Type 2 diabetes mellitus (DM2) is a prevalent chronic disease worldwide. Around 9% of adults have DM2, increasing to more than 20% of those aged 65 years or older [1, 2]. DM2 and its complications are an important cause of morbidity, and people with DM2 have substantially reduced life expectancy [3]. Duration of DM2 and degree of metabolic control are important factors determining the prognosis for people with DM2 [4]. However, the use of drugs for managing DM2 has been associated with preventable drug-related causes of admission to emergency units in older populations [5,6,7].

Metformin is one of the most widely prescribed first and second line oral glucose-lowering drugs. While it has low risk for hypoglycaemia, the risk for gastrointestinal effects is higher and it is contraindicated in patients with renal insufficiency [8,9,10]. Renal function declines with age and, therefore, should be monitored closely in older adults who are prescribed metformin [11, 12]. Clinical guidelines (in the United Kingdom, Canada and Australia) have advised that the use of metformin is contraindicated, or that lower doses be used, depending on renal function [13]. The use of metformin has also been associated with a higher risk of lactic acidosis but this has not been widely reported [14].

Currently, there is little empirical data about patient safety and effectiveness data on to the use of oral antidiabetics including metformin among older adults. Evidence-based clinical guidelines for the treatment of DM2 have acknowledged the lack of direct evidence in older people [10]. STOPP/START criteria version 2 considered metformin as a potentially inappropriate medication for older people with severe renal failure [15]. Inappropriate prescribing may involve the prescription of a wrong dose, the lack of a clear indication or the lack of evidence-base, among others [16].

The objectives of this systematic review (SR) are:

  • to identify and collect existing literature on the risks and benefits of use of metformin in the treatment of DM2 in older adults,

  • to assess the quality of the evidence identified, and develop recommendations when to discontinue or to adjust the dose of metformin in the treatment of DM2 in older adults.

This evidence was used to develop recommendations on discontinuation or dose adjustment of metformin in older people for the management of DM2 in order to reduce potentially inappropriate prescribing. These recommendations will be used in the electronic decision support tool of the “Polypharmacy in chronic diseases: Reduction of Inappropriate Medication and Adverse drug events in elderly populations by electronic Decision Support” (PRIMA-eDS) project [16].


This systematic review was developed following the methods proposed by both the Cochrane Handbook for Systematic Reviews of Interventions [17] and the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) [18]. A full description of the methods has been published previously [19].

Study inclusion criteria

Types of studies

We included systematic reviews, meta-analyses, controlled interventional studies and observational studies reporting on risks and benefits of the use of metformin in the treatment of DM2 in older adults. We excluded abstracts, pooled analyses, editorials, opinion papers, case reports, case series, narrative reviews, letters, and qualitative studies.

Types of participants

The population of interest were older people with DM2. We considered the age of 65 as cut-off point for defining older people, which has been traditionally used because of its association with retirement age in some developed countries [20,21,22]. The criteria for inclusion in this systematic review were:

For existing systematic reviews and meta-analyses:

  • overall mean or median age ≥ 65 years; or

  • overall mean or median age < 65 with subgroup analysis reporting on participants ≥65 years; or

  • overall mean or median age not reported but 80% or more of the included studies reporting a mean or median age ≥ 65 years.

For individual controlled interventional studies and observational studies:

  • ≥80% of participants ≥65 years; or

  • <80% of participants ≥65 years with subgroup analysis reporting on participants ≥65 years.

Types of interventions

Studies reporting on the efficacy and/or safety of metformin for the management of DM2 were included irrespective of whether metformin was prescribed as monotherapy or in combination with any other. Included studies compared metformin versus placebo, usual or no treatment, and other drugs to treat DM2 or a non-pharmacological intervention.

Types of outcomes

The following clinically relevant endpoints were included either as primary or secondary outcomes:

  • Quality of life

  • Mortality

  • Life expectancy

  • Hospitalisations

  • Cognitive impairment or cognitive status

  • Functional impairment or status

  • Cardiovascular event including stroke

  • Renal failure

  • Composite end points including any of the above (extraction of individual outcomes was undertaken if reported by original studies)

  • Adverse drug event including hypoglycaemia

  • Any of the above evaluated as safety endpoints.

