figure b

Introduction

Type 2 diabetes has both environmental and genetic contributors, the global epidemic consistently following obesity. Its onset is primarily driven by weight gain to an excessive level for that individual, in a complex disease process involving gut hormones, low-grade inflammation and metabolites, possibly including some from the gut microbiota [1]. Ectopic fat accumulation in liver, pancreas and muscle impairs organ functions, resulting in hyperglycaemia, commonly associated with hypertension and dyslipidaemia [2, 3]. Type 2 diabetes requires lifelong management, but disabling and life-shortening complications occur despite treatment [4]. Without strategic commitment, internationally, to effective preventive actions, type 2 diabetes will affect 629 million people worldwide by 2045 [5].

Weight loss improves all weight-related risk factors and reduces medication load. During an intensive weight loss programme, or early after bariatric surgery, there are already significant improvements in hepatic and muscle insulin sensitivity, and pancreatic first-phase insulin secretion, with rapid loss of ectopic fat from skeletal muscle and liver [2, 3, 6]. A non-diabetic state can be restored for 2 years for 70–80% of people with type 2 diabetes by interventions that maintain over 10 kg weight loss (36/149, 24% of participants in the Diabetes Remission Clinical Trial [DiRECT]) [7], which sustains loss of ectopic fat, reversing the pathophysiology and normalising pancreas morphology [8].

Awareness of the benefits of weight loss for type 2 diabetes is high, but both patients and healthcare practitioners currently lack authoritative guidance over diets [9]. Current guidelines state that various dietary strategies may be effective, and stress personalising weight management, to take account of social situations, but do not provide guidance over diet compositions [9, 10]. Consequently, practice can be led by distorted evidence and claims.

Adhering to any energy-reduced diet will inevitably generate and sustain weight loss, whether defined by restriction of energy, of food groups or of specific nutrients, provided that there is incomplete compensation in energy intake and expenditure. In practice, adherence and weight losses vary widely within the same programme, and comparisons between diets often appear to have conflicting results [11]. Metabolic diversity in response to specific nutrient contents has been postulated, but possibly overwhelmed or confounded by mixed behavioural responses to dietary advice. Unless carefully designed, some diets may achieve negative energy imbalance but lack essential micronutrients [12, 13] or introduce adverse health effects through other pathways [14,15,16]. Furthermore, short-term results may not be sustained, potentially requiring additional behavioural approaches for long-term maintenance. While different strategies may work better for some individuals (or some practitioners) than others, there may be preferred diet compositions to optimise weight control [17].

Guideline development has been difficult because systematic reviews and meta-analyses of diet types, themselves open to bias, have appeared conflicting [11]. To resolve these uncertainties and to inform clinical decision making and guideline development as part of a programme of work to update the EASD dietary recommendations, we conducted an umbrella review, to collate and critically appraise all available systematic reviews with meta-analyses of dietary interventions for weight loss in people with type 2 diabetes. As remission of diabetes is now an important goal for weight management, we also conducted a new systematic review and quality appraisal of published intervention studies of non-surgical dietary approaches for type 2 diabetes remission.

Methods

Protocol and registration

This paper focuses on dietary strategies for weight loss and type 2 diabetes remission and includes two systematic reviews: (1) a systematic ‘umbrella review’ of published meta-analyses of RCTs of diets for weight loss in people with type 2 diabetes (PROSPERO CRD42020169258); (2) a systematic review of any intervention studies which report type 2 diabetes remission (PROSPERO CRD42020208878). Our paper is written in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 [18] and the Synthesis Without Meta-analysis in systematic reviews: reporting guideline [19].

Detailed methods of both systematic reviews are presented in the electronic supplementary material (ESM) Methods and summarised in Fig. 1. The search strategy is in ESM Table 1.

