Introduction

Sodium-glucose co-transporter 2 inhibitors (SGLT2is) are effective in lowering glycated hemoglobin (HbA1c) levels and also have beneficial effects on weight and blood pressure [1, 2]. These potential benefits have been recognized, as evidenced by the increasing prominence of this class of prescription medicine in the latest clinical practice guidelines [3]. Ertugliflozin is the newest SGLT2i to become available, approved as an adjunct to diet and exercise to improve glycemic control in adult patients with type 2 diabetes mellitus (T2DM), with recent regulatory approvals in the United States (USA) (December 2017), European Union (EU) (March 2018) and elsewhere. Regulatory approval of ertugliflozin was based on multiple phase III trials that found ertugliflozin lowered HbA1c, weight and blood pressure [4,5,6,7,8].

The clinical and safety profile of individual SGLT2is has been investigated in clinical trials in comparison to placebo and other oral anti-diabetic drug classes, as mono-therapy and add-on therapy among patients with T2DM. The SGLT2i drug class as a whole has been also compared to other anti-diabetic drug classes in research using indirect comparison methodology [9]. However, comparisons within a class of medicines are also of interest to clinical practitioners and health policy decision-makers. Because no head-to-head randomized clinical trials (RCTs) assessing efficacy have been conducted among SGLT2is, indirect comparisons synthesized from network-meta analysis (NMA) can provide alternative evidence of comparative efficacy. NMAs are an extension of pairwise meta-analysis that bring together evidence from RCTs and use common comparators to indirectly make comparisons across multiple treatments either within or outside of a drug class. Although this approach allows for an observational synthesis of prior evidence, careful consideration of RCTs included in an NMA is required to minimize potential bias in observed results due to differences in study characteristics (e.g. quality, baseline patient characteristics, disease severity) that can influence the treatment effects of the interventions [10].

Several indirect comparisons within the SGLT2i class have been published comparing canagliflozin, dapagliflozin and empagliflozin. These have either compared these drugs across all trial populations or focused on patients inadequately controlled on either diet and exercise or metformin [11, 12]. These comparisons have not included ertugliflozin as data from trials with ertugliflozin were not yet available at the time of the studies.

Thus, the primary objective of the research reported here was to conduct a systemic literature review (SLR) and NMA to assess the efficacy and safety of ertugliflozin relative to other SGLT2is for the treatment of T2DM in three patient populations defined by inadequate control of HbA1c during prior treatments with (1) diet and exercise alone, (2) metformin alone and (3) metformin plus a dipeptidyl peptidase-4 inhibitor (DPP4i). Given the established role of insulin and glucagon-like peptide-1 receptor agonists (GLP-1 RAs) as options for third-line therapy, a secondary objective was to include these as comparators in the population uncontrolled on combination therapy with metformin and a DPP4i.

NMAs require that heterogeneity between studies be minimized; therefore, the present analyses focused on each population separately. As there is no published evidence synthesis specifically evaluating SGLT2i performance in those inadequately controlled on combination therapy with metformin plus a DPP4i, a secondary objective was to compare the effectiveness and safety of ertugliflozin added to dual therapy with metformin + sitagliptin (a combination regimen consisting of three oral antihyperglycemic therapies) to the addition of insulin or dual therapy (metformin plus a GLP-1 RA) in this population.

Methods

This article is based on previously conducted studies and does not contain any new studies with human participants or animals performed by any of the authors.

Data Sources and Searches

This study was conducted according to a pre-specified protocol and followed the standard guidelines for conducting and reporting SLRs and NMAs [13,14,15]. PubMed, Embase and Cochrane databases were searched using the Ovid platform through to December 2016 for English-language RCTs reporting outcomes at 24–26 weeks of follow-up [see Electronic Supplementary Material (ESM) for search strategy]. In addition to the primary search strategy, desk research was performed to access relevant gray literature [e.g. National Institute for Clinical Excellence (NICE) technology assessments, clinical treatment guidelines from 2010 to the present, European Public Assessment Reports (EPAR) and Food and Drug Administration label documents] for missing variables and to ensure all relevant trials were identified via the search. See ESM Table 1 for search terms.

