A systematic review and meta-analysis of the protective effects of metformin in experimental myocardial infarction

Metformin improves cardiovascular prognosis in patients with diabetes mellitus, compared to alternative glucose-lowering drugs, despite similar glycemic control. Direct cardiovascular protective properties have therefore been proposed, and studied in preclinical models of myocardial infarction. We now aim to critically assess the quality and outcome of these studies. We present a systematic review, quality assessment and meta-analysis of the effect of metformin in animal studies of experimental myocardial infarction. Through a comprehensive search in Pubmed and EMBASE, we identified 27 studies, 11 reporting on ex vivo experiments and 18 reporting on in vivo experiments. The primary endpoint infarct size as percentage of area at risk was significantly reduced by metformin in vivo (MD -18.11[-24.09,-12.14]) and ex vivo (MD -18.70[-25.39, -12.02]). Metformin improved the secondary endpoints left ventricular ejection fraction (LVEF) and left ventricular end systolic diameter. A borderline significant effect on mortality was observed, and there was no overall effect on cardiac hypertrophy. Subgroup analyses could be performed for comorbidity and timing of treatment (infarct size and mortality) and species and duration of ischemia (LVEF), but none of these variables accounted for significant amounts of heterogeneity. Reporting of possible sources of bias was extremely poor, including randomization (reported in 63%), blinding (33%), and sample size calculation (0%). As a result, risk of bias (assessed using SYRCLE’s risk of bias tool) was unclear in the vast majority of studies. We conclude that metformin limits infarct-size and improves cardiac function in animal models of myocardial infarction, but our confidence in the evidence is lowered by the unclear risk of bias and residual unexplained heterogeneity. We recommend an adequately powered, high quality confirmatory animal study to precede a randomized controlled trial of acute administration of metformin in patients undergoing reperfusion for acute myocardial infarction.

k for roval cardiovascular morbidity and mortality, compared with alternative glucoselowering drugs. These observations suggest that metformin exerts direct protective effects on the heart, independent of its glucose-lowering action.
A number of animal studies (mostly performed between 2002 and 2011) indeed show protective effects of metformin in animal models of myocardial ischaemia-reperfusion injury and cardiac remodeling. However, recent randomized clinical trials have shown no effect of metformin on cardiovascular outcomes in non-diabetic patients after myocardial infarction or CABG. This raises the question why the protective effects of metformin appear to have translated from bedside to bench, but not back to bedside. The preclinical evidence has not yet been systematically reviewed. The internal and external validity of the preclinical studies, as well as possible publication bias, may have influenced the translational value of the animal studies. We aim to address these matters in the present systematic review. Specify (a) the number of reviewers assessing the risk of bias/study quality in each study and (b) how discrepancies will be resolved At least two reviewers will assess the risk of bias and study quality of all studies reporting on one of the outcome measures selected for meta-analysis. Discrepancies will be resolved by discussion.

38.
Define criteria to assess (a) the internal validity of included studies (e.g. selection, performance, detection and attrition bias) and/or (b) other study quality measures (e.g. reporting quality, power) □By use of SYRCLE's Risk of Bias tool 4 X By use of SYRCLE's Risk of Bias tool, adapted as follows: additional scoring of reporting of study quality indicators "reporting of any randomisation", "reporting of any blinding" , "reporting of temperature regulation", "reporting of a power calculation" and "reporting of a conflict of interest statement". □By use of CAMARADES' study quality checklist, e.g 22 □By use of CAMARADES' study quality checklist, adapted as follows: □Other criteria, namely: Collection of outcome data 39.
For each outcome measure, define the type of data to be extracted (e.g. continuous/dichotomous, unit of measurement)  IS/AAR %, continuous in %  (HS) Troponin I, continuous in ng/ml  LVEF, continuous in %  LVESD, continuous in mm3  Mortality, incidence  Cardiac hypertrophy (various units of measurement possible, method of extraction to be determined)

40.
Methods for data extraction/retrieval (e.g. first extraction from graphs using a digital screen ruler, then contacting authors) 1. Direct extraction of data from tables, text and figures 2. Extraction from graphs using digital screen ruler 3. Contacting authors by e-mail for original data if data not reported or unclear All data will be collected as mean and standard deviation (SD). Standard error of the mean will be recalculated to SD. In case the number of animals is unclear and cannot be retrieved, a conservative estimate will be made. In case the data are reported as median and interquartile range, the authors will be contacted for raw data. In case an outcome was measured at multiple time points, the measurement of greatest efficacy will be chosen. Multiple treatment regimes from one study will be extracted as separate comparisons.
In case of missing data and no author contact details, or no response from authors within 3 weeks including a reminder, the study will be omitted from analysis.

41.
Specify (a) the number of reviewers extracting data and (b) how discrepancies will be resolved One reviewer will extract the data, a second reviewer will check the extracted data for inconsistencies.
Specify (per outcome measure) how you are planning to combine/compare the data (e.g. descriptive summary, meta-analysis) Meta-analysis will be performed for all selected outcomes reported in three or more studies, but in case of high heterogeneity studies will not be pooled. If less than three studies report on a selected outcome, a descriptive summary will be provided.

43.
Specify (per outcome measure) how it will be decided whether a metaanalysis will be performed A meta-analysis will be performed if ≥3 studies report on a specific outcome measure. For subgroup analysis a minimum of 3 studies per subgroup is required. If a meta-analysis seems feasible/sensible, specify (for each outcome measure):

44.
The effect measure to be used (e.g. mean difference, standardized mean difference, risk ratio, odds ratio) For IS/AAR and LVEF, the raw difference in means will be used, since these are relative outcome measures expressed as a %. For troponin and LVESD, we aim to use a normalized mean difference, if there are sham or baseline data available for the selected outcome measures. If (for any of the OMs) such data are not reported in the majority of studies, we aim to use a standardized mean difference. For mortality, an odds ratio will be calculated.

45.
The statistical model of analysis (e.g. random or fixed effects model) Random effects model for all outcome measures 46.
The statistical methods to assess heterogeneity (e.g. I 2 , Q) (residual) I 2 and adjusted R 2 for all outcome measures 47.
Which study characteristics will be examined as potential source of heterogeneity (subgroup analysis) -Animal species (stratified per species) -Sex (stratified m vs f vs mixed vs not reported) -Dose of metformin (linear or stratified) -Timing of metformin treatment (linear or stratified) -Co-morbidity (stratified y/n) -Injury model (stratified by type, e.g. in vitro vs in vivo) -reporting of randomisation (stratified y/n) -reporting of blinding (stratified y/n)

48.
Any sensitivity analyses you propose to perform Choose 1 specific time-point for outcome measure, instead of choosing the time-point of greatest efficacy. Perform SMD analysis instead of MD or NMD if applicable.

49.
Other details meta-analysis (e.g. correction for multiple testing, correction for multiple use of control group) We will perform a Holm-Bonferroni correction on the pvalue, depending on the number of subgroup analyses performed. Correction for multiple use of control group will be performed by dividing the number of animals in the control group by the number of comparisons performed with this control group.

50.
The method for assessment of publication bias Produce funnel plots and perform visual analysis of these plots. We are aware that funnel plots of SMD are susceptible to distortion and will omit the assessment of publication bias if this is suspected for our dataset. In addition, we aim to perform Trim and Fill analysis and Egger's test for small study effects for outcome measures containing 20+ studies