Therapeutic Use of Metformin in Prediabetes and Diabetes Prevention

The defects in glucose metabolism that underlie type 2 diabetes begin many years before the diagnosis of diabetes is made [6, 7]. The development of insulin resistance, in which the action of insulin on glucose metabolism is blunted, occurs early in the pathogenesis of dysglycaemia. Increased secretion of insulin initially compensates for the presence of insulin resistance; however, a simultaneous and progressive loss of β-cell mass and β-cell function limits the ability of the pancreas to maintain euglycaemia by increasing insulin secretion [8].

The early manifestations of prediabetic dysglycaemia represent one or both of [9]:

Table 1 shows the usually accepted diagnostic criteria for the diagnosis of prediabetes based on measurements of glycaemia [4, 8‐10]. A simple blood test is sufficient to diagnose IFG, while a 75 g oral glucose tolerance test (OGTT) is required for the diagnosis of IGT. The originally used cutoff level for IFG (110 mg/dL [6.1 mmol/L]) was reduced to that shown in Table 1 (100 mg/dL [5.6 mmol/L]) by an Expert Committee of the American Diabetes Association in 2003, in order to equalise the prognostic impact of diagnosis of IFG or IGT, in terms of the future risk of diabetes in a subject with either condition [11]. It should be noted that the World Health Organization (WHO) diagnostic criteria for IFG retains the 110 mg/dL (6.1 mmol/L) cutoff value for diagnosing IFG [12].

Diagnostic criteria from the American Diabetes Association also support a diagnosis of prediabetes (but not a specific diagnosis of IFG or IGT) in subjects with elevated, but non-diabetic, levels of HbA_1c (5.7–6.4 % [39–46 mmol/mol]), as this approach identifies a group at high risk of diabetes and other adverse outcomes who may benefit from counselling to identify ways to mitigate these risks [4]. A study from the US National Health and Nutrition Examination Survey (NHANES), published in 2010, found that the use of fasting glucose identified twice as many subjects as having prediabetes compared with the use of HbA1c (28 vs. 13 %), while only 8 % satisfied both criteria [13]; see below for a fuller discussion of the prevalence of prediabetes. Moreover, HbA1c does not correlate accurately with glucose levels in individuals and may therefore over- or underestimate the current level of glycaemic control [14]. Accordingly, while HbA1c is a convenient measurement for evaluating long-term glycaemic control in diabetes, the impact of its use as a diagnostic criterion for prediabetes will require further study.

The loss of glycaemic control as insulin resistance and β-cell dysfunction develops is insidious, progressive and continuous [6, 7]. Indeed, clinically normoglycaemic subjects may already have lost at least half of their β-cell function by the time they progress to IGT, and subjects with IGT and post-OGTT glucose near the cutoff for diagnosis of type 2 diabetes will have lost at least 80 % of their β-cell function [15].

The prevalence of both type 2 diabetes and prediabetes has increased rapidly in recent years. A pooled analysis included data from non-diabetic subjects in three population-based observational cohorts in the US [16]: the Screening for Impaired Glucose Tolerance (SIGT) study enrolled 1581 subjects, and other subjects were from the nationally representative NHANES III (N = 2014) or NHANES 2005–2006 (N = 1111) cohorts. For consistency between cohorts, analyses were restricted to subjects of White/Caucasian or non-Hispanic Black heritage, with lower cutoffs for age ≥18 years for SIGT or NHANES 2005–2006, and ≥40 years for NHANES III (OGTTs were performed for this age group only). Overall, more than 30 % of the population demonstrated one or both forms of prediabetic dysglycaemia (Fig. 1). Similar data are available from the Augsburg (Germany) cohort of the large, international Monitoring Trends and Determinants on Cardiovascular Diseases (MONICA) study, conducted in subjects aged 25–74 years. Among a control group matched for age and gender with a parallel diabetic cohort, 23 % had IGT and 36 % had IFG [17]. According to the International Diabetes Federation, 8 % of the world’s population, comprising approximately 470 million individuals, will have IGT by the year 2035 [18].

