Cardiovascular outcomes in trials of new antidiabetic drug classes: a network meta-analysis

MEDLINE, EMBASE, the Cochrane database, ClinicalTrials.gov, and congress proceedings from recent cardiology conferences were searched up to April 20, 2019 without applying any language and publication status restrictions. Our search threads were limited to large outcome trials evaluating new antidiabetic drug classes, namely, GLP-1 RAs, SGLT-2 or DPP-4 inhibitors in patients with T2DM. The search keywords included “glucagon-like peptide-1 receptor agonist”, “sodium-glucose co-transporter 2 inhibitor”, “dipeptidyl peptidase-4 inhibitor”, “major adverse cardiovascular event (MACE)”, “cardiovascular risk”, “cardiovascular event”, “type 2 diabetes mellitus”, and their synonyms and related keywords.

Studies were eligible to be included if they met the following inclusion criteria for this network meta-analysis: (1) phase III/IV CVOTs or renal outcome trials evaluating cardiovascular outcomes; (2) allocation of GLP-1 RAs, SGLT-2 or DPP-4 inhibitors; (3) patients with T2DM (≥ 18 years of age); (4) patients with established cardiovascular disease or cardiovascular risk factors; (5) report of the rate of MACE and deaths.

Two investigators (YF and MFT) independently performed literature search and selection. Discrepancies were resolved by consensus. For eligible studies, information about the year of publication, age, gender, sample size, body weight, ethnic group, smoking status, intervention, duration of follow-up, glycated haemoglobin, baseline comorbidities of cardiovascular disease or cardiovascular risk factors, duration of diabetes, and duration of follow-up were extracted. Quality of the included studies and risk of trial bias were assessed using the Cochrane risk of bias assessment tool.

The primary outcome in our network meta-analysis was MACE, defined as the composite of cardiovascular mortality, nonfatal MI, and nonfatal stroke. Secondary outcomes included the individual components of the primary outcome, all-cause mortality, hospitalisation for heart failure (HF), and renal composite outcome, defined as a composite of adjudication-confirmed end-stage renal disease, death due to renal failure, or a sustained decrease of at least 40% in estimated glomerular filtration rate (eGFR) from baseline to less than 60 ml per minute per 1.73 m² of body-surface area [19]. Definitions of the renal composite outcome in the included trials are listed in Additional file 1: Table S1. We followed the definitions of outcomes used in each trial.

We used both a frequentist approach [20] and a Bayesian framework [21] with non-informative priors to compare the effect of different antidiabetic drug classes on outcomes at the trial level. Count data were extracted to analyse study outcomes. Pooled random-effects odds ratios (ORs) and 95% confidence intervals (95% CIs) were used as the summary statistics. A 95% CI not including 1.00 or a two-tailed p value less than 0.05 was considered statistically significant. Forest plots comparing relative treatment effects were generated with a frequentist approach using the statistical package ‘netmeta’ (version 0.9–8, https://cran.r-project.org/web/packages/netmeta/index.html↗) in R (version 3.3.2).

Sensitivity analysis was performed using the fixed-effects model to avoid small-study effects and was conducted by assessing the effect of excluding individual trials. Inconsistency in the network was evaluated using loop-specific heterogeneity estimates; τ² estimate of 0.04, 0.14, and 0.40 represented a low, moderate, and high degree of variance, respectively. Heterogeneity within each pairwise meta-analysis of the direct comparisons of antidiabetic drug classes was assessed by I² statistic. I² < 25%, 25% to 50%, and > 50% corresponded to mild, moderate, and severe heterogeneity, respectively. Potential publication bias was assessed using funnel plots, Begg’s, Egger’s, and trim-and-fill tests for direct comparisons with three or more studies.

We evaluated the consistency of inferential estimates from hierarchical modelling using Markov chain Monte Carlo simulations, which were performed with 1000 tuning iterations and 5000 simulation iterations within a Bayesian framework using R statistical package ‘gemtc’ (version 0.8–2, https://cran.r-project.org/web/packages/gemtc/index.html↗) and ‘rjags’ (version 4–6, https://cran.r-project.org/web/packages/rjags/index.html↗) to minimise Monte Carlo error. The relative probability of each antidiabetic drug class having the largest effect size on each outcome was generated using P-rank scores.

