Introduction

The Diabetes Prevention Program (DPP; 1996–2001), an RCT to prevent or delay the onset of type 2 diabetes, was designed in the mid 1990s. Metformin was selected as one of the interventions, based on its mechanism of action and acceptable safety and tolerability profiles, with lifestyle intervention or placebo comprising the other treatment arms [1, 2]. The possibility of preventing or delaying diabetes in adults without diabetes but at high risk had been hypothesised for decades. Small randomised clinical trials using type 2 diabetes treatment drugs (phenformin or tolbutamide) for diabetes prevention were performed in the 1960s/70s, but were inconclusive [3,4,5]. They were followed by larger clinical trials testing lifestyle interventions that proved to be effective [6, 7]. The DPP was the first major diabetes prevention trial using metformin [2].

The DPP/Diabetes Prevention Program Outcomes Study (DPPOS) represents the largest controlled clinical trial of metformin in a population at high-risk of developing diabetes, and also the longest trial of metformin for any indication. The effects of intensive lifestyle (ILS) intervention in the DPP and several other major trials, and the effects of other medications have been described elsewhere and are summarised in Table 1 [2, 8,9,10]. In this review, we focus on the effects of metformin on diabetes prevention, its long-term glycaemic and cardiometabolic effects, and its safety in the DPP/DPPOS.

Table 1 Summary of select RCTs evaluating the prevention of progression to diabetes
Full size table

Overview of the DPP/DPPOS

Design of the DPP/DPPOS

The DPP enrolled 3234 participants aged 25 years or older who were at high risk of developing diabetes, defined as impaired glucose tolerance, with elevated fasting plasma glucose (FPG) (5.3–6.9 mmol/l [≤6.9 mmol/l in Native Americans]) and a BMI of 24 kg/m2 or higher (≥22 kg/m2 in Asian-Americans). Participants were randomly assigned to placebo (n = 1082), metformin (n = 1073) titrated to 850 mg twice daily, or ILS intervention (n = 1079), which aimed for 7% weight loss through a low-energy, low-fat diet (based on recommendations for health) and ≥150 min/week of moderate-intensity physical activity [2]. Interventions were discontinued if there were safety concerns. Diagnosis of diabetes was based on annual OGTTs or semi-annual FPG tests, using the ADA diagnostic criteria, with the diagnosis requiring confirmation with repeat testing [11]. Diagnosis of diabetes and FPG ≥7.8 mmol/l resulted in discontinuation of study medication and referral to the participant’s own physician for further treatment [2].

The DPP was stopped in 2001, 1 year ahead of schedule, owing to demonstrated efficacy of both metformin and the lifestyle intervention [2]. Given the demonstrated effects of ILS, all participants were offered a group-administered version of the lifestyle curriculum at the end of the DPP. Eighty-eight per cent (n = 2776) of eligible DPP participants continued follow-up in the DPPOS, in which placebo was discontinued, those previously assigned to metformin received metformin 850 mg twice daily (now unmasked) and lifestyle messages were intermittently reinforced. The study-provided metformin was discontinued if diabetes was diagnosed and HbA1c was ≥7% (≥53 mmol/mol), hence requiring management by the participant’s physician [8]. Outcomes in the DPPOS from 2002 to 2013 centred on the long-term effects of the interventions on diabetes prevention, diabetes-associated microvascular complications [9] and cardiovascular disease (CVD) risk factors.

Participant characteristics

By intention, the DPP enrolled a heterogeneous population, with 45% from racial or ethnic minorities, 20% aged 60 years or older and 68% women, including 350 women with a history of gestational diabetes (GDM). The mean age at randomisation was 51 years and mean BMI was 34 kg/m2. Mean FPG was 5.9 mmol/l and baseline HbA1c was 5.9% (41 mmol/mol). Sex, ethnic distribution and risk factors for diabetes were similar among the randomised treatment arms [2].