The Diabetes Prevention Program (DPP), which recruited a population of subjects at elevated risk of diabetes due to IGT plus high–normal FPG, demonstrated that approximately 10 % of subjects progressed to clinical diabetes each year over an average of 2.8 years of follow-up in the study [19]. In general, some 70 % of individuals with IFG and/or IGT can expect to go on to develop clinical type 2 diabetes at some time in the future [8]. People diagnosed with combined IFG and IGT are at increased risk of developing diabetes compared with individuals with only one of these conditions. For example, a pooled analysis of two diabetes prevention studies in India showed that the 3-year incidence of type 2 diabetes in subjects with both IGT and IFG was 56 % compared with 34 % for subjects with isolated IGT [20]. Conversely, only 18 % of subjects with combined IFG and IGT at baseline reverted to normal glucose tolerance compared with 32 % with isolated IGT [20].

Even modest, long-term elevations of plasma glucose, consistent with those associated with IGT or IFG have been associated with damage to the vasculature [21‐26]. Retinopathy was diagnosed in 7.9 % of subjects in the DPP who did not progress to type 2 diabetes (compared with 12.6 % of those who did) [27]. In the German control cohort from the MONICA study described above, polyneuropathy was already present in 13 % of subjects with IGT and 11 % with IFG compared with 28 % for subjects with a diagnosis of diabetes and 7 % with normal glucose tolerance [17].

Prediabetic dysglycaemia also increases the risk of adverse cardiovascular events (such as myocardial infarction, stroke or cardiovascular death; see reviews elsewhere [28, 29]). In addition, as an insulin resistant state, prediabetes frequently coexists with other cardiovascular risk factors associated with the metabolic syndrome, such as elevated blood pressure and dyslipidaemia [30]. Insulin resistance, which appears early in the pathogenesis of dysglycaemia and is likely to be present in most prediabetic subjects, is itself a powerful prognostic indicator for an increased risk of diabetes or cardiovascular disease [6‐8].

Intervention in prediabetes is known to prevent or delay the onset of diabetes, a state known to be associated with markedly elevated cardiovascular risk [31]. The appearance in the circulation of the cardiac-specific isozyme of troponin is indicative of damage to the myocardium, and is associated with an adverse long-term prognosis [32]. A recent study has demonstrated elevated circulating cardiac troponin T in 11 % of diabetic subjects, 6 % of subjects with IGT and 4 % of subjects who remained normoglycaemic over 6 years of follow-up [33]. Accordingly, the presence of prediabetes may be associated with subclinical damage to the myocardium.

It is clear that the increased risk of adverse cardiovascular outcomes associated with type 2 diabetes does not begin at the diagnostic cutoff for plasma glucose (or HbA1c) at which the condition is diagnosed. Rather, there appears to be a continuum of increased microvascular and macrovasular risk that extends to levels of glycaemia well below these cutoffs. While it seems reasonable to hypothesise that correction of prediabetic dysglycaemia might also reduce the future risk of adverse cardiovascular outcomes, further evidence from clinical trials is needed to demonstrate improved long-term outcomes in this setting [31].

Prediabetes (IGT and IFG) and clinically established type 2 diabetes are each characterised by insulin resistance and β-cell dysfunction, and represent a continuum of increasing severity of dysglycaemia, as described above. Accordingly, the overall principles of managing diabetes and prediabetes are similar [4]. People at risk of diabetes, particularly those who are overweight or obese, or women with a history of gestational diabetes, should be tested for the presence of prediabetes or diabetes alongside other cardiovascular risk factors.

Lifestyle intervention remains the cornerstone of care for individuals with prediabetes or diabetes, based on an improved diet and regular moderate physical exercise with the aim of achieving weight loss in overweight or obese subjects (Table 2) [4, 34]. Pharmacologic therapy with interventions used to promote weight loss (e.g. orlistat or bariatric surgery) or with drugs usually used for the management of type 2 diabetes (metformin, thiazolidinediones, α-glucosidase inhibitors or basal insulin) have also been shown to effectively delay or prevent the conversion of prediabetes to diabetes (described below).

Currently, metformin is the only pharmacologic agent recommended for the prevention or delay of type 2 diabetes in at-risk subjects due to its effectiveness as demonstrated in well-designed trials (see below), its generally good tolerability (aside from the well-understood gastrointestinal side effects associated with this agent) and its low cost [4, 34]. At present, metformin has no formal indication for this purpose in most countries (Turkey, Poland, and The Philippines are exceptions), although such an indication may become established in many countries in the future. The remainder of this review will focus in detail on the therapeutic profile of metformin for the management of prediabetes and other insulin-resistant states that predispose to the subsequent development of type 2 diabetes.