What is new in this network meta-analysis is the confirmation that SGLT-2 inhibitors are now superior to the other antidiabetic drug classes in terms of cardiovascular and renal endpoints, especially deaths, making it arguably the drug class of choice. Our findings on SGLT-2 inhibitors are consistent with recent meta-analyses of CVOTs on this drug class [17, 39], in which SGLT-2 inhibitors were found to reduce MACE, cardiovascular mortality, hospitalisation for HF, and progression of renal disease in patients with atherosclerotic cardiovascular disease. However, through network meta-analysis, we could draw comparisons with other drug classes. By including all the relevant trials completed recently, we extended the findings of previous meta-analyses by ourselves and other researchers [11, 17, 39–45]. The significant benefits of SGLT-2 inhibitors in reducing MACE, hospitalisation for HF, and renal events were not only confirmed when compared to placebo but also identified when compared to the other new antidiabetic drug classes. There are several possible mechanisms behind these benefits. The natriuretic response through SGLT2 inhibition in the proximal tubule and the inhibition of the tubuloglomerular feedback may play an important role in the reduction in hospitalisation for HF and the progression of diabetic kidney disease [46]. The weight loss and reduction in blood pressure may contribute to the cardiovascular protection of SGLT-2 inhibitors [47]. Left ventricular (LV) diastolic dysfunction is strongly associated with HF in T2DM patients. The improved LV diastolic function [48] and reduced tissue sodium content [49] with dapagliflozin lend support to these possible mechanisms. In addition, SGLT-2 inhibitors inhibit the Na +/H + exchanger (NHE) 1 in the myocardium results and NHE3 in the proximal tubule, resulting in decreased cytoplasmic sodium and calcium levels while increasing mitochondrial calcium level and inhibiting sodium reabsorption, all of which help to explain the reduction in HF with this drug class [46, 50]. The cardiac antifibrotic effects of SGLT-2 inhibitors found in rat models with dapagliflozin [51] and in human cardiac fibroblasts with empagliflozin [52] may also explain the protective effect of SGLT-2 inhibitors on HF.

DECLARE-TIMI 58 [14] confirmed the benefit of reduction in HF hospitalisation with SGLT-2 inhibitors, but did not show a significantly lower rate of MACE which contrasted with the findings from the EMPA-REG OUTCOME [10], CANVAS [15] and CANVAS-R [15] trials (Table 1, Additional file 1: Table S15). The differences in patient characteristics and drug used across trials could be possible explanations. More than half of the T2DM patients enrolled in DECLARE-TIMI 58 did not have cardiovascular disease but only had risk factors. On the contrary, almost all the patients included in EMPA-REG OUTCOME and more than half of the patients included in CANVAS and CANVAS-R were T2DM patients with established cardiovascular disease. It was more difficult to demonstrate a reduction in MACE in patients without pre-existing cardiovascular disease in DECLARE-TIMI 58. It is possible that dapagliflozin used in this trial has less protection against MACE compared to empagliflozin and canagliflozin. Nevertheless, the reduction in HF hospitalisation with dapagliflozin was consistent across multiple subgroups regardless of history of atherosclerotic cardiovascular disease, HF and CKD. This was also found in other studies of dapagliflozin in humans [48, 49], in animals [51], and in vitro [52]. These diverse pieces of evidence point to the specific protection of HF by dapagliflozin.

All the CVOTs on SGLT-2 inhibitors except EMPA-REG OUTCOME [28] showed robust reductions in renal composite outcomes regardless of baseline factors. The renal benefits of SGLT-2 inhibitors were further confirmed in our network meta-analysis after including CREDENCE [18], the first renal outcome trial. Now the application of SGLT-2 inhibitors can be expanded to T2DM populations with CKD in addition to cardiovascular diseases. The renal protection of SGLT-2 inhibitors needs to be further confirmed in the ongoing renal outcome trial DAPA CKD [36], which recruited CKD patients with or without diabetes. This trial together with CREDENCE will answer whether SGLT-2 inhibitors could be considered a promising option for renal protection in either T2DM or nondiabetic patients.

