glycemia and insulinemia: a systematic review and meta-analysis of randomized controlled trials

Abstract Whole grain (WG) intake has been associated with reduced risk of type 2 diabetes (T2D) and may protect against T2D by lowering postprandial glycemia and insulinemia and improving insulin sensitivity. The objective of this systematic review and meta-analysis was to evaluate the effect of WG intake, compared to refined grain (RG) intake, on postprandial glycemia and insulinemia and markers of glycemic control and insulin resistance in randomized controlled trials (RCTs) in adults. A search of PubMed and Scopus yielded 80 relevant RCTs. Compared to RG, WG intake significantly reduced postprandial glycemia (SMD: −0.30; 95% CI: −0.43, −0.18; P < 0.001), insulinemia (SMD: −0.23; 95% CI: −0.35, −0.10; P < 0.001) and glycated hemoglobin (HbA1c) (SMD: −0.21; 95% CI: −0.37, −0.06; P = 0.007). There was no effect of WG on fasting glucose, fasting insulin, or homeostatic model assessment of insulin resistance (HOMA-IR). These results suggest WG foods improve short-term glycemia and insulinemia, which may improve HbA1c, a marker of long-term glycemic control. This may partially explain the inverse association between WG intake and risk of T2D, but further investigations are needed to understand if short-term reductions in glycemia translate to longer term benefits in reducing the risk of T2D. Systematic Review Registration: PROSPERO Registration CRD42020180069. Supplemental data for this article is available online at https://doi.org/10.1080/10408398.2021.2017838


Introduction
Whole grains (WG) have been defined as intact, ground, cracked or flaked grain kernels that contain all three anatomical components -endosperm, bran, and germ -in the same relative proportions as they exist in the intact kernel (American Association of Cereal Chemists International 1999; Van Der Kamp et al. 2014). Compared to RGs that typically have most or all of the bran and germ removed, WGs have more fiber, B vitamins and several minerals. Consequently, many health authorities recommend WG as part of a healthy eating pattern (Health Canada 2013; Public Health England 2016; U.S. Department of Agriculture and U.S. Department of Health and Human Services 2020), but around the globe, many individuals continue to consume more RG foods than WG foods.
Observational studies have consistently shown higher intake of WGs to be associated with lower risks for several chronic diseases, particularly type 2 diabetes (T2D) (Al Essa et al. 2015;Hu et al. 2020;Parker et al. 2013;Reynolds et al. 2019;Schwingshackl et al. 2017;Seal and Brownlee 2015). In fact, a recent umbrella review found the observational evidence for WG intake and reduced risk of T2D was stronger than for any other metabolic disease (Tieri et al. 2020). RCTs have been conducted to assess possible mechanisms that may account for the reduction of T2D risk with increasing WG intake. Most studies have focused on the effect of WG on postprandial glycemia and insulinemia as well as long-term markers of glucose control, such as insulin sensitivity and glucose tolerance. Dietary fiber and resistant starches found in WG have the potential to influence digestion and absorption of glycemic carbohydrates (Jovanovski et al. 2019). Additionally, fermentation of fiber and resistant starch can lead to the formation of short chain fatty acids (SCFA) which may influence gastrointestinal hormone secretions. These incretin hormones can amplify the insulin response to ingested carbohydrates and slow gastric emptying which may improve glycemic and insulinemic responses at subsequent meals or across the day (Canfora, Jocken, and Blaak 2015;Müller et al. 2019). WGs also have high antioxidant capacity, and some have proposed these antioxidants may protect against oxidative damage to pancreatic beta-cells, although data from humans to support this hypothesis are limited (Pereira et al. 2002;Seal and Brownlee 2015;Slavin 2003). Results from in vitro studies have suggested phenolics found in grains may also be able to influence digestion and absorption of starches (M. Li et al. 2017).
Recent meta-analyses have summarized RCTs on WGs and glycemia with a focus on types of WG (Musa-Veloso et al. 2020;Musa-Veloso et al. 2018) or specific populations (e.g., healthy individuals) (Marventano et al. 2017). However, there are no meta-analyses examining multiple types of WG in different populations (healthy and with metabolic dysfunction). Therefore, the purpose of this systematic review and meta-analysis was to evaluate the impact of WG foods, compared to RG foods, on short-term and longer-term markers of glycemia and insulinemia in RCTs in healthy adults and adults with metabolic dysfunction (e.g., T2D, metabolic syndrome). The primary outcomes were postprandial glycemia and insulinemia and the secondary outcomes were postprandial glycemia and insulinemia at a subsequent meal, fasting glucose, fasting insulin, glycated hemoglobin (HbA1c), homeostasis model assessment of insulin resistance (HOMA-IR), and glucose tolerance as assessed by an oral glucose tolerance test (OGTT).

