Diet in the management of type 2 diabetes: umbrella review of systematic reviews with meta-analyses of randomised controlled trials

Literature search

A systematic literature search was conducted in PubMed, Embase, Cochrane, and Epistemonikos, with predefined search terms, from inception to 24 April 2021, with an update on 23 May 2022. None of the studies included in the in 2021 screening were exclusively from Embase, and therefore this database was omitted from the update in 2022. No filters were applied during the literature search process. Online supplemental table S1 shows the detailed search strategies. We also reviewed studies received from PubMed Alert at weekly intervals up to 5 June 2022 and screened the reference lists of all included publications. The literature search originally identified meta-analyses of prospective cohort studies and randomised controlled trials, but for this study, we focused on meta-analyses of randomised controlled trials. The literature screening was conducted by two authors independently. Disagreements were resolved by consensus with a third author.

Eligibility criteria

The systematic reviews with (network) meta-analyses of randomised controlled trials included in our study met the following criteria: the effects of dietary factors (dietary patterns, food groups, and dietary supplements) were investigated; the intervention in randomised controlled trials lasted ≥12 weeks for any health outcome (eg, mortality, incidence of cardiovascular disease, body weight, levels of HbA_1c, and heath related quality of life); only individuals with type 2 diabetes were included; effect estimates were reported (eg, for continuous outcomes, mean difference or standardised mean difference with corresponding 95% confidence intervals, or standard error; or for dichotomous outcomes, risk ratio, including hazard ratio or odd ratio, with 95% confidence intervals).

We excluded studies that were not systematic (including narrative reviews), not meta-analyses, not based exclusively on populations with type 2 diabetes, and single study findings. We also excluded publications reporting on alternative dietary treatments or medical nutrition (eg, Chinese medicine, plant extracts, herbs, and Ramadan fasting), and meta-analyses of randomised controlled trials with exclusively active control groups (comparing two intervention arms with each other). We did not exclude studies based on the publication language.

If we identified more than one meta-analysis on the same effect, we chose one meta-analysis for each intervention to avoid inclusion of duplicate results. In this case, we included the meta-analysis with the largest number of randomised controlled trials and participants. If more than one meta-analysis had the same number of randomised controlled trials, we selected the meta-analysis with the largest number of participants. Because we only included meta-analyses that were based on a systematic literature search, we assumed that newer and larger meta-analyses included the same randomised controlled trials as older and smaller meta-analyses looking at the same research question. We always prioritised evidence from direct comparisons over network meta-analyses.

Data extraction

Data were extracted by one author and double checked by a second author. Discrepancies were discussed between the authors, involving a third author if necessary. Data extracted from each publication were name of the first author, publication year, intervention type, comparison, outcome, duration of the study, number and study design of randomised controlled trials, and number of participants in the intervention groups. Online supplemental table S2 lists the definitions and cut-off values for categorising dietary patterns in terms of the amount of energy derived from macronutrients. We also extracted summary effect estimates (mean difference, standardised mean difference, or risk ratio) with corresponding 95% confidence intervals, the statistical model used (fixed effect or random effects model), a measure of inconsistency between randomised controlled trials (I²), publication bias, risk of bias, and certainty of evidence (if assessed according to the GRADE approach). Online supplemental table S2 has further details on data extraction.

Assessment of methodological quality and certainty of evidence

The methodological quality of each included publication was assessed with the validated A MeaSurement Tool to Assess systematic Reviews (AMSTAR) 2 by two reviewers independently.23 Online supplemental table S3 lists a detailed description of the tool, included domains, and grading system.

The certainty of evidence was evaluated with the GRADEpro tool.24 25 GRADE rates the certainty of evidence as high, moderate, low, or very low. A high certainty of evidence means that it is very likely that the true effect lies on one side of a specified threshold or within a chosen range, a moderate certainty of evidence indicates moderate confidence that the true effect lies on one side of a specified threshold or within a chosen range and further studies might change the results, and low or very low certainty of evidence means that confidence in the meta-evidence available is limited.26 27 Three reviewers, in pairs of two, independently rated the certainty of the evidence. Online supplemental table S4 lists a detailed description of the grading of the certainty of evidence.

Data analysis

For meta-analyses that used a random effects model and provided I² values and forest plots, we extracted the reported summary effect estimates with 95% confidence intervals. For meta-analyses that used a fixed effect model, or did not provide I² values or forest plots, we recalculated the mean difference, standardised mean difference, and summary risk ratio with corresponding 95% confidence intervals with the random effects model by DerSimonian and Laird.28 Information on small study effects and publications bias was extracted from the included publications or, if not reported, was assessed by funnel plots and Egger's test (if ≥10 studies were included).29 30 Online supplemental table S5 presents further details on the statistical methods. All analyses were conducted with Stata version 14.1.

