Metformin for non-diabetic patients with coronary heart disease ( the CAMERA study ) : a randomised controlled trial

Methods We did a single-centre, double-blind, placebo-controlled trial at the Glasgow Clinical Research Facility (Glasgow, UK). We enrolled patients taking statins who did not have type 2 diabetes but who did have coronary heart disease and large waist circumferences. Participants were randomly assigned (1:1) by computer to either metformin (850 mg twice daily) or matching placebo in block sizes of four. Patients, investigators, trial staff , and statisticians were masked to treatment allocation. The primary endpoint was progression of mean distal carotid intima-media thickness (cIMT) over 18 months in the modifi ed intention-to-treat population. Secondary endpoints were changes in carotid plaque score (in six regions), measures of glycaemia (HbA1c, fasting glucose, and insulin concentrations, and Homeostasis Model Assessment of Insulin Resistance [HOMA-IR]), and concentrations of lipids, high sensitivity C-reactive protein, and tissue plasminogen activator. The trial was registered at ClinicalTrials. gov, number NCT00723307.


Introduction
Cardiovascular disease is a leading cause of morbidity and mortality. Although type 2 diabetes increases cardiovascular risk, people with non-diabetic dysglycaemia also have increased risk 1 and some researchers regard insulin resistance as the most important cardiovascular risk factor. 2 The proatherogenic phenotype characterising dysglycaemia includes dyslipidaemia, vascular dysfunction, infl ammation, and abnormal haemostasis, each of which are associated with cardiovascular events. 3,4 Metformin-a safe and inexpensive glucose-lowering drug-is sometimes used to treat non-diabetic people with polycystic ovarian syndrome, for aiding weight loss, and for some people with impaired glucose tolerance, partly on the basis of its purported cardiovascular benefi ts. [5][6][7] Metformin improves dyslipidaemia and reduces concentrations of infl ammatory and haemostatic biomarkers in non-diabetic individuals, 8,9 suggesting possible cardio vascular benefi t. The eff ect of metformin on cardiovascular outcomes of patients with type 2 diabetes has been studied in two randomised controlled trials. In the UKPDS, overweight patients taking metformin had a 39% lower risk of myocardial infarction over 10 years than did patients on conventional dietary therapy, and post-trial observational data showed continuing benefi t. 10,11 In the HOME trial, 12 which included 390 patients with type 2 diabetes on insulin, metformin reduced the composite cardiovascular endpoint by 40%. Metformin treatment also reduces the risk of developing type 2 diabetes. 13 However, its eff ects on cardiovascular endpoints in non-diabetic people remain unknown.
We designed the Carotid Atherosclerosis: MEtformin for insulin ResistAnce (CAMERA) study to assess the eff ect of metformin on progression of mean carotid intima-media thickness (cIMT) in non-diabetic individuals with coronary heart disease.

Study design and participants
CAMERA was a randomised, placebo-controlled, doubleblind trial done at the Glasgow Clinical Research Centre (Glasgow, UK). Inclusion criteria were: age 35-75 years, proven coronary heart disease (previous acute coronary syndrome, coronary artery bypass surgery, or angiographically proven coronary heart disease), large waist circumference as per International Diabetes Foundation criteria (≥94 cm in men, ≥80 cm in women), 14 and prescribed a statin (dose and type were not adjusted 15 ). Exclusion criteria were: pregnancy or lactation at screening; premenopausal woman not taking adequate contraception (daily oral hormonal con traception or regular injectable hormonal con traception); type 2 diabetes or people with either HbA 1C of 7·0% (53 mmol/mol) or more, or fasting plasma glucose of 7·0 mmol/L or more at screening; acute coronary syndrome within the previous 3 months; New York Heart Association functional class 3 or 4 heart failure; uncontrolled angina; hepatic impairment (based on assessment of available liver function tests and liver imaging by the study physician but not on biochemical thresholds); renal impairment (estimated glomerular fi ltration rate <45 mL/min per 1·73 m² at screening); hypersensitivity to metformin; acute illness (dehydration, severe infection, shock, acute cardiac failure); and suspected tissue hypoxia. People with HbA 1C 6·0-6·9% (42-52 mmol/mol) and fasting plasma glucose less than 7·0 mmol/L at screening had an oral glucose tolerance test; those with post-challenge glucose of 11·1 mmol/L or more were excluded. The trial was designed before the adoption of HbA 1c as a diagnostic test for type 2 diabetes. 16 All participants provided written informed consent and were followed up for 18 months. This study was approved by the Medicines and Healthcare Products Regulatory Agency and West Glasgow Research Ethics Committee, and done in accordance with the principles of the Declaration of Helsinki and good clinical practice guidelines.