Literature searches
This systematic review and meta-analysis followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (Moher et al. 2009). The PubMed and SCOPUS® databases were utilized to conduct a comprehensive literature search which covered studies published from 1946 through April 2020. The search was designed to identify publications of RCTs that examined WG intake from intact WGs (e.g., oats, wheat, brown rice, etc.) or foods made with WGs (e.g., muffins, ready-to-eat breakfast cereals, etc.) and outcomes related to glycemia and insulinemia, including postprandial glucose and insulin, fasting glucose and insulin, insulin resistance measures (e.g., HOMA-IR, Matsuda Index), glucose tolerance measures, and HbA1c. Full details on search terms are provided in (Supplemental Table 1). An updated literature search was performed again in December 2020 in PubMed only to identify relevant studies published since the initial search. No additional studies were identified.

Inclusion and exclusion criteria
Inclusion criteria consisted of RCTs conducted in adult humans (≥18 y of age), published in English language, WG or WG-containing foods as the main intervention compared to RG foods as a control, documented (or the ability to determine) quantitative intake of WG, and a measurement of glycemia or insulinemia. Exclusion criteria included observational studies (cross-sectional, retrospective or prospective cohorts), case-control or single-arm studies with no control condition, studies in animals or in vitro, studies without a RG control, multicomponent interventions where the effect of WG cannot be determined, interventions on individual grain components (e.g., bran) or dietary supplements, studies in children (<18 y of age) or pregnant/lactating women, and interventions administered via tube feeding or enteral nutrition. Except for T2D mellitus, obesity, or metabolic syndrome, studies in subjects with a chronic disease were also excluded.

Screening and data extraction
After removal of duplicate publications, first level screening of titles and abstracts was completed independently by a member of the research team using Abstrackr (http:// abstrackr.cebm.brown.edu/). Full texts of potentially eligible publications were obtained for further review. Publications that were unclear with respect to eligibility were discussed among the research team to determine inclusion or exclusion. Reference lists from relevant publications were reviewed to determine any additional studies for inclusion. One reviewer extracted PICO (population, intervention, comparator and outcome) data from the eligible studies into a database which was then verified for accuracy independently by a second reviewer. Discrepancies were resolved by discussion among the research team by referencing the original publication. Based on the original PROSPERO registered protocol, outcomes extracted from eligible studies included acute postprandial glycemia and insulinemia, fasting glucose and insulin, markers of insulin resistance and HbA1c. During the data extraction several publications measuring postprandial glycemia and insulinemia at subsequent meals (i.e., WG fed at evening meal, postprandial glycemia measured next morning after standardized breakfast) were identified, thus, these two outcomes were also included in the data extraction.
In publications where outcomes were reported in bar graphs, means and standard deviation (SD) or standard error of the mean (SEM) were estimated using Engauge Digitizer software version 2.14 (http://digitizer.sourceforge.net/) and included in the database. If studies reported measuring glycemic or insulinemic response but did not report the data or the variability, the corresponding author was contacted by email to request the quantitative data. One author responded with additional data for inclusion in the meta-analysis. One publication (Breen et al. 2013) did not report SDs or SEMs for postprandial insulin and the email address of the corresponding author was no longer valid, therefore, the SDs for the outcomes were estimated as the third quartile SD reported by other studies of similar duration.
For studies that did not include the amount of WG in the final food, the test food recipes or commercial label information was reviewed. In the case of low moisture foods, such as flakes or pasta, the % WG in the dry ingredients of the recipe was estimated as the % WG of the final food. For higher moisture foods, such as bread, the % WG in the dry ingredients of the recipe was estimated as the % WG of the final food after adjustment for the moisture content. The moisture content of the final food was determined by values provided in the publication, when available. When not provided in the publication, moisture content was estimated using Food Data Central from the USDA Agriculture Research Service (US Department of Agriculture ARS 2019).