Exposure to metformin

Integral to understanding the effects of metformin over time is the separation of study and non-study exposure, keeping in mind that DPP/DPPOS participants who developed diabetes were subsequently managed by their own physicians, often with non-study metformin. By 15 years after randomisation, 37% of the original placebo participants had been treated with metformin by their healthcare providers, the vast majority associated with diabetes diagnosis (Fig. 1). The mean exposure, including study- and non-study metformin, from 1996 to 2013, remained widely separated, at 10.7 vs 2.3 metformin-years in metformin vs placebo groups [9].

Fig. 1
figure 1

Exposure to metformin throughout the DPP/DPPOS. AUCs represent total metformin-years of exposure, including study-provided metformin (blue), non-study-provided metformin for diabetes treatment (green), and non-study-provided metformin prescribed to individuals without diabetes (red)

Full size image

Throughout the DPP and DPPOS, pill counts and structured interviews were used to promote adherence [12]. During the DPP, adherence to metformin, defined as taking at least 80% of assigned study pills, was 72% (Fig. 1) [2]. An additional 10–15% of participants took some metformin, albeit at less than 80% of pills assigned. Adherence to metformin (at the >80% threshold) fell to an average of 49% over the DPPOS (2002–2013) [9].

Results

Effects of metformin on diabetes prevention

In 2002, the DPP published its primary findings from the masked-treatment phase, showing that the ILS and metformin groups had a respective 58% and 31% lower incidence of diabetes than the placebo group [2]. Subsequently, the DPPOS addressed the longer-term effects of metformin, showing a decline in risk reduction by 18% compared with placebo over 10 and 15 years post-randomisation (Fig. 2a) [8, 9]. Although the differences in incidence rates over the entire follow-up remained significant, the observed diabetes incidence rates during the DPPOS period (i.e. in the period after the DPP completed) were not significantly different between the original randomised groups. Diabetes incidence rates during the DPP were 7.8 cases per 100 person-years in the metformin group and 11.0 cases per 100 person-years in the placebo group [2], and these decreased in the DPPOS (2002–2008) to 4.9 cases per 100 person-years for metformin and 5.6 cases per 100 person-years for placebo [8], remaining stable thereafter. This reduced diabetes incidence approximates the five cases per 100 person-years rate observed in the lifestyle group during the DPP, which has remained nearly constant throughout the DPP/DPPOS [9]. The average genetic risk score, derived from 34 type 2 diabetes-associated genetic variants, declined over time among participants who remained without diabetes in the DPP/DPPOS, in both the metformin and placebo groups [13]. This suggests that the lower annual incidence rate of diabetes seen in the DPPOS was not entirely due to an effect of the lifestyle intervention offered during the transition to the DPPOS, but, in part, due to ‘exhaustion of susceptibles’, or that diabetes developed in the people who were most susceptible to diabetes during the DPP and that remaining participants in the DPPOS were less susceptible to diabetes [13].

Fig. 2
figure 2

(a) Cumulative incidence of diabetes and (b) weight change over 15 years in the DPP/DPPOS, in metformin (blue line) and placebo (red line) groups. (a) Adapted from The Lancet Diabetes & Endocrinology 3:866–875 [9]; Diabetes Prevention Program Research Group, Long-term effects of lifestyle intervention or metformin on diabetes development and microvascular complications over 15-year follow-up: the Diabetes Prevention Program Outcomes Study, Copyright (2015), with permission from Elsevier

Full size image

Effects of metformin on diabetes prevention/delay in subgroups of interest

The DPP was not powered to assess the significance of effects within subgroups. Nonetheless, examination of treatment effects in cohort subgroups revealed significant heterogeneity. For example, obese participants with a BMI ≥35 kg/m2 were more responsive to metformin than to placebo, with a 53% risk reduction for diabetes but only a 3% reduction in those with BMI 22 to <30 kg/m2. In addition, those with a higher fasting glucose (6.1–6.9 mmol/l) had a greater risk reduction with metformin (48%) compared with those with a fasting glucose of 5.3–6.1 mmol/l (15% risk reduction). Although not significant for heterogeneity across strata, metformin appeared more effective in younger participants compared with placebo, reducing diabetes onset by 44% (95% CI 21%, 60%) in those 25–44 years old vs 11% (95% CI −33%, 41%) for those ≥60 years of age at study entry. Of note, no such differences were observed by sex, race/ethnicity, or tertiles of baseline 2-h plasma glucose (2-hPG) [2].

