Effects of whole grain intake on glycemic traits: A systematic review and meta-analysis of randomized controlled trials

Abstract Whole grains (WGs) may have various health benefits, including lowering blood glucose and improving insulin sensitivity. To conduct a meta-analysis of the effects of WGs compared with non-WGs on changes in fasting glucose, fasting insulin, glycated hemoglobin (HbA1c), and homeostasis model assessment of insulin resistance (HOMA-IR). A systematic literature search was performed for all published randomized controlled trials on the effects of WG intake on fasting glucose, fasting insulin, HbA1c and HOMA-IR response up to February 2021. Weighted mean differences (WMD) were calculated. Pre-specified subgroup and univariate meta-regression analyses were explored to identify the sources of heterogeneity. Sensitivity analysis and bias analysis were conducted to appraise study quality. Among 12,435 articles screened for eligibility, data were extracted from 48 articles. Meta-analysis of 4,118 participants showed that WG consumption resulted in a significant reduction in fasting glucose by −0.15 mmol/L, fasting insulin by −2.71 pmol/L, HbA1c by −0.44%, and HOMA-IR by −0.28, respectively. Compared with mixed grains, brown rice, and wheat, oats were significantly lower on marker of glycemic. Besides, multiple interventions per day consolidated effectiveness of WGs. WG consumption decreased the levels of fasting glucose, fasting insulin, HbA1c, and HOMA-IR compared with non-WG consumption. Supplemental data for this article is available online at https://doi.org/10.1080/10408398.2021.2001429 .


Introduction
Whole grains (WGs) are defined as both intact and pulverized forms containing the expected proportion of bran, germ, and endosperm for the specific grain type (De Munter Jeroen et al. 2007). Plant cell walls are composed of complex carbohydrates such as cellulose, hemicellulose, and pectin; therefore, WGs are a particular source of dietary fiber (DF). In addition, WGs are rich in myriad vitamins, minerals, and phytochemicals (Liu 2003).
Previous researchers favored studies on DF rather than WGs. Indeed, cereal fibers have shown beneficial effects on human health. In WGs, DF coexists with numerous micronutrients and bioactive compounds (Liu, Wang, Hao, et al. 2020), which are likely to have significant health benefits. Therefore, the focus of research has shifted toward observational studies of WG consumption. Increasing evidence on WGs in relation to health outcomes has accumulated over the past 10 or 15 years (Kyrø and Tjønneland 2016). Evidence indicates that the consumption of WGs reduces the risk of several major chronic diseases, such as obesity (Tighe et al. 2010), type 2 diabetes mellitus (T2DM) (Aune et al. 2013), breast cancer (Xie et al. 2019), and colon cancer (Egeberg et al. 2010). WGs were reported to lower the risk of diabetes, possibly because they contained higher amounts of fiber, trace elements, and bioactive compounds, which showed the potential of WGs to improve glucose metabolism and insulin sensitivity (Malin et al. 2019;Dayib, Larson, and Slavin 2020).
Although WGs are considered to have a variety of health benefits, their effects on human glucose metabolism remain controversial. Some researchers pointed out that no significant differences were found between the effect of WG and non-WG diets on fasting glucose (Kazemzadeh et al. 2014;MacKay et al. 2012), insulin (Giacco et al. 2013;Andersson et al. 2007) and glycated hemoglobin (HbA1c) (Kondo et al. 2017). It has also been reported that WG intake does not affect cardiovascular disease risk, inflammatory markers, blood biochemistry, body composition, or intestinal microbiology (Ampatzoglou et al. 2016;Ampatzoglou et al. 2015;Brownlee et al. 2010). However, some studies have found that WG diet has a positive regulatory effect on healthy people, patients with diabetes and obesity, resulting in improvements in measures of glycemia control, blood lipids, body weight, and inflammation, as well as a reduction in premature mortality (Reynolds, Akerman, and Mann 2020;Maki et al. 2019). To date, three meta-analyses have examined the relationship between WGs and blood glucose (Reynolds, Akerman, and Mann 2020;Musa-Veloso et al. 2018;Marventano et al. 2017). Of these, one study investigated the changes in blood glucose levels in healthy people who ate WGs, and did not involve subjects with diabetes or impaired glucose tolerance (Reynolds, Akerman, and Mann 2020). The other meta-analysis focused on the changes in the postprandial blood glucose in T2DM who ate WGs (Musa-Veloso et al. 2018). The last meta-analysis examined WG consumption on postprandial blood glucose (PBG) with 20 studies, while the glycemic traits, including fasting glucose, HbA1c and HOMA-IR, were not evaluated (Marventano et al. 2017). At present, no study has systematically analyzed the effects of WGs on subjects with different health status. Accordingly, it was necessary to explore the significance of WGs on glycemic traits through a comprehensive meta-analysis with randomized controlled trials (RCTs).
The aim of the present study was to assess evidence from RCTs of the effects of WGs on fasting glucose, fasting insulin, HbA1c and HOMA-IR in healthy people, diabetics, overweight or obese, and other people consuming WGs compared with non-WG consuming controls. Importantly, we analyzed the effects of various factors on fasting glucose, fasting insulin, HbA1c and HOMA-IR in various populations after long-term intake of WGs, such as the type and frequency.