Fig. 1
figure 1

Summary of the methodological processes of both systematic reviews. Detailed methods are presented in the ESM Methods. aThese types of NRSs provided intervention to participants and assessed outcomes at designated specific time points (baseline and at the end of intervention), although they could suffer from selection bias and confounding bias. bAMSTAR 2 level of quality assessment: high quality—the meta-analysis provides an accurate and comprehensive summary of the results of the available studies that addresses the question of interest; moderate—the meta-analysis has more than one weakness, but no critical flaws. It may provide an accurate summary of the results of the available studies; low—the meta-analysis has a critical flaw and may not provide an accurate and comprehensive summary of the available studies that address the question of interest; or critically low—the meta-analysis has more than one critical flaw and should not be relied on to provide an accurate and comprehensive summary of the available studies. CENTRAL, Cochrane Central Register of Controlled Trials; ROBINS-I, Risk Of Bias In Non-randomised Studies – of Interventions; T2D, type 2 diabetes

Full size image

(1) Umbrella review of published meta-analyses

We searched MEDLINE (Ovid), PubMed, Web of Science and Cochrane Database of Systematic Reviews, up to 7 May 2021, for eligible meta-analyses of RCTs of dietary advice for weight loss.

Data synthesis

Synthesised findings (weight loss and HbA1c) from each meta-analysis included are grouped by diet type, ranked by overall methodological quality using A Measurement Tool to Assess Systematic Reviews (AMSTAR) 2 (ESM Tables 2, 3) and categorised into four levels: high, moderate, low and critically low. Grading of Recommendations, Assessment, Development and Evaluations (GRADE) evaluates the certainty of evidence of pooled results (ESM Table 4).

Planned analysis of associations between changes in energy intake and weight changes from baseline, to differentiate effects of energy restriction and dietary regimen, proved impossible from the published information.

(2) Systematic review of diets for type 2 diabetes remission

We searched MEDLINE (via PubMed), Embase and Cochrane Central Register of Controlled Trials, up to 10 May 2021, for any intervention studies reporting type 2 diabetes remission with weight loss dietary advice. We first included RCTs reporting type 2 diabetes remission as the primary outcome, the design most likely to provide trustworthy evidence. However, as few such RCTs have been conducted, we also evaluated non-randomised studies (NRSs) to capture the totality of the evidence for ‘best available advice’ to inform practice and policy [20]. Cochrane Risk of Bias tool 2.0 [21] and Risk Of Bias In Non-randomised Studies – of Interventions (ROBINS-I) [22] were used for quality assessment of RCTs and NRSs, respectively.

Data synthesis

Remission of diabetes was reported as percentage from intention to treat (ITT), including all participants. If only completers were reported, we computed an ITT figure assuming participants lost to follow-up all failed to achieve remission (as in the published RCTs). We summarised effect estimates (e.g., median and interquartile ranges), without performing meta-analysis, due to the limited number and heterogeneity of studies [23]. GRADE assesses the certainty of synthesised findings [24].

For the main synthesis, priority was set to RCTs reporting 1 year outcome and low risk of bias. If there was no RCT for a particular diet, synthesis findings were drawn from NRSs with low, followed by high, risk of bias. If both RCTs and NRSs were available for a diet, RCTs were used for synthesised findings and NRSs as supportive evidence [20]. Heterogeneity was explored according to hypothesised effect modifiers: study design, duration of type 2 diabetes and ethnicity.

Results

(1) Umbrella review of published meta-analyses of RCTs of diets for weight loss and glycaemic control

Identification of meta-analyses

We retrieved 1064 records, including all languages. After removing duplicates, we screened 690 titles and abstracts, and assessed 59 full texts for eligibility. Excluded full texts, with reasons, are shown in ESM Table 5. We included a total of 21 systematic reviews (with 19 meta-analyses) for data synthesis and quality assessment (ESM Fig. 1).

Characteristics of included meta-analyses

Of the 19 meta-analyses (Table 1, ESM Table 6), 18 reported direct comparisons of specific diets. Control diets varied, either usual/routine care or a particular dietary regimen. One meta-analysis used a network method to consider both direct and indirect comparisons between multiple diets (Mediterranean diets, low-carbohydrate diets [LCDs], low-fat diets [LFDs], high-carbohydrate diets and usual diets) [25]. Most meta-analyses were of critically low (n = 7; [25,26,27,28,29,30,31]) to low quality (n = 5; [32,33,34,35,36]). Only seven meta-analyses (LCDs, n = 5 [37,38,39,40, 41]; liquid meal replacement, n = 1 [42]; very low energy diet (VLED), n = 1 [43]) were assessed as high quality. The ESM Results and ESM Tables 7–10 present methodological quality, heterogeneity and overlaps in source trials of meta-analyses included in the umbrella review.