Patient and public involvement

No patients or public were involved in setting the research question, the design of the study, the interpretation of the results, or the writing of the manuscript. The results of this study will be spread to the public via social media or press release.

Literature search

Of 25 221 publications, 764 eligible articles were retrieved for screening of the full text of the article (online supplemental figure S1). Online supplemental table S6 lists the excluded studies and the reasons for exclusion. We identified 88 eligible publications of systematic reviews with meta-analyses of randomised controlled trials for inclusion in our umbrella review.

Description of published meta-analyses

Online supplemental table S7 presents the characteristics and results of all of the included systematic reviews with meta-analyses of randomised controlled trials. Eighty eight publications, with 310 meta-analyses of randomised controlled trials, investigated the effectiveness of dietary patterns (liquid meal replacement, energy, carbohydrate, and fat restriction, low glycaemic index and glycaemic load, high protein, Mediterranean, plant based, vegetarian, and ketogenic diets, and intermittent fasting), foods (mixed nuts, tree nuts, walnuts, almonds, blueberries and cranberries, olive oil, cocoa products, cinnamon, curcumin, ginger, soy and soy isoflavones, and non-nutritive sweeteners), nutrients (monounsaturated fatty acids, omega 3 fatty acids, α-lipolenic acid, L-carnitine, and salt), micronutrients (magnesium, chromium, and zinc), vitamins (niacin, and vitamins E, C, and D), phytochemicals (anthocyanins, flavonoids, polyphenols, and resveratrol), fibre (psyllium, soluble fibre, and guar gum), and probiotics (online supplemental table S7).

Outcomes included anthropometric, glycaemic, and lipid markers, blood pressure, kidney, liver, and inflammation parameters, as well as quality of life, reduced use of drug treatments, constipation, and major cardiovascular events. The number of participants in meta-analyses of randomised controlled trials ranged from 44 to 28 199, and intervention time from 12 weeks to 96 months. In total, 293 of the included meta-analyses were pairwise meta-analyses and 17 were network meta-analyses,12 31 and 15% of the meta-analyses included ≥10 randomised controlled trials. Publication bias was detected for 2.2% of the results of meta-analyses of randomised controlled trials, but 59.4% of meta-analyses included ≤5 studies (online supplemental table S7).

Methodological quality was rated as high for 18% (n=16), moderate for 5% (n=4), low for 22% (n=19), and critically low for the remaining 55% (n=49) of publications (online supplemental table S8). The most common reasons for downgrading the methodological quality were missing study protocol, inappropriate methods for the search strategy (literature search only by one investigator, search strategy not shown, or included reports restricted to the English language), and lack of evaluation or discussion, or both, of the effect of risk of bias on the results.

Evidence for the effect of diet on managing type 2 diabetes: meta-analyses of randomised controlled trials

The certainty of evidence was rated as high for seven (2%), moderate for 40 (13%), low for 120 (39%), and very low for 143 (46%) outcomes (online supplemental tables S9–S32). Table 1 summarises the results of the meta-analyses with high to moderate certainty of evidence and with clinically meaningful changes. Figures 1–6 show the summary effect estimates of the included meta-analyses of randomised controlled trials with high and moderate certainty of evidence. Online supplemental figures S2–S10 show the results of meta-analyses of randomised controlled trials with low and very low certainty of evidence.

Request permission

Request permission

Request permission

Request permission

Request permission

Request permission

Main findings

In this umbrella review, we included 88 publications, with 310 meta-analyses of randomised controlled trials for different dietary factors and health outcomes, mainly surrogate markers of disease and health, in populations with type 2 diabetes. We found a high certainty of evidence and clinically important changes for liquid meal replacement diets in reducing body weight and body mass index, as well as for a low carbohydrate diet (<26% total energy) in lowering levels of HbA_1c and triglycerides. A moderate certainty of evidence with clinically important changes was found for plant based diets and reduced anthropometric measures; liquid meal replacement diets, and supplementation with fibre, ginger, anthocyanins, or probiotics improved glycaemic measures; and a high protein diet, soya, or fibre improved blood lipids. For other results, no clinically relevant differences were found or the certainty of evidence was low or very low.

Comparison with existing dietary guidelines

Dietary interventions based on energy restriction for weight loss are the focus of current guidelines for the management of type 2 diabetes.48–53 Clinically important weight loss of 5-10% can improve insulin sensitivity, glycaemic control, blood pressure, and dyslipidaemia.48 49 53 Also, the guidelines suggest that low calorie meal replacement plans are relatively easy to adhere to and allow for clinically significant weight loss in people with type 2 diabetes.48 50 53 We found a reduction in body weight of 2.4 kg after at least 12 weeks of liquid meal replacement diet.