During the DPP, women with a history of GDM randomised to placebo had a 71% higher risk of diabetes than parous women without such a history, despite similar FPG and 2-hPG values at baseline [14]. Significant heterogeneity was observed in response to metformin with a 50% reduction in incidence of diabetes in women with a history of GDM compared with 14% in parous women with no such history. Ten-year follow-up in the DPPOS confirmed these effects, demonstrating a sustained and relatively greater risk for diabetes in women with a history of GDM, which was reduced by 40% with metformin [15].

Insights from the DPP/DPPOS on how metformin prevents or delays diabetes

Acute pharmacological effect or amelioration of pathophysiology?

During the DPP, evaluations were carried out without interruption of study medication (placebo or metformin), except for withholding study medicine the morning of glycaemic testing. Thus, some (or all) of metformin’s effect could have been a transient pharmacological treatment effect (‘masking of diabetes’), rather than a true delay in the onset of diabetes. The DPP addressed this issue by retesting participants who had not developed diabetes by study end, 1–2 weeks after stopping metformin. After this washout period, the incidence of diabetes was still reduced by 25%, compared with the 31% reduction seen in the primary analysis, suggesting a more durable effect of metformin treatment on glucose metabolism [16].

Explanation of metformin-induced effects

Some of metformin’s diabetes prevention effect is attributed to weight loss, which was durable over time in the DPP/DPPOS (Fig. 2b, Table 2) [2, 8, 9]. Weight loss with metformin explained 64% of the its beneficial effect on diabetes risk at the end of the DPP [17]. Favourable changes were also seen in other measures of adiposity (waist circumference, waist-to-hip ratio), and in fasting insulin and proinsulin [17]. No differences were seen in self-reported physical activity or diet, or insulin secretion measured by the insulinogenic index between the metformin and placebo groups. While no single covariate completely explained the beneficial effect of metformin vs placebo, the combination of weight, fasting insulin and proinsulin levels, and other metabolic factors explained 81% of the beneficial outcomes with metformin [17]. Improvements in FPG and estimated insulin sensitivity with metformin may be owing to a combination of weight loss and other direct effects on the liver and, perhaps, other tissues.

Table 2 Effect of metformin on diabetes risk and CVD risk factors at baseline and at the end of each phase of the DPP and DPPOS
Full size table

Effects of metformin on blood glucose measures

The effects of the DPP interventions on FPG and HbA1c were examined in all participants, regardless of whether they had developed diabetes. During the DPP, metformin and ILS were similarly effective in restoring normal FPG values [2]. Despite metformin and ILS having similar effects on FPG, diabetes incidence was more significantly reduced by ILS than by metformin, reflecting the fact that most diabetes diagnoses in the DPP were triggered by the 2-hPG rather than FPG, and that ILS was more effective than metformin at restoring a normal 2-hPG. This latter observation was likely because, while both active interventions improved beta cell function, this effect was greater with ILS [18]. Consistent with metformin’s known ability to suppress hepatic glucose production during fasting [19, 20], its reduction of diabetes incidence compared with placebo was much greater in those entering the study with a FPG 6.1–6.9 mmol/l than in those with a FPG 5.3–6.1 mmol/l [2]. Metformin also lowered HbA1c relative to placebo, but to a lesser extent than did ILS [2].

After the DPP had been completed, an International Expert Committee and the ADA expanded the diagnostic criteria for diabetes to include HbA1c ≥6.5% (≥48 mmol/mol) [21, 22]. Although HbA1c was measured during the DPP/DPPOS, eligibility and diabetes diagnoses were based on fasting and/or 2-hPG. Thus, a secondary analysis using HbA1c ≥6.5% (≥48 mmol/mol) as an alternative definition of diabetes was performed, excluding the 13% of participants with HbA1c ≥6.5% (≥48 mmol/mol) at study entry. Although ILS was more effective than metformin in reducing the incidence of diabetes defined by the FPG and OGTT criteria, the effect of metformin was no longer significantly different from ILS when diabetes was diagnosed based on HbA1c (44% vs 49% reduction in the DPP, 38% vs 29% reduction throughout DPP/DPPOS; metformin vs ILS) [23]. In summary, metformin was as effective as ILS in preventing diabetes by some measures (i.e. HbA1c), but not by 2-hPG, in the DPP/DPPOS population.