Assessment of study quality
Risk of bias was assessed with the Cochrane risk of bias tool for randomized trials (Sterne et al. 2019), using the appropriate tool for parallel and crossover studies. The Grading of Recommendations Assessment, Development and Evaluation (GRADE) method (Guyatt et al. 2008) was utilized by the research team to evaluate the quality of evidence for each outcome.

Statistical analysis
Meta-analyses were completed using Comprehensive Meta-Analysis Software version 3 (Biostat, Englewood, NJ). All glycemia and insulinemia measures were identified as the main outcomes, but as acute postprandial measures were the most frequently measured outcome, postprandial glycemic and insulinemic response were selected after the data extraction, but prior to data analysis, to be the primary outcomes. The primary analysis for postprandial glycemia and insulinemia used pooled standardized mean difference (SMD) estimates (WG versus RG control) and 95% confidence intervals (CI) for area under the curve (AUC) of postprandial glucose (mmol × timeframe/L) and insulin (nmol × timeframe/L). For studies reporting multiple timeframes for AUC measurement (e.g., 120 min and 180 min), only the longest time frame was included in the main analysis. SMDs express the effect size relative to the standard deviation and do not have units. This approach enables pooling of measurements taken using different scales or measurement timeframes (e.g., 120 min, 180 min, etc.). SMDs should not be interpreted as true differences in means between the test and control because the difference in means is divided by the standard deviation. The primary analysis for other outcomes used pooled SMD estimates (WG versus RG control) and 95% CIs for change from baseline in fasting values of glucose (mmol/L) and insulin (pmol/L), AUCs for OGTT (mmol x timeframe/L) and means of HOMA-IR and HbA1c (%). Individual study SMDs were computed using the mean and variance (or SEM) for within-treatment changes, when reported. Otherwise, the mean change within each treatment was computed as the mean of the treatment values minus the mean of the baseline value. The SEM for mean change was computed using an imputed within-treatment correlation of 0.59 (Balk et al. 2012). Statistical significance for individual study and pooled SMDs was declared when the 95% CI does not include the null value of 0 (i.e., p-value < 0.05). Studies were weighted according to the inverse of the variance of each study's effect using random effects models. Random effects models were chosen for the primary analyses due to differences across studies in key design elements such as subject characteristics, type of WG and length of test period. Fixed effects models were also evaluated as sensitivity analyses. The magnitude of effect sizes were interpreted as < 0.40 = small, 0.40 − 0.70 = moderate and > 0.70 = large (Schünemann et al. 2019).
Data from parallel and crossover RCTs were entered into the software using the following data entry formats: "Independent groups (means, SD)" and "Paired groups (means, SD)", respectively. SMDs and corresponding SEMs for individual studies were computed by the software using the methods for independent groups and matched groups described by Borenstein et al. (2009a) and an imputed between-treatment correlation of 0.50 for matched groups. Sensitivity analyses were performed for the primary analyses assuming no between-treatment correlation for matched groups and the results were similar. For multiple comparisons within a study (i.e., comparisons that shared a common active or control group or comprised data from the same subjects), individual effect size estimates and variances were computed for all comparisons within a study. A pooled effect size estimate was then computed as the weighted average of the individual effect size estimates. The corresponding pooled effect variance was computed as the variance of a linear combination of two or more variables (i.e., average of two or more effect size estimates) using between-comparison correlations equivalent to the weighted average of the between-WG correlation and the between-control correlation (Borenstein et al. 2009b). A single pooled effect size estimate and variance was then included in the analysis for each study with multiple comparisons. Effect size estimates were pooled in a stepwise manner, pooling estimates for comparisons that shared an intervention first followed by those that comprised data from the same subjects. When multiple comparisons included in a subgroup or sensitivity analysis varied from those included in the primary analysis, the above steps were repeated prior to running the subgroup or sensitivity analysis.
Sensitivity analyses were completed for postprandial glycemia and insulinemia measurements to determine if different timeframes for AUC measurement may have impacted the results. This was achieved by analyzing AUC measurements ≤120 and >120 min separately. For studies with multiple time frames, all time frames were included in the sensitivity analyses. Multiple timeframes ≤120 min or >120 min were handled similarly to multiple comparisons using a within-treatment correlation of 0.59. Sensitivity analyses removing studies with high-amylose varieties of grains were also performed. For other outcomes, sensitivity analyses were completed using a one-study-removed analysis.