Current guidelines for type 2 diabetes recommend vegetarian or plant based diets and Mediterranean diets for reducing body weight.48 49 52 Our umbrella review confirmed the beneficial effect of plant based diets on reducing body mass index and waist circumference. Also, some of the current guidelines conclude that short term low carbohydrate interventions have a beneficial effect on anthropometric measures, but no clear recommendations exist for the amount of carbohydrates that should be restricted.48 49 Our study showed that a decrease in body weight was greater in interventions with low (<26% total energy) or very low (<15% total energy) carbohydrate intake than in interventions with moderate (<45% total energy) carbohydrate intake for at least 12 weeks.

We found a low certainty of evidence for the effects of a Mediterranean or plant based diet on glycaemic measures. Our findings of the beneficial effect of low carbohydrate diets on levels of HbA_1c, however, are in line with current guidelines.49 50 Increased intake of dietary fibre is emphasised in all guidelines, and our results are consistent with the positive effect of fibre rich foods, especially psyllium supplements, on fasting blood glucose levels.49 53 Adding to the current guidelines, we identified a moderate certainty of evidence for the beneficial effects of ginger supplements, as a plant, capsule, or powder, on fasting blood glucose levels, and the potential importance of polyphenols or polyphenol rich products (eg, cocoa) on HOMA-IR.

Our umbrella review supports evidence of the beneficial effects of a Mediterranean diet on triglyceride levels, but we found only a low certainty of evidence for its effect in reducing levels of high and low density lipoprotein cholesterol. In agreement with current guidelines, we also found beneficial effects of low carbohydrate and low glycaemic index diets on levels of triglycerides and low and high density lipoprotein cholesterol, and of higher fibre intake on levels of total cholesterol.48–50 53 Current Canadian, German, and European guidelines recommend no change in protein intake, which for most people with normal kidney function is 15-20% and 10-25% of total energy.48 50 53 Thus our findings of a beneficial effect of a high protein diet (>25% total energy) on levels of total cholesterol and low density lipoprotein cholesterol provide novel evidence in this context. A previous meta-analysis, however, showed that the health benefits are from an increased intake of plant based rather than animal based protein sources.54 Furthermore, our study provides evidence of the beneficial effects of soy proteins on levels of total cholesterol and low density lipoprotein cholesterol. Pulses are considered a good alternative to meat and a good source of fibre and are therefore recommended for controlling levels of blood lipids.48 50

In common with current guidelines, we found insufficient evidence to recommend reducing total fat intake to improve cardiometabolic measures in people with type 2 diabetes.48 50 American and European guidelines recommend a higher intake of monounsaturated and polyunsaturated fatty acids to improve glucose metabolism, but the evidence supporting this recommendation, according to our results, is weak.49 53 We found that studies of supplementation with monounsaturated or polyunsaturated fatty acids showed improvements in levels of HbA_1c and insulin, and in HOMA-IR, but the estimates were imprecisely estimated and the number of participants included was small (≤800). Also, a subgroup analysis comparing different doses of omega 3 fatty acid supplements was performed in one meta-analysis,45 and a stronger risk reduction of major cardiovascular events was seen in the group receiving >3 g/day of omega 3 fatty acids, whereas no precisely estimated effect was found in the group receiving lower doses. This finding might explain the observed effect of omega 3 fatty acids on the incidence of major cardiovascular events in people with type 2 diabetes in our umbrella review. Moreover, a recent systematic review and meta-analysis of randomised controlled trials with an intervention duration of at least 3.9 years showed that supplementation with omega 3 fatty acids reduced cardiovascular events in people with type 1 and type 2 diabetes (risk ratio 0.93, 95% confidence interval 0.90 to 0.97; n=8 randomised controlled trials). The effect was marginal but precisely estimated.55

Strengths and limitations

Our study has several strengths. Our umbrella review included meta-analyses of all possible dietary interventions and their effect on many health outcomes in populations with type 2 diabetes. We summarised the evidence from randomised controlled trials that lasted at least 12 weeks, giving a more reliable assessment of the effectiveness of the interventions. We recalculated the results of meta-analyses that used a fixed effect model, or did not provide I² values or forest plots, to ensure valid meta-estimates with 95% confidence intervals to provide an evaluation of the certainty of evidence based on the GRADEpro approach. Finally, we detected gaps in the evidence that indicate the need for future research.