A brief summary of the pharmacologic mechanism and clinical actions of metformin is provided in this review as these are relevant to its therapeutic effects on prediabetic subjects. Preventing or reversing the progressive insulin resistance and/or β-cell dysfunction associated with dysglycaemia holds the key to preventing or delaying the conversion of prediabetes to clinical type 2 diabetes.

Metformin acts primarily by enhancing the action of insulin in the liver to reduce the rate of hepatic glucose production [35]. Improvements in insulin action in skeletal muscle also contribute to the therapeutic actions of metformin, mainly resulting in increased non-oxidative glucose disposal [36]. Together, these actions reduce blood glucose in the setting of hyperglycaemia, with very little potential for inducing hypoglycaemia [37].

An increase in anaerobic metabolism in the intestinal wall is also probably a clinically significant antihyperglycaemic mechanism of metformin [38, 39]. In addition, metformin has been shown to increase circulating levels of glucagon-like peptide-1 (GLP-1) by increasing the secretion of GLP-1 itself and/or by decreasing the activity of dipeptidyl peptidase-4 (DPP4), the enzyme principally responsible for inactivating GLP-1 in the tissues and circulation [40‐42]. Metformin may also induce upregulation of the expression of GLP-1 receptors on the surface of pancreatic β-cells [41]. As GLP-1 enhances glucose-dependent insulin release from the pancreas, this mechanism may provide modest support to the function of the β cell [40, 43]. An effect of metformin on the gut microbiome has also been postulated [44].

Mechanistically, metformin appears to inhibit mitochondrial respiration at the level of respiratory chain complex I [45]. The resulting shift in cellular energy balance increases the activity of AMP kinase, which promotes the action of insulin and reduces hepatic gluconeogenesis [45]. An increase in circulating cyclic adenosine monophosphate (cAMP) also opposes the hyperglycaemic action of glucagon [45, 46]. Other studies have shown that metformin enhances the action of DPP4 inhibitors by either reducing the activity of DPP4 or enhancing secretion of GLP-1 [47, 48]. Metformin relies on transport into cells via the organic cation transporter-1 (OCT1) for its clinical action, and polymorphisms of this transporter influence the efficacy of metformin in type 2 diabetes [49, 50]. The relevance of this mechanism for diabetes prevention has yet to be determined and further research is required.

In the UK Prospective Diabetes Study (UKPDS), metformin was the first antidiabetic drug shown to improve cardiovascular prognosis [51]; clinically significant reductions in the risk of a range of adverse outcomes were observed, relative to the standard diet-based treatment of the time, including all-cause mortality, diabetes-related death and myocardial infarction, which were greater than those expected from improved blood glucose control per se. Long-term epidemiologic follow-up of this trial beyond the randomised phase demonstrated ‘legacy benefits’, in which cardiovascular benefits persisted in patients previously randomised to metformin despite the cessation of randomised treatment and rapid equalisation of mean HbA_1c between patients previously in the metformin and control arms of the study [52]. A second trial in insulin-treated patients also demonstrated improved cardiovascular outcomes in patients randomised to metformin relative to placebo [53]. Improved cardiovascular outcomes with metformin is included in the European labelling of this drug. The forthcoming Glucose Lowering In Non-diabetic hyperglycaemia Trial (GLINT) trial is expected to definitively demonstrate the extent to which metformin protects the vasculature in a prediabetic population at high risk of adverse cardiovascular events [54].

Improvements in glycaemia during treatment with metformin were insufficient to explain the improved cardiovascular outcomes observed in the UKPDS [52]. A recent uncontrolled study in 390 insulin-treated type 2 diabetes patients showed that long-term treatment with metformin (average 4.3 years) reduced levels of a range of circulating markers of endothelial dysfunction; these observations were consistent with a protective effect on the vasculature, as endothelial dysfunction is an early marker of atherosclerosis [55]. Numerous other mechanisms have been proposed to explain the protective effect of metformin on the vasculature in the UKPDS, including improved haemostasis (reduced potential for atherothrombotic disease), reduced vascular inflammation, amelioration of oxidative stress, inhibition of the formation of advanced glycation end-products, improved function of the microcirculation and modification of the cellular processes that occur during atherogenesis [56].

