Author + information
- Received February 14, 2018
- Revision received May 1, 2018
- Accepted May 10, 2018
- Published online September 24, 2018.
- Caroline K. Kramer, MD, PhDa,b,
- Chang Ye, MSca,
- Sara Campbell, MDa and
- Ravi Retnakaran, MDa,b,c,∗ ()
- aLeadership Sinai Centre for Diabetes, Mount Sinai Hospital, Toronto, Canada
- bDivision of Endocrinology, University of Toronto, Toronto, Canada
- cLunenfeld-Tanenbaum Research Institute, Mount Sinai Hospital, Toronto, Canada
- ↵∗Address for correspondence:
Dr. Ravi Retnakaran, University of Toronto, Leadership Sinai Centre for Diabetes, Mount Sinai Hospital, 60 Murray Street, Suite L5-025, Mailbox-21, Toronto, Ontario M5T 3L9, Canada.
Objectives The authors conducted a systematic review and network meta-analysis of placebo-controlled, randomized clinical trials in the post–Food and Drug Administration (FDA) guidance era to formally compare the effects of 3 new classes of glucose-lowering drugs on hospitalization for heart failure (HF) in type 2 diabetes mellitus.
Background The 2008 FDA Guidance for Industry launched an era of cardiovascular outcome trials for new glucose-lowering drugs in T2DM, including glucagon-like peptide (GLP)-1 agonists, dipeptidyl peptidase (DPP)-4 inhibitors, and sodium glucose co-transporter (SGLT)-2 inhibitors.
Methods We searched Embase, PubMed, Cochrane Library, and clinicaltrials.gov between December 1, 2008, and November 24, 2017, for randomized placebo-controlled trials, and performed network meta-analyses by Bayesian approach using Markov-chain Monte Carlo simulation method to compare the effects of glucose-lowering drugs on risk of HF hospitalization and estimate the probability that each treatment is the most effective.
Results Nine studies were identified, yielding data on 87,162 participants. In the network meta-analysis, SGLT-2 inhibitors yielded the greatest risk reduction for HF hospitalization compared with placebo (relative risk [RR]: 0.56; 95% CrI [credibility interval]: 0.43 to 0.72). Moreover, SGLT-2 inhibitors were associated with significant risk reduction in pairwise comparisons with both GLP-1 agonists (RR: 0.59; 95% CrI: 0.43 to 0.79) and DPP-4 inhibitors (RR: 0.50; 95% CrI: 0.36 to 0.70). Ranking of the classes revealed 99.6% probability of SGLT-2 inhibitors being the optimal treatment for reducing the risk of this outcome, followed by GLP-1 agonists (0.27%) and DPP-4 inhibitors (0.1%).
Conclusions Current evidence suggests that SGLT-2 inhibitors are more effective than either GLP-1 agonists or DPP-4 inhibitors for reducing the risk of hospitalization for HF in type 2 diabetes mellitus.
- heart failure
- dipeptidyl peptidase-4 inhibitor
- glucagon-like peptide-1 agonist
- sodium glucose co-transporter-2 inhibitor
- type 2 diabetes
In 2008, the Food and Drug Administration (FDA) issued a Guidance for Industry mandating the demonstration of cardiovascular safety for any new glucose-lowering drugs, thereby launching an era of cardiovascular outcome trials in type 2 diabetes (T2DM) (1). Accordingly, the 3 new diabetes medication classes that were introduced over the past decade (glucagon-like peptide [GLP]-1 agonists, dipeptidyl peptidase [DPP]-4 inhibitors, and sodium glucose co-transporter [SGLT]-2 inhibitors) have undergone cardiovascular evaluation to a degree that is unprecedented for glucose-lowering drugs. Indeed, although the FDA guidance mandated evaluation of a primary outcome of major adverse cardiovascular events (MACE) (typically defined as a composite of cardiovascular death, nonfatal myocardial infarction, and nonfatal stroke), the resultant trials have yielded additional vascular insights beyond these events. Most notably, in the SAVOR-TIMI 53 (Saxagliptin Assessment of Vascular Outcomes Recorded in Patients with Diabetes Mellitus-Thrombolysis In Myocardial Infarction 53) trial, the DPP-4 inhibitor saxagliptin was unexpectedly associated with an increased incidence of hospitalization for heart failure (HF) (2). As SAVOR-TIMI 53 was one of the first completed cardiovascular outcome trials of the post-FDA guidance era (published alongside the EXAMINE [Examination of Cardiovascular Outcomes with Alogliptin Versus Standard of Care] trial in 2013) (3), HF has become a secondary outcome of particular interest in the trials that have followed.
