What Is the Impact of Different Glucose-lowering Agents for Atrial Fibrillation Risk: A Systematic Review and a Network Meta-analysis

Background: The emergence of new glucose-lowering agents has brought revolutionary changes to the treatment of cardiovascular diseases. Diabetes is associated with atrial brillation (AF) and atrial utter (AFL) progression, while whether or not glucose-lowering agents would bring a reduction of AF/AFL is not clear. We therefore evaluate the effect of different glucose-lowering agents on AF/AFL and made this network meta-analysis to identify the optimal treatment for diabetes patients to reduce AF/AFL events. Methods: We searched PubMed, Embase, and the Cochrane Library until September 30 2020, and Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines were used in this network meta-analysis. The primary endpoint for our study was AF or AFL events. Only studies with a follow-up period of at least 12 months and reporting AF/AFL as clinical endpoints were included. Results from trials were presented as odds ratios (ORs) with 95% condence intervals (CIs) and were pooled using a bayesian random-effects model. Results: 5 eligible studies (9 glucose-lowering agents were analyzed including thiazolidinedione[TZD], metformin[Met], sulfonylurea[SU], insulin[Insu], dipeptidyl peptidase-4 inhibitor[DPP-4i], glucagon-like peptide-1 receptor agonist[GLP-1RA], sodium-glucose cotransporter 2 inhibitor[SGLT2i], alpha glucosidase inhibitor[AGI], and non-sulfonylurea[nSU]) consisting of 263583 patients with type 2 diabetes mellitus were included. Pooled results show that GLP1-RA, when compared to Met (OR 0.17, 95% CI 0.04-0.61), SU (OR 0.23, 95% CI 0.07-0.73), Insu (OR 0.20, 95% CI 0.07-0.86), and nSU (OR 0.18, 95% CI 0.04-0.66) signicantly reduce AF/AFL events. In addition, DPP-4i could also reduce AF/AFL events when compared with nSU (OR 0.33,

to exert potential direct cardiac protective effects. [7] Because aforementioned effect are also risk factors for the occurrence and development of AF/AFL, we hypothesized that new glucose-lowering agents with various characteristic would exert different effect on reducing the risk for AF/AFL events, and we made this network meta-analysis to identify the optimal treatment for diabetes patients to reduce AF/AFL events.

Methods
We have registered in PROSPERO (international prospective register of systematic reviews; (CRD42020212994) for this network meta-analysis which was designed according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (Supplement 1).

Study Selection Criteria
Clinical research comparing different kind of glucose-lowering agents in diabetes patients in which clinical outcomes (primary or secondary endpoints) were AF/AFL were considered for our present metaanalysis. The inclusion criteria for our study included (1) Clinical research with at least 2 comparator arms; (2) study population of patients with diabetes; (3) AF/AFL were reported as primary or secondary endpoints; (4) AF/AFL events were reported in individual glucose-lowering agents arm. (5) follow-up of at least 12 months. We excluded review article, single arm studies, duplicate studies, mechanism research and article where AF/AFL were reported as side effects. No language, publication date, or publication status restrictions were applied. We also screened references of prior systematic reviews and metaanalyses for related studies.

Search Strategy and Information Sources
We used keywords related to diabetes, atrial brillation, antihyperglycemic medications (thiazolidinedione, metformin, sulfonylurea, insulin, dipeptidyl peptidase-4 inhibitor, glucagon-like peptide-1 receptor agonist, sodium-glucose cotransporter 2, alpha glucosidase inhibitor, and non-sulfonylurea) to searched PubMed/MEDLINE, Ovid/Embase and Cochrane databases from database inception through the nal search date of September 30, 2020. The key words for search strategy are provided in the Supplement 2.
Two reviewers performed a systematic review, and disagreements were resolved in a panel discussion of 3 reviewers . Study selection involved screening of titles and abstracts followed by full-text evaluation of possible eligible studies.

Outcome Measures
The primary endpoint was AF/AFL de ned by the diagnosis of AF/AFL based on the ICD-9-CM code of 427.31; ICD-8-CM code of 427.93, 427.94, 427.3 or ICD-10-CM code of I48, in either an in-patient or outpatient department at least once.

Data Collection Process
Two reviewers independently extracted data on the study design, baseline characteristics, interventions, and outcomes. Any disagreements of collected information between the 2 reviewers were reconciled through discussion.

Quality Assessment and Publication Bias
Two independent reviewers performed the qualitative assessment using Cochrane Collaboration tool (in the Supplement 3). Given the limited number of publications, we did not assess the risk of publication bias.

Statistical Analysis
We tted a bayesian random-effects network meta-analysis model to simultaneously compare multiple glucose-lowering agents. We estimated odds ratios (ORs) of the treatment effects of the 2 individual glucose-lowering agents and the associated 95% credible intervals (CrIs) using Markov chain, Monte Carlo algorithms. All analyses were conducted using the gemtc package (version 0.8-7) and rjags package (version 4-10) in R, version 3.6.1 (The R Foundation). We assumed that the direct and indirect evidence for a treatment comparison had no discrepancy, called evidence consistency. To account for effect heterogeneity across trials, we allowed random effects to net-work meta-analysis and measured the magnitude of heterogeneity. We used the package's default setting including noninformative prior distributions with 4 parallel chains, where each chain consists of 50 000 samples after a 20 000-sample burn-in.
To evaluate and rank regimens, we calculated rank probabilities (ie, probability of an AGM being the best, second-best, or worst for an outcome) and the Surface under the Cumulative Ranking (SUCRA). The SUCRA is a numerical summary that accounts for both magnitude and uncertainty of the estimated effect for each regimen. A larger SUCRA value indicates better performance for the outcome We also made several sensitive analysis to show an class-effect of new oral glucose-lowering agents, detailed results of sensitive analysis can be found in Supplement 4.