Metformin’s interaction with genetic factors

The DPP investigated several genetic variants previously associated with risk of type 2 diabetes or metformin action. For example, homozygosity for the major diabetes risk variant rs7903146 in the TCF7L2 gene was associated with an 81% higher diabetes incidence in the placebo group that was reduced to a 62% increased risk in the metformin group [24]. In addition, a genetic risk score predicted diabetes incidence in the DPP, but with no significant interaction between the score and treatment group. That is, the interventions were equally effective, regardless of genetic susceptibility [25]. There was, however, a nominal interaction with metformin (p = 0.006) with the variant rs8065082 in the metformin transporter gene SLC47A1, with the minor allele being associated with lower incidence of diabetes in the metformin arm (HR 0.78 [95% CI 0.64, 0.96]; p = 0.02) [26].

Effects of metformin on microvascular complications

At the end of the DPP, the only microvascular outcome assessed was microalbuminuria. There was no effect of treatment intervention on the percentage of participants with elevated albumin:creatinine ratio (ACR) levels, although those who developed diabetes had a 59% increased risk of developing an elevated ACR (≥3.39 mg/mmol) [27]. One of the main goals in the longer-term follow-up of the DPPOS is to determine if treatments effective in preventing diabetes also affect the development of microvascular complications, specifically retinopathy, nephropathy and neuropathy. A composite of these microvascular outcomes at 15 years in the DPPOS was 28% less frequent in those who did not progress to diabetes, but there was no difference between the original treatment arms [9]. The very small difference in HbA1c levels among the treatment groups, limited power, and early referral to care providers for treatment of hypertension and dyslipidaemia have been considered reasons for the lack of an effect of the active treatments on microvascular outcomes thus far, despite the reduction in diabetes incidence [9]. It is still possible that treatment effects may emerge with longer follow-up and longer diabetes duration in the cohort.

Effects of metformin on cardiovascular disease risk factors

In the DPP, metformin had favourable effects on several cardiovascular risk factors, including lipoprotein subfractions [28], C-reactive protein and tissue plasminogen activator [29]. It also reduced the incidence of the metabolic syndrome by 17% compared with placebo [30]. No significant effects on lipid levels or blood pressure were seen [31] (Table 2). Over longer-term follow-up (10 years), no significant differences in traditional cardiovascular disease (CVD) risk factors have been noted between the metformin and placebo groups [32] (Table 2).

An average of 14 years after randomisation, subclinical atherosclerosis was assessed in 2029 participants using coronary artery calcium (CAC) measurements, according to the original randomisation group. There was a significant interaction between sex and the effects of metformin vs placebo on CAC presence (p = 0.01) and CAC severity (p = 0.08). Compared with placebo, metformin significantly lowered the presence and severity of CAC in men, with no effect in women. Of interest, no reduction in the prevalence of clinically significant plaque (Agatston score >100) was observed, suggesting the possibility that metformin affects smaller, more recently calcified plaques, rather than well-established plaques. There was no difference in CAC between ILS and placebo groups, suggesting a possible long-term differentiation between metformin and ILS [33]. Longer-term follow-up with ascertainment of CVD outcomes is underway.

figure a

Long-term safety and tolerability of metformin in the DPP/DPPOS

The long-term use of metformin within the context of a closely-monitored clinical trial has provided additional information on metformin safety and tolerability. Minor gastrointestinal symptoms were reported by 9.5% of those randomised to metformin, compared with 1.1% in the placebo group, but these were generally mild and tended to wane over time [34]. The risk of lactic acidosis with metformin use has recently been shown to be much lower than previously suspected [35] and there have been no reported cases of lactic acidosis in over 15,000 person-years of exposure to metformin in the DPP/DPPOS.