Study Selection

The patient populations, interventions, comparators, outcomes and study design (PICOS) statement are presented in Table 1. In brief, patients were adults with T2DM, uncontrolled HbA1c (all patients with HbA1c ≥ 7.0%) and either managed with diet and exercise, on metformin alone; or on metformin plus a DPP4i prior to intervention. Studies evaluating treatments in special populations, such as those with renal impairment, were excluded.

Table 1 Patient, population, inteventions, comparators, outcomes and study design (PICOS) statement
Full size table

Comparators to ertugliflozin 5 mg and 15 mg were other commercially available SGLT2i products in the USA and the EU, including canagliflozin 100 mg and 300 mg, dapagliflozin 5 mg and 10 mg and empagliflozin 10 mg and 25 mg, and placebo for all populations. For those inadequately controlled on combination therapy with metformin plus a DPP4i, GLP-1 RAs and insulins were also included in the scope of the SLR.

Data Extraction and Quality Assessment

Two researchers performed the literature search and conducted an initial abstract review. Uncertainty regarding whether to include studies was resolved either through reconciliation or via consultation with a third reviewer. Data extraction forms were developed to capture the RCT evidence, with data extracted for each outcome measure. Data were extracted by one reviewer and checked by a second reviewer for quality assurance and completeness.

To minimize heterogeneity, outcomes were extracted from primary analyses using last observation carried forward (LOCF) where available. Study quality was assessed via the Grades of Recommendation, Assessment, Development and Evaluation (GRADE) working group guidelines [16]. The assessment of study quality was used to determine feasibility of networks and sensitivity analyses.

Data Synthesis and Analysis

The existence of a connected network of studies per outcome as well as study design, patient characteristics and sufficient non-zero frequency of events for safety outcomes was used to assess the feasibility of a valid NMA [17]. Efficacy outcomes included change from baseline in HbA1c, weight, systolic blood pressure (SBP), and proportion of patients with HbA1c of < 7%. Safety outcomes were the incidence of urinary tract infections (UTIs), genital mycotic infections (GMIs), severe and non-severe hypoglycemia events (SHEs, NSHEs) and patients with one or more adverse events (AEs). NMA feasibility analysis was performed on all outcomes. However, due to low frequency of reported events among placebo participants, NMAs were not possible for GMIs and hypoglycemic events in any of the populations of interest.

For remaining outcomes, evidence was synthesized in WinBUGS software [18] using both fixed effect (FE) and random effects (RE) Bayesian NMA models. The FE model was selected unless there was a significant difference in the deviance information criteria (DIC) of ≥ 3 [19]. The DIC is a measure for how well the model would predict a replicate dataset, as per guidelines. The appropriate statistical models were used based on the nature of the outcomes. An identity link and a normal likelihood were used for continuous outcomes (e.g. HbA1c change from baseline). Logit link with binomial likelihood distribution was used for dichotomous outcomes (e.g. AEs). Non-informative prior distributions of relative treatment effects were used for all outcomes of interest, per published expert guidelines [19].

Results are presented as the median of mean difference (MD) for continuous outcomes and median odds ratio (OR) for binary outcomes. Credible intervals (CrI), analogous to 95% confidence intervals, were used to determine significance. Bucher tests [20] of inconsistency were performed on all closed loops with direct and indirect evidence for change in HbA1c, weight and SBP to ensure that the results calculated in the NMA (indirect evidence) were consistent with the evidence from RCTs (direct evidence) [21].

Sensitivity Analysis

Formal meta-regression sensitivity analyses may be used to control for differences in baseline characteristics that may represent effect modifiers of the treatment effect. However, in this research meta-regressions were not feasible for any population due to the insufficient number of studies meeting the inclusion criteria. Therefore, to assess the impact of study heterogeneity on the results, we performed sensitivity analyses for key outcomes (change in HbA1c and weight) for each population.