Table 1 Characteristics of included meta-analyses of RCTs of dietary weight management in type 2 diabetes
Full size table

Dietary advice for weight loss

Weight loss outcomes from published meta-analyses are presented in Figs. 2, 3 and ESM Table 11.

Fig. 2
figure 2

All published meta-analyses of intervention diets vs control diets on weight loss (kg) stratified by overall quality in each diet type using AMSTAR 2 quality (green, high quality; orange, low quality; red, critically low quality). WMDs are presented alongside 95% CIs (error bars). Pooled results of McArdle et al., 2019 [34], Fan et al., 2016 [27], Zafar et al., 2019 [36] and Zhao et al., 2018 [30] are standardised mean differences. aComplete case data. GRADE level for certainty of evidence is rated as follows: ‘high’ indicates that we are very confident that the true effect lies close to that of the estimate of the effect; ‘moderate’ indicates that we are moderately confident in the effect estimate (the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different); ‘low’ indicates that our confidence in the effect estimate is limited (the true effect may be substantially different from the estimate of the effect); and ‘very low’ indicates that we have very little confidence in the effect estimate (the true effect is likely to be substantially different from the estimate of effect)

Full size image
Fig. 3
figure 3

Meta-analyses with source RCTs of 12 months or longer on weight loss (kg) outcome. WMDs are presented alongside 95% CIs (error bars). Different colours indicate meta-analysis quality: green, high quality; red, critically low quality. GRADE level for certainty of evidence: ‘high’ indicates that we are very confident that the true effect lies close to that of the estimate of the effect; ‘moderate’ indicates that we are moderately confident in the effect estimate (the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different); ‘low’ indicates that our confidence in the effect estimate is limited (the true effect may be substantially different from the estimate of the effect); and ‘very low’ indicates that we have very little confidence in the effect estimate (the true effect is likely to be substantially different from the estimate of effect)

Full size image

LCDs

Ten meta-analyses reported on LCDs compared with higher-carbohydrate diets. Not all reported whether source RCTs were ad libitum or hypocaloric prescriptions, with results often pooled from both trial types. Definitions of LCDs varied, including <130 g/day, and <26% or <45% of energy intake from carbohydrate. Duration of interventions ranged from 8 weeks to 4 years.

Four high-quality meta-analyses [37,38,39,40] reported that LCDs and higher-carbohydrate diets were equally effective for weight loss, with mean difference ranging between <1 and <2.5 kg, at all durations. GRADE assessment ranged from low to high certainty of evidence. Just one meta-analysis reported greater weight loss with LCD, by 3.5 kg, using complete case data for pooled results [41]. The remaining critically low- to low-quality meta-analyses showed differences of <1 kg between the two diets [26, 27, 31, 33, 34].

Very low-carbohydrate diets (21–70 g of carbohydrate daily) showed no greater weight loss than higher-carbohydrate diets over durations of 3–36 months (weight mean difference [WMD] −0.7 kg; 95% CI −2.0, 0.7; I2 = 46%, p = 0.10) in a subgroup analysis [37]. A subgroup analysis of RCTs with low risk of bias reported no difference (WMD 0.9 kg; 95% CI -1.9, 3.6), while RCTs with high risk of bias showed greater weight loss for LCDs than higher-carbohydrate diets (WMD −1.8 kg; 95% CI −2.8, −0.7) [37].

High-protein diets

All meta-analyses (n = 3) of high-protein diets were of critically low quality [26, 28, 30]. Critical domains unmet were presence of a review protocol and assessing risk of bias in synthesised findings (ESM Table 3). Only one provided a definition of ‘high protein’ (>20% of energy intake), reporting significantly greater weight loss (−1.2 kg; 95% CI −2.17, −0.24; I2 = 5%, p = 0.38) than with lower-protein diets (<20% of energy from protein) [28].