The principal side effects of metformin occur in the gastrointestinal tract (mostly diarrhoea); these can be minimised by starting metformin at a low dose and increasing the dose cautiously and infrequently cause treatment discontinuation [57]. Prolonged-release formulations of metformin are available, which appear to improve gastrointestinal tolerability compared with the immediate-release formulation [58]. Biguanide antidiabetic agents have long been associated with an increased risk of lactic acidosis but it is now clear that the risk of lactic acidosis with metformin is extremely low when this agent is prescribed correctly [59, 60]. Contraindications to metformin intended to reduce the risk of lactic acidosis, as described in its labelling, generally reflect cardiovascular morbidity and renal dysfunction that might provoke accumulation of metformin in the body—these conditions are less likely to be prevalent in a prediabetic population compared with a population with established type 2 diabetes at higher risk of long-term complications of the disease.

Treatment with metformin has also been associated with clinically significant vitamin B₁₂ deficiency in some patients, where neuropathy arising from low levels of B₁₂ may mimic diabetic neuropathy [61]. A 4.3-year study demonstrated an increase in the risk of clinical B₁₂ deficiency (<150 pmol/L) of 7 %, with an average reduction in B₁₂ levels of 19 %, each compared with placebo [62]. A meta-analysis of 29 studies that included a total of 8089 patients found that metformin increased the risk of B₁₂ deficiency, with an odds ratio (OR) of 2.45 versus comparators [63]. Monitoring B₁₂ levels, with supplementation where necessary, may be useful in all subjects receiving long-term treatment with metformin.

The principal features and main results of major trials that evaluated metformin for diabetes prevention are shown in Table 3 [19, 64‐69]. Significant reductions in the risk of progressing from prediabetes (principally IGT) to type 2 diabetes in subjects treated with metformin were observed in populations in:

These trials will be described in greater detail in the following sections. Details of trials that did not evaluate metformin are also shown in Table 3, for comparison [70‐78]. Lifestyle intervention was consistently effective in reducing the risk of diabetes and should be recommended for all subjects at risk of diabetes or cardiovascular disease, irrespective of other therapies prescribed. It is also clear that thiazolidinediones, α-glucosidase inhibitors and weight-reducing interventions also have the potential to prevent or delay the onset of diabetes in prediabetic subjects. A post hoc analysis of the STOP-NIDDM (Study to Prevent Non-Insulin-Dependent Diabetes Mellitus) trial [72] suggested a reduced incidence of hypertension and cardiovascular disease for acarbose versus placebo in subjects with IGT, although the number of events was low [73].

The effects of metformin on diabetes outcomes in the DPP, and also on cardiometabolic endpoints relevant to diabetes prevention, are discussed in detail here. Health economic analyses are discussed separately in a later section, alongside comparable analyses from other trials.

The IDDP differed from the DPP in that an intensive lifestyle intervention and metformin were evaluated separately and in combination, each in comparison with a control group given standard healthcare advice [65]. For this trial, subjects involved in physical labour or regular exercise, or who needed to walk or cycle for more than 30 min each day, were exempt from additional exercise targets; other subjects in this group were advised to walk briskly for at least 30 min/day. Metformin was initially given at a dose of 250 mg, titrated to 250 mg twice daily after 2 weeks, according to tolerability. The dose was lower than that used in the DPP to account for the smaller average body size of South Asian versus American subjects.

Similar reductions in the risk of diabetes were observed in active treatment groups, with no sign of synergy between the lifestyle intervention and metformin (Table 3). Numbers needed to treat ranged between 6.4 and 6.9 to prevent one case of incident diabetes. There were no significant changes in body weight at 3 years in any group. The proportions of patients with elevated LDL-cholesterol decreased in all active treatment groups but not in the control group, while the prevalence of hypertension increased significantly in all groups, irrespective of treatment [120].

A cutoff value for HbA_1c of 43 mmol/mol (6.05 %) predicted new-onset diabetes with 67 % sensitivity and 60 % specificity in the IDDP [121]. As in the DPP, subjects with more severe insulin resistance or β-cell dysfunction at baseline were more likely to develop diabetes, and diabetes prevention was associated with improvements in these parameters [122]. However, although insulin resistance (HOMA-IR) increased in line with the number of criteria for the metabolic syndrome, the presence or absence of metabolic syndrome criteria per se did not influence the risk of developing diabetes (HR 1.02; 95 % CI 0.78–1.35; p = 0.88) [123].