Two sensitivity analyses were performed for the diet and exercise population. In the first, Kaku [22] was added to the network. This study was excluded initially as the low threshold of HbA1c for inclusion was 6.5% HbA1c, versus our inclusion criteria of ≥ 7.0%. However, despite potential heterogeneity previous SGLT2i NMAs have included this study, and therefore a sensitivity analysis was considered appropriate [23, 24]. In the second sensitivity analysis, studies that were not linked to the network via a common comparator (placebo) arm (Hadjadj [25]; Lewin [26]; Rosenstock [27]) were dropped from the network and analyses run. For the uncontrolled on metformin population, one sensitivity analysis was run that excluded studies that were not linked to the network via a placebo arm (VERTIS FACTORIAL randomized trial [28] and DeFronzo [29]).

One sensitivity analysis was conducted for patients inadequately controlled with a combination therapy of metformin plus a DPP4i that included a GLP-1 RA (liraglutide) [30] for HbA1c and weight change (SBP not reported). Liraglutide was included as a sensitivity analysis as opposed to the base case due to heterogeneity in this study [30]. Specifically, consistent with clinical practice, the liraglutide arm discontinued sitagliptin, resulting in a indirect comparison of triple oral therapy with a SGLT2i with a GLP-1 RA on a background of metformin. Further, a substantial reduction in HbA1c was observed post-randomization in the placebo arm. Compared to other included studies this study was identified as a source of heterogeneity that could confound the analysis and was therefore included in a sensitivity analysis only.

Results

Systematic Literature Review

The literature review identified 10,566 non-duplicate titles. After a review of the abstracts, 198 papers were identified for full text review. Ultimately, 18 published studies met the inclusion criteria, with two additional studies extracted for sensitivity analyses (Fig. 1). At the time of the literature search, the results of four ertugliflozin clinical trials (VERTIS MONO, VERTIS MET, VERTIS FACTORIAL, VERTIS SITA2) had not been published. Therefore, SLR results were supplemented with the manufacturer’s clinical study reports for the evidence synthesis. The results from the VERTIS MONO, VERTIS MET, VERTIS FACTORIAL and VERTIS SITA2 trials were subsequently published by Terra et al. [8], Rosenstock [7], Pratley [6] and Dagogo-Jack [4], respectively. No additional studies were identified via gray literature searches. Figure 1 displays the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) diagram.

Fig. 1
figure 1

PRISMA diagram. BC Base case, CSR clinical study report, DPP4i dipeptidyl peptidase-4 inhibitor, SA sensitivity analysis, Diet and Exercise inadequately controlled on diet and exercise, Metformin Alone inadequately controlled on metformin alone, Metformin plus DPP4i inadequately controlled on metformin plus a DPP4i

Full size image

Study Characteristics and Quality Assessment

Study characteristics for all three populations are described in Table 2, and network diagrams showing feasible comparisons are presented in Fig. 2. Outcome data is available in ESM Table 2. Quality assessment determined that all studies across the three populations were RCTs with clearly described interventions; all were assessed to be of high quality.

Table 2 Study design and mean baseline characteristics
Full size table
Fig. 2
figure 2

Network diagrams for type 2 diabetes mellitus (T2DM) patients inadequately controlled on diet and exercise (a), inadequately controlled on metformin alone (b) and inadequately controlled on combination therapy with metformin (MET) plus a dipeptidyl peptidase-4 inhibitor (DPP4i; name listed where relevant) (c). Superscripts: 1 Study added for sensitivity analysis only, 2 study dropped from sensitivity analysis (no connection via placebo), 3 study does not report systolic blood pressure, 4 data from VERTIS MONO, subsequently published by Terra et al. 2017 [8], 5 study does not report non-severe hypoglycemic events, genital mycotic infections (GMIs), severe hypoglycemic events, 6 data from the VERTIS MET trial, subsequently published in Rosenstock et al. 2018 [7], 7 data from the VERTIS FACTORIAL trial, subsequently published in Pratley et al. 2018 [6], 8 data from VERTIS SITA2 trial, subsequently published in Dagogo-Jack et al. 2018 [4], 9 study does not report target glycated hemoglobin, GMIs, urinary tract infections, hypoglycemic events, adverse events. Circles represent sodium-glucose co-transporter-2 inhibitor treatments, hexagons represent glucagon-like peptide-1 receptor agonists. Dark gray shapes are on a background of MET, light grey shapes are on a background combination therapy of MET plus a DPP4i (specific DPP4i stated for clarity). CANA Canagliflozin, DAPA dapagliflozin, EMPA empagliflozin, ERTU ertugliflozin, LINA linagliptin, LIRA liraglutide, PBO placebo, SAXA saxagliptin, SITA sitagliptin