Mediterranean diets

Two meta-analyses, of low and critically low quality, considered weight loss from Mediterranean diets [26, 32]. The control interventions combined no diet (usual care) and specified diets, including LFD and LCD. Pooled results indicated significantly greater weight loss with Mediterranean diets than in control groups, by 0.3 kg (low quality; [32]) to 1.8 kg (critically low quality; [26]), over durations of 4–24 weeks. A network meta-analysis also reported that Mediterranean diets were marginally more effective than LFDs for weight loss (−1.2 kg; 95% CI −1.99, −0.37; four RCTs, low quality; p-heterogeneity = 0.08; ESM Table 12) [25].

Formula meal replacement

One high-quality meta-analysis [42] of nine RCTs including 931 participants reported that replacing one to three main meals daily (replacing 13–47% of total energy) produced significantly greater weight loss than low-energy diets over 12–52 weeks (−2.4 kg; 95% CI −3.3, −1.4; I2 = 84%, p < 0.001; GRADE moderate certainty of evidence).

VLEDs

One high-quality meta-analysis [43] of two RCTs reported that VLEDs (1.7–2.1 MJ/day for 8–12 weeks) achieved greater weight loss at 3 months (−6.6 kg; 95% CI −9.5, −3.7; I2 = 58%, p = 0.12) and at 6 months (−5.7 kg; 95% CI −11.1, −0.4; I2 = 58%, p = 0.12), compared with an energy-restricted diet (4.2–6.3 MJ/day). These data were from participants who completed the trials.

High-monounsaturated-fatty-acid, vegetarian and low-glycaemic-index diets

High-monounsaturated-fatty-acid (MUFA) [29] and vegetarian diets [35] showed greater weight losses, by −1.6 to −2 kg, than the control diets. Low-glycaemic-index diets [26, 36] were not associated with greater weight loss than control diets. Published meta-analyses of these diets were of low to critically low quality.

Intermittent fasting

We did a post hoc analysis to evaluate all systematic reviews without meta-analyses (no pooled weight loss; n = 10) that were excluded from our main analysis (as intended protocol). Eight were systematic reviews whose source RCTs were already pooled in meta-analyses identified in this umbrella review. The remaining two systematic reviews compared altered eating patterns with conventional energy-restricted diets (ESM Table 13) [44, 45]. From these two reviews, three RCTs were identified: two for 5:2 diets reported no difference in weight loss (high-risk-of-bias RCTs) [46, 47], and one for time-restricted dieting reported 1.4 kg greater weight loss than conventional energy restriction (high-risk-of-bias RCT) [48].

Adherence

Some meta-analyses offered assessed dietary adherence separately from weight change. Adherence assessed up to 1 year was poorer with very low-carbohydrate diets (<50 g of carbohydrate) than with LCDs (<130 g of carbohydrate) [34, 37, 40], possibly because most of the trials allowed increased carbohydrate intake for later weight loss maintenance. High adherence to VLEDs (up to 6 months), judged from rapid early weight loss and dietary assessment, led to better long-term results [43].

Effects of weight-loss diet intervention on HbA1c

Among published meta-analyses, HbA1c reduction broadly followed weight loss, and differences between diet types assessed over 3–12 months were small. The published data do not permit an individual-level regression analysis to quantify weight loss-independent effects on HbA1c (ESM Results, ESM Table 14).

(2) Systematic review of intervention studies (either RCTs or NRSs) of diets for remission of type 2 diabetes

Identification of studies

From 373 records identified, we included 16 papers for data synthesis and quality assessment (ESM Fig. 2; excluded studies with reasons in ESM Table 5). These reported on 14 studies (six RCTs, eight NRSs), of seven diet types: total diet replacement (n = 4), formula meal replacement (n = 2), VLED (n = 2), very low-carbohydrate ketogenic diet (n = 1), Mediterranean diet (n = 2), LFD (n = 4) and the ADA diet (n = 1). Five studies compared diet interventions with usual care according to clinical guidelines, without providing foods or dietary products for participants [49,50,51,52,53]. Among these, three provided diabetes education or advice (Table 2) [49, 50, 52]. Included studies were conducted in Barbados, India, Italy, Qatar, South Africa, Spain, Thailand, the UK and the USA. Detailed characteristics and methodological quality are in the ESM Results, ESM Tables 15–17 and ESM Fig. 3.