Identification of potentially relevant studies
Three independent researchers (SL, RA, and HW) conducted electronic search in PubMed, Embase, and the Cochrane library from studies published between January 1970 and February 2021. After deleting duplicate results, two investigators (JL and XG) first scanned the studies' titles and abstracts independently. The full text was obtained independently by the other two investigators (SL and RA). Differences were decided by consensus or by a third investigator (TX). The search strategy included following key words: ("whole grain" OR "wholegrain" OR "wholemeal" OR "wholewheat" OR "brown rice" OR "wheat" OR "rice" OR "barley" OR "maize" OR "corn" OR "rye" OR "oat" OR "millet" OR "sorghum" OR "tef " OR "triticale" OR "canary seed" OR "Job's tears" OR "fonio" OR "wild rice" OR "Amaranth" OR "buckwheat" OR "Quinoa" OR "spelt" OR "emmer" OR "faro" OR "einkorn" OR "kamut" OR "durum" OR "bread" OR "cereal" OR "flour") and ("glucose" OR "plasma glucose" OR "diabetes" OR "diabetic" OR "fasting glucose" OR "blood glucose" OR "blood sugar" OR "insulin" OR "iletin" OR "Ins" OR "glycated hemoglobin" OR "HbA1c" OR "glycosylated hemoglobin" OR "GHb" OR "HbAlc"). The search was limited to human studies with the full text in English. Filters were used in the retrieval process. Duplicates were removed. The reference lists of the relevant studies were inspected to identify any additional published studies not identified by the literature searches.

Inclusion and exclusion criteria
The inclusion criteria were 1) The study design was RCT, including parallel rows and crossover interventions. 2) The WG was defined as a complete, crushed, or flaked caryopsis, and its basic components included the amyloid endosperm, germ, and bran. The main types of WGs were included in the search term. The WG forms in the experimental groups included WG products, WG diets, and whole wheat products. 3) Participants aged between 18 to 75 years old and had unlimited medical conditions, including health, overweight, obesity or diabetes. 4) The duration of the intervention was five days or more; however, the blind method was not restricted because WGs taste significantly different. 5) In studies with ≥ three intervention arms, of which two or three were eligible, only the eligible arms were included. 6) Controlled trials of WGs versus non-WGs or refined grains were included regardless of energy control. 7) Data were presented as means with standard deviations (SDs) or standard errors (SEs) at baseline or endpoint or both for the outcomes investigated. 8) Outcomes included fasting glucose, fasting insulin, HbA1c or HOMA-IR.
The exclusion criteria were 1) Studies were not RCTs. 2) There were other interfering factors and multi-component interventions that did not clarify the effects of WGs. 3) Studies on foods that were based on individual grain components, such as bran or germ, were excluded.