The low-dose combination therapy with rosiglitazone and metformin to prevent type 2 diabetes mellitus study (CANOE) sought to avoid the common adverse events associated with rosiglitazone (oedema with increased risk of incident heart failure) and metformin (gastrointestinal side effects, principally diarrhoea) by using rosiglitazone and metformin in a low-dose combination in a 4-year, randomised, placebo-controlled study in subjects with IGT [64]. The reduction in the risk of new-onset type 2 diabetes in the CANOE study was at least comparable to the risk reductions observed in the other studies using metformin, and was also comparable to the risk reductions observed with intensive lifestyle interventions or rosiglitazone in other studies (Table 3). However, the combination treatment did not alter the rate of progression of decreases in insulin sensitivity and β-cell function [124].

Current prescribing restrictions relating to safety concerns with thiazolidinediones [125] potentially limit any future role for thiazolidinedione-based combinations in diabetes prevention. However, the CANOE study established an important principle in that effective diabetes prevention can be achieved using low-dose combinations of drugs that minimise the potential for side effects. Another study found that targeted combinations of metformin with pioglitazone or exenatide (based on results from oral glucose tolerance tests) induced more marked improvements in indices of glycaemic function than lifestyle intervention in 105 subjects with IGT and/or IFG, although diabetes outcomes were not measured here [126].

A meta-analysis of three diabetes prevention studies [19, 65, 66] presented data in two ways: (a) as presented by the original authors; and (b) including all subjects enrolled in each study as a true intention-to-treat analysis with a ‘worst-case’ scenario, where subjects lost to follow-up on metformin or control treatment were assumed to have progressed/not progressed to diabetes, respectively [127]. Metformin significantly prevented diabetes when analysed in either of these ways, and also when only considering placebo-controlled data or the use of low doses of metformin in Asian subjects (Fig. 4). A further systematic review, which did not attempt a meta-analysis, concluded that while there is substantial evidence that lifestyle or pharmacologic interventions effectively delay or prevent diabetes, there was insufficient evidence to compare their effectiveness [128].

A 12-month observational study in 366 overweight/obese subjects in Greece (all received a lifestyle intervention and 95 received metformin) demonstrated a reduced frequency of diabetes in the metformin-treated cohort (absolute risk difference 7 %; p = 0.012) [129]. Larger effects were observed in subjects with prediabetic dysglycaemia (absolute risk difference 18.5 %; p = 0.01) or the metabolic syndrome (absolute risk difference 12.9 %; p = 0.04) at baseline. The reduction in diabetes incidence was associated with reduced FPG, total cholesterol and LDL-cholesterol, and increased HDL-cholesterol.

Further studies, described below, did not include new-onset diabetes in their endpoints but evaluated the effect of metformin on parameters relevant to diabetes prevention (primarily insulin resistance) in non-diabetic populations. The IDDP [53] and the studies by Wenying et al. [68] and Li et al. [66] did not demonstrate an additional effect of combining a lifestyle intervention with metformin, relative to that of metformin alone (see Table 3). A study in 32 subjects with IGT showed that exercise training, metformin, and exercise training with metformin all increased insulin sensitivity relative to a placebo control, with no significant difference between active treatment groups [130]. An observational study in Thailand showed that the proportion of subjects with IGT who reverted to normal glucose tolerance was similar for subjects receiving lifestyle intervention alone (83 %) or with metformin (85 %) [131]. A post hoc analysis of the 1-year, randomised Biguanides and the Prevention of the Risk of Obesity (BIGPRO1) study demonstrated significant improvements for metformin vs. placebo in FPG, systolic BP, and total and LDL-cholesterol in abdominally obese subjects with IGT or IFG [132]. Similar benefits of metformin were observed in subjects who met the inclusion criteria for the DPP in this analysis. Randomisation of 40 participants in the Botnia cohort in Sweden, who had IGT and a first-degree relative with type 2 diabetes, resulted in improved glucose homeostasis that was sustained over 12 months [133]. In addition, little information is available on optimal dietary management for use alongside metformin: the RESIST (Researching Effective Strategies to Improve Insulin Sensitivity in Children and Teenagers) study is currently comparing two diets in obese, metformin-treated children and adolescents with IFG, IGT or elevated insulin resistance [134].