Randomisation and masking
Participants were randomly assigned to metformin or placebo (1:1) on the CAMERA website. The randomisation sequence was generated independently by computer (by the Robertson Centre for Biostatistics) with permuted blocks of four without stratifi cation. Patients, investigators, trial staff , and statisticians were masked to treatment allocation.

Procedures
Participants were advised to take one study tablet (850 mg metformin or matching placebo) daily for 1 week before titrating up to 2 tablets daily (one with the morning meal, one with the evening meal). Participants unable or unwilling to take the medication twice daily could take one tablet daily from any point. Masked study medication was supplied in numbered bottles and compliance (defi ned as taking >80% of study medication during the trial) was assessed by tablet count. Study medication was reduced to one tablet daily if estimated glomerular fi ltration rate fell below 45 mL/min per 1·73 m², and stopped below 30 mL/min per 1·73 m². For assessment of mean cIMT, carotid artery scans were done at baseline, 12 months, and 18 months. The imaging protocol consisted of B-mode ultrasound image acquisition of the right and left far walls of the distal 10 mm of the common carotid arteries, using an Acuson Sequoia C512 (Siemens Medical Solutions; Erlangen, Germany) with an L8 5-12 MHz linear array broadband transducer, and electrocardiogram gating. 17 All images were obtained from a single ultrasound machine by one doctor (DP). Participants were positioned recumbent with the head tilted 10° to the contralateral side. On each side of the neck, three digital clips were recorded from lateral, posterolateral, and anterolateral angles, each clip including roughly three QRS complexes. Depending on image quality, up to 18 cIMT images (in end-diastolic frame) were available from each visit for analysis of mean cIMT. Mean left-side and right-side cIMT were calculated separately and averaged to give the overall mean cIMT. Participants in whom only one side could be assessed at baseline had the same side analysed throughout. Images were analysed when all study visits had been completed by a single researcher using semiautomated artery measurement software 18 or a combination of semiautomated and manual approaches depending on image quality. Pretrial intra-individual cIMT reproducibility was 7%.
Carotid plaque score-a value of 0-6 depending on plaque presence in six regions-was assessed by transverse and longitudinal transducer positioning. Presence of plaque was assessed in three regions for both left and right carotid arteries (ie, six regions): distal 10 mm of the common carotid artery, carotid bulb (from the widening of the common carotid artery to the fl ow divider), and proximal 10 mm of the internal carotid artery (distal from the fl ow divider). Plaque was defi ned as cIMT of 1·5 mm or more, or 0·5 mm or more focal encroachment into the arterial lumen. Plaques in the entire common carotid artery and internal carotid artery were also recorded for a sensitivity analysis to avoid any eff ect of variations in the selection of the 10 mm boundaries. Other plaque surrogate markers (carotid plaque area and carotid plaque volume) have been proposed; 19 however, we did not use them because of concern about potential subjectivity of carotid plaque area and the requirement for experience in operating specialised equipment to assess carotid plaque volume.
Laboratory assessments measured at every visit included renal and liver function tests and-at every 6 month visit-fasting plasma glucose, HbA 1c , fasting lipid profi le, high sensitivity C-reactive protein, fasting insulin, vitamin B12, tissue plasminogen activator, and high sensitivity troponin T. With the exception of insulin, vitamin B12, tissue plasminogen activator, and high sensitivity troponin T, all biochemical and haematology analyses were done at Gartnavel General Hospital, Glasgow, UK, with strict quality control procedures and where all assays (Abbott Diagnostics, North Chicago, IL) performed acceptably according to UK National External Quality Assessment Service, our external quality assurance scheme. Vitamin B12 and high sensitivity troponin T were analysed from stored plasma samples with an automated clinically validated analyser (Roche Diagnostics; Burgess Hill, UK) and manufacturer standards and quality control material; the low control coeffi cient of variation was 8·1% for vitamin B12 and Chronic heart failure (NYHA I or II) 9 (10%) 7 (8%) Estimated glomerular fi ltration rate (mL/min per 1·73 m²) 85 (16) 81 (17) Alanine Data are mean (SD) or n (%) unless stated otherwise. NYHA=New York Heart Association. HOMA-IR=homoeostatic model assessment insulin resistance. *Geometric mean (SD). †Lovastatin ≥80 mg, atorvastatin ≥20 mg, simvastatin ≥40 mg, or any dose of rosuvastatin daily. ‡All doses lower than intense doses of lovastatin, atorvastatin, simvastatin, or any dose of pravastatin or fl uvastatin daily. §In the carotid bulb, distal 10 mm of the common carotid artery, and proximal 10 mm of the internal carotid artery. Vital signs (blood pressure, pulse, bodyweight, body fat by bio-impedance with a Tanita BIA body fat analyser [Tanita Corporation, Tokyo, Japan]), waist circumference (measured midway between lowest rib and iliac crest), and hip circumference (measured around widest part of the buttocks) were measured at each visit. Participants were asked about adverse events and serious adverse events at each visit. Serious adverse events were recorded from paper and electronic hospital records but were not independently reviewed.
The primary endpoint was progression of mean distal common carotid artery cIMT at 18 months. Secondary endpoints were progression of carotid plaque score and change in HOMA-IR, HbA 1C , total cholesterol, HDLcholesterol, non-HDL-cholesterol, triglycerides, tissue plasminogen activator, and high sensitivity C-reactive protein. Additional (not prespecifi ed) endpoints were changes of adiposity (bodyweight, body fat, body-mass index (BMI), and waist circumference), vitamin B12 con centrations, high sensitivity troponin T concentrations, and surrogate markers of liver fat-ie, γ glutamyl transferase and alanine aminotransferase concentrations.