Studies evaluating only glycaemic control or lactate levels. To aid interpretation of findings outcomes were classified into two tiers according to their anticipated impact on longer-term health and quality of life: Tier 1 outcomes have shorter-term impact including hypoglycaemia and adverse events (including serious adverse events); tier 2 outcomes have longer-term impact including, but not limited to, cardiovascular and cerebrovascular events, related admissions, and death.


All settings were included.


Language restrictions were not applied for study searches. However, the inclusion of studies was restricted to languages that could be read by the research team English, German, Finish, Italian, and Spanish.

Search method

Database searches were conducted by YVM and AW following staged methodology comprising four sequential literature searches. Each search being performed only if the preceding one yielded high quality results or if evidence insufficient to enable any evidence based recommendations to be made. Each search was conducted on 09 December 2015 using the OVID interface for each database. The searches included the following databases and types of studies:

  • Search 1: Systematic reviews and meta-analyses in the Cochrane database of Systematic Reviews (2005 to 2015) and the Database of Abstracts or Reviews of Effects (1991 to 2015).

  • Search 2: Systematic reviews and meta-analyses in MEDLINE and MEDLINE (R) In-Process & Other Non-Indexed Citations (1946 to 2015), EMBASE (1974 to 2015), Health Technology Assessment (HTA) (2001 to 2015) and International Pharmaceutical Abstracts (IPA) (1970 to 2015).

  • Search 3A: Interventional and observational studies meeting eligibility criteria included in systematic reviews which did not meet the inclusion criteria for searches 1 and 2

  • Search 3B: Additional controlled interventional and observational studies identified from MEDLINE, MEDLINE (R) In-Process & Other Non-Indexed Citations, EMBASE, HTA and IPA published since 2011.

References of included studies were checked to identify further articles for inclusion, and we also considered studies identified from manual searches and snowballing. Protocols for yet-to-be published studies were collected to inform future updates of this systematic review. Studies excluded after full-text check are listed in Additional file 1 together with reasons for exclusion.

The PICOS-framework was used to develop the search terms (population: older people, intervention: metformin, comparison: no limits, outcomes: see list above “Types of outcomes” and study design: systematic reviews, meta-analyses, controlled interventional studies and observational studies). We also developed search filters specific for different study designs, described in detail in the protocol [19]. Additional file 2 lists the full search terms for each search (i.e. search 1, 2 and 3B).

Data management

Search results were uploaded to Endnote X7 reference management software where search results were retrieved and de-duplicated.

Selection of studies

Two reviewers assessed titles and abstracts from each search independently to identify studies to consider for inclusion. Full manuscripts were then obtained for all titles and abstracts meeting the inclusion criteria or where there was any uncertainty about inclusion. YVM, ARG, CA, BF, CS and LS were involved in this task.

Reviewers discussed any disagreement about studies to include. AS was consulted when YVM and ARG could not reach an agreement on whether or not to include a study. YVM and ARG were consulted when CA, CS, BF and LS could not reach agreement.

Data extraction

YVM, ARG, CA, BF, CS and LS (reviewers) independently conducted data extraction of the included studies using a standardised and piloted data collection form previously published with the protocol [19]. This extraction form included information related to the study design and aim, characteristics of participants (i.e. age, sex, setting, comorbidity, use of concomitant medications, functional status, and cognitive status), the intervention (i.e. metformin) and comparison, time to follow-up, and reported outcomes. Completeness and accuracy of data extraction was then double-checked by a second reviewer.

Quality appraisal

For each study design we used separate validated assessment tools to evaluate quality (AMSTAR) [23, 24] was used for systematic reviews/meta-analyses, for intervention studies the Cochrane Collaboration’s tool for assessing risk of bias [17] was used, and for observational studies the Critical Appraisal Skills Programme (CASP) [25, 26].

Dealing with duplicate and companion publications

All relevant data from publications relating to a single primary study were included. The staged approach carries a risk of ‘double counting patients whose trials are included in a systematic review. Any such instances have been identified, reported and corrected for in our data synthesis.

