Triple therapy combinations for the treatment of type 2 diabetes – a network meta-analysis

Aim: To estimate and compare the results from all randomized trials of triple combinations of anti-diabetes therapies that reported the reduction of glycated haemoglobin (HbA1c) and associated effects on body weight and hypoglycaemia. Methods: PubMed and the Cochrane Library were searched for trials with at least one study arm on triple therapy and which reported the differences in mean change in HbA1c between two study arms. These were included in a network meta-analysis. Results: Altogether, 15182 participants from 40 trials with treatment duration of 6 to 12 months were included. Compared with none/placebo added to dual therapy, the addition of a drug therapy from six of eight drug classes to existing dual therapy resulted in significant additional mean reductions in HbA1c from -0.56% (-6.2 mmol/mol; dipeptidyl peptidase 4 inhibitors) to - 0.94% (-10.3 mmol/mol; thiazolidinediones). Of the six drug classes, three were associated with less favourable weight change and two were associated with more favourable weight change when compared with none/placebo added to dual therapy. Furthermore, five drug classes were associated with greater odds of hypoglycaemia. Similar results were observed in analyses of studies with 6 months treatment duration and after excluding study arms that contained insulin. Conclusions: Overall triple therapy combinations were similar in improving diabetes control although there were some differences in adverse effects. By balancing the risks and benefits of each therapy, the estimates of pairwise comparisons of triple therapies for HbA1c, body weight and hypoglycaemia provided in this study may further inform evidence based practice.


Introduction
In clinical practice, glucose-lowering pharmacotherapy is prescribed when lifestyle modification is not effective in the management of type 2 diabetes.If glycaemic control becomes inadequate with a single therapy, a second and then a third therapy may be added to the treatment, while reinforcing the importance of lifestyle modification [1].A network meta-analysis published in 2011 compared the effects of a number of therapies added to metformin (MET) and sulfonylurea (SU) for the treatment of type 2 diabetes [2].However, triple therapy combinations other than those that include both MET and SU are increasingly used and may result in greater reduction in glycated haemoglobin (HbA1c).A randomised trial that compared the triple combinations MET/SU/DPP-4 (dipeptidyl peptidase 4 inhibitors) and MET/DPP-4/INS (insulin) reported a -0.40% [95% confidence intervals: -0.66, -0.15] (-4.4 [-7.3, -1.7] mmol/mol) greater reduction in HbA1c with MET/DPP-4/INS after a six month treatment period [3].Moreover, a third therapy is usually added to the existing dual therapy treatment and because of individualized treatment plans, patients may not necessarily be taking the MET/SU dual therapy combination.Therefore, we aimed to estimate and compare the effect of all triple therapy combinations that have been studied in clinical trials on glycaemic control as assessed by HbA1c and to examine the effect on weight changes and hypoglycaemia to further inform evidence based practice in the management of type 2 diabetes.

Materials and methods
We consulted the PRISMA extension statement for reporting of systematic reviews incorporating network meta-analyses checklist in preparation of this review [4].

Study selection, data extraction, and quality assessment
Studies were included if they fulfilled the following criteria: randomised trials in adults (aged ≥ 18 years) with type 2 diabetes; at least one study arm involved triple therapy; at least two study arms were on different drug class combinations; and reported the mean change and its variability (i.e. standard deviation, standard error, or 95% confidence interval) in HbA1c from baseline for each study arm or the difference in mean change and its variability between two study arms.Attempts were made to include studies that did not report the required summary statistics on HbA1c by contacting the corresponding authors of these studies.Drug classes available for the treatment of type 2 diabetes include MET, SU, DPP-4, INS, glucagon-like peptide-1 receptor agonist (GLP-1), sodium-glucose linked transporter protein 2 inhibitors (SGLT2), alpha glucosidase inhibitors (AGI), thiazolidinediones (TZD), and meglitinides (MEG).
Studies were excluded if the treatment period was less than 20 weeks, since adjustments to dosages, such as for INS, may take place in the first few weeks of the treatment period and change in HbA1c may not be noticeable in the first three months due to the 120 day life span of red blood cells.Studies with treatment duration greater than 54 weeks that did not report interim results during the first 6 to 12 months of treatment, sample size less than 30 per study arm, or compared a triple therapy arm with a monotherapy arm were also excluded.Studies that combined participants on different background therapy combinations during the treatment period were excluded unless mean changes in HbA1c were reported for the subgroups that were on the same background therapy combinations.
Literature search and data extraction were conducted by C.M.Y.L in consultation with S.C.
Information extracted from each study included study characteristics (name of primary author, year of publication, location of trial, drug class combinations, sample size and details of medications used for each study arm, treatment duration, and analysis set used), baseline characteristics of studied populations (proportion of females, mean age, mean body mass index, mean duration of diabetes, and mean HbA1c), and study outcomes (change in HbA1c, change in body weight, and number of participants experienced at least mild hypoglycaemia during treatment period).For studies that have multiple follow-up visits, data collected closest to 6 months after the start of treatment were included in the analysis.For multi-arm studies that included study arms that were assigned the same drug class combinations but at different dosages for one of the drugs, the study arm allocated the dosage that was most commonly used in other studies for that drug was included in the analysis.The Jadad Scale was used to assess the quality of the included studies [5].