Statistical analysis
Metformin reduced cIMT by 0·032 mm at 12 months in a single-centre randomised trial of non-diabetic participants (n=60). 21 In the single-centre ARBITER trial (n=161), 22 cIMT decreased by 0·059 mm in 12 months in patients taking atorvastatin 80 mg compared with those taking pravastatin 40 mg. Our sample size calculations suggested that 180 participants completing the trial would be needed to detect a diff erence of 0·021 mm (SD 0·050 mm) between groups for cIMT progression with 80% power at 18 months. This estimate was based on two cIMT readings per participant whereas our study included three. We initially intended to follow up participants for 24 months but this was reduced to 18 months to enable completion within the time available. This change occurred in June 2010, when 115 participants had been randomly assigned to treatment but none had yet completed 18 months of follow-up.
Non-normally distributed variables were logtransformed with data presented as geometric means. We used repeated measures regression models to assess progression of mean cIMT and progression of carotid plaque score. These models allowed for participantspecifi c random intercepts and random slopes. We estimated the rate of change (slope) for each treatment group and the diff erence in these slopes. We also adjusted the models for pre-defi ned baseline cardiovascular risk factors (age, sex, smoking status, systolic blood pressure, diastolic blood pressure, HDLcholesterol, LDL-cholesterol, and estimated glomerular fi ltration rate). We did sensitivity analyses for progression of mean cIMT for right and left cIMT separately; an analysis based on a minimum number of cIMT values per participant per visit (when a participant had 5-9 measurements per side and when average values could be calculated for both sides); an analysis of only those participants who completed the study, did not reduce the dose, and took more than 80% of study medication; and a carotid plaque score analysis of the carotid bulb and entire internal carotid artery and common carotid artery.
All analyses were done for the modifi ed intention-totreat population (had baseline and at least one later cIMT or plaque measurement) apart from a pre-planned perprotocol analysis, which was done for the primary outcome including only patients with no major or minor protocol deviations. Major protocol deviations included errors in the recording of informed consent, failure to comply with inclusion or exclusion criteria, and failure to amend study drug prescription based on biochemical data. Minor protocol deviations included compliance less than 50%, failure to attend all visits, failure to record study measurements, and stopping study drug before the end of the study. We did analyses for the primary outcome at 18 months comparing subgroups of tertiles for cIMT, fasting insulin, HbA 1c , BMI, and non-HDLcholesterol at baseline plus sex. Diff erence in slope (95% CI; mm per year) 0·007 (-0·006 to 0·020); p=0·29 ·· Data are mean (SD) unless otherwise stated. Analyses are for modifi ed intention-to-treat analyses of participants with at least one measurement after baseline. Figure 1 shows number of patients attending the relevant visits. *Adjusted for age, sex, smoking status, systolic and diastolic blood pressure, HDL-cholesterol, LDL-cholesterol, and estimated glomerular fi ltration rate. Other secondary endpoints and additional endpoints were analysed with repeated measures models that allowed for random intercepts only and assumed a general covariance structure; the visit was fi tted as a fi xed categorical variable. Treatment-by-visit interactions were evaluated to assess whether overall treatment eff ects or treatment eff ect by visit were appropriate analyses. Analyses were done with SAS (version 9.3). p values were two-sided and 0·05 was the threshold for statistical signifi cance.