The importance of considering the impact of new glucose-lowering drugs on HF is underscored by several lines of reasoning. First, the presence of diabetes confers a marked increase in the risk of HF (2-fold higher in men and 5-fold higher in women) (4). Second, among patients with HF, the rate of mortality is approximately 2-fold higher in those with concomitant diabetes (5). Third, the pathophysiologic bases of coronary artery disease and HF are not identical, such that medications may have differential effects on these processes (6). Indeed, in the cardiovascular trials of the post-FDA guidance era, the impact of a glucose-lowering drug on MACE has not always been concordant with its effect on HF, which has amplified interest in the latter outcome owing to its singular prognostic significance in this patient population.
These trials have all had a similar design in which the respective new glucose-lowering drug under study has been compared with placebo on the background of existing diabetes treatment and modern vascular risk-reduction therapies (i.e., lipid-lowering and blood pressure control). As such, we reasoned that the shared placebo-controlled design of the FDA-mandated trials can provide an opportunity to objectively rank the new classes of glucose-lowering medications (GLP-1 agonists, DPP-4 inhibitors, and SGLT-2 inhibitors) according to their impact on HF risk by applying network meta-analysis methodology. Thus, we have conducted a systematic review and a network meta-analysis of placebo-controlled RCTs in the post-FDA guidance era to formally compare the effects of these 3 classes of glucose-lowering drugs on hospitalization for HF in patients with T2DM.
This systematic review and meta-analysis is reported in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement and was registered at International Prospective Register of Systematic Reviews (CRD42017082376) (7). The authors are experienced in meta-analyses (8–11).
Data sources and searches
We selected relevant studies published between December 1, 2008, and November 24, 2017, by searching Embase, PubMed, Cochrane Library, and clinicaltrials.gov to reflect a systematic review of the published literature after the FDA issued the 2008 Guidance for Industry (1). The following combined text and Medical Subject Heading terms were used: type 2 diabetes mellitus, randomized controlled trial, and heart failure. The complete search used for PubMed was: ((“Diabetes Mellitus, Type 2”[Mesh]) AND “(Randomized Controlled Trial” [Publication Type]) AND (“Heart failure” [Text])). All potentially eligible studies were considered for review, regardless of primary outcome or language. Manual search was also performed using references of key articles published in English.
Studies were considered eligible for inclusion if they: 1) were randomized, placebo-controlled trials conducted in adults with T2DM; 2) compared add-on therapy of any current antidiabetic medication to placebo; and 3) reported incidence of HF or hospitalization for HF as primary or secondary outcome. Exclusion criteria were as follows: observational and retrospective studies, studies completed before the FDA guidance, and studies that did not report heart failure as an outcome.
Data extraction and quality assessment
Two investigators (C.K. and R.R.) independently reviewed study titles and abstracts, and studies that satisfied the inclusion criteria were retrieved for full-text evaluation. Trials selected for detailed analysis and data extraction were analyzed by 2 investigators with agreement value (κ) 95.3%.