Search Results
We have searched EMBASE, PubMed/Medline databases, and the Cochrane library for Randomized Controlled Trials (RCTs) and observational studies for our study purpose, and 199 individual studies were identi ed by using key words (Supplement 2) in our search strategies. After excluding duplicates, 182 studies were enrolled in next assessment. Of those, 170 studies were deemed irrelevant based on title and abstract screening. Then, investigators (S.W.C and Z.W.C) viewed all full text copies of potential 12 relatively studies. A third investigator (D.C.H) resolved any discordance in assessments. However, 2 of them did not provide available data for further analysis and 10 studies were included for qualitative synthesis. Finally, the main analysis included 5 articles after excluding another 5 studies by carefully group-discussion. (Figure 1

Sensitive analysis
As shown in the Supplement 4, we performed a number of additional analyses. We found similar outcomes when using an NMA in which GLP1-RA were compared with Met or insulin-providing therapy (Insu, SU, or nSU).

Discussion
To our knowledge, this manuscript is the rst net-work meta-analysis focusing on the effect of glucoselowering agents on AF/AFL prognosis. Our results indicated that new oral glucose-lowering agents, especially GLP-1RA, showed property to lower AF/AFL events among diabetes population. The association between diabetes and atrial brillation has been demonstrated for decades. [3,8,9] The potential mechanism is complex with different pathophysiology, such as in ammation [10] and oxidative stress, which would affect structural, electromechanical and mechanical myocardial remodeling. [11] According to latest evidence, [12][13][14][15][16] new oral glucose-lowering agents are recommended as rst-line prescription for diabetes patients with high risk for cardiovascular disease. [17] Mechanism beyond glucose-lowering might related to the reduction of cardiovascular risk.
The pooled results of our network meta-analysis comparing 9 glucose-lowering agents showed that GLP-1RA possessed the best property to prevent AF/AFL evets among diabetes patients. Glucagon-like peptide-1 (GLP-1) is an incretin hormone secreted in the gut in response to meal ingestion, which increases insulin secretion and inhibits glucagon production, targeting pancreatic b-cells. [18] Therefore, GLP-1RA could improve hyperglycaemia. Some research show that GLP-1RA could reduce matrix metalloproteinase-2 levels [19], inhibit proliferation [20] and paly role in anti-in ammatory effects. [21] In addition, GLP-1 receptor expression in ventricular cardiac myocytes [22] suggests that cardiac-direct actions of GLP-1RA may account for the reduction for AF/AFL events. However, in a separate study, [23] we did not nd that GLP-1RA has a similar effect in reducing AF/AFL events, which may be due to the more prominent edge effect caused by the meta-analysis.
Our analysis indicated DPP-4i play a protective role against AF/AFL only when compared with Nsu, although Chang's [24]research showed an association between DPP-4i use and AF/AFL events reduction. DPP-4i is incretin-based glucose-lowering therapy, it can inhibit the degradation of GLP-1 and increase the serum levels of GLP-1, which indirectly stimulate insulin secretion and enhance beta-cell function.
Research showed that DPP-4i exert antiarrhythmic effects by increasing the threshold of ventricular brillation. [25,26] This mechanism might related to the reduction of AF/AFL events in diabetes patients.
Our analysis did not show that SGLT2i perform an extra AF/AFL protective effect when compared with any other glucose-lowering agents. Signi cant reduction in re-hospitalization rate has been well clari ed, and SGLT2i is recommended as rst-line medication for diabetes patients. Although heart failure is associated with AF/AFL, SGLT2i did not decrease AF/AFL risk according to our results. This unexpecting results indicated that diuresis might paly critical role in improve cardiovascular outcomes, therefore protective effect on AF/AFL could be detected. [27] We noticed that the SCURA values of Met, SU, Insu, nSU were the lowest (ie, worst performance) which were associated with increased risk for AF/AFL compared with GLP-1RA. Both SU, Insu and nSU are insulin-providing therapies, and they seemed inferior to new oral glucose-lowering drugs (DPP-4i, GLP-1RA, SGLT2i) in reducing adverse cardiovascular endpoints. [28,29] This might be related to their single glucose-lowering effect.
We noticed that former studies concerned more on the bene ts of AGM on major adverse cardiovascular events (death, myocardial infarction and stroke), few of them pay attention to arrhythmic endpoints. Our pooled results provide a perspective that diabetes patients might bene t from different glucose-lowering agents in reducing AF/AFL events, and clinicians should choose glucose-lowering agents according to the risks classi cation for arrhythmia.
There are some limitations in our manuscript: (1) Because few studies focus on arrhythmia endpoints, only 1 RCT was enrolled in our network meta-analysis, this may affect the results. (2) Although all patients in our study suffered diabetes, the glucose-lowering therapy was decided by clinician experience and local medical policy, which leads to differences in patient baselines between studies.

Conclusion
The nding of our study indicated that GLP1-RA could be optimal glucose-lowering agent for diabetes patients to prevents AF/AFL. Met and insulin-providing therapy (insulin, sulfonylurea, or nonsulfonylurea) should be avoided to patients with high risk of AF/AFL.