Statistical analysis
To perform the comparison, the glucose level was converted from mg/dL to mmol/L, according to the conversion formula (1 mmol/L = 18 mg/dL) and the insulin level was converted from mU/L or μU/mL to pmol/L, according to the conversion formulae (1 pmol/L = 6.965 mU/L; 1 pmol/L = 6.965 μU/mL) (Liu, Wang, Hao, et al. 2020). Changes from baseline to endpoint were used for the analysis of fasting glucose, fasting insulin, HbA1c and HOMA-IR. When the SD was not reported, it was derived from the available data (95% confidence interval (CI), p-values, or SE) using the method suggested by the Handbook for Systematic Review of Interventions (Crowther, Lim, and Crowther 2010). When needed, the SD for changes from baseline was imputed using a pooled correlation coefficient according to a published procedure, where the constant R = 0.5. If trials compared multiple intervention groups with the same control group, the shared control group was considered as two or more groups (Hollaender, Ross, and Kristensen 2015). Effect size consisted of weight mean difference (WMD) and 95% CI between the outcomes of intervention and the control groups using the generic inverse-variance random effects model. A two-sided p-value of 0.05 was considered statistically significant. Heterogeneity between trial results was tested using the I 2 statistic (Higgins et al. 2003) and I 2 ≥50% indicated a significant level of heterogeneity. The stability of the results and the possible sources of heterogeneity were explored using sensitivity analyses. Moreover, subgroup analysis was used to evaluate the influence of certain factors, including study design, region, duration of intervention, weight status, the type of WGs, and the WG form. Quality assessment was conducted by two researchers (AZ, LL) using the guidelines of Cochrane's handbook for systematic review of interventions to evaluate the included trials. Articles were judged as high, low, or unclear in the following domains: Randomization sequence generation, allocation concealment, blinding of participants and study personnel, blinding of outcome assessors, incomplete outcome data, selective reporting, and other biases. The quality of the methodology for each study was regarded as high (low risk of bias in one or more items), unclear (low or unclear risk of bias in all items), or low (high risk of bias in all items). In addition, the new Jadad scale was also used to assess the quality of each study, to answer the following four questions: whether random sequence generation was used, whether randomized concealment was used, whether blindness study was performed, and whether the number and reasons for participant loss or quit were described. A total of seven points were generated from the four abovementioned questions, with 1 to 3 points classified as low-quality literature and 4 to 7 points classified as high-quality literature. Publication bias was examined by visual inspection of funnel plots using Begg's test and Egger's test, and significant publication bias was defined as a P value < 0.05, no bias was defined as a P value > 0.1. Trim was used to explain the publication bias of the results. In addition, we formed a panel to determine the bias of each study. Sensitivity analysis was used to assess the source of heterogeneity. Sensitivity analysis was used to recalculate its effect by deleting each study.
Stata version 12 (Statacorp LP, College Station, TX, USA) was used to combine the individual study results. Moreover, regression analysis was used to determine whether the dose and duration of intervention were significantly linearly associated with fasting glucose, fasting insulin, HbA1c, and HOMA-IR.

Patient and public involvement
We did not involve patients or the public in this research at any stage. No patients were involved in setting the research question or the outcome measures, nor were they involved in developing plans for the design or implementation of the study. No patients were asked to advise on the interpretation or writing up of results; however, we do have plans to disseminate the results of the research to the relevant patient community.

Characteristics of the included studies
The basic characteristics of the subjects are listed in Table 1. The duration of the studies ranged from 5 days to 1 year; most of the studies lasted 6-12 weeks. Eight studies contained multiple intervention and control measures. In total, the results including 53 fasting glucose, 37 fasting insulin, 18 HbA1c, and 29 HOMA-IR were reported from 56 individual treatment arms. The publication dates for the included studies ranged from 2001 to 2020. The intervention groups in all the trials consumed WGs, including oats, brown rice, mixed, and others. Ten trials offered a full diet for intervention, while the others offered only WGs. Trials were conducted across three regions, including Europe (Tighe et al. 2010;Giacco et al. 2013;Andersson et al. 2007;Ampatzoglou et al. 2016;Ampatzoglou et al. 2015;Brownlee et al. 2010;Schutte et al. 2018;Dinu et al. 2017;Liangkui et al. 2018;Vetrani et al. 2016;Connolly et al. 2016;Saltzman et al. 2001;Kristensen et al. 2012;Roager et Table 2 showed a summary of the risk of bias, as determined using the Cochrane risk of bias tool and the new Jadad scale. In the RCT, information about random sequence generation and concealment was unclear for most trials. Most of the studies were assessed as having an unclear risk of bias across the three domains, including random sequence, allocation concealment, and blinding.

Study quality
No observation bias and selective reporting bias were identified among the included studies. No other sources of bias were identified among most of the studies.
According to the new Jadad scale, high-quality research accounts for more than one-third.