Full size image

Patients Inadequately Controlled on Diet and Exercise

All included trials were double-blind RCTs (n = 11) and all but two were placebo controlled [26, 27]. Most were multinational studies, and two RCTs involved only an Asian population [31, 32]. Mean baseline HbA1c was generally similar, clustering in the range of 7.9–8.3%. However, in two studies the mean baseline HbA1c level was slightly higher, respectively at 8.7 [25] and 8.8% [27]. Other potential effect modifiers, such as SBP, and patient age were found to be relatively similar across studies, while baseline weight was lower in the studies with an Asian population [31, 32]. Mean baseline estimated glomerular filtration rate (eGFR) varied from a low of 83 mL/min/1.73 m2[31] to a high of 93 mL/min/1.73 m2 [25] for studies included in the base case, with an outlier of 67 mL/min/1.73 m2 for the study included in a pre-specified sensitivity analysis [22]. The percentage of females in each patient population varied from a low of 35% [31, 32] to a high of 56% [33].

Patients Inadequately Controlled on Metformin Alone

All trials were double-blind RCTs (n = 7), and all but two studies were placebo controlled [28, 29]. All but one RCT were multinational (n = 6), and one comprised exclusively an Asian population [34]. Mean baseline HbA1c was 8.1%, with all studies falling in the range of 7.9 to 8.1% with the exception of one study (8.6%; VERTIS FACTORIAL trial, which was removed in sensitivity analysis) [28]. Baseline weight was similar in all studies, with the exception of the study of Yang et al. [34], which had a lower mean baseline weight of 71 kg. Potential effect modifiers (SBP, patient age, percentage female, eGFR) were otherwise found to be relatively similar across studies.

Patients Inadequately Controlled on combination therapy with Metformin Plus a DPP4i

The combinations identified (plus background therapies containing metformin and a DPP4i) were dapagliflozin 10 mg (metformin + sitagliptin), dapagliflozin 10 mg (metformin + saxagliptin), empagliflozin 10 mg and 25 mg (metformin + linagliptin), titrated canagliflozin (metformin + sitagliptin) and ertugliflozin 5 mg and 15 mg (metformin + sitagliptin). No RCTs that included insulin as a comparator met the inclusion criteria for investigation as add-on to combination therapy with metformin  plus a DPP4i. Only one RCT of a GLP-1 RA (liraglutide on background of metformin) met the inclusion criteria; as such, this study was included in a sensitivity analysis to explore the NMA’s secondary objective. All included trials were placebo-controlled double-blind RCTs. Limited variability was observed in the baseline characteristics of the included studies.

Given the limited number of studies meeting the inclusion criteria of our study, it is important to note the elements of heterogeneity in study designs that should be considered when interpreting the findings. In addition to metformin, study participants had differing DPP4i background therapies, requiring an assumption of a non-differential impact of underlying DPP4i therapy. Further, in the study of Rodbard et al. [35] the patients underwent dose titration of canagliflozin (84.5% titrated from 100 to 300 mg), complicating the ability of the authors to interpret a distinct low or distinct high dose. Finally, patients randomized into the liraglutide arm in the study of Bailey et al. [30] (included in sensitivity analysis 1) discontinued sitagliptin and commenced liraglutide with weekly dose escalation until the maintenance dose of 1.8 mg/day was reached.