Most importantly, SGLT-2 inhibitors were found to be significantly better in reducing cardiovascular and all-cause mortality regardless of existing cardiovascular disease or CKD in patients with T2DM. The superiority of SGLT-2 inhibitors over GLP-1 RAs and DPP-4 inhibitors in reducing deaths can, therefore, be considered a class effect. However, the difference in overall mortality did not reach statistical significance (OR 0.83, 95% CI 0.68–1.02) in the CREDENCE trial [18], as with other trials on SGLT-2 inhibitors [14, 15], which differed from the findings in the EMPA-REG OUTCOME trial [10] and this network meta-analysis. Despite the high rate of drug discontinuation (27% of participants), the CREDENCE trial was stopped early because of the overwhelming benefits in lowering the risk of kidney failure and cardiovascular events. As a result of the shortened follow-up period, the study lacked power to demonstrate any mortality reduction. In contrast, EMPA-REG OUTCOME showed a reduction in mortality, so there was heterogeneity in this outcome, although the indirect evidence also supported a reduction in mortality. The protection against deaths observed with empagliflozin but not the other SGLT-2 inhibitors needs to be further explored in future trials to understand whether it is a specific effect of empagliflozin.

The favourable cardiovascular safety of GLP-1 RAs, especially their positive effects in reducing MACE, nonfatal stroke, and nonfatal MI, were revealed in our network meta-analysis after including the latest evidence. The reported significant decrease in cardiovascular and all-cause mortality in GLP-1 RA recipients in the LEADER trial [9] was not found in our network meta-analysis or other CVOTs on GLP-1 RAs. Although the improvement in cardiovascular mortality was reported in a recent meta-analysis of the CVOTs on GLP-1 RAs, Harmony Outcomes trial was not included while the results were largely driven by the LEADER trial [53]. The shorter duration of CVOTs besides LEADER reduced their power to show an effect on cardiovascular deaths. Moreover, trial results are also affected by patient characteristics and the different GLP-1 RAs used across the trials.

Liraglutide reduces postprandial glucose without increasing insulin concentration and improves beta-cell function in T2DM patients [54]. Its cardiovascular protective effects have been studied. Reduction in arterial stiffness, LV myocardial strain, twisting and untwisting, N-terminal pro-brain natriuretic peptide, and oxidative stress with liraglutide could offer myocardial protection and prevention of diabetic heart disease [55]. Moreover, liraglutide reduces postprandial non-esterified free fatty acids and suppresses soluble vascular cell adhesion molecule-1 compared to metformin [56]. It protects against acute liver injury in the mouse [57]. However, cardiovascular benefits have not been observed in studies of exenatide [13, 58]. In low-risk patients undergoing scheduled coronary artery bypass grafting surgery, postoperative intravenous exenatide provided no additional cardio-protective effect compared to intravenous insulin [58]. Therefore, it is still necessary to study the cardiovascular efficacy and safety profile of GLP-1 RAs so as to confirm their place in clinical practice.

There are several ongoing CVOTs evaluating the efficacy and safety of GLP-1 RAs in different patient populations with different drug formulations. The FREEDOM-CVO trial is the first trial studying ITCA 650, the first injection-free GLP-1 RA with a full year of treatment delivery from a single placement of an osmotic mini-pump [31]. In the REWIND study on dulaglutide, 70% of patients did not have established cardiovascular disease at baseline [32]. The PIONEER 6 trial investigating oral semaglutide recently announced an insignificant reduction in MACE but a significantly lower risk of cardiovascular and all-cause deaths [33]. These trials will add to the evidence on the cardiovascular effects of GLP-1 RAs, and clarify their effects on mortality.

The cardiovascular benefits observed in GLP-1 RAs were not found in another incretin-based drug class, namely DPP-4 inhibitors. Our network meta-analysis including the latest evidence from CARMELINA [16] still failed to demonstrate cardiovascular benefits with DPP-4 inhibitors, like previous meta-analyses [11, 41, 59–61]. Although a few meta-analyses reported substantial reductions in MACE with DPP-4 inhibitors [62, 63], their conclusions were based on studies with small sample sizes and limited numbers of cardiovascular events, which might be underpowered.

Although no increase in the risk of cardiovascular events was found with DPP-4 inhibitors, a red flag was raised in terms of increased HF. Saxagliptin was observed to result in a significant increase in hospitalisations for HF in the SAVOR-TIMI 53 trial [26]. Alogliptin in the EXAMINE trial [25] showed a trend toward an increased risk for HF hospitalisation while sitagliptin in the TECOS trial showed no increased risk [23]. In the latest CARMELINA, no increase in HF hospitalisation and renal composite outcome, and a significant reduction in microvascular events driven primarily by the reduction in albuminuria progression (p = 0.003) were found with linagliptin. It established the cardiovascular safety profile of linagliptin in diabetic patients at high risk of cardiovascular and kidney disease, those most vulnerable to developing HF. The conflicting effects of DPP-4 inhibitors on hospitalisation for HF need to be reconciled when more evidence becomes available.