Data analysis
We estimated the mean difference in HbA1c between each drug class added to an existing dual therapy compared to adding a nothing or adding a placebo (none/placebo) to an existing dual therapy to determine if any of the third therapy added to existing dual therapy is superior in reducing HbA1c.. Since not all triple therapies have been compared in randomised trials, a multivariate network meta-analysis was employed instead of a traditional pairwise metaanalysis.Multivariate network meta-analysis can provide estimates for all pairwise comparisons that are linked to a network of trials through utilising both direct evidence obtained from studies directly comparing drug class combinations and indirect evidence estimated through a common comparator [6].Furthermore, information from multi-arm studies can be included in the network meta-analysis, thus increasing the number of studies that can be included in the analysis.Estimates of all pairwise comparisons of triple therapies were provided to assist in the selection of the drug class that appears most appropriate as add on to existing dual therapy.In order to assess the adverse effects of the triple therapy combinations, the analyses were repeated for difference in mean change in body weight reported as kilograms and for difference in the proportion of participants who experienced hypoglycaemia reported as odds ratios.Since the level of HbA1c may be different at 6 months and 12 months after the start of treatment, a sensitivity analysis was conducted restricted to studies with about 6 months (20-30 weeks) treatment duration.Analyses were also repeated after excluding study arms that contained insulin, and after removing studies that were rated "poor" based on the Jadad Scale.
Multiple-outcomes meta-analysis was conducted using network, a suite of Stata commands for network meta-analysis within a frequentist framework [7].A network plot was used to assess the geometry of the network [4].Each node represents a drug class added to existing dual therapy and direct comparisons between drug classes are represented by connections between the nodes.The number of studies available per direct comparison is provided in the network.
The size of the node reflects the number of studies available for the drug class.Estimates derived from consistency models are reported.Overall inconsistency across comparisons between direct and indirect evidence were tested using the design-by-treatment interaction inconsistency model.Relative rankings of the therapy combinations were estimated using the surface under the cumulative ranking curve (SUCRA).SURCA is the cumulative probabilities of a treatment to achieve each of all possible ranks out of all competing treatments [8].For instance, if three treatments were compared, SUCRA for treatment 1 would be the cumulative probabilities of treatment 1 ranking first out of all three treatments, ranking second out of all three treatments, and ranking third out of all three treatments.All statistical analyses were performed using Stata/IC 12.0 for Windows (Stata Corp LP., College Station, TX, USA).

Results
A total of 11666 titles and abstracts were screened after duplicates were removed (Supplementary Figure 1).Of these, 137 articles were retrieved to assess their eligibility and 87 articles were removed based on our inclusion and exclusion criteria.Nine authors were contacted for missing data; three authors provided additional information and missing data from two studies were obtained from a published meta-analysis.A further six articles were removed; four trials had sample size less than 30 per arm and two trials were included in other publications.Therefore, 40 studies were included in the network meta-analysis.

Characteristics of included studies
Altogether, 15182 participants with type 2 diabetes from 40 trials with treatment duration ranging from 20 to 54 weeks (90% of trials with treatment duration of 6 months (20-30 weeks)) were available for analysis (Table 1).The characteristics of the studied populations varied with baseline mean values ranging from 52.6 to 65.3 years (median 56.6 years) for age, 7.2 to 10.2% (55 to 88 mmol/mol) (median 8.3% (67 mmol/mol)) for HbA1c, 26.2 to 34.4 kg/m 2 (median 31.2 kg/m 2 ) for body mass index, and 4.7 to 13.0 years (median 9.0 years) for duration of diabetes.
Analyses were conducted on the full analysis set or intention-to-treat in 36 studies, per protocol in 3 studies, and not reported in 1 study.Details of medications used in each study are provided in supplementary table 1.Of the 33 studies with a dual therapy arm, 27 included a placebo.The quality of studies was rated "good" in 93% of the trials (Supplementary table 2).