Subgroup analyses were performed on acute postprandial glycemia and insulin for type of WG, same grain in treatment and control, similar processing for the treatment and control, amount of WG consumed (≤ median or > median), feeding approach (available carbohydrates matched or not matched), health status (healthy, overweight/obese, T2D, metabolic syndrome), and sex (male or female). Subgroup analyses for fasting glucose and insulin included type of WG, amount of WG consumed (≤ median or > median), study design (crossover, parallel), study duration (≤ median or > median) and health status (healthy, overweight/obese, metabolic dysfunction). Subgroup analyses were not possible for HbA1c, HOMA-IR or OGTT due to a small number of studies in the main analysis.
Statistical heterogeneity across studies was assessed using Cochran's Q test and the I 2 statistic. An I 2 value of ≥40% was used to designate moderate or higher heterogeneity, as recommended by the Cochrane Handbook (Deeks, Higgins, and Altmann 2019). Treatment effects in subgroups were compared by using Cochran's Q test to assess heterogeneity between subgroup estimates. Chi-square test statistics and p-values are reported for Cochran's Q tests for heterogeneity across studies and between subgroups. The presence of publication bias was assessed by Egger's test and examining funnel plots measuring the SEM as a function of the SMD.

WG intake and postprandial glycemia
Overall, 127 comparisons reported in 52 studies were included in the analysis of WG on postprandial glycemia. Intake of WG foods resulted in significantly lower postprandial glycemia compared to RG foods (Table 1, SMD: −0.30; 95% CI: −0.43, −0.18; P < 0.001) although the effect size was small in magnitude and the data displayed substantial heterogeneity (Q = 183; P < 0.001; I 2 = 68.8). A sensitivity analysis (Supplemental Table 3) showed a larger effect size when AUC measurements ≤ 120 min and slightly smaller effect size at > 120 min but the effect of WG remained significant in both time frames and heterogeneity remained substantial. Additionally, removal of studies utilizing high amylose varieties did not substantially impact the results.
Subgroup analyses are shown in Table 2. The effect of WG on postprandial glycemia remained significant for all types of WG (P < 0.05) except rice, and there was significant heterogeneity between the subgroups (Q = 14.5, P = 0.043). Quinoa and buckwheat showed large effect sizes, but there were few studies on these grains and one study demonstrated unusually large SMDs . The effect of WG remained significant at intakes above and below the median (79.1 g) with a similar effect size (−0.37 and −0.31, respectively, P ≤ 0.002). When the same grain was included in the test and control conditions (e.g., WG barley vs RG barley) the effect of WG remained significant for rice and wheat (p < 0.05), but not barley or rye. The impact of processing was most substantial for wheat with no significant difference in postprandial glycemia when WG wheat was provided as a flour. The feeding approach was also critical as studies that did not match available carbohydrate (e.g., matched energy or volume) showed a non-significant reduction in postprandial glycemia with WG intake.
Interestingly, the effect of WG on postprandial glycemia was significant in healthy individuals (P < 0.001) but not in overweight/obese individuals or those with T2D. WG intake also significantly reduced postprandial glycemia in males (P = 0.016), but not females (P = 0.223), but very few studies limited inclusion to only men or women.