We extracted the following data from each selected study: number of participants, age, sex, trial duration, background cardiac and antidiabetic therapies, changes in glycosylated hemoglobin (HbA1c) and weight compared with placebo, and number of participants with incident HF or hospitalization for HF. Two reviewers (R.R. and C.K.) evaluated risk for bias according to PRISMA recommendations (Online Table 1).
Data synthesis and analysis
We calculated pooled estimates of the relative risk (RR) for incident HF/hospitalization for HF by random-effects model. To evaluate whether the impact of novel glucose-lowering medications on incident HF (RR) was associated with HbA1c reduction, we performed meta-regression analyses (Figure 1). Next, individual meta-analyses of the impact of GLP-1 agonists, DPP-4 inhibitors, and SGLT-2 inhibitors, compared with placebo, were performed to evaluate the effect of each drug class on HF (Figure 2).
We assessed the possibility of publication bias by constructing contour-enhanced funnel plots of the effect size in each trial against the SE. The Cochran Q test was used to evaluate heterogeneity between studies (12). I2 testing was also performed to evaluate the magnitude of the heterogeneity between studies, with values >50% considered indicative of moderate-high heterogeneity (13). All of these analyses were performed using Stata 11.0 software (StataCorp LP, College Station, Texas).
To further assess the effect of each class of antidiabetic medication on HF, we performed a network meta-analysis, which enables multiple treatment comparisons even in the absence of head-to-head trials. Network analysis can provide estimates of the effect of each treatment relative to every other treatment and also the probability that each treatment is the most effective. Analyses were conducted by Bayesian approach using the Markov-chain Monte Carlo (MCMC) simulation method with minimally informative prior distributions. We assumed that relative treatment effect for each pair comparison follows a common distribution and specified a random-effect model with a Poisson likelihood and log link function. The estimated RRs and probabilities in the ranking of each treatment were calculated from the mean of their posterior distributions simulated from MCMC and the corresponding 95% credibility intervals (CrIs) (the Bayesian equivalent of confidence intervals [CIs]). We assessed goodness-of-fit of the model to the data by the posterior mean of the residual deviance, which should be approximately similar to the number of data points used in the model. Because the network in our analysis is star-shaped and does not contain a closed loop, inconsistency of the network cannot be assessed. All the analyses were performed in WinBUGS version 1.4.3 (MRC Biostatistics Unit, Cambridge, United Kingdom).
Literature search and study characteristics
We identified 1,173 studies through electronic searches (Online Figure 1). Of these, 1,161 were excluded on the basis of the title and abstract, leaving 12 studies for further evaluation. Nine trials fulfilled our inclusion criteria, providing data on 87,162 participants.
Table 1 summarizes the 9 trials that were published between 2013 and 2017. As per the inclusion criteria, they were placebo-controlled trials that evaluated the following classes of antidiabetic medication: GLP-1 agonists (4 studies evaluating lixisenatide , liraglutide , semaglutide , and once-weekly exenatide , respectively, in 33,457 participants), DPP-4 inhibitors (3 studies evaluating alogliptin , saxagliptin , and sitagliptin , respectively, in 36,543 participants), and SGLT-2 inhibitors (2 studies evaluating empagliflozin  and canagliflozin  in 17,162 participants). All studies reported hospital admission for HF as a secondary outcome, defined as new hospitalization and/or presentation to acute care facility due to congestive HF. The study populations had a mean age ranging from 60.3 to 65.5 years, with a male preponderance (60.7% to 71.6%) and mean duration of T2DM ranging from 7.2 to 13.9 years. Reflecting contemporary management of cardiovascular risk factors, the majority of the participants were receiving statins (71.5% to 92.7%), aspirin (63.3% to 97.5%), and angiotensin-converting-enzyme inhibitors (33.7% to 85%).
Online Table 1 shows the assessment of risk of bias in the trials. All trials reported adequate randomization, none were stopped early, and all were multicenter. All were funded by industry.