Table 2 Type 2 diabetes remission (%) and mean weight loss (kg) from baseline according to different dietary regimens/patterns
Full size table

Definition of type 2 diabetes remission

All included studies defined remission as a diagnostic test result, without glucose-lowering medication, below the WHO threshold for diagnosis of type 2 diabetes (HbA1c < 48 mmol/mol [6.5%], or fasting plasma glucose <7 mmol/l), but they differed in the duration prior to assessment of remission (ESM Tables 15–16). Some studies [54,55,56,57] subdivided results as previously proposed by Buse et al. [58]. Glucose-lowering medications were not routinely withdrawn at the beginning of diets in some of the studies, so only minimum remissions can be reported.

Effects of diets on type 2 diabetes remission and weight at 1 year

Remission rates and weight changes at 1 year are summarised in Fig. 4 and Table 2, with GRADE certainty of evidence in Table 3.

Fig. 4
figure 4

Percentage of remissions of type 2 diabetes at 12 months after intervention with different diet types, stratified by study design and risk of bias. Each dot, with varying shapes to reflect risk of bias, indicates the data point for each of the studies mentioned in the main text which provided data in this form at 12 months. The column represents the mean for the diet type. Remission was defined as either HbA1c < 48 mmol/mol (<6.5%) or fasting plasma glucose <7 mmol/l, with no glucose-lowering medication. Total diet replacement programmes included an initial low-energy formula diet, prescribed for an 8–12 week induction phase, followed by stepped food re-introduction aimed to achieve energy balance for weight loss maintenance. VLED advised a 2.9 MJ (700 kcal) food-based diet for 1 week, then dietary advice for energy intake that matched for ideal body weight. Very low-carbohydrate ketogenic diet was ad libitum intake, carbohydrate <30 g/day to achieve ketosis and 3–5 servings of non-starchy vegetables. Usual diet or standard diet interventions included diabetes education support, but no new diet intervention

Full size image
Table 3 Summary of findings of type 2 diabetes remission at 1 year after diet intervention compared with baseline with GRADE certainty of a body of evidence
Full size table

Programmes that included an induction phase of formula ‘total diet replacement’ were studied in two RCTs with low risks of bias. Compared with remissions of 4–12% in well-matched usual care control arms, the interventions generated median 54% remission at 12 months from baseline (N = 445, two RCTs; GRADE high certainty of evidence), with diabetes durations <6 or <2 years, and mean weight loss of 10 and 12 kg. These two RCTs were designed with remission as the primary outcome [51, 53].

Among trials reporting post hoc analyses for remission, one using two meal replacements/day during 0–20 weeks and one per day thereafter reported 11% (247/2157) remission at 1 year (prevalence estimates), with mean weight loss 8.6 kg, compared with 2% (43/2170) in standard care (N = 4503; GRADE low certainty of evidence), with some concern over risk of bias [49].

A single RCT of Mediterranean diet over 12 months reported a remission prevalence of 15% (15/102), with mean weight loss 6.2 kg, compared with 4% (4/97) with weight loss 4.2 kg in the control LFD arm (N = 215; GRADE low certainty of evidence; some concern over risk of bias) [59].

No RCT has evaluated LCDs/ketogenic diets for type 2 diabetes remission. A non-randomised, controlled study of a very low-carbohydrate (ketogenic) diet reported 20% remission (52 out of 262 who started treatment [ITT] who had HbA1c < 48 mmol/mol [6.5%] without diabetes medication), with mean weight loss 13.8 kg, at 1 year, compared with no remission in a control arm (N = 349; GRADE very low certainty of evidence; serious risk of bias) [50]. The dropout rate was 17% (44/262) and 22% had incomplete outcome data. This study primarily focused on HbA1c lowering, not remission, so glucose-lowering medications were not routinely withdrawn.