Metformin group Placebo group
This trial is registered with ClinicalTrials.gov, number NCT00723307.

Role of the funding source
The sponsors had no role in study design, data collection, analysis, or interpretation, or writing of the report. DP, SML, IF, and NS had access to the raw data. The corresponding author had full access to all of the data and the fi nal responsibility to submit for publication.

Results
We identifi ed roughly 3000 potential participants from electronic searches of Glasgow general practice databases, supplemented by patients from hospital cardiology clinics. Of those invited, 805 replied and 356 were screened. 173 were enrolled and randomly assigned (86 to metformin, 87 to placebo; fi gure 1). Trial visits were done between 2009 and 2012. Baseline characteristics did not diff er substantially between treatment groups (table 1). Average age was 63 years, mean cIMT was 0·717mm (SD 0·129), and mean carotid plaque score was 2·43 (SD 1·55). Of the patients for whom complete data were available based on drug return, 64 of 82 (78%) assigned to metformin and 68 of 85 (80%) assigned to placebo were compliant. More participants taking metformin (n=19) than placebo (n=10) reduced to one tablet daily. 12 participants assigned to metformin and four assigned to placebo did not attend the fi nal visit. Those participants lost to followup (n=6) could not be reached by post or telephone. However, according to electronic hospital records, none had a serious adverse event during follow-up. During follow-up, four participants stopped statin treatment  Diff erence in slope (95% CI; points per year) -0·07 (-0·28 to 0·13); p=0·48 ·· Data are mean (SD) unless otherwise stated. Analyses are for modifi ed intention-to-treat analyses of participants with at least one measurement after baseline; baseline data missing for one patient in placebo group, therefore not analysed. Figure 1 shows number of patients attending the relevant visits. *Adjusted for age, sex, smoking status, systolic and diastolic blood pressure, HDL-cholesterol, LDL-cholesterol, and estimated glomerular fi ltration rate. †Sensitivity analysis.  (three taking metformin, one taking placebo), while estimated glomerular fi ltration rate fell to 30-44 mL/min per 1·73m² in fi ve participants (two taking metformin, three taking placebo) but did not fall lower than 30 mL/min/ per 1·73m² in any participants. Over 18 months, mean cIMT increased signifi cantly in both groups (0·024 mm per year [95% CI 0·014 to 0·033] for metformin, 0·017mm per year [95% CI 0·008 to 0·026] for placebo). Rates of cIMT progression in the metformin and placebo groups did not diff er signifi cantly (diff erence in slope 0·007 mm per year, 95% CI -0·006 to 0·020; p=0·29; table 2, fi gure 2A). This null cIMT fi nding was consistent in all sensitivity analyses (data not shown), in the per-protocol population (n=140; diff erence in slope 0·008 mm per year, 95% CI -0·005 to 0·022; p=0·23), and when restricted to participants fi ve or more images analysed from each side (data not shown). For 101 participants who completed the trial without reducing dose and who were compliant with study drug (45 in the metformin group and 56 in the placebo group), results were also much the same (diff erence in slope 0·014 mm per year, 95% CI -0·003 to 0·031; p=0·11). Analyses using thirds of baseline cIMT and thirds of selected baseline variables did not identify benefi t in any subgroups (appendix). Similarly, there was no diff erence according to sex. Scan reproducibility, assessed as SD for change in cIMT in the placebo group, was 0·046 mm at 12 months and 0·059 mm at 18 months.
Change in carotid plaque score in the carotid bulb, distal 10 mm of the common carotid artery, and proximal 10 mm of the internal carotid artery combined did not diff er between groups (0·01 per year, 95% CI -0·23 to 0·26; p=0·92; table 3, fi gure 2B). In the sensitivity analysis of carotid plaque score in the carotid bulb, entire common carotid artery, and internal carotid artery, treatment groups did not diff er signifi cantly (table 3).
HbA 1c , log fasting insulin concentration, log HOMA-IR, and tissue plasminogen activator were reduced in the metformin group compared with placebo, whereas we detected no signifi cant diff erence for total cholesterol, HDL-cholesterol, non-HDL-cholesterol, fasting plasma glucose, triglycerides, and high sensitivity C-reactive protein (table 4).
We recorded 258 adverse events: 138 in 64 participants assigned to metformin and 120 in 60 participants assigned to placebo (table 5, appendix). More patients  assigned to metformin developed diarrhoea and nausea or vomiting than those assigned to placebo (28 vs 5). 41 serious adverse events occurred, 14 in 11 participants assigned to metformin and 27 in 18 participants assigned to placebo. Few participants had cardiovascular events (n=10), were diagnosed with cancer (n=5), or developed type 2 diabetes (n=8). One participant had an out-ofhospital cardiac arrest and died from presumed cardiovascular causes in the fi rst few months of follow-up and therefore no data were available for analyses of the primary and secondary endpoints.

