Metformin Use and Risk of All-Cause Mortality and Cardiovascular Events in Patients With Chronic Kidney Disease—A Systematic Review and Meta-Analysis

Background To evaluate whether metformin use assuredly alters overall all-cause death in patients with type 2 diabetes mellitus (T2DM) and chronic kidney disease (CKD). Methods Pubmed, Web of Science, Embase, and Cochrane Central Register of Controlled Trials were systematically searched from inception to Feb. 29, 2020 with no language restriction. All related articles comparing all-cause death of T2DM and CKD patients after metformin use (monotherapy or combination) versus non-metformin treatment were identified. Pooled risk ratios (RR) and 95% confidence intervals (CI) were computed using random-effects models regardless of the heterogeneity quantified by Cochrane χ2 and I2 statistics. Results Totally 13 studies (9 cohort studies [CSs], 3 subanalyses or post-hoc analyses of randomized controlled trials [RCTs], and 1 nested case-control article) involving 303,540 patients were included. Metformin-based treatments relative to any other measure displayed significantly lower risks of all-cause mortality (Pooled RRs 0.71, 95%CI 0.61 to 0.84; I2 = 79.0%) and cardiovascular events (Pooled RRs 0.76, 95%CI 0.60 to 0.97; I2 = 87.0%) in CKD patients at stage G1-3, with substantial heterogeneity. Metformin use was not significantly related with these end points in advanced CKD patients. Conclusions Metformin use is connected with significantly less risks of all-cause mortality and cardiovascular events in patients with T2DM and mild/moderate CKD. However, RCTs with large sample sizes are warranted in the future to assess whether these key benefits extend to later stages of CKD by dose adjustment.


INTRODUCTION
Chronic kidney disease (CKD) is a global public health issue and gradually becomes more prevalent, mainly due to an increment in the diagnosis of type 2 diabetes mellitus (T2DM) (1,2). CKD attacks about 43% of patients with type 2 diabetes, and its incidence is up to 61% in the diabetic people of US at age 65 years and above (3). Metformin is a biguanide diabetic drug that delays intestinal glucose absorption and glucose generation in the liver and intensifies insulin sensitivity, since not all effects of metformin are clear so far (4). Metformin is commonly used in the first-line for type 2 diabetic patients because of its low cost, effectiveness, weight independence, and cardiovascular benefits (5,6). However, metformin has been banned from kidney dysfunctional patients because of the suspected risk of lactic acidosis. Pharmacologically, after 3b CKD patients took 1,000 mg/d and G4 CKD patients used 500 mg/d for 4 months, the serial blood metformin levels were always below the upper limit of normal level, and lactate levels were kept at ≤ 5.0 mmol/L (7). Similarly, two recent systematic reviews reported a lack of evidence that metformin relative to other antidiabetic drugs can raise the occurrence of lactic acidosis (8,9). In another systematic review, metformin intake decreased the all-cause deaths of moderate CKD patients (10), but this review had many limitations, including the lack of subgroup analysis, and ignorance of heterogeneity and publication bias. Evidence now warns that gentle to mild CKD people should use metformin carefully (9,11).
Nevertheless, metformin is yet prescribed to elderly advanced-CKD patients with DM in Canada (12). It is unknown whether metformin is safe for advanced CKD patients (13,14), and more recent instructions suggest careful use for these patients until more confirmed evidence about its safety is reported (15). Recent proofs also give rise to conflicting results about metformin use with the presence of both type 2 diabetes and advanced CKD. As reported, supplementation with metformin is probably unlinked to a lower all-cause death rate among advanced CKD patients (15,16). However, it is also believed that mortality rates will considerably differ after metformin use (17,18). In particular, some undetected confounders (e.g., study design, renal dysfunction criteria, ignorance of model adjustment, dissimilar comorbidity profile) may result in bias. For these reasons, we planned to explore how metformin will impact the risk of death rates reported among the existing and up-to-date studies.

Search Scheme and Inclusion Criteria
Bibliographical searching on EMBASE, Pubmed, Web of Science, and Wiley Cochrane Central Register of Controlled Trials was finished independently by Y.H. and M.L. (any date up to 29 February, 2020) as per relevant guidance (19). Disagreement was resolved by consensus or by consultation with a third investigator. The study types of interest were clinical trial, observational study (OS), post-hoc analysis or subanalysis of randomized controlled trial (p-h/sa of RCT), and review. No language limitation was applied. The broad key words included "metformin", "Glucophage", "dimethylbiguanidine", "dimethylguanylguanidine", "dimethylbiguanidium"; "chronic kidney disease", "CKD", "chronic renal failure", "CRF", "chronic renal dysfunction", "renal insufficiency"; "dialysis" and "NIDDM", "type 2 diabet*", "type II diabet*", "T2D", "T2DM"; "death", "die", "dead", "decease", "mortality", and "mortalit*". The search scheme was elaborated in Additional file 1 and the consistency with PRISMA Statement for reporting Systematic Review and Meta-analysis was illustrated in Additional file 2. When detailed information needed for the analysis was unavailable, the original authors will be contacted through e-mail to obtain the missing information.