Another very small uncontrolled study evaluated a 1 week 2.9 MJ/day (700 kcal) food-based diet, finding 22% (2/9) remission at 1 year, with mean weight loss 4.2 kg (N = 9, one single-arm intervention; GRADE very low certainty of evidence; critical risk of bias) [60].

Sources of heterogeneity

Single-arm intervention studies reported higher remission than RCTs. Participants with shorter type 2 diabetes duration, and Asian ethnicity, were more likely to achieve remission than those with longer type 2 diabetes duration or another ethnicity (Table 2, ESM Tables 15,16).

Discussion

Dietary weight reduction for people with type 2 diabetes

This study was conducted to inform practice and policy over dietary advice for weight management of people with type 2 diabetes. It has therefore focused on interventions in free-living individuals, with a view to long-term management. Based on methodological quality and certainty of the evidence, our umbrella review of meta-analyses found that VLEDs and formula meal replacements produce greater weight losses than conventional low-energy diets. The evidence does not favour LCDs above higher-carbohydrate diets, nor other dietary approaches, i.e., high-protein, Mediterranean, high-MUFA, vegetarian and low-glycaemic-index diets, above control diets. Currently popular intermittent fasting was only captured in systematic reviews without meta-analysis (high-risk-of-bias RCTs) [44, 45]. The evidence, albeit of variable ‘quality’, is rather consistent such that no one diet type is superior over others for weight management in type 2 diabetes.

While the evidence does not suggest important differences between macronutrient compositions in effectiveness, there may be differences in cost-effectiveness. The evidence on relative cost-effectiveness of weight-loss diet programmes is limited from head-to-head diet comparison trials, but one RCT showed that LCD was not more cost-effective than the standard weight-loss diet [61]. In the Doctor Referral of Overweight People to Low Energy total diet replacement Treatment (DROPLET) trial among people without diabetes, in routine practice, a total diet replacement programme (formula diets) with behavioural support proved more cost-effective than nurse-led dietary advice for long-term prevention of obesity-related diseases [62]. For diabetes remission, a total diet replacement programme was estimated to be both more cost-effective and cost-saving than standard care in the DiRECT trial, reflecting reduced need for medications and fewer diabetes complications [63].

Health benefits from weight management depend largely on long-term control of body weight. Most of the evidence cited relates to short-term outcomes, relevant to the initial weight loss induction phase of weight management. Few trials have reported data beyond 12 months, to reflect weight loss maintenance, which may demand different behavioural strategies. One large RCT of high-protein diet suggested benefit for weight loss maintenance, increasing satiety and energy expenditure, albeit for a maintenance phase of only 6 months after completing weight loss [64]. Nutrient-specific effects have been postulated, but are likely to be overwhelmed by variable behavioural responses to dietary advice [65]. Behavioural programmes help to sustain new behaviours, relationships with foods and adherence to dietary advice [66,67,68]. Consistent evidence is also accruing that long-term weight loss maintenance is better after more rapid early weight loss [69]. Thus, treatments effective for weight loss only in the short term may have long-term value if complemented with a good weight loss maintenance strategy. Practitioners can therefore be confident that a variety of diet types can all achieve the intended weight losses, and potentially remissions of type 2 diabetes, if their patients are able to adhere to the programme sufficiently.

The analyses contradict some popular claims about specific diets: in particular, ‘low-carb’ diets hold no overall advantage for weight loss when compared with higher-carbohydrate diets. However, we cannot conclude that any individual with type 2 diabetes, in any context, will do equally well with any diet advice, or that a skilled practitioner may not have greater success advising one diet type. The skills and empathy of practitioners may overcome any diet-specific effects on weight loss by providing consistent evidence-based support [70]. Realistic trials are required, in which individuals are offered choices, perhaps using n = 1 randomised trial designs.