Discussion
Metformin-a drug already used for patients without type 2 diabetes-did not reduce progression of mean cIMT or carotid plaque score at 18 months in statintreated non-diabetic participants with established coronary heart disease and large waist circumferences. Metformin generated expected reductions in markers of diabetes risk together with notable reductions in measures of adiposity. Gastrointestinal adverse events were common in patients taking metformin. The results of twwo previous randomised trials-UKPDS 10 and HOME 12 -suggest that metformin reduces cardiovascular risk for patients with type 2 diabetes (panel). In UKPDS, the cardiovascular benefi t was greater than expected for the diff erences in glycaemic control achieved, suggesting that metformin might reduce cardiovascular risk for non-diabetic individuals by mechanisms independent of glucose-lowering. However, less than 2% of participants in UKPDS and 32% in HOME were treated with statins. All participants in CAMERA were taking statins, which might have limited the ability of metformin to show a reduction in cIMT. Similarly, although small randomised trials of metformin have shown reductions of total cholesterol and LDLcholesterol 8 in people who did not have diabetes, participants in CAMERA had no improvement for these predictors. A meta-analysis has also cast doubt on metformin's purported cardiovascular benefi ts. 24 Metformin can modestly reduce weight. 13 In CAMERA, participants who were selected partly on the basis of large waist circumference had sustained and progressive reductions for all measures of adiposity, similar to that achieved by weight loss drugs, 25 even in this non-diabetic group. Metformin can also reduce hepatic steatosis and biochemical markers of liver fat; 26 we recorded a reduction in γ glutamultransferase but not alanine aminotranferase concentrations. As in previous studies, 27 a moderate fall in vitamin B12 concentrations occurred in the metformin group. Tissue plasminogen activator was also lower with metformin than with placebo, consistent with previous studies, 8 suggesting that its purported cardiovascular benefi t could partially be a result of a reduction in prothrombotic potential. Unexpectedly, high sensitivity troponin T concentrations were higher in the metformin group than in the placebo group (though not signifi cantly) in the absence of change of renal function. This fi nding should be studied further.
CAMERA had several strengths, including its randomised placebo-controlled double-blind design; inclusion of patients with high cardiovascular risk, all taking statins as per best practice; ultrasound scans were analysed with semiautomated software and yearly cIMT progression rates were entirely consistent with large pooled studies, 28 suggesting that our method was robust;