Data synthesis

A narrative synthesis describing all included systematic reviews, meta-analyses, intervention and observational studies, participants and findings was carried out. The included studies were highly heterogeneous in comparison treatments, length of follow-up, type of design, and definition of outcomes; therefore no additional meta-analyses were performed. The quality of the included studies is also reported.

Identification of “references of interest” for the development of recommendations

During the search process, reviewers identified additional references which did not fulfil the inclusion criteria of the SRs but which they considered of interest for the development of recommendations according to the methodology described by Martinez-Renom Guiteras (2016) [19].

Development of recommendations

Included studies and references of interest were summarised in a document used by the research team to develop and discuss recommendations to discontinue the use of metformin for the management of DM2 in older people including: a) study design or type of reference, target population and sample size, metformin dose (if available) and comparison groups, outcomes, main results, subgroup analysis if applicable; b) quality appraisal ratings of included studies; and c) proposed recommendations. Each recommendation was given a strength (weak or strong) and quality (low, moderate or high) rating following the Grading of Recommendations Assessment, Development and Evaluation (GRADE) methodology [27,28,29]. Recommendations were written following a standardised schema and reflecting the strength and the quality of the evidence. The Finnish team of editors from Duodecim Medical Publications Ltd. participated in the later stages and approved the recommendations.


Results of the search

Searches 1, 2, 3A and 3B were all conducted. Search 1 identified one relevant meta-analysis, by Lamanna et al. [30] which did not provide summary results for our SR targeted at old age and was excluded. However eligible individual studies were identified from it and added to search 3A. The date of the search by Lamanna (2011) [30] was used as the start date for our search for additional individual studies under search 3B. No relevant meta-analyses were identified from search 2.

In total 2185 records were found through initial database searching (126 from search 1, 175 from search 2, 1884 from search 3B). Additionally, we identified 66 records from search 3A (individual studies from excluded systematic reviews and meta-analyses) and 461 records from reference lists of included studies. After removing duplicates, we screened 2318 records and excluded 1878 checking titles and abstracts. We assessed 440 full-texts for eligibility and excluded 425. Main reasons for exclusion were wrong population, wrong intervention and wrong outcome. We included 15 studies reported by one publication each. The PRISMA flow diagram is presented in Fig. 1.

Fig. 1
figure 1

Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) flow diagram

Included studies

Fifteen studies were included [31,32,33,34,35,36,37,38,39,40,41,42,43,44,45] including 426,549 participants of all ages of which 230,229 were 65 years and older. However, analyses of efficacy and risks of antidiabetic drugs did not always include all participants. Table 1 shows the summary of the study characteristics of included studies.

Table 1 Summary of study characteristics

Study designs

Only four studies were randomised controlled trials (RCTs) [31, 35, 36, 43]. Most of the included studies were observational (11 studies), three of these prospective [32, 38, 42], seven retrospective [33, 34, 37, 39, 41, 44, 45], and one cross-sectional [40]. Data on outcomes was extracted for the end of the follow-up period of each included study.


In ten studies at least 80% of participants were aged 65 years or older [32, 34,35,36,37,38,39,40, 43, 44]. In these studies, the lowest mean age was 69.3 years and the highest 80.5 years. The remaining five studies provided subgroup analyses of older people (≥65 or ≥71 or >80 years) [31, 33, 41, 42, 45]. Length of follow-up varied from none (cross-sectional study) [40] to 11 years [38].

Representation of males ranged from 24.1% [44] to 97% [41]. Eight studies reported ethnicity with most participants being white/Caucasian (up to 81%) [37]. Participants from four different continents were involved in the studies including America (n = 8), Europe (n = 7), Asia (n = 3) and Australia (n = 2). Comorbidities were reported in 11 studies, and congestive heart failure, myocardial infarction, hypertension, chronic kidney disease and hypothyroidism were the most commonly reported. Functional status was reported in one study [37]. No studies reported on the frailty level of the participants. Cognitive status was reported in four studies [37,38,39,40]. The percentage of participants with dementia was low in most studies ranging from 2.8% [38] to 67% [37]. Participant characteristics are summarised in Additional file 3: Table S1.