WG intake and postprandial insulinemia
In 68 comparisons from 30 studies, WG intake resulted in a significantly lower postprandial insulinemic response compared to RG foods, although the effect size was small (Table 3, SMD: −0.23; 95% CI: −0.35, −0.10; P < 0.001) and there was moderate heterogeneity (Q = 54.5; P < 0.006; I 2 = 43.1). A sensitivity analysis showed small changes in effect size when AUC measurements ≤ 120 min and > 120 min, but the effect of WG remained significant in both time frames and heterogeneity was moderate to substantial (Supplemental Table 3). a pooled effect size and variance was used for studies with multiple comparisons when either active or control groups comprised data from the same subjects. abbreviations: Ci = confidence interval, Ha = high amylose, lg = large, on = after overnight fast, rG = refined grain, sm = small, sMd = standardized mean difference, t2d = type 2 diabetes, wG = whole grain Subgroup analyses are shown in Table 4. Only barley and rice significantly reduced the postprandial insulinemic response with a moderate effect size (P < 0.03), although rye and wheat approached significance (P < 0.06 and P < 0.07, respectively). The effect of barley remained significant when WG barley was processed to flour (P = 0.03). In studies with the same grain in the WG and control conditions, only WG rice reduced postprandial insulinemia (P = 0.025), and there was significant between subgroup heterogeneity (Q = 7.4, P = 0.025). This heterogeneity was likely driven by a small, non-significant increase in postprandial insulinemia in the rye subgroup, while wheat and rice directionally decreased postprandial insulinemia. Similar to postprandial glycemia, the effect of WG remained significant at intakes above and below the median (82.4 g, P < 0.01) and was only observed in studies with matched available carbohydrate. Additionally, the effect of WG was significant in healthy individuals (P = 0.002), but not in individuals with overweight/obesity or T2D. Although there were few studies in men or women only, the impact of WG on postprandial insulinemia did not differ by sex. Unlike postprandial glycemia, WG intake significantly reduced postprandial insulinemia at a subsequent meal (Supplemental Table 5, SMD: −0.24; 95% CI: −0.39, −0.08; P = 0.002), with minimal heterogeneity (Q = 5.37; P = 0.615; I 2 = 0.00).

WG intake and fasting glucose and insulin
Twenty-three comparisons from 22 studies showed no significant impact of WG compared to RG on fasting glucose (Figure 2, SMD: −0.11; 95% CI: −0.24, 0.03; P = 0.123). Similar results were observed for fasting insulin (Figure 3, SMD: −0.03; 95% CI: −0.12, 0.07; P = 0.563). A one-studyremoved sensitivity analysis revealed a strong influence of Vitaglione et al. (2015) on fasting glucose results, with the results demonstrating a benefit of WG (P = 0.006) when the comparison from this study is removed (Supplemental Table  6). The one-study-removed sensitivity analysis for fasting insulin did not find any one study to substantially influence the results (Supplemental Table 6).
Subgroup analyses for fasting glucose and insulin are in Table 5. Wheat was the only WG tested in isolation and did not significantly impact fasting glucose or insulin (P = 0.121 and 0.646, respectively). However, a significant reduction in fasting glucose was observed in studies with mixed WGs (P = 0.013) and there was significant heterogeneity between the wheat and mixed WG subgroups (Q = 4.7, P = 0.031). The amount of WG consumed and the duration of the testing did not impact the effect of WG on fasting glucose or insulin. However, there was an impact of study design on fasting glucose, with a significant effect of WG in crossover studies (P = 0.010), but not parallel studies. This effect of study design was not observed for fasting insulin. Similar to findings with postprandial glycemia and insulinemia, a significant benefit of WG for fasting glucose and insulin was only observed in healthy individuals (P = 0.027), but not individuals with overweight/obesity or metabolic dysfunction.

WG intake and markers of glycemic control, insulin resistance and glucose tolerance
Few studies compared WG intake to RG for other glycemic and insulinemic markers, such as HbA1c, HOMA-IR or glucose tolerance tests. Six studies reported HbA1c, ten studies reported HOMA-IR and three studies conducted a pooled effect size and variance was used for studies with multiple comparisons when either active or control groups comprised data from the same subjects. significance of bold values is p ≤ 0.05. abbreviations: CHo = carbohydrate, sMd = standardized mean difference, rG = refined grain, wG = whole grain a pooled effect size and variance was used for studies with multiple comparisons when either active or control groups comprised data from the same subjects. abbreviations: Ci = confidence interval, Ha = high amylose, lg = large, on = after overnight fast, rG = refined grain, sm = small, sMd = standardized mean difference, t2d = type 2 diabetes, wG = whole grain. a pooled effect size and variance was used for studies with multiple comparisons when either active or control groups comprised data from the same subjects. abbreviations: CHo = carbohydrate, sMd = standardized mean difference, rG = refined grain, wG = whole grain. OGTTs. There was a small, but significant improvement in HbA1c (Figure 4; SMD: −0.21; 95% CI: −0.37, −0.06; P = 0.007) and all but one of the studies (J. Li et al. 2003) were in individuals with T2D or with risk factors for T2D. WG intake, compared to RG, did not significantly impact HOMA-IR or OGTT outcomes (Supplemental Figures 1-2). There was no significant heterogeneity in any outcomes. A one-study-removed sensitivity analysis showed no influence of individual studies on HOMA-IR, but a small influence of the study by Roager et al. (2019) on HbA1c (Supplemental Table 5).