Panel: Research in context
Systematic review Treatment with metformin was shown to reduce cardiovascular events in patients with type 2 diabetes by the initial 10 year follow-up of UKPDS, 10 with continued benefi t shown in a subsequent analysis of long-term follow-up over 25 years. 11 Systematic reviews and meta-analyses of the cardiovascular eff ects of metformin have produced mixed results depending on the inclusion criteria used. For example, a systematic review 23 published in 2008 found that metformin treatment reduced cardiovascular mortality by 26% in an analysis of 11 385 patients with type 2 diabetes from six trials. By contrast, a meta-analysis 24 published in 2012 with 13 110 participants with type 2 diabetes from ten trials found no change in cardiovascular mortality. Given that a generic form of metformin is available and has also been shown to reduce the risk of developing type 2 diabetes, 13 there is interest in its cardiovascular eff ects for individuals at high cardiovascular risk but without type 2 diabetes.

Interpretation
In the placebo-controlled CAMERA study, we evaluated the eff ect of metformin treatment on the progression of carotid intima-media thickness, carotid plaque score, and other surrogate markers of cardiovascular disease and type 2 diabetes in 173 participants without diabetes but with established coronary heart disease already taking statins. Over 18 months, neither carotid intima-media thickness nor carotid plaque score improved for patients taking metformin. However, metformin did lead to a signifi cant reduction of weight together with improvements in other risk factors for development of type 2 diabetes. Major cardiovascular outcome trials are needed to conclusively assess metformin's cardiovascular eff ects in people without type 2 diabetes-such trials are underway at present.  and we used metformin doses commonly used to treat type 2 diabetes, although some participants did reduce their dose. Potential weaknesses include use of cIMT-a surrogate marker albeit approved by the US Food and Drug Administration-as the primary endpoint; some drugs have reduced cIMT progression but not improved cardiovascular outcomes in large trials 29,30 whereas others have shown improvements, 31,32 and the relationship between cIMT progression and cardiovascular events in epidemiological studies has been questioned. 33 In addition, fewer participants than needed completed the trial (158 completed vs 180 needed according to our sample size calculation), although this calculation was based on two carotid ultrasound scans per participant, rather than three. Participants were followed up for 18 months rather than 24 months, although previous cIMT studies 22,34,35 have shown treatment eff ects over even shorter periods. Substantially larger multicentre multinational cIMT trials have been done 36 but CAMERA is similar in size to other single-centre cIMT trials 22 and benefi ted from a single sonographer and a single scan reader throughout. The number of participants who discontinued or reduced treatment, together with the use of several statistical tests, might have reduced our ability to detect meaningful eff ects. GLINT (ISRCTN34875079), with a feasibility phase commencing in 2013, is a double-blind randomised trail planned to assign roughly 12 000 people with non-diabetic hyperglycaemia and increased cardiovascular risk to metformin or placebo for 5 years with a cardiovascular primary outcome. Metformin's eff ect on cardiovascular surrogate markers is also being tested in other trials such as REMOVAL (assessing change of cIMT over 3 years in patients with type 1 diabetes; NCT01483560) and GIPS-III (assessing change in left ventricular ejection fraction for 4 months after acute myocardial infarction in non-diabetic patients; NCT01217307). 37 Further evidence is needed from large trials of cardiovascular outcomes before metformin can be recommended for cardiovascular benefi t for non-diabetic patients with high cardiovascular risk who are being treated with statins.
Contributors DP had the idea for and designed the study, collected, analysed, and interpreted data, wrote the fi rst draft and revised later drafts of the article, and obtained funding for the study. SML collected, analysed, and interpreted data, did the statistical analysis, and revised the article. IF analysed and interpreted data, did the statistical analysis, revised the article, and obtained funding. JJM analysed and interpreted data, revised the article, obtained funding, and supervised the study. RRH, MF, and CJP analysed and interpreted data, revised the article, and obtained funding. PW analysed and interpreted data, revised the article, and provided administrative, technical, and material support. NS had the idea for and designed the study, analysed and interpreted data, revised the article, obtained funding, and supervised the study.