Covariates in models

Adjustment for relevant covariates is important in observational studies to reduce confounder bias. The use of covariates indicates that a study’s authors have considered this issue, although by no means implies that bias has been eliminated. Many of the studies used long lists of covariates, therefore Additional file 3: Table S2 is given for online publication.

Interventions and outcomes

Most of the included studies investigated the effect of metformin as monotherapy [33, 37, 39,40,41,42, 44, 45]. We classified comparison treatments as either “non-specific active treatments” (the comparator was not a single specific drug or treatment e.g. usual care, no insulin sensitizer, not on metformin), or as “specific active treatments” (the comparator was a specified treatment such as insulin, sulfonylureas, dipeptidyl peptidase-4 inhibitors, thiazolidinediones, and other specific drugs and combinations).

One RCT addressed both tier 1 and tier 2 outcomes [31]; the remaining RCTs investigated tier 1 outcomes only [35, 36, 43]. All the included observational studies addressed tier 2 outcomes only [32,33,34, 37,38,39,40,41,42, 44, 45]. We did not find any studies reporting on the following relevant endpoints: quality of life, life expectancy, functional impairment or status, and renal failure.

Main findings

Table 2 summarises the results for each study for both metformin and comparison groups, with estimated risk ratios with 95% confidence intervals, together with statistical comparisons from the study. The results are organised by type of outcome (safety or efficacy), and then by type of comparator within each type of outcome (metformin against non-specific active treatments; metformin against other specific active treatments). A Additional file 3: Table S2 is available with the covariates that were taken into account in the statistical models of the included studies.

Table 2 Summary of study findings

Tier 1 outcomes (hypoglycaemia and adverse events)

Tier 1 outcomes were investigated by all the included trials but none of the observational studies. In one large trial metformin was not significant different than usual care (non-specific active treatment) for serious adverse events [31]. Compared against other specific active treatments, a combination of insulin glargine plus glimepiride and metformin demonstrated significantly fewer hypoglycaemic events (both confirmed and unconfirmed hypoglycaemia) compared to premixed insulin [35]. However, in other trials, participants taking tolbutamide reported fewer side effects than participants on metformin [36] and vildagliptin outperformed metformin on all safety outcomes reported [43], but in neither case was any formal statistical comparison reported.

Tier 2 outcomes: Metformin compared to other non-specific active treatments

One trial and five observational studies compared metformin as monotherapy with non-specific active treatments for efficacy-related outcomes [31, 34, 38,39,40, 42]. Three large observational studies reported significantly fewer deaths in participants taking metformin compared to participants taking either no insulin sensitizer [39], no antidiabetic drugs [38], or no metformin [42]. However, there was no significant difference in mortality for patients aged over 80 years, and so was it for those patients with GFR ≤60 [42]. Inzucchi et al. (2005) [34] in another large study also reported that there was no significant difference in mortality in the metformin group. Admissions for various types of causes were evaluated by three studies [31, 34, 39] without significant differences between metformin and other active treatments, except for re-admission for heart failure which was significantly different favouring metformin [39].

Tier 2 outcomes: Metformin compared to other specific active treatments

Six observational studies compared metformin with specific active treatments on their effect on efficacy outcomes [32, 33, 37, 41, 44, 45]. One of these studies [32] reported substantially reduced mortality in the metformin monotherapy group (16% of the 422 participants) compared to groups taking metformin and sulfonylureas in combination (32%) and sulfonylurea monotherapy (51%). The remaining studies compared monotherapy with metformin against a range of other mostly monotherapy drug treatments. Metformin outperformed sulfonylureas with significantly fewer hospital admissions for acute myocardial infarction, stroke or death [41]; fewer hospitalisations for hypoglycaemia [37]; fewer events of non-fatal cardiovascular disease [33]; fewer fractures [44] and fewer events of myocardial infarction, congestive heart failure and all-cause mortality [45]. Other study results favoured metformin with significant differences in comparisons against thiazolidinediones [34] (mortality and all-cause and HF readmissions). There were no significant differences when metformin was compared to rosiglitazone and pioglitazone [45] (myocardial infarction, congestive heart failure and all-cause mortality).

Excluded studies

Additional file 1 has the full list of reasons for exclusion of studies after full text analysis.