Metformin use has been associated with impaired intestinal absorption of vitamin B12 and increased risk of vitamin B12 deficiency. This risk was recognised in the design of the DPP and annual testing was performed to detect anaemia as a potential manifestation of low vitamin B12 levels. In addition, vitamin B12 levels were directly measured at two time points in the DPPOS. Biochemical vitamin B12 deficiency levels (<150 pmol/l) occurred more often in individuals in the metformin group than the placebo group at 5 years (4.3% vs 2.3%; p = 0.02); a similar pattern was observed but was not significant at 13 years (7.4% vs 5.4%; p = 0.12) [36]. Low or ‘borderline’ vitamin B12 (defined as levels <220 pmol/l) is accepted by some as evidence of inadequate vitamin B12 stores and was more common in those in the metformin group at 5 years vs placebo (19.1% vs 9.5%; p = 0.01) and 13 years (20.3% vs 15.6%; p = 0.02). In a multivariate model, years of metformin use, including metformin prescribed outside of the study, were associated with increased risk of vitamin B12 deficiency with the odds ratio for vitamin B12 <150 pmol/l per year of metformin use being 1.13 (95% CI 1.06, 1.20). Anaemia prevalence was higher in the metformin group but, importantly, did not differ by vitamin B12 level, suggesting that haematological monitoring may not be sufficient to detect metformin-associated vitamin B12 deficiency [36]. Given these findings in this large cohort, current guidelines now recommend consideration of periodic measurement of vitamin B12 levels and supplementation as needed in patients treated with metformin [37].

Looking to the future

The impact of prediabetes and diabetes worldwide is enormous, with 415 million adults currently having diabetes and a projected increase to 642 million by 2040 [38]. Both lifestyle intervention and metformin are effective in the prevention or delay of diabetes. Originally used for the treatment of type 2 diabetes, metformin, now proven to prevent or delay diabetes, may serve as an important additional tool in battling the growing diabetes epidemic. As detailed in this review, metformin had sustained benefit in preventing/delaying diabetes for at least 15 years. Further, while lifestyle intervention was uniformly effective across subgroups [2], the DPP identified significant benefit from metformin in those who were more obese, had a higher fasting glucose or a history of GDM, and a suggestion of greater effect than lifestyle intervention in those who were younger. Although not specific to treatment assignment, lack of progression to diabetes was associated with lower risk of microvascular complications [9] and, among men, metformin reduced atherosclerosis development [33]. Furthermore over 10 years, metformin treatment was estimated to be cost-saving, decreasing the cumulative costs of medical care received outside the DPP/DPPOS, compared with placebo [39]. Guidelines consistently recommend either lifestyle intervention or metformin therapy for the prevention of diabetes, with considerations for metformin in subgroups in which metformin had a relatively greater effect in the DPP [37, 40]. Given our current understanding of the beneficial effects of metformin to prevent or delay diabetes, a concerted global effort to translate this evidence may help redirect the continuing increase in the prevalence of type 2 diabetes.

The potential for additional benefits of metformin extends beyond diabetes prevention and represents the next phase of study for the DPPOS. As the largest and longest clinical trial of metformin treatment, uniquely in a population initially without diabetes, the DPP/DPPOS is now poised to evaluate whether starting metformin early in high-risk individuals impacts the development and risk for even later-stage comorbidities, notably CVD and cancer. Although decreasing the incidence of diabetes would be expected to decrease CVD risk, the effect of metformin and diabetes delay/prevention on CVD is unproven. In addition, based on experimental and epidemiological data, metformin has recently received attention as a potential anti-cancer agent. Prospective intervention studies with treatment of long duration and follow-up are needed to address these important questions. DPP/DPPOS, with over 15 years of randomised metformin experience, now aims to address this need.

In conclusion, the DPP/DPPOS clearly demonstrated a role for metformin in the prevention of diabetes. Looking to the future, understanding whether translation of these findings into routine clinical care improves current trends in the development of diabetes is of critical importance. The possibility that metformin can further impact additional complications of dysglycaemia that have not yet been investigated remains an exciting area of study.