Fasting glucose
Fifty-three studies, including 3,990 subjects, considered fasting glucose as an outcome. The meta-analysis showed that the fasting glucose levels in the experimental group (WMD: −0.15 mmol/L, 95% CI: −0.24 to −0.71 mmol/L; P < 0.001) were significantly decreased compared with control group (Figure 2). The overall test for heterogeneity resulted in a significant heterogeneity with I 2 = 79.0% (P < 0.001). Subgroup analysis was used to determine the effect of WG consumption on fasting glucose (Table 3). The effects of various factors that might affect the experimental results for fasting glucose concentration were analyzed, including     Table 3. (Continued) more effective at reducing fasting glucose (WMD = −0.42 mmol/L; 95% CI: −0.67 to −0.19 mmol/L; P < 0.001), compared with the other regions. The effect of the intake of oats (WMD = −0.81 mmol/L; 95% CI: −1.22 to −0.40 mmol/L; P < 0.001) was better than that for the other WGs. No drugs were used during the intervention, and the results were significantly reduced (WMD = −0.071 mol/L; 95% CI: −0.13 to −0.01 mmol/L; P = 0.016). When the dose of intervention whole grains is lower than 100 g, it reduced fasting glucose level more significantly (WMD = −0.33 mmol/L; 95% CI: −0.50 to −0.16 mmol/L; P < 0.001). By contrast, there was no difference in fasting glucose results based on intake of cereal products. Based on the grouping of research quality, the intervention results of high-quality research (WMD = −0.23 mmol/L; 95% CI: −0.36 to −0.10 mmol/L; P < 0.001) were significantly better than low-quality research. In addition, there was significant difference in the overall fasting glucose results among the other subgroups.
After excluding the studies one by one and repeating the meta-analysis, the final results were not reversed in any direction in any of the included studies, indicating that the data from the included studies was stable and consistent.
Bias analysis of included data was conducted by Begg's test and Egger's test, and the results showed that fasting glucose (PBegg = 0.006, PEgger = 0.019) had significant bias ( Supplementary Figures 1 and 2). After the trim analysis, the WMD changed slightly (-0.15 mmol/L to −0.49 mmol/L) (Supplementary Figure 3).
Meta-regression analysis was used to quantify whether there was a linear relationship between the continuity indicators (including intake dose, intervention time) and the effect of WG consumption. The duration of the included studies ranged from 5 days to 1 year, significant association was found when a meta-regression analysis was performed to investigate whether duration could influence changes in fasting glucose (P = 0.043) (Supplementary Figure 4). No significant association was observed between intervention dose and fasting glucose (Supplementary Figure 5).
Similarly, after excluding the studies one by one and repeating the meta-analysis, the final results were not reversed in any direction in any of the included studies, indicating that the data from the included studies was stable and consistent.
Begg's test and Egger's test were used for bias analysis of the included data, which showed that fasting insulin (PBegg= 0.123, PEgger= 0.113) had no bias ( Supplementary  Figures 6 and 7). There was no difference in the effect size after the analysis of trim, which indicated that the results were robust (Supplementary Figure 8).
Regression analysis of the two continuity indicators of fasting insulin did not find significant linear relationship ( Supplementary Figures 9 and 10).
After excluding the studies one by one and repeating the meta-analysis, all the estimated values of the included studies fell within the total effect value, indicating that the data from the included studies was stable and consistent.
Similarly, regression analysis of the two continuity indicators of HbA1c did not find significant linear relationship ( Supplementary Figures 14 and 15).
After excluding the studies one by one and repeating the meta-analysis, all the estimated values of the included studies fell within the total effect value, indicating that the data from the included studies was stable and consistent.
Begg's test and Egger's test were used for bias analysis of the included data, which showed that HOMA-IR (PBegg = 0.680, PEgger = 0.137) had no bias ( Supplementary  Figures 16 and 17). After the trim analysis, the WMD changed from −0.277 to −0.070 (Supplementary Figure 18).
Similarly, a regression analysis of the two continuous measures of HOMA-IR showed a significant linear relationship between insulin resistance levels and whole grain doses (P = 0.013) (Supplementary Figures 19 and 20).