Inclusion and Exclusion Criteria
Inclusion criteria were: (I) RCT, p-h/sa of RCTs, or OSs; (II) provision of endpoints for all-cause mortality and cardiovascular events in patients with T2DM and CKD (regardless of CKD stage or dialysis type) with or without metformin use. The tested outcomes were assured by physical tests and hospital records, or recognized from links of administrative records. The exclusion criteria were: (I) kidney transplant; (II) case report, comment, editorial, letter, quasi-experiment (non-random subject assignment), or unpublished study; (III) abstract or conference proceeding. Of two or more articles from the same team or organization, only the one with latest publication or largest sample size was selected.

Data Isolation and Quality Assessment
Two investigators independently extracted all information of interest in standardized form, including demographic characteristics of patients (age, gender, race/ethnicity), stage of CKD, duration of DM, and follow-up duration. Other data of concern included study information (selection criteria), type, intervention, and clinical outcomes (all-cause death, cardiovascular events), analysis scheme (statistical models, adjusting variables), and effect levels (hazard ratio [HR], risk ratio[RR], odds ratio [OR] or raw data for re-computation). We acquired effect estimates from models with minimal or full adjustment and reporting the adjusting variables.
The quality of each p-h/sa of RCT or OS was assessed by Y.H. and M.L. with a 9-star Newcastle-Ottawa scale (20,21), and high quality was implied by a score > 6 stars. The evidence quality of each outcome among the enrolled articles was assessed as deficient, low, modest or high by YH and ML independently according to Grading of Recommended Assessment, Development and Evaluation (22). Any inconsistency between them was addressed via discussion.

Statistical Analysis
The first and second end points were all-cause mortality and cardiovascular events, respectively, as per the staging of CKD. Dichotomous outcomes were synthesized using HR, RR or OR with 95% confidence interval (CI). If an article presented more than one result of the relationship between metformin intake and risk of key clinical outcomes as per subtypes (e.g., by weaker kidney function), a within-article overall estimate was acquired. Natural logarithm of RRs (logRRs) and standard errors were pooled by DerSimonian and Laird's approach in a randomeffects model. The 95%CI with null "1" indicated no clinical significance even if P < 0.05.
Given the a-priori dissimilarity of the enrolled studies, stratified analyses by study design (p-h/sa of RCT, OS) was pre-specified if the data were enough. I 2 ≥50%, 25%-50%, and <25% indicated high, gentle and low inter-study heterogeneity respectively. Any identifiable source of dissimilarity was found out by sensitivity test, in which the articles were removed one by one.
Publication bias was examined by Begg's and Egger's tests when 5 or more studies were present for analysis, and by visually testing the dissymmetry of funnel plots of estimated effects versus standard errors (23). Any publication bias (P<0.10) was processed by Duval & Tweedie's trim-and-fill approach. Other statistical analyses were finished on STATA 12.0 (STATA College Station, USA) at significance level of two-sided P < 0.05.
Sensitivity analysis uncovered that heterogeneity did not disappear after the deletion of single studies. Significant publication bias was found by Egger's test (P=0.043), but not by funnel plot inspection or Begg test (P=1) ( Figure S1). During publication bias exploration based on the trim-and-fill approach, the probable missing data were not replaced, so the results were    Six trails reported all-cause mortality in advanced CKD patients (15-18, 25, 32). Metformin use had no significant therapeutic effect on all-cause death (pooled RR: 0.80; 95%CI: 0.49, 1.31), with heavy between-study heterogeneity (I 2 = 86.7%; P<0.001; Figure 2). The effect was still insignificant in the stratified analysis by study design ( Figure S2). The heterogeneity did not disappear after the exclusion of any single study.
The funnel plots showed no evident systematic bias between all-cause death and advanced CKD (Begg test, P= 0.707, Egger's test, P=0.130; Figure S3).
The evidence quality that metformin prevented cardiovascular events at stage G1-3 of CKD as determined with GRADE was "low".