In pooled analysis of the 9 trials, addition of the new glucose-lowering medication has a neutral effect in the incidence of hospitalization for HF (RR: 0.89; 95% CI: 0.74 to 1.08) compared with placebo, with significant between-study heterogeneity (I2 = 85%; p < 0.0001) (Online Figure 2). The contour-enhanced funnel plot is shown in Online Figure 3. Meta-regression analyses showed that there was no association between the risk reduction for HF and the lowering of HbA1c in the trials (p = 0.59) (Figure 1).
Given the heterogeneity of this analysis pooling all 3 classes of medication, we next proceeded to meta-analyses restricted to trials that assessed the impact of GLP-1 agonists, DPP-4 inhibitors, and SGLT-2 inhibitors, respectively (Figure 2). In the pooled analysis of the 4 trials involving GLP-1 agonists, this class of medications was not associated with reduction in hospitalization for HF (RR: 0.94; 95% CI: 0.84 to 1.04) with no between-study heterogeneity observed (I2 = 0%; p = 0.74) (Figure 2A). Similarly, pooled analysis of the 3 trials evaluating DPP-4 inhibitors also showed a neutral effect on incident hospitalization for HF (RR: 1.11; 95% CI: 0.95 to 1.30) with no significant between-study heterogeneity (I2 = 42.2; p = 0.17) (Figure 2B).
In contrast to the neutral effects observed for GLP-1 and DPP-4 inhibitors, pooled analyses of the 2 trials involving SGLT-2 inhibitors showed a significant reduction in the incidence of hospitalization for HF (RR: 0.56; 95% CI: 0.41 to 0.77) (Figure 2C). Although some heterogeneity was noted (I2 = 70.2%; p = 0.067), both studies individually showed a significant reduction in the risk of this outcome.
As shown in the network of randomized controlled trials (Online Figure 4A), there are currently no head-to-head trials comparing different medication classes on hospitalization for HF. As such, we performed a network meta-analysis to obtain pairwise comparisons between drug classes, which enabled a ranking of the classes according to their respective effects on HF (Online Figure 4B). In the network meta-analysis, SGLT-2 inhibitors yielded the greatest risk reduction compared with placebo (RR: 0.56; 95% CrI: 0.43 to 0.72). Moreover, SGLT-2 inhibitors were associated with a significant reduction in the risk of hospitalization for HF in pairwise comparisons with both: 1) GLP-1 agonists (RR: 0.59; 95% CrI: 0.43 to 0.79); and 2) DPP-4 inhibitors (RR: 0.50; 95% CrI: 0.36 to 0.70). Ranking of the classes revealed 99.6% probability of SGLT-2 inhibitors being the optimal treatment for reducing the risk of this outcome, followed by GLP-1 agonists (0.27%), and DPP-4 inhibitors (0.1%).
In this study, we show that there are distinct differences in the effects of new classes of antidiabetic medications on the risk of HF in patients with T2DM. Specifically, SGLT-2 inhibitors appear to markedly reduce the incidence of hospitalization for HF in this patient population, whereas GLP-1 agonists and DPP-4 inhibitors have a neutral effect. Accordingly, from the perspective of HF risk, the network meta-analysis yields a clear hierarchy among these classes of medications, with SGLT-2 inhibitors to be preferred over GLP-1 agonists and DPP-4 inhibitors.
Whereas a recent network meta-analysis has compared these medication classes on mortality and other outcomes (21), the current study focused specifically on HF in the 9 cardiovascular outcome/safety trials of the post-FDA guidance era thus far and evaluated the impact of glucose-lowering in this context. Although the United Kingdom Prospective Diabetes Study had reported an association between chronic glycemia and the risk of HF in recently diagnosed T2DM (22), this relationship was not supported by subsequent trials of intensive glucose-lowering strategies (23–26). Similarly, our meta-regression analysis showed no association between A1c reduction and the risk of HF in the placebo-controlled trials of GLP-1 agonists, DPP-4 inhibitors and SGLT-2 inhibitors of the post-FDA guidance era (Figure 1). Moreover, there was significant heterogeneity in the direct meta-analysis of these trials (Online Figure 2). Taken together, these data suggest that glycemia is not likely a mechanistic basis for the diverse effects of these antidiabetic medication classes on HF risk (as glucose-lowering is a feature that they all share), thereby implicating nonglycemic effects.