Quality appraisal of included studies

Randomised trials

Four randomised trials were included and assessed for risk of bias (Table 3). Two of these trials did not provide enough information to assess the risk of bias [36, 43]. Another trial was judged to be of high risk of selection, performance and other bias [31]. One trial was judged to be of high risk of performance and detection bias [35].

Table 3 Quality appraisal for intervention studies

Observational studies

Quality appraisal was assessed with the CASP tool for the included 11 observational studies (Table 4). Most of the included studies reported sufficient detail to assess their quality. All studies addressed a clearly focused issue and all but one [33] used an appropriate method to answer their research question. Selection bias was not a problem in 8 of the 11 included studies as the recruitment method was adequate for the design (either cohort or case-control study) [32,33,34, 37,38,39, 44, 45]. Potential confounding factors were taken into account in the design or analysis in 10 of the 11 included studies [33, 34, 37,38,39,40,41,42, 44, 45]. It was clear that the results of all studies (apart from the cross-sectional study) could be applied to our population of interest (older people). Nearly half the studies did not report accurately how the exposure or outcome were measured, which could lead to high risk of measurement or classification bias [34, 40,41,42, 45]. For example, the comparison group was not clearly defined in one study [34]. It was unclear whether the follow-up was sufficiently long in 5 studies [32, 33, 39, 41, 42] and it was considered that the follow-up insufficient in 3 studies [34, 37, 44].

Table 4 Quality appraisal for observational studies according to the Critical Appraisal Skills Programme (CASP)

Additional references of interest for the development of recommendations

One further reference was incorporated as additional reference of interest for the development of the recommendations [46]. A clinical guideline from the American Geriatrics Society which included recommendations about the management of DM2 in older people with renal insufficiency [46] was included as an additional reference of interest for the development of recommendations.


Four recommendations about stopping the use of metformin in older people with DM2 (Table 5) were developed related to halting In order to discuss and agree on these recommendations three meetings took place between YVM (researcher) and ARG (researcher and clinician). IK (senior clinician and researcher) and AS (senior clinician and researcher) participated in one of these meetings. The whole body of evidence identified in the SR was taken into consideration for the development of the recommendations. However, each recommendation was specially supported by the following specific studies included in the SR or considered as additional references of interest: a clinical guideline, an observational study without high quality [42] and two randomised trials with insufficient information to assess their risk of bias [36, 43]. All recommendations were considered to be weak and based on evidence of low quality, and the reasons for this are reported in Table 5. The recommendations were included in the Comprehensive Medication Review (CMR) tool developed as part of the PRIMA-eDS project, and they were formulated according to their strength and the quality of their evidence [19].

Table 5 Recommendations to stop the use of metformin in older people with type 2 Diabetes mellitus


Our aim was to systematically review the existing evidence on the risks and benefits of the use of metformin for the management of DM2 in older people. We therefore included only those studies where a high proportion of participants were aged 65 years or older, as specified in our inclusion criteria. No systematic review or meta-analysis fulfilled our inclusion criteria, and we finally included 4 RCTs and 11 observational studies, with most observational studies being retrospective.

When comparing metformin with sulfonylureas, results suggest that metformin may be better than sulfonylureas in reducing several outcomes such as cardiovascular outcomes, mortality, hospitalisation for hypoglycaemia, or risk of falls in people aged 65 and older with DM2 [32, 33, 37, 41, 45]. When comparing metformin with no insulin sensitizer antidiabetic drugs, divergent results were found depending on the study population [34, 39] Generally, these results are in line with clinical guidelines recommending metformin as the first-line drug treatment for adults with DM2 [10]. Guidelines also suggest that if initial drug treatment with metformin fails to control levels of glycated haemoglobin, dual therapy should be considered [10]. Only one study was identified which specifically analysed risks and benefits of combining metformin with other antidiabetic drugs, where risk of hypoglycaemia with the combination of metformin, sulfonylurea and insulin glargine compared to premixed insulin was decreased [35]. Thus, there seems to be a lack of evidence analysing the benefits and risks of combined therapy including metformin in older people with DM2.