Quality of evidence
The quality of evidence as assessed by GRADE criteria is summarized in Table 6. Evidence for outcomes were rated as moderate to low. The evidence rating for acute and second meal postprandial glucose was downgraded due to concerns about serious inconsistency, moderate imprecision and possible publication bias. Publication bias was also evident for postprandial insulin (P = 0.023). Long term markers of glycemic control and insulin resistance did not have significant publication bias but did show moderate to serious imprecision. Risk of bias assessment on outcomes of individual studies and Egger's statistics for publication bias are included in Supplemental Table 7 and 8.

Discussion
The results of this meta-analysis of RCTs suggest that WG intake reduces acute postprandial glycemic and insulinemic response and improves HbA1c compared to RG, but does not impact fasting glucose, fasting insulin, HOMA-IR or glucose tolerance. Several reviews and observational studies have shown a consistent inverse association of WG intake with risk and incidence of T2D (Al Essa et al. 2015;Hu et al. 2020;Parker et al. 2013;Tieri et al. 2020) and results from this meta-analysis suggests reducing postprandial glycemia and insulinemia may be one mechanism by which WG could protect against T2D. This agrees with results from other meta-analyses that have found a significant effect of WG on postprandial glycemia measures (Marventano et al. 2017;Musa-Veloso et al. 2020;Musa-Veloso et al. 2018;Tosh 2013). WGs typically contain more fiber than RGs so the replacement of digestible (available) carbohydrate with non-digestible fiber may help reduce the glycemic response. However, since most studies match available carbohydrate between the WG and RG treatments, there appear to be benefits beyond substitution of digestible carbohydrate for fiber. The additional benefit may be due to the ability of fiber and phenolics to slow the rate of digestion of glycemic carbohydrates or the potential impact of fiber fermentation metabolites on incretin hormones (Jovanovski et al. 2019;M. Li et al. 2017;Müller et al. 2019). Grains rich in beta-glucan, such as oat and barley, may increase viscosity of the intestinal contents during digestion which can limit enzyme accessibility to carbohydrates as well as limit the rate of absorption of free glucose through the intestinal wall.
The results also demonstrate the importance of the type of WG and degree of processing on postprandial glycemic response. In the current study, all WGs, regardless of type, reduced postprandial glycemia (although this did not reach significance for rice). However, for wheat, the impact on postprandial glycemia was lost when the WG was processed to flour. Milling to flour disrupts the structure of the whole grain wheat, making the starchy endosperm more accessible and less protected by the bran (fiber) layer. Depending on the degree of milling, particle size of the grain may also be smaller providing more surface area for digestive enzymes. Interestingly, processing to flour did not impact the benefit of WG barley and corn. However, studies with WG barley flour utilized RG wheat or rice flour as the control Liljeberg and Bjorck 1994; and two of the studies with corn utilized high amylose varieties Luhovyy et al. 2014). Therefore, these results should be interpreted with caution as it is difficult to differentiate the effects of processing from the effects of different comparator grains or varieties. Unfortunately, there were insufficient studies to evaluate the same grain with the same degree of processing in the heterogeneity (Q = 4.8, P = 0.090) 1 a pooled effect size and variance was used for studies with multiple comparisons when either active or control groups comprised data from the same subjects. 2 subjects had one or more risk factors for metabolic syndrome, hypertension or slightly elevated total cholesterol. abbreviations: CHo = carbohydrate, rG = refined grain, wG = whole grain, wks = weeks. same food matrix. In the subgroup where the same grain was used in the treatment and control, there was not a significant difference between the WG and RG versions of barley and rye. This agrees with other meta-analyses (Musa-Veloso et al. 2018) and may be partially due to the fiber content of the endosperm and the structure of the starch granules in these two grains. Barley studies compared WG barley kernels to RG pearled barley kernels (Aldughpassi, Abdel-Aal el, and Wolever 2012). Pearling is a process which removes the outer bran portion of a WG but does not disrupt the native structure of the endosperm, possibly contributing to slower digestion and yielding little difference in postprandial glycemia between WG and RG. The endosperm of rye also has a higher fiber content than some other grains, such as wheat. For example, one study reported 6.7 g dietary fiber/100 g endosperm rye bread compared to 9.6 g dietary fiber/100 g in WG rye bread and only 1.8 g dietary fiber/100 g in RG wheat bread ). The microstructure of WG and RG rye bread is also less porous than other breads, limiting the accessibility of hydrolytic enzymes. Cooking rye leads to leakage of amylose from the starch granule to form a coating which may further slow enzymatic hydrolysis . Thus, RG rye and barley may not differ from their WG counterparts in glycemic response due to these unique structural components.