Network Meta-analysis

The forest plots for key efficacy outcomes of HbA1c, weight loss and SBP for ertugliflozin compared to the respective low and high doses for other SGLT2i for each population are shown in Fig. 3. At the international level there is some variability in the starting dose or availability of specific SGLT2i doses. Therefore, we provide supplementary results that include additional dose comparisons as well as outcomes not shown in this article (ESM Tables 3, 4). Note that differences between treatments with credible intervals overlapping zero (for median of MD) or one (for median OR) were not considered to be significant.

Fig. 3
figure 3

Forest plots for T2DM patients inadequately controlled on diet and exercise, inadequately controlled on metformin alone and inadequately controlled on combination therapy of metformin plus a DPP4i. CrI Credible interval, HbA1c glycated hemoglobin, MD mean difference. DPP4i in parentheses for clarity, single asterisk indicates results from sensitivity analysis (switch from SITA), double asterisk indicates titrated CANA

Full size image

Inadequately Controlled on Diet and Exercise

For the population on diet and exercise, FE models were selected for all analyses. Ertugliflozin 15 mg was significantly more effective in reducing HbA1c level than either dapagliflozin 10 mg (MD − 0.36%; CrI − 0.65, − 0.08) or empagliflozin 25 mg (MD − 0.31%; CrI − 0.58, − 0.04). There were no significant differences between ertugliflozin 5 mg and dapagliflozin 10 mg in the base case (MD − 0.19; CrI − 0.48; 0.09) (ESM Table 3). At high doses only, canagliflozin was significantly more effective than ertugliflozin in reducing SBP. There were no significant differences between SGLT2is for other efficacy outcomes (weight loss, HbA1c below target) for low and high SGLT2i doses, respectively, and there were no significant differences among SGLT2is for safety outcomes (AEs and UTIs). Finally, there was no evidence of inconsistency in the examined outcomes (i.e., direct evidence from the RCTs and indirect results from the NMA did not differ significantly).

Addition of the study of Kaku et al. [22] (i.e. lower HbA1c threshold study) to the sensitivity analyses resulted in ertugliflozin 5 mg becoming significantly more effective than dapagliflozin 5 mg (MD − 0.41%; CrI − 0.67, − 0.15) and dapagliflozin 10 mg (MD − 0.38; CrI − 0.64, − 0.11) for HbA1c change, while ertugliflozin 15 mg remained more effective than dapagliflozin 10 mg and empagliflozin 25 mg. There were no other changes to the findings between ertugliflozin and other respective low and high doses of other SGLT2is for remaining HbA1c change and weight comparisons. The second sensitivity analysis (dropping studies not connected by placebo [25,26,27]) did not impact base case findings for ertugliflozin 5 mg and ertugliflozin 15 mg for HbA1c and weight, with the significance for dapagliflozin 10 mg and empagliflozin 25 mg remaining for HbA1c change (ESM Table 5).

Inadequately Controlled on Metformin Alone

The FE model was selected for all analyses except for weight change. Ertugliflozin was significantly more effective in reducing HbA1c than low and high doses, respectively, of dapagliflozin (ertugliflozin 5 mg vs. dapagliflozin 5 mg MD − 0.22%; CrI − 0.42, − 0.02). Further, ertugliflozin 15 mg was more effective than dapagliflozin 10 mg (MD − 0.26%; CrI − 0.46, − 0.06) and empagliflozin 25 mg (MD − 0.23%; CrI − 0.44, − 0.03). Consistent with the population inadequately controlled on diet and exercise, there were no significant differences between low-dose ertugliflozin 5 mg and high-dose dapagliflozin 10 mg on the background of metformin (MD − 0.14%; CrI − 0.34; 0.06) (ESM Table 3). There were no additional significant differences for other efficacy outcomes (e.g. weight loss and SBP) and safety outcomes (AEs, UTIs). Similar to the population inadequately controlled on diet and exercise, in this population there was again no evidence of inconsistency in examined outcomes (i.e. direct evidence from the RCTs and indirect results from the NMA did not differ significantly).