DISCUSSION
All existing studies with 303,540 patients were searched, but evident positive relationship between metformin intake and allcause death was found only at stage G1-3 of CKD. The risk reduction may be partially ascribed to the cardiovascular protective effects of metformin on coherent improvement in hypertension, dyslipidaemia and body mass index, despite the absence of HbA1c reduction found in one study (30). Metformin may decrease mortality following several biological mechanisms (33,34): by lowering the insulin and insulin-like growth factor 1 signaling, mechanistically inhibiting rapamycin, weakening inflammation, reducing reactive oxygen species generation or DNA damage, or stimulating AMP-activated protein kinase. Metformin can experimentally initiate AMP-activated protein kinase and thus prevent diabetic lipoapoptosis of human coronary artery endothelial cells. Metformin can remove glycocalyx barrier in db/db mice, and relieve Ang-II-caused atheromatous plaque generation and aortic aneurysm in ApoE (−/−) mice. Metformin intake can remarkably weaken neointimal hyperplasia in fructose-caused insulin-resistant rats and myocardial reshaping and neutrophil recruitment after myocardial infarction in rats. Metformin also leads to a less risk of atrial fibrillations in T2DM patients probably by relieving myolysis and oxidative stress after atrial cell tachycardia.   Metformin versus non-metformin treatment resulted in 17% and 24% significant drop in all-cause death in CKD patients at stage G3A and G3B respectively. The risk reduction may not be ascribed to the cardiovascular protective effects of metformin use, which was seen in this review. A recent systematic review on health outcomes and conflicts related to metformin therapy summarized that metformin use can improve key clinical outcomes in T2DM patients with moderate CKD (10), which is similar to our result. This finding was partially based on the results from two OSs that presented death by CKD stage, which, however, was rated as possessing a modest risk of bias. According to Swedish National Diabetes Register, metformin-based measures resulted in a less risk of death in people with G3a CKD, but not with G3b CKD (HR 1.02, 95% CI 0.84-1.24) (28). Moreover, a cohort study enrolling American patients registered in Veterans Health Administration found that metformin use largely lowered the risk of all-cause mortality with the presence of eGFR at 45-59 mL/min/1.73 m 2 (HR 0.80, 95%CI: 0.71-0.91), but not at 30-44 ml/min/1.73m 2 (0.72, 0.51-1.01) (31). Whitlock et al. found metformin use was unrelated with smaller risk for allcause death in CKD patients at stage of either G3A (HR 0.85, 95%CI: 0.46-1.57) or G3B (0.92, 0.53-1.61) (32). The deletion of one study (28) a bit changed the overall risk result in the random-effects model, and almost no heterogeneity was found. The differences can be ascribed to the discrepancies in study populations and comparator therapy.
No significant relationship was identified between metformin use and the end points of all-cause death or cardiovascular events in advanced CKD patients. The reasons why the mortality benefit from metformin versus any other measure is not considerable at stage of advanced CKD are unclear. One probable reason is that the cause of advanced CKD may be other than diabetic nephropathy and thus this comorbid condition may offer extra death risk factors that do not respond to OHA treatment, therefore restricting the death benefit of metformin seen in this population. Another possible reason is that the follow-up duration was too short to correctly uncover the actual chronic death benefits. A third potential reason might be that the number of patients with advanced CKD or ESKD on metformin is too small, and most studies involved less than 200 of patients with advanced CKD or ESKD on metformin except for one study (15). It is well-known that studies with smaller sample sizes might, very likely, lack sufficient statistical power to reveal the true association.
Heterogeneity of OSs cannot be avoided in any review and metaanalysis, owing to the differences in study design, patient characteristics and statistical approaches. Even strict criteria were used, and the enrolled articles stand for a comprehensive try to pool together published and unpublished studies of interest. Hence, single article effects were judged at the summary-level. Furthermore, the statistical approaches used in the OSs did not well solve the impact of unmeasured confounding elements on the overall effect estimation. Sensitivity analyses revealed the origins of heterogeneity were ascribed to the differences in study populations and comparator use.
This review has some merits, such as the large sample size, broad search scheme, and the accordance to Cochrane guidelines. Nevertheless, it also has some shortcomings. First, the absence of RCTs or unpublished articles may have biased our conclusions. Second, advancing patients who more likely to switch to glucosereducing therapy during the study period were not excluded, which diluted the observations. Third, comparator uses were diverse and some new antidiabetic agents (e.g., sodium-glucose cotransporter-2 inhibitors, glucagon-like peptide 1 agonists) were basically not used by low-eGFR patients, so we cannot assess the impacts of metformin relative to these agents. Fourth, data were only about the dose and duration effect of metformin, case numbers, and person-years for FIGURE 5 | Forest plot of the risk of cardiovascular events in patients with type 1 diabetes mellitus (T2DM) and moderate chronic kidney disease (CKD) according to estimated glomerular filtration rate (eGFR). T2DM, type 1 diabetes mellitus; CKD, Chronic kidney disease; eGFR, estimated glomerular filtration rate.
dose-reaction meta-analyses. Fifth, we cannot well clarify the cause of CKD. As probably expected, the risks of these key clinical outcomes observed in diabetic CKD patients were largely different among the OSs, but we cannot conduct subgroup analysis because of the limited number of trails. Furthermore, Egger's test for all-cause death in G1-3 CKD patients uncovered possible publication bias, which cannot be easily ascertained. In all, we may overestimate the real effect if insignificant articles were missed.
In conclusion, metformin use is related to significantly less risks of all-cause mortality and cardiovascular events in mild/ modest CKD patients. However, RCTs with large sample sizes are warranted in the future to assess whether these key benefits extend to later stages of CKD by dose adjustment.

DATA AVAILABILITY STATEMENT
All datasets presented in this study are included in the article/ Supplementary Material.