While the benefit of WG on postprandial glycemia was observed only immediately after consumption, the benefit of WG on postprandial insulinemia was present acutely as well as at a subsequent meal. This suggests the effects of WG consumption on postprandial insulin may be longer term and go beyond the digestibility of carbohydrate. Fibers and resistant starches within WG can be fermented in the large intestine, leading to the production of SCFA. These SCFA have been shown in pre-clinical models to increase glucose oxidation in the liver, stimulate GLP-1, and suppress the production of free fatty acids in the late postprandial period and during an overnight fast, all of which may contribute to improvements in insulin sensitivity (Canfora, Jocken, and Blaak 2015;DeFronzo 2009;Müller et al. 2019). Furthermore, studies in humans have shown a positive association of circulating SCFA with GLP-1 concentrations and an inverse association with free fatty acid concentrations (Müller et al. 2019). Thus, the fermentation of fiber and resistant starch in WG may explain part of the improvement in insulin response at subsequent meals.
Beyond acute and subsequent meal effects, the benefit of WG intake on longer term glycemic and insulinemic outcomes was less clear. There was an improvement in HbA1c with WG intake, which suggests the acute benefits of WG on postprandial glycemia may provide a cumulative benefit for glycemic control over time. However, more studies are needed to confirm this finding since the sensitivity analysis revealed this effect may be driven by one influential study . WG intake did not show a benefit for fasting glucose, fasting insulin, HOMA-IR, and glucose tolerance. While these findings are similar to other meta-analyses (Marshall et al. 2020;Marventano et al. 2017), it is unclear why short-term improvements in postprandial glucose and insulin did not translate to clear long-term benefits in studies with chronic consumption. One consideration for fasting glucose, is the presence of one influential study with surprising results. Vitaglione et al. (2015) found a 6 mg/dL increase in fasting glucose with WG treatment compared to RG treatment. Although it was reported as not significant in the publication and was confirmed through email communications with the author, it appears statistically significant in this meta-analysis and influences the results. The removal of this study from the analysis resulted in a significant improvement in fasting glucose with WG intake. Similarly, a meta-analysis by Marventano et al. (2017) also found an increase in the effect size of WG on fasting glucose when excluding the Vitaglione et al. study. The high degree of influence of this one study suggests it may be overshadowing a small but significant effect of WG on fasting glucose. Study durations may have also been insufficient to produce measurable long-term changes. However, with a median duration of 8 weeks, most of these studies were of sufficient length to predict changes in fasting glucose, fasting insulin, and insulin sensitivity measures. The study design and type of WG may also explain some of the lack of effect of WG intake on long-term measures of glycemia and insulinemia. Most studies used a mix of WGs which resulted in a significant reduction in fasting glucose, but studies using only wheat did not. Although postprandial analyses show most WGs significantly reduce postprandial glycemia, the effect size is larger for some (e.g., oat, barley) which may offer a greater cumulative benefit over time. Furthermore, the WG foods in three of the four chronic wheat-based studies used WG wheat flour, which did not demonstrate a difference from RG wheat in postprandial glycemia. Additionally, crossover studies resulted in a small, but significant impact on fasting glucose not observed in parallel trials. The lack of effect in the analysis of parallel studies may be partially due to the unusual results in the Vitaglione et al. (2015) study and possibly the substantial heterogeneity in parallel studies (I 2 = 62.5). In contrast, heterogeneity was low in crossover studies (I 2 = 0.0) with seven of the nine studies reporting a decrease in fasting glucose with WG intake, although only one study reached statistical significance. Taken together, these findings suggest additional studies on the glycemic impact of long-term WG consumption are necessary before stronger conclusions can be drawn.