The main limitation of our umbrella review was that recently published randomised controlled trials not yet included in systematic reviews and meta-analyses were not considered in our report.56–58 Also, we did not explore subgroup analyses (eg, by sex, race, or duration of diabetes) or sensitivity analyses (eg, excluding studies with a high risk of bias). We used the DerSimonian and Laird approach28 for recalculation of extracted estimates with the random effects meta-analyses, but the Hartung and Knapp method might perform better in terms of more adequate error rates, especially when the number of studies was small.59

In terms of the limitations of the publications included in our umbrella review, studies of interventions lasting at least 12 weeks provide more reliable estimates than shorter trials, but these studies do not represent long term interventions. Also, the number of randomised controlled trials in most meta-analyses was small (13 meta-analyses with ≥10 randomised controlled trials). Consequently, the number of participants was also low (n<800 for 83% of meta-analyses of randomised controlled trials). This limitation was one of the main reasons for downgrading the certainty of evidence because of imprecision. Other reasons for downgrading the certainty of evidence were the high risk of bias in the randomised controlled trials and inconsistency of the results. Furthermore, the methodological quality of most included publications was low to very low.

In meta-analyses of randomised controlled trials, a high risk of bias caused by lack of blinding of participants to the dietary intervention is a concern. For dietary interventions based on health promoting changes in diet (eg, advice on a Mediterranean diet, where participants are actively instructed and encouraged to change their diet), however, blinding to the intervention might not be feasible. Furthermore, low compliance with the assigned dietary regimens and high dropout rates were commonly seen in randomised controlled trials, especially for low carbohydrate and ketogenic diets,60 potentially resulting in underestimation of their actual effect. For example, one meta-analysis found greater clinically significant weight loss in those with greater adherence than in those with lower adherence to a very low carbohydrate diet,37 and another meta-analysis reported inadequate compliance with the ketogenic diet, as assessed by urinary measurements of ketones.13 Systematic reviews of other dietary interventions lacked a description of compliance with the intervention. Also, study arms did not always receive isocaloric diets, making it difficult to differentiate between the effects of change in dietary patterns, foods, or nutrients on the reduction in energy intake. Some of the meta-analyses conducted subgroup analyses, however, and did not find differences when calories were restricted or matched with controls in the trials.33 34 37

Study implications

Robust evidence exists that in people with type 2 diabetes, liquid meal replacements decrease energy intake and thus body weight. For people who prefer a dietary approach, plant based or carbohydrate restricted diets are also effective in reducing anthropometric measure. Other dietary regimens, such as a ketogenic diet or intermittent fasting, reduced body weight, but the certainty of evidence was low or very low.

Many dietary interventions effectively improved glycaemic measures, but the certainty of the evidence was robust only for liquid meal replacement diets and restricting carbohydrates. Intake of fibre rich foods, polyphenols, probiotics, and ginger was also beneficial for glycaemic control. For control of blood lipids, a low carbohydrate (<26% total energy), Mediterranean, or high protein diet is recommended for people with type 2 diabetes, with the advice to change the protein source to plant based alternatives. Energy restriction was the only effective approach in reducing blood pressure with robust evidence. Restriction of salt intake also showed beneficial effects, but the certainty of evidence was very low and thus recommendations cannot be made from the current body of evidence.

For clinical outcomes, robust evidence exists only for a low carbohydrate diet and reduction in the use of drug treatments. For other outcomes, including kidney function parameters, inflammatory markers, liver enzymes, and patient relevant outcomes (eg, remission of diabetes, health related quality of life, and incidence of cardiovascular disease), meta-analyses are already available, but the certainty of evidence was rated as low or very low, and definitive conclusions on these findings cannot be drawn. In summary, people with type 2 diabetes can be advised to reduce their energy intake if they have obesity or overweight, decrease their carbohydrate intake, or increase their consumption of foods from plant sources (especially plant based proteins) or from Mediterranean-style diets (foods high in polyunsaturated fatty acids, such as fish and nuts, or foods high in polyphenols, such as fruit, vegetables, and legumes).

To strengthen the certainty of evidence for many of the findings on diet and the management of type 2 diabetes and its complications, future studies with a low risk of bias are needed. More randomised controlled trials should investigate the effects of long term (at least six months) dietary interventions, as well as considering isocaloric comparisons between intervention and control arms. Also, more research on dietary patterns (eg, DASH (dietary approaches to stop hypertension), Nordic, and portfolio diets), fasting approaches, single foods (eg, dairy products, fish, or meat), and single nutrients (eg, specific fatty acids) is needed. Finally, future systematic reviews and meta-analyses should follow current guidelines for conducting and reporting on included studies (eg, reports should be transparent about the methods used and critically examine the effect of risk of bias on meta-findings).