Weighing benefits against risks

Food is fundamental for personal and social wellbeing, and diets can be psychologically testing. Patient preferences, culture, context and lifestyle demand open conversation and shared decision making between practitioners and patients. For either medication or diet, weighing benefits against risks is vital: treatment benefits are often overestimated but harms underestimated [71]. Although all diet types are similarly effective for weight control, health risks were not systematically reported across the studies, and could differ [72]. More rapid early weight loss with more intensive programmes is associated with better longer-term weight outcomes [69], but severe caloric restriction without attention to nutrient content can have unwanted effects. Blood pressure falls with weight loss, and postural hypotension, common in older people and those with diabetes, is aggravated during rapid weight loss if diuretic or antihypertensive drugs are taken concurrently [73]. Hypoglycaemia is possible if hypoglycaemic drugs are also taken [74]. Diets other than nutritionally complete formula diets could incur vitamin and mineral deficiencies [75]. With ketogenic diets, heart failure and neurological problems from thiamine deficiency have been reported [76, 77], as well as reduced intakes of folate, iron and magnesium [12]. Replacing high-carbohydrate foods with red or processed meat (high animal protein and fat) increases sodium and long-chain saturated fat intakes, elevating LDL-cholesterol [15, 16] and potentially increasing cardiovascular disease risk [78,79,80]. High protein intake has been associated with kidney diseases in several observational studies [81]. Metabolic ketoacidosis with ketogenic diets is a hazard, particularly with sodium−glucose cotransporter 2 (SGLT2) inhibitors [82,83,84,85,86,87]. Meanwhile, extreme fat avoidance provokes cholelithiasis [88].

Remission of type 2 diabetes

Current evidence on diets for type 2 diabetes remission is more limited. Only two RCTs had remission as the pre-specified outcome, both relatively large and using almost identical designs and diets, with very similar results, but in very different populations, notably with different durations of diabetes [51, 53, 89]. A large majority can achieve remission if they maintain sufficient weight loss.

NRSs (non-RCT, single-arm intervention) reported remission rates ranging from 3% to 75% by ITT, over various follow-up durations. The highest remission rates, up to 75%, were in people with newly diagnosed diabetes or with <2 years of type 2 diabetes duration. Much lower 20–22% remissions were reported with longer type 2 diabetes duration (8 years) or a very brief diet period (1 week). However, these studies did not all fully ascertain remission status, and they had critical risks of bias due to lack of comparator groups and/or randomisation. NRSs reflect performance among those who select and can adhere to a particular diet, and so usually reported better results than those featuring random assignment. In some cases, remission rates were reported for completers only, rather than using the ITT population to properly guide healthcare practice and policy. Despite extracting baseline data for ITT analysis, residual bias/confounding may remain with these study designs.

The main contributor to HbA1c reduction and remission appears to be weight loss, irrespective of diet type. From the high-quality studies with high GRADE certainty, structured programmes with an intensive induction phase with total diet replacement were effective. Remission of diabetes occurs when a patient no longer satisfies the diagnostic criteria, without receiving glucose-lowering medication. To ascertain remission for those already prescribed glucose-lowering drugs, a therapeutic trial of withdrawing medication is necessary, with an appropriate protocol for re-introduction if necessary. Confirmation over a defined duration (e.g., 6 or 12 months) will be required for re-classifying individuals, and for legal or insurance purposes. The diagnostic HbA1c cut-off for diabetes of 48 mmol/mol (6.5%) was defined by WHO as broadly the level where diabetes-specific microvascular complications start to emerge [90, 91]. However, many people in remission from type 2 diabetes remain in the pre-diabetes range of HbA1c, where cardiovascular disease risk begins to rise [92, 93]. Lowering HbA1c to very low levels with multiple medications among people with longstanding disease is associated with increased mortality rate, possibly by relative hypoglycaemia provoking arrythmias [94]. No such concerns have been reported in the small numbers who achieved and sustained HbA1c < 42 mmol/mol (<6.0%) from diet restriction [89].