In this context, the nonglycemic effects of SGLT-2 inhibitors are of great interest because of the apparent beneficial effect of these medications on HF risk. There were striking similarities in the impact of empagliflozin and canagliflozin on HF in the EMPA-REG OUTCOME (BI 10773 [Empagliflozin] Cardiovascular Outcome Event Trial in Type 2 Diabetes Mellitus Patients) and CANVAS (Canagliflozin Cardiovascular Assessment Study) trials, respectively. First, both medications yielded a marked reduction in the risk of hospitalization for HF. Second, in both trials, this effect was apparent in individuals with and without a history of HF (20,27). Third, in EMPA-REG OUTCOME and CANVAS, this benefit appeared to emerge within the first few months of treatment (20,28). This rapid effect potentially may be suggestive of acute effects on renal/systemic hemodynamic function or direct effects on cardiovascular physiology (29). Indeed, in promoting natriuresis and osmotic diuresis, SGLT-2 inhibitors may reduce plasma volume and decrease preload, coupled with reduction of blood pressure, arterial stiffness, and afterload (29). Alternatively, by inducing a shift in metabolic substrate from fatty acids to ketone bodies, it has been suggested that SGLT-2 inhibition may improve myocardial energy usage and thereby improve cardiac function (29–31). At present, the precise mechanistic basis underlying the beneficial effect of SGLT-2 inhibition on HF risk remains to be elucidated.
Although the relevant mechanisms may be uncertain, the clinical implications of these data are nevertheless emerging. Current clinical practice guidelines for T2DM recommend patient-centered management through consideration of the patient’s relative prioritization of factors such as goals of treatment, implications for comorbidities, and risks of side effects (32). In this context, our network meta-analysis provides insight for incorporating HF risk into this decision-making process by formally comparing SGLT-2 inhibitors, GLP-1 agonists, and DPP-4 inhibitors and yielding a clear ranking of these classes in this regard. Although the evidence base will change in the coming years due to ongoing cardiovascular outcome/safety trials of other members of these classes, it is nevertheless encouraging that the current evidence from 9 trials involving >80,000 participants decisively differentiates the HF implications of SGLT-2 inhibitors (favored), GLP-1 agonists (neutral), and DPP-4 inhibitors (neutral) at this time. Moreover, the apparent role of nonglycemic effects in the beneficial impact of SGLT-2 inhibitors has led to studies evaluating this medication class in patients with HF in the absence of diabetes (29).
A limitation of this analysis is the absence of a closed loop within the network, thereby precluding the evaluation of inconsistency. However, this concern is attenuated by the concordance between the rankings and the findings from the direct meta-analyses within each class (each of which yielded fairly consistent findings). A second limitation is that the trials were generally conducted in patients who were considered to be at risk of cardiovascular events, which may limit their generalizability to those at lower risk. In addition, there were only 2 SGLT-2 inhibitor trials, with HF as a secondary outcome and differences in its definition.
The beneficial effects of empagliflozin and canagliflozin were apparent in individuals both with and without a history of HF (20,27). In addition, empagliflozin was beneficial in both those at low risk and those at high risk for incident HF (28). However, it is important to recognize that the baseline characterization of cardiac function in the trials to date has generally been modest and limited to history alone. Deeper characterization of baseline HF risk is recommended for future studies. In addition, as empagliflozin, canagliflozin, and dapagliflozin differ in their degree of SGLT-2 selectivity (33), comparative effectiveness studies will be of interest. Indeed, recognizing that the lower-limb amputation risk reported with canagliflozin has not been seen with empagliflozin (34), such studies may help guide the choice between SGLT-2 preparations in specific clinical scenarios. Finally, mechanistic studies are needed to better understand the physiological basis for the beneficial effect of these medications on HF risk.