Again, the studies described here did not include new-onset diabetes as an endpoint, and effects on clinical parameters relevant to diabetes prevention are described here. Table 5 summarises details of clinical evaluations of metformin in paediatric patients without diabetes at baseline [135‐141]. A 6-month crossover study compared the effects of metformin 2000 mg/day and placebo in 28 obese, insulin-resistant young adolescents (mean age 12.5 years) [135]. Metformin was associated with significant reductions versus placebo in mean weight (treatment difference –4.35 kg; p = 0.02), BMI (treatment difference −1.26 kg/m²; p = 0.002), waist (treatment difference −2.8 cm; p = 0.003) and fasting plasma insulin (treatment difference −2.2 µU/L; p = 0.011). Significant improvements also occurred in the metformin group in subcutaneous abdominal fat measured using CT scanning. A second 6-month study randomised 29 obese hyperinsulinemic adolescents to placebo or metformin 1000 mg/day [136]. Significant improvements were observed in the metformin group for BMI (p < 0.02 vs. placebo), while improvements from baseline in plasma insulin and FPG occurred during metformin treatment (p < 0.02) but not during treatment with placebo. Another study randomised 24 obese, hyperinsulinemic adolescents to receive placebo or metformin 1700 mg/day for 8 weeks [137]. Metformin markedly reduced body weight (by 6.5 vs. 3.8 % on placebo; p < 0.01), and improved indices of body-fat content and insulin resistance versus placebo.

The Metformin in Obese Children and Adolescents trial in the UK randomised 151 obese children or adolescents to metformin or placebo for 6 months [138]. Significant improvements in the metformin group were observed for BMI (the primary endpoint) at 6 months, with a mean change of −0.1 SD (95 % CI −0.18 to −0.02; p = 0.02), and at 3 months for FPG (−0.16 mmol/L; 95 % CI −0.31 to −0.00; p = 0.047); alanine aminotransferase levels (−19 %; 95 % CI −5 to −36 %; p = 0.008) and adiponectin:leptin ratio (32 %; 95 % CI 4–67; p = 0.02). Another randomised, placebo-controlled study in 52 glucose-intolerant paediatric subjects showed that 12 weeks of treatment with metformin 850 mg twice daily significantly reduced levels of resistin (a hormone associated with insulin resistance), HOMA-IR and HbA1c [139]. An uncontrolled, observational study in children in China reported similar results in that the clinical features of the metabolic syndrome, HOMA-IR and adiponectin levels were improved in 20/30 subjects who completed 3 months of treatment with metformin and a lifestyle intervention [140]. Finally, a meta-analysis of four randomised studies of at least 2 months’ duration in children or adolescents showed that treatment with metformin versus placebo decreased fasting insulin (mean change −9.6 µU/mL; 95 % CI 6.3–13.0), HOMA insulin resistance (mean change −2.7; 95 % CI 1.7–3.6) and BMI (mean change −1.7 kg/m²; 95 % CI 1.1–2.3) [141].

Obesity and insulin resistance are recognised side effects of some antipsychotic drugs, and 37 % of a large cohort of 783 adult psychiatric inpatients receiving this treatment were recently found to have prediabetic dysglycaemia [142]. A number of randomised, controlled trials have demonstrated benefits in terms of improved glycaemic regulation and weight loss associated with coadministration of metformin with antipsychotic gents [143‐150]. In one 16-week, double-blind study in 39 children and adolescents who gained at least 10 % of initial body weight within 1 year of starting treatment with an atypical antipsychotic agent, randomisation to metformin was associated with weight stabilisation, while patients receiving placebo continued to gain weight at a rate of 0.31 kg/week [150].

Meta-analyses found a weight benefit for metformin versus placebo of 5 kg [151], 3.2 kg [152] or 4.8 % [153] of initial weight. One meta-analysis found a non-significant reduction in the risk of type 2 diabetes (relative risk 0.30; p = 0.13), although the studies included were not designed as outcomes trials [151]. Three studies showed no benefit for metformin on body weight in this setting [154‐156].