In our study, GLP-1 RAs were better than DPP-4 inhibitors in reducing hospitalisation for HF. A retrospective cohort study of T2DM patients reported a significant reduction in HF with DPP-4 inhibitors compared to GLP-1 RAs. The association was consistent in patients with or without prior cardiovascular disease, but not statistically significant in those with prior HF [64]. The DPP-4 inhibitors used in this study were saxagliptin and sitagliptin, while the GLP-1 RAs compared were not specified. Whether the association was due to different effects within the same drug class remained unclear. Besides, the superiority of GLP-1 RAs to DPP-4 inhibitors found in our network meta-analysis was driven largely by indirect evidence due to the absence of head-to-head trials between these two classes. Until such evidence is available, observational studies provide real-world evidence of efficacy.

The two incretin-based drug classes, DPP-4 inhibitors and GLP-1 RAs have different effects on cardiovascular outcomes. GLP-1 RAs directly activate GLP-1 receptors in the arteries and kidneys. The improved endothelial function through nitric oxide-induced vasodilation and the reduced oxidative stress and natriuretic effect by inhibiting the renin–angiotensin–aldosterone system all confer their effects on blood pressure lowering. Also, GLP-1 RAs could protect against endothelial dysfunction and have direct effects on the myocardium, which might also account for their beneficial effects in reducing vascular disease [65].

A retrospective cohort study on second-line antidiabetic drugs showing similar efficacy and safety profiles of GLP-1 RAs, SGLT-2 and DPP-4 inhibitors on cardiovascular events has been published [66]. However, observational studies may be prone to selection bias. The imbalance of the patient numbers in each drug class, in which DPP-4 inhibitors recipients (28,898) were five times more than SGLT-2 inhibitors recipients (5677) and twice more than GLP-1 RAs recipients (11,351) limited its statistical power without using propensity score matching. In addition, death was not assessed in this study. Although those findings differed from our network meta-analysis, real-world data from observational studies are still valuable.

Previous meta-analyses on each drug class showed conflicting results of their effects on death [43, 44, 53, 63, 67, 68], in which, the neutral effects of GLP-1 RAs and DPP-4 inhibitors identified in our study were contradicted by their protection in all-cause deaths [63, 67]. We therefore performed a sensitivity analysis to analyse the risk of all-cause mortality by including small trials excluded in our analysis but had been included in these meta-analyses. The overall conclusions did not change, albeit that GLP-1 RAs became significantly better than DPP-4 inhibitors in reducing all-cause mortality (Additional file 1: Table S16 and Online-only reference).

Choosing an antidiabetic drug class with proven benefit in reducing cardiovascular risk and mortality in patients with T2DM is now advocated in guidelines. The newly published CVOTs and renal outcome trials reinforced the confidence in SGLT-2 inhibitors, and to a lesser extent, GLP-1 RAs, as effective treatments. Ongoing trials will provide further evidence on cardiovascular safety and efficacy and potential renal benefits of these drug classes, the validity of the concept of a class effect, and help to refine future guidelines. Head-to-head CVOTs are difficult and expensive to conduct. Therefore, observational studies based on real-world data with sufficient length of follow-up and statistical power compared these new antidiabetic drug classes could be useful to demonstrate benefits at the population level.

Our network meta-analysis including the latest, powerful, and many more CVOTs than previous network meta-analyses has profound and unequivocal clinical implications. They support the new decision pathways of the American College of Cardiology (ACC) [69], which were announced before the release of the results of DECLARE-TIMI 58, Harmony Outcomes, and CARMELINA. Our analysis including all these latest trials is consistent with the ACC report. SGLT-2 inhibitors and GLP-1 RAs should, therefore, be considered evidence-based treatments for T2DM after metformin. SGLT-2 inhibitors can be considered before GLP-1 RAs because of the reduction in mortality. However, whether to choose an SGLT-2 inhibitor or a GLP-1 RA should still be tailored for different patients. The former is oral while the latter is mostly injected. The route of administration may, therefore, influence physician and patient preference. They also have different side effects, such that a patient intolerant of one can be switched to the other. Indeed, the two drug classes can both be given if the glycaemic target is not attained.

Not applicable.