In the sensitivity analysis, when studies not connected to the network via a placebo arm (VERTIS FACTORIAL trial [6]; Defronzo [29]) were dropped, ertugliflozin 5 mg was no longer significantly more effective than dapagliflozin 5 mg for change in HbA1c due to the slight widening of the credible interval (MD − 0.19%; CrI − 0.40, 0.02) while ertugliflozin 15 mg remained significantly more effective than empagliflozin 25 mg and dapagliflozin 10 mg, respectively. These sensitivity analyses did not otherwise significantly impact other SGLT2i comparisons for weight and HbA1c change (ESM Table 6).

Inadequately Controlled on Metformin Plus a DPP4i

Among patients inadequately controlled on combination therapy with metformin plus a DPP4i, FE models were selected for all analyses. All ertugliflozin and also other SGLT2i combinations were significantly more effective than dapagliflozin 10 mg added to metformin + sitagliptin. There were no further significant differences for other efficacy outcomes (e.g. weight loss, SBP, HbA1c below target) between ertugliflozin and low and high SGLT2i doses, respectively. Analyses of AEs and UTIs did not identify any significant differences.

Sensitivity analysis including liraglutide [30] for change in HbA1c did not identify any significant differences between liraglutide and ertugliflozin treatments (ertugliflozin 5 mg MD − 0.07%; CrI − 0.35, 0.22; ertugliflozin 15 mg MD − 0.15%; CrI − 0.43, 0.14).

Discussion

This evidence synthesis represents the first published NMA that compares the effectiveness and safety of ertugliflozin to other SGLT2is. Previous NMAs of the SGLT2i class conducted prior to the availability of ertugliflozin suggested differences in effectiveness, generally favoring canagliflozin [11, 23, 24].

The current analysis has several findings that may be informative to prescribers and healthcare policy-makers. Indirect comparisons for HbA1c reduction found that ertugliflozin 5 mg was associated with greater reductions in HbA1c than dapagliflozin 5 mg in patients uncontrolled on metformin alone, whereas ertugliflozin 15 mg was associated with greater HbA1c reductions compared to dapagliflozin 10 mg and empagliflozin 25 mg when added to the diet and exercise regimen and to the metformin regimen. HbA1c reduction associated with both doses of ertugliflozin were no different from that with the respective canagliflozin doses across all populations.

In this first study to compare SGLT2is as add-on to patients inadequately controlled on combination therapy with metformin plus a DPP4i, the results indicate that both ertugliflozin 5 mg and 15 mg were more effective in lowering HbA1c than dapagliflozin 10 mg when added to a combination of sitagliptin + metformin. However, no significant differences were observed for the comparisons of ertugliflozin (5 mg or 15 mg) added to sitagliptin + metformin versus dapagliflozin 10 mg added to saxagliptin + metformin.

With the exception of a lower SBP for canagliflozin 300 mg in patients inadequately controlled with diet and exercise, ertugliflozin doses did not differ significantly from other respective low and high doses of other SGLT2is for the remaining outcomes across each population. In a sensitivity analysis among patients uncontrolled on combination therapy with metformin plus a DPP4i, no significant differences for efficacy outcomes were found between adding ertugliflozin to the regimen compared to a switch from sitagliptin to liraglutide.

Lastly, as dapagliflozin 5 mg is not universally approved as the starting dose worldwide, comparisons between ertugliflozin 5 mg and dapagliflozin 10 mg are relevant to healthcare decision-makers in those countries. In terms of HbA1c reduction, ertugliflozin 5 mg was no different than the highest available dose of dapagliflozin in patients inadequately controlled by either diet and exercise or by therapy with metformin alone, but it was associated with significantly greater reductions in HbA1c among patients inadequately controlled on combination therapy with metformin + sitagliptin.