Most type 2 diabetes is treated in primary care, the setting for both published remission trials using an intensive ‘total diet replacement’ induction phase with formula diets [51]. Simpler food-based programmes may be effective. A service evaluation from one UK general practice reported weight loss and remission in 59 out of 128 patients who opted for, and persisted with, LCD advice for a mean 23 months [95]. This completers’ analysis omits information about numbers who declined the diet, who started but failed to persist and who did not provide outcome data at designated times. The LCD was routinely offered since 2013, and the total number of patients with type 2 diabetes was 473 at the time of evaluation, so these data imply that 12.5% of the practice achieved remission [95]. A population-based cohort study from 49 general practices in the UK (the Anglo–Danish–Dutch Study of Intensive Treatment in People with Screen-Detected Diabetes in Primary Care [ADDITION-Cambridge]) included 867 participants with newly diagnosed type 2 diabetes; after being followed up for 5 years, there was an overall 30% remission (n = 257/867; ITT analysis). Loss of >10% of baseline body weight in the first year after diagnosis was associated with 70% higher chance of remission at 5 years [96]. Every 1 kg of weight loss was associated with 7% higher chance of remission at 5 years, regardless of specific diet regimens or lifestyle interventions [96]. There is therefore consistent evidence that remission should be attempted as early as possible from diabetes diagnosis [70, 96].

Limitations

AMSTAR 2 assesses the quality of meta-analyses, prioritising critical domains, where errors and bias can impact pooled findings (ESM Table 2). Only one or two flaws can label a meta-analysis ‘low’ or ‘critically low’, with some criteria potentially subjective (e.g., adequacy of the literature search; ESM Table 2). In the umbrella review, many meta-analyses were of ‘low’ and ‘critically low’ AMSTAR 2 quality, predominantly through ‘no protocol reported’ (despite clear and sound methods) and no assessment of publication bias. Many meta-analyses had fewer than ten RCTs to permit assessment of publication bias by funnel plot [97]. If AMSTAR 2 criteria are relaxed for protocol reporting and publication bias, the meta-analyses allow some confidence in the consistent findings of little/no difference in weight loss between any diets.

Although the search strategy was wide and not language-restricted, most studies included European participants; results may not be equally applicable to other ethnic and/or deprived communities. South Asians develop type 2 diabetes at younger ages, more rapidly and with lower BMI, so may be more sensitive to weight loss, with physiological differences in insulin resistance, body composition and fat oxidation [98, 99].

The criteria used in the reported meta-analyses and studies focused on specific diet types. However, not all reported sufficient detail about macronutrient or micronutrient contents, or prescribed and reported energy intakes, including energy intake of nutrient-restricted ad libitum diets, which limits interpretation and transferability of results. Control diets used in the meta-analyses and source RCTs also varied, including ‘usual’ diets in different countries, as well as specified dietary regimens (Table 1). Despite this, differences in weight loss between intervention and control diets, commonly 0–2 kg, are of little clinical significance. Durations of interventions varied: as weight regain is frequent over a longer period, heterogeneity might be expected. However, duration did not introduce heterogeneity, probably because trials with longer follow-up tended to be evaluating more intensive interventions with greater initial weight loss, such that the net weight changes at endpoint are similar to short-term trials.

Given the extent of literature concluding that differences in weight control or HbA1c from different diet compositions are not clinically significant, future trials of similar diet comparisons are unlikely to add useful information. Instead, evidence from clinical practice is needed to identify safe and effective approaches to achieve and maintain weight loss with available skills and training, to assess long-term outcomes from high-quality trials and prospective audits of practice with different diets. Interpretating the existing data might be enhanced through individual patient data meta-analysis. Alternatively, the very large amount of work entailed in conducting repeated meta-analyses, and the limitations of different inclusion criteria and detailed methods, support a prospective meta-analysis approach [100, 101]. All primary studies for inclusion should use an RCT design, with data analyses conducted ‘blind’. They should define the intervention clearly (e.g., diets, physical activity, and behavioural and psychological support), and address separately the induction (usually 3–6 months) and maintenance (≥12 months) phases of weight management, potentially employing different methods within a treatment programme.