Few studies provided data on adverse events other than hypoglycaemia and falls with the use of metformin in this population such as bloatedness, nausea, and diarrhoea [31, 36, 39, 40, 43]. Furthermore, only two studies analysed the adverse event of lactic acidosis, and no increased risk was found for metformin [31, 39]. All these studies had considerable methodological limitations. We did not identify any study reporting on renal failure as adverse event of metformin. Thus, further prospective studies should evaluate the adverse events of the use of metformin in older people.

The benefits of metformin on the mortality of very old people (aged 80 years over) were investigated by only one included study [42]. Here, mortality was significantly decreased with the use of metformin in people aged 65–80 but the effect was not significant for the population aged 80 and older. The study could been underpowered for this subgroup analysis, but evidence this may also suggest that the benefits of metformin on mortality may be non-existent for very old people, especially those with limited life expectancy, as suggested by other authors [47].

The included studies rarely reported on the functional level and cognitive status of the participants; the use of concomitant drugs and the presence of other diseases were more frequently reported but focused mostly on cardiovascular drugs and diseases. Thus, the present systematic review demonstrates that not only very old people, but also cognitively and functionally impaired people and old people with multimorbidity are underrepresented or at least underreported in existing studies, which limits the generalisability of already scarce evidence for this heterogeneous group of older people. A growing body of literature presents functional and cognitive status as well as multimorbidity as predictors of mortality among older people independently from their chronological age [48,49,50], which supports the idea that further studies analysing these aspects are necessary.

Although the lack of evidence has been previously commented on by several authors [1, 47, 51], to the best of our knowledge, ours is the first study to systematically review all available evidence on the risks and benefits of metformin for the management of DM2 in older people.

Current guidelines suggest that for older people with limited life expectancy or functional limitation, intensive glycaemic control is not recommended [1, 47, 52]. However, older people who are functionally and cognitively intact and have significant life expectancy should be treated with goals similar to those developed for younger people [1]. In the present systematic review, six studies [32, 35, 36, 41, 43, 45] reported on mean baseline levels of glycated haemoglobin which ranged between 7.0% [41] and 10.2% [36]. One of these studies considered a glycated haemoglobin level of 7% or less without experiencing nocturnal hypoglycaemia to represent successful therapy [35].

None of the other studies reported their target glycaemic control level clearly. It would be useful for future studies on the management of DM2 to report the target glycaemic control level, especially for frail older people.

Our research team developed four recommendations using both the results of the systematic review and the additional references identified [36, 42, 43, 46]. The recommendations advise clinicians to consider discontinuing metformin in people aged 80 and older, those with gastrointestinal complaints during the last year, and/or those with Glomerular Filtration Rate (GFR) ≤60 ml/min. These recommendations have been incorporated in the trial version of an electronic decision support tool that aims to help general practitioners to reduce inappropriate prescriptions for older people with multimorbidity. Decisions on prescription or de-prescription of metformin should be made taking the symptoms and individual characteristics of each patient into account and clinicians receive instructions on that. Currently, the tool is being tested in a cluster randomised controlled trial in four European countries [53].

This systematic review has limitations. The search strategy and inclusion criteria were designed to identify studies focusing on older people; studies on the general population that may have contained relevant information for the older population might have been overlooked. However, using independent reviewers for study selection and our peer reviewed process of development of recommendations should have minimised this problem. Our recommendations focus only on the discontinuation of metformin, as it was not the aim of the PRIMA-eDS project to develop recommendations when to use metformin. Nevertheless, this systematic review aims at providing an overview of the existing evidence on both the benefits and risks of the use of metformin in older people.


This study highlights the lack of good quality evidence on the risks and benefits of metformin for the management of DM2 in older people. The use of metformin seems associated with benefits to lower mortality risk in older people, and may also be associated with a reduced risk of adverse events such as hypoglycaemia and non-fatal cardiovascular events, than other antidiabetic drugs, especially sulfonylureas. However, no prospective studies focussing on very old (80 and older) and functionally and cognitively impaired older people are available. In very old people, those with renal insufficiency (GFR ≤60 ml/min) and those with gastrointestinal complaints during the last year, the discontinuation of metformin should be considered, especially for those with limited life expectancy or functional impairment. There is an urgent need for studies on the risks and benefits of metformin for the management of DM2 in these populations in order to guide clinicians in planning of individualised patient care.