Triple therapy in type 2 diabetes; a systematic review and network meta-analysis

Systematic literature review

The Cochrane Handbook for Systematic Reviews of Interventions (Higgins & Green, 2011) and the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) reporting guidelines (Moher et al., 2009) were used in the development, execution and reporting of this review. A review protocol was not registered; however, the review was an update to a predetermined report on the Comparative Safety and Effectiveness of Type 2 Diabetes Medicines (Australian Government DoH, 2013).

The PICO for the systematic review was as follows:

Population: All patients with type 2 diabetes

Intervention: Any triple therapy combination for treatment of type 2 diabetes

Comparator: metformin plus sulphonylurea dual therapy, and other triple therapy combinations.

Outcome: HbA1c, body weight, hypoglycaemia and adverse events

The systematic literature review was performed in three stages: (1) identify the most relevant systematic literature reviews on the efficacy and safety of medications for glycaemic control; (2) update the literature search using the relevant systematic literature reviews as identified in step 1 as a starting point; and (3) identify the relevant randomized controlled trials (RCTs) from steps 1 and 2. Ovid MEDLINE and The Cochrane Library Database were used for searching the literature.

Stage 1: identifying systematic reviews

Each bibliographic database was systematically searched using search terms for type 2 diabetes medications included in Table 1 on 5th March 2014. The key inclusion criteria were systematic reviews, English language publications, patients with type 2 diabetes, reviews published since 2010, reviews that include relevant outcomes, i.e., change in body weight, frequency of hypoglycaemic events, and/or total number severe adverse events. The key exclusion criteria was reviews that reported only on short-term treatment outcomes (<24 weeks). Supplementary Document—Stage 1, contains full details of the searches and terms used in each database. The searches were carried out by MJD and the results extracted and imported into the bibliographical software Endnote X7 (Thomson Reuters, New York, NY, USA). Duplicates were removed, articles that did not meet the inclusion criteria or met the exclusion criteria were also removed. The remaining articles were checked independently by two staff members for inclusion. If two systematic reviews were published based on the same data, the most comprehensive version of the paper was included. The review which was the most recent, included most therapeutic groups and outcomes was preferred.

Bennett et al. (2011) was considered the most relevant systematic literature review. Bennett et al. (2011) did not include SGLT2-i, DPP-4-i or insulin. Therefore for the SGLT2-i and DPP-4-i two further systematic reviews were included; Berhan & Barker (2013) (SGLT2-i), and Monami et al. (2010) (DPP-4-i). We did not identify a systematic review on insulin in triple therapy that matched the inclusion and exclusion criteria. The clinical trials from the systematic literature reviews were extracted for further consideration.

Stage 2: updating identified systematic reviews

Database searches were carried out to update the three identified systematic reviews in November 2014. Each bibliographic database was systematically searched using different search terms for each search strategy. Supplementary Document—Stage 2 contains full details of each search and the terms used in each database. The flow chart of the selection process is presented in Fig. S1.

Stage 3: identifying relevant RCTs

The articles identified through Stage 1 (included in relevant systematic literature reviews) and 2 (updated searches) were title checked by one author MJD and then independently by two authors (MJD and ET) and included in the final analysis if they met all of the following inclusion criteria: randomised controlled trial (RCT), English language publication, patients with type 2 diabetes, at least 100 participants in the trial, duration of treatment at least 24 weeks, published after 2002, trials that include any of the following relevant outcomes: HbA_1c, body weight change, frequency of hypoglycaemic events, or frequency of other (serious) adverse events. Publications were excluded if they were not randomised control trials or if they were post hoc analysis of randomised control trials. If two publications used the same data, the most comprehensive version of the paper was included. The flow chart of the selection process is presented in Fig. 1.

Statistical analysis

Our primary outcome was mean difference in change in HbA_1c at approximately six months. To test the hypothesis that all triple therapies were superior to dual therapy, we considered the minimal clinically important difference (MCID) to be −0.3% (3.3 mmol/mol) as used by the Federal Drug Agency (FDA) (CDER, 2008) and the European Medicines Association (CHMP, 2011) and also quoted in Australia (Australian Government DoH, 2010). HbA_1c assays suffer from some variability, and a smaller MCID would be within that variability in an individual (Cohen, Haggerty & Herman, 2010). To demonstrate that triple therapies had similar efficacy the lower and the upper boundary of the confidence intervals needed to be within 0.3% (3.3 mmol/mol) using the FDA and the European Medicines Agency’s guidelines (CDER, 2008; CHMP, 2011).

Secondary outcomes were the difference in mean change in body weight (Kg) and difference in the hypoglycaemia odds ratio within six months. For those outcomes, we did not identify any MCIDs. Other outcomes investigated but not reported included adverse events, serious adverse events, severe hypoglycaemia and mortality.

Excel was used to consolidate and standardise the outcome measures and measures of variability; standard deviations, standard errors and 95% confidence intervals were imputed where necessary (Follmann et al., 1992). The data was then imported into STATA for meta-analysis and network meta-analysis. Where multiple trials were available head-to-head meta-analyses were performed using a random effects model.