Differences in acute and chronic study design may also explain divergent findings between long-and short-term outcomes. Most postprandial studies matched available carbohydrate between the WG treatment and RG control, while chronic studies primarily asked subjects to substitute RG foods for WG foods and did not control the remainder of the diet. Thus, the amount of digestible carbohydrate was not controlled in chronic studies as it was in postprandial studies and it is unknown if postprandial glycemic responses were reduced over the study duration. This may be an important consideration as postprandial studies not matching available carbohydrates did not show a significant impact of WG on postprandial glycemia or insulinemia. Nevertheless, the substitution approach is more practical and is aligned to recommendations (e.g., choose whole wheat bread instead of white bread) (Health Canada 2013; U.S. Department of Agriculture and U.S. Department of Health and Human Services 2020). While matching available carbohydrate may reveal a mechanistic action of WG, it is not the way most people eat. Thus, it would be beneficial to design studies that feed WG over several weeks, but also monitor acute glycemic and insulinemic response to determine if acute improvements in postprandial glycemia and insulinemia contribute to long term benefits of WG on glycemic control. Continuous glucose monitoring may be one way to achieve this. One chronic feeding study implemented 24 hour continuous glucose monitoring at week 2 and week 8 of the WG intervention and found no difference in daily glucose AUC compared to the RG group, but they did not specifically assess postprandial responses .
Differences in health status also impacted the current findings, with studies in healthy individuals showing a benefit of WG intake on postprandial glycemia, insulinemia, and fasting glucose, while studies in overweight/obese individuals or those with metabolic dysfunctions, such as T2D, showed non-significant effect sizes. However, there was a significant reduction in HbA1c with WG intake and all but one of the studies were in individuals with T2D or risk factors for T2D. This supports a potential benefit of long-term WG consumption for glycemic control, particularly for individuals with T2D or at risk of T2D. More acute studies of glycemia and insulinemia are needed in these populations, as there were very few included in this meta-analysis and there is greater baseline variability in glycemia and insulinemia in individuals with T2D or risk factors, such as impaired fasting glucose. Several studies in the current meta-analysis also defined healthy as only an absence of disease with screening criteria that included overweight and normal weight individuals. Future studies should consider evaluating normal weight, overweight, and obese subjects separately as well as evaluating baseline insulin resistance to better understand the potential benefits of WG in healthy and metabolically at-risk populations.
While the body of evidence, particularly for postprandial measures, is large, the quality of the evidence was rated as low to moderate due to issues with publication bias, inconsistency, and imprecision. Publication bias may have been an issue with studies of postprandial glycemia and insulinemia, possibly because these studies typically have a smaller average sample size. Therefore, studies that do not confirm the hypothesis and are small are less likely to be published. Publication bias was not an issue for longer term markers of glycemia and insulinemia, possibly because they utilize large sample sizes and are more likely to be published regardless of outcome. Inconsistency was considered serious for postprandial glycemia, and it is likely several factors contribute to significant heterogeneity. Even after subgrouping by type of WG, degree of processing, health status, sex and other factors, heterogeneity remained substantial.
The strengths of the current systematic review and meta-analysis include a broad scope to cover multiple types of WG, health conditions and outcomes related to short-term and long-term glycemic control. This allowed for several subgroup and sensitivity analyses that identified factors that may impact the effect of WG on glycemic and insulinemic outcomes. However, the systematic review and meta-analysis was also limited by poor reporting of the amount of WG in the studies, resulting in the exclusion of several studies. There was also an indication of publication bias for postprandial outcomes and significant heterogeneity for postprandial glucose that could not be fully explained. There were also an insufficient number of studies to evaluate detailed subgroups, such as similar grain in treatment and control with a similar degree of processing and a similar food matrix.
In summary, the results of this meta-analysis show that intake of WG foods moderately, but significantly reduce postprandial glycemia and insulinemia compared to RG foods. However, the type and variety of WG, degree of processing, and food matrix may have substantial influence that deserve further investigation. WG intake also resulted in a small but significant reduction in HbA1c, but other long-term glycemic and insulinemic markers were not impacted. The relationship of WG intake to reduced risk of T2D may be partially attributable to short term benefits in glycemic control, but more long-term intake studies are needed with individual WGs and with improved measurement of acute postprandial effects to determine whether short-term benefits for postprandial glycemia will translate to longer-term glycemic benefits such as lower incidence of prediabetes and T2D.