A sound explanation for these differences in efficacy between SGLT2is is not readily apparent in the literature. In the past, differences in selectivity for the SGLT2 receptor have been postulated [23]. However, within this drug class ertugliflozin is one of the most selective approved agents and canagliflozin is the least selective approved agent, but they were among the most effective in terms of HbA1c reduction. Another potential consideration for differences between some SGLT2is is dose selection during clinical development. For example, while dose–response modeling for ertugliflozin indicated that doses at both 5 and 15 mg achieved near maximal urinary glucose excretion (UGE) [36], dapagliflozin doses higher than 10 mg were not selected for phase III trials despite achieving greater UGE (EPAR for Forxiga [37]). Given the lack of definitive explanations for the differences in HbA1c lowering among the individual SGLT2is in this NMA, additional clinical investigations exploring possible mechanistic differences across the SGLT2i class may help address this question.

Interpretation of the results and their significance should consider that NMAs are not equivalent to head-to-head trial evidence, thus the role of chance cannot be discounted in explaining some or all of these findings. NMA methods have a number of inherent limitations. First, the analysis is observational in nature since patients can be randomized ‘within’ but not ‘between’ trials. Second, despite searching secondary sources, such as labels, not all outcomes were available across all trials, which resulted in exclusion of some comparators for affected outcomes. In addition, evidence synthesis outcomes for GMIs, NSHE and SHE could not be interpreted due to the scarcity of events that prevented network convergence (i.e. unstable results prevented robust comparisons). Nevertheless, visual inspection of the results indicates that these outcomes were similar across clinical trials and active interventions. Third, between-study heterogeneity may have been present. For example, relatively small differences in potential effect modifiers, such as mean baseline HbA1c, existed between studies in the data synthesis and could have influenced study findings. However, the observed consistency of results from the base case and sensitivity analyses that suggest additional HbA1c lowering with ertugliflozin versus dapagliflozin and empagliflozin, in both mono- and dual-therapy populations, makes this explanation less likely.

In addition, regarding the population inadequately controlled on metformin plus a DPP4i, where a mix of DPP4is were used in the background therapy (sitagliptin, saxagliptin and linagliptin) in combination with metformin. By including combinations with different underlying DPP4i regimens, the networks assume that the impact of each placebo + DPP4i combination is comparable. Due to the limited number of studies identified for this population the impact of this assumption on results could not be evaluated. Similarly, switching sitagliptin to liraglutide (consistent with clinical practice) as opposed to adding liraglutide to sitagliptin gave rise to a comparison of a regimen consisting of three oral anti-hyperglycemic agents (ertugliflozin added on to metformin + sitagliptin) versus a dual-therapy regimen with subcutaneous injection (liraglutide added on to metformin) in the sensitivity analysis. Fourth, although the overall quality of included studies was high, relatively few studies met the strict inclusion criteria. The use of strict inclusion and exclusion criteria for RCTs helped to minimize heterogeneity from baseline characteristics. However, minor differences in baseline characteristics (such as HbA1c) need to be recognized, and thus residual confounding cannot be ruled out as a source of bias. Unfortunately, a formal approach of using meta-regression to control for differences in baseline characteristics was not possible due to the small number of studies. Despite these limitations, conducting a SLR and following best practice guidance [13,14,15] ensured that the findings of this NMA are relevant for healthcare decision-making.

Conclusion

In summary, the current NMA found that ertugliflozin may be more effective in lowering HbA1c than dapagliflozin and empagliflozin in patients uncontrolled on diet and exercise as well as those on therapy with metformin alone. The analysis did not find differences in the HbA1c reductions associated with ertugliflozin compared to canagliflozin across all populations evaluated. The degree to which the efficacy and safety found in short-term studies translate to effects on long-term complications of T2DM, most notably cardiovascular and renal events, is of considerable interest. Completed studies for empagliflozin [38], canagliflozin [39] and dapagliflozin [40] have shown reductions in cardiovascular outcomes. VERTIS CV, an ongoing outcomes trial, will evaluate the effects of ertugliflozin versus standard of care for cardiovascular and renal outcomes [41].