A multiple-treatments network meta-analysis was undertaken to summarise the results of triple therapy for each of the outcomes where common treatment arms existed (HbA_1c, body weight, and hypoglycaemia) using the trial data in the clinical evidence base. Different medications from the same class combinations were pooled, using the assumption that all these medications would have similar efficacy and safety. The network meta-analysis was conducted using STATA network package and mvmeta (Higgins et al., 2012; White, 2011) (The STATA .do file for HbA_1c is presented in Supplementary Document—Statistical Analyses). The network meta-analysis allowed for heterogeneity between studies during calculation (random effects). An inconsistency model was also applied to test for disagreement between direct and indirect evidence (Higgins et al., 2012). Possible covariates (Baseline HbA_1c, age and Body Mass Index) were examined prior to carrying out the network analysis to ensure similarities in baseline characteristics. The measurements of treatment effect calculated were mean differences and their 95% confidence intervals (CI) for continuous data, and odds ratios and their 95% CI for dichotomous outcomes. Differences between treatments were considered statistically significant if there were no overlap in 95% CI. Some trials only compared treatments from the same combination group (e.g., MET+ SU+ INS vs MET+ SU+ INS). These trials were excluded from the network meta-analyses as they were considered to compare the same treatment and there did not randomise two different treatment regimens.

Systematic literature review

The literature search identified 27 publications covering 26 trials with triple therapy; Fig. S2 provides a network for the direct comparisons identified in the search. Nine trials (10 publications) were excluded from the network meta-analyses as they were comparisons of the different drugs from the same treatment group e.g.,: MET+ SU+ INS vs. MET+ SU+ INS. One trial was excluded as it did not have a common comparator arm (Table S1). The majority of the remaining trials were of 24–26 weeks (11/17 trials) (Bergenstal et al., 2009; Dailey et al., 2004; DeFronzo et al., 2012; Heine et al., 2005; Hermansen et al., 2007; Liu et al., 2013; Lukashevich et al., 2014; Moses et al., 2014; Owens et al., 2011; Rosenstock et al., 2006; Russell-Jones et al., 2009), with only five trials longer than or equal to one year (Bosi et al., 2011; Derosa et al., 2013; Nauck et al., 2007; Schernthaner et al., 2013; Wilding et al., 2013). The included trials and comparisons are presented in Table 2.

Of the 17 trials included in the network analysis, three trials were identified as having a high risk of bias (Dailey et al., 2004; Rosenstock et al., 2006; Kendall et al., 2005), nine trials as having an unclear risk of bias (Bergenstal et al., 2009; DeFronzo et al., 2012; Heine et al., 2005; Liu et al., 2013; Lukashevich et al., 2014; Owens et al., 2011; Nauck et al., 2007; Schernthaner et al., 2013; Wilding et al., 2013), and five trials as having a low risk of bias (Hermansen et al., 2007; Moses et al., 2014; Russell-Jones et al., 2009; Derosa et al., 2013) (Table S2). Only two of the studies were not funded by industry (Liu et al., 2013; Derosa et al., 2013).

Overall, the key features of the triple therapy trials varied (see Tables S3 and S4), which may limit the comparability of these trials. When reported, the trials recruited patients between 2002 and 2011 and most of the trials were performed in an international setting. The key features were that patients were adult patients with HbA_1c of 7% (53 mmol/mol) or higher. The primary outcome of interest for each study was mean difference in HbA_1C from the control group and the trials were powered to assess this outcome, although Rosenstock et al. (2006) did not provide a power calculation.

Efficacy results

Two efficacy results were reported in the majority of the trials, change in HbA_1c and change in body weight. Seventeen RCTs were identified for inclusion in the network analysis for HbA_1c at six months (N = 9, 144) (Bergenstal et al., 2009; Bosi et al., 2011; Derosa et al., 2013; Nauck et al., 2007; Schernthaner et al., 2013; Wilding et al., 2013; Kendall et al., 2005) (Fig. S2). Sixteen RCTs were identified for inclusion in the network analysis for body weight change at six months (N = 8, 341) (Bergenstal et al., 2009; Dailey et al., 2004; DeFronzo et al., 2012; Heine et al., 2005; Hermansen et al., 2007; Liu et al., 2013; Lukashevich et al., 2014; Moses et al., 2014; Owens et al., 2011; Rosenstock et al., 2006; Russell-Jones et al., 2009; Derosa et al., 2013; Nauck et al., 2007; Schernthaner et al., 2013; Wilding et al., 2013; Kendall et al., 2005). Table S5 summarizes the raw data from the included trials. The network models were tested for consistency and the direct comparison results were similar to the indirect comparison for both HbA1c (p = 0.996) and body weight (p = 0.431).

Adverse events results

Hypoglycaemia was the only adverse event reported in the majority of the trials. Ten RCTs were identified for inclusion in the network analysis for hypoglycaemia (N = 4, 458) (Bergenstal et al., 2009; Dailey et al., 2004; Hermansen et al., 2007; Liu et al., 2013; Lukashevich et al., 2014; Moses et al., 2014; Owens et al., 2011; Rosenstock et al., 2006; Russell-Jones et al., 2009; Kendall et al., 2005). Data for MET+ SU+ SGLT2-i were only available for the 12 month time point (Schernthaner et al., 2013; Wilding et al., 2013) and not available at six months; therefore, this triple therapy was not included in the network analysis.

All adverse events, serious adverse events and serious hypoglycaemia were not systematically reported in the trials, and therefore no further network analyses are able to be presented.