This network meta-analysis shows that, for reducing the risk of HF, current evidence favors the choice of SGLT-2 inhibitors over either GLP-1 agonists or DPP-4 inhibitors. Moreover, their divergent respective effects in this regard yield a clear hierarchy among these classes from this perspective. Thus, an evidence base has now emerged that practitioners might consider when reviewing HF risks in the patient-centered management of T2DM in clinical practice.
COMPETENCY IN MEDICAL KNOWLEDGE: This study shows that current evidence strongly favors the choice of SGLT-2 inhibitors over either GLP-1 agonists or DPP-4 inhibitors for reducing the risk of HF hospitalization in patients with T2DM. Thus, the impact on HF risk has now emerged as a factor to consider in the selection of glucose-lowering drugs in the personalized management of T2DM.
TRANSLATIONAL OUTLOOK: As glucose-lowering is not the mechanistic basis for the diverse effects of these antidiabetic medication classes on HF risk, further study of the nonglycemic effects of SGLT-2 inhibitors may yield insight into the pathophysiology and targeted treatment of HF in T2DM.
This study was supported by intramural funds from Mount Sinai Hospital. Dr. Kramer holds the Banting and Best Diabetes Centre New Investigator Award and the Canadian Diabetes Association Clinician-Scientist Award; and has received grants from Boehringer Ingelheim. Dr. Retnakaran is supported by a Heart and Stroke Foundation of Ontario Mid-Career Investigator Award; holds the Boehringer Ingelheim Chair in Beta-cell Preservation, Function and Regeneration at Mount Sinai Hospital; has received grants from Novo Nordisk, Boehringer Ingelheim, and Merck; and has received personal fees from Novo Nordisk, Eli Lilly, Takeda, Sanofi, and Merck. All other authors have reported that they have no relationships relevant to the contents of this paper to disclose.
- Abbreviations and Acronyms
- confidence interval
- credibility interval
- dipeptidyl peptidase
- Food and Drug Administration
- glucagon-like peptide
- major adverse cardiovascular event(s)
- Markov-chain Monte Carlo
- Preferred Reporting Items for Systematic Reviews and Meta-Analyses
- sodium glucose co-transporter
- type 2 diabetes
- Received February 14, 2018.
- Revision received May 1, 2018.
- Accepted May 10, 2018.
- 2018 American College of Cardiology Foundation
- ↵Food and Drug Administration. Guidance for industry: diabetes mellitus—evaluating cardiovascular risk in new antidiabetic therapies to treat type 2 diabetes. Silver Spring, Maryland: Food and Drug Administration, December 2008. Available at: www.fda.gov/downloads/Drugs/Guidances/ucm071627.pdf. Accessed November 23, 2017.
- MacDonald M.R.,
- Petrie M.C.,
- Varyani F.,
- et al.
- Kramer C.K.,
- Zinman B.,
- Gross J.L.,
- et al.
- Kramer C.K.,
- Zinman B.,
- Retnakaran R.
- Higgins J.P.,
- Thompson S.G.,
- Deeks J.J.,
- Altman D.G.
- Zheng S.L.,
- Roddick A.J.,
- Aghar-Jaffar R.,
- et al.
- Stratton I.M.,
- Adler A.I.,
- Neil H.A.,
- et al.
- Lytvyn Y.,
- Bjornstad P.,
- Udell J.A.,
- Lovshin J.A.,
- Cherney D.Z.I.
- Ferrannini E.,
- Mark M.,
- Mayoux E.
- Mudaliar S.,
- Alloju S.,
- Henry R.R.
- American Diabetes Association
- Inzucchi S.E.,
- Iliev H.,
- Pfarr E.,
- Zinman B.