Discussion
To our knowledge, this is the first meta-analysis to examine the effects of WGs on glycemic traits. WGs had shown health benefits, as they were rich in nutrients and antioxidants, and were strongly associated with improved outcomes for chronic diseases, including T2DM (Marshall et al. 2020). The present study suggested that intakes of WGs could significantly lower fasting glucose, fasting insulin, HbA1c and HOMA-IR levels. The findings from this analysis have also demonstrated the potential benefits of WG oats, multiple intakes of WGs in various populations. Besides, the benefits of WGs intake were shown in people with diabetes, health and obese subjects. Intakes of WGs diets were better than WG products and whole wheat products. Moreover, WGs intake showed a better effect on lowering fasting glucose, fasting insulin, HbA1c and HOMA-IR levels in the non-medicated participants.
Using meta-regression analysis, significant linear relationship between intervention time with fasting glucose and intervention dose with HOMA-IR were found, indicating that duration of intervention and dose could be the sources of heterogeneity of fasting glucose and HOMA-IR, respectively. However, there was no significant linear relationship between HbA1c both with intervention time and dose. HbA1c is mainly to assess the blood glucose in the body in the past three months, and there are various confounding factors when measuring HbAlc (Nitin 2020). This is probably why no linear relationship between intervention time and HbA1c was found in the regression analysis.
The subgroup analysis found that low intake of WGs (0-100 g) exerted more obvious health effect with low degree of heterogeneity. Hence, 100 g/d intervention of WGs was recommended. Both the short term and long-term intervention has attenuated better effect on the glycemic index. Combined with the heterogeneity analysis, more than 8 weeks intervention was recommended.
Oats were more effective than other types of WGs, especially in diabetes. In contrast, the brown rice, mixed WGs, and other WGs had only a small effect on fasting glucose. Studies have shown that oats are rich in beta-glucan, which could increase satiety, reduce appetite, slow down the circulation and absorption of sugar, and decrease the fasting glucose level (Davidson et al. 1990). A low-glycemic index diet containing mainly intact oats significantly reduced C-peptide concentrations (by 32%). In addition, oats could acutely lower fasting glucose levels, which might improve insulin sensitivity and beta-cell function (Zafar et al. 2019).
Compared with a single meal, multi-intake of WGs had a greater effect on the participants' fasting glucose, which might reflect delayed absorption of carbohydrates and thus increased insulin sensitivity. Fiber and WG consumption are likely to correlate highly. In addition, previous studies showed that dietary fiber-rich foods have longer gastric retention time compared to foods with high energy density, which can stimulate the body's metabolism (Dayib, Larson, and Slavin 2020).
Surprisingly, WG intake showed a better effect on lowering fasting glucose level in the non-medicated participants than in the other participants in various populations. Considering that people with impaired glucose tolerance experience a small change in their fasting glucose even with drug intervention (Rosenstock et al. 2008;Hemmingsen et al. 2016;Liu et al. 2017;Holman, Cull, and Turner 1999), these changes caused by WGs were relatively meaningful. WGs regulated gluconeogenic enzymes in the liver (phosphoenolpyruvate carboxykinase, glucose kinase, and glucose 6-phosphatase), improved abnormal glucose metabolism, inhibited glycogen synthesis by regulating the protein kinase B (AKT)/glycogen synthase kinase 3 and c-Jun N-terminal kinase path ways, lowered fasting glucose, inhibited inflammation, and improved insulin resistance and glucose metabolism disorders.
In addition, a comprehensive WG diet had a more pronounced effect on participants' fasting glucose levels, demonstrating the important role of WGs in a nutritious diet. The fasting glucose levels of participants in the WG groups decreased significantly, regardless of whether their energy intake was controlled (Liu et al. 2002).

Strengths and limitations
Our systematic review and meta-analysis has several strengths. Firstly, we did a comprehensive and reproducible search and selection process of the literature examining the effect of food sources of WGs on glycaemic control. Secondly, collation and synthesis of all the available evidence from 48 studies and 4,118 subjects of controlled intervention studies was performed. Thirdly, we included an assessment of overall quality of study using Cochrane Collaboration risk of bias tool and the new Jadad scale.
Several of our analyses presented limitations. Firstly, despite the inclusion of many studies, but the results showed that the number of each indicator was inconsistent. Secondly, it was found that the heterogeneity (I 2 ) between studies for fasting glucose, HbA1c and HOMA-IR was more than 50%. Therefore, regression analysis and subgroup analysis were conducted to disentangle potential determinants of heterogeneity and the removal of individual studies during sensitivity analyses explained this heterogeneity. Third, unmeasured or residual confounding cannot be completely ruled out.

Conclusion
In summary, the present meta-analysis demonstrated that intakes of WGs significant reduced the fasting glucose, fasting insulin, HbA1c, and HOMA-IR levels, especially oats. Besides, multiple meals were more effective than single meals. However, this conclusion requires further research for confirmation. Therefore, further study is needed to better understand the medium and long-term effects and biological mechanisms of the health benefits of WG consumption.