HbA1c
A network plot of diabetes treatment combinations for all trials which reported HbA1c and had a treatment duration of 6 to 12 months (20-54 weeks) is presented in Figure 1.Four of the 40 trials included had three arms but two of the trials had two dual therapy arms, hence, 46 direct comparisons were available for comparison among nine therapy combinations (one dual (none/placebo added) and eight triple).Where available, the number of studies per direct comparison between therapy combination pairs ranged from 1 to 13.The triple therapies available for comparison comprised SU, DPP-4, GLP-1, SGLT2, INS, AGI, TZD or MEG added to existing dual therapy.
Compared with none/placebo added to dual therapy, the addition of a third therapy to existing dual therapy resulted in significant additional mean reductions in HbA1c, which ranged from -0.56% [95% confidence intervals: -0.70, -0.42] (-6.2 mmol/mol [-7.7, -4.6]) for DPP-4 to -0.94% [-1.18, -0.70] (-10.3 mmol/mol [-13.0,-7.7]) for TZD (Figure 2).Non-significant mean reductions in HbA1c were observed when AGI or MEG was added to dual therapy.When a third therapy was compared with other third therapies added to existing dual therapy, significant difference was observed between DPP-4 and GLP-1, INS, and TZD (Supplementary table 3).TZD was ranked the most effective in the reduction of HbA1c (SUCRA = 89.6%).There was limited evidence to suggest inconsistency between studies (p = 0.80).Estimates of all pairwise comparisons of triple therapies for HbA1c are provided in Supplementary table 3.
In the 34 studies with around 6 months treatment duration, 38 direct comparisons were available for nine therapy combinations.The overall conclusion did not differ when compared with the 6 to 12 months results.Removal of three studies that were rated "poor" also did not alter the conclusion.

Body weight
Body weight was reported in 27 two-arm studies and 4 three-arm studies (2 with two dual therapy arms), which resulted in 37 direct comparisons available for eight therapy combinations (data were not available for AGI).Compared with none/placebo added to dual therapy, a more favourable weight difference was observed for GLP-1 (-1.85 kg [-2.81, -0.89]) and SGLT-2 (-1.79 kg [-3.03, -0.55]) (Figure 3).There was, however, evidence of inconsistency between direct and indirect evidence (p < 0.0001).According to the SUCRA values for body weight and HbA1c, the two drug classes (TZD, INS) that ranked highest for their effectiveness in the reduction of HbA1c ranked poorly for body weight due to their effect on weight gain (Figure 4).GLP-1 (73.7% and 91.9%) and SGLT-2 (57.3% and 90.6%) have relatively high SUCRA values for both HbA1c and body weight.Estimates of all pairwise comparisons of triple therapies for body weight are provided in Supplementary table 4. Network meta-regression was not performed to investigate whether baseline mean weight was a possible explanation for inconsistency due to the limited studies available for each comparison.However, there was no systematic ordering (higher to lower) of baseline mean weight weighted by sample size according to the order of mean difference in body weight (weight reduction to weight gain) between triple therapy combinations when compared with none/placebo added to dual therapy.
Similar results were obtained when study arms containing insulin were removed from analysis.
For sensitivity analysis of studies with about 6 months treatment duration, similar results were observed except that the mean difference in body weight became non-significant for SU (0.90 kg [-0.40, 2.20]) and there was limited evidence to suggest inconsistency (p = 0.53).

Hypoglycaemia
Twenty-seven studies reported the number of participants that experienced at least one episode of mild, or worse, hypoglycaemia during the treatment period, which resulted in 33 direct comparisons available for nine therapy combinations.Estimates of all pairwise comparisons of triple therapies for hypoglycaemia are provided in Supplementary table 5.
When study arms containing insulin were removed from analysis, the odds of hypoglycaemia also became significantly higher for SU (11.54 [4.75, 27.99]).For studies with about six months treatment duration, only GLP-1 (2.46 [1.24, 4.88]) and INS (3.80 [1.62, 8.89]) were associated with greater odds of hypoglycaemia than none/placebo added to dual therapy.

Discussion
This is the first study which estimated and compared the effectiveness of all triple therapy combinations that have been studied in randomised trials, not limited to those that included both MET and SU [2], on HbA1c and the associated effect on body weight and hypoglycaemia.
There is general consensus that additional reduction in HbA1c can be achieved by adding a third therapy to existing dual therapy treatment.Gross et al reported a 0.70% (7.7 mmol/mol; acarbose) to 1.08% (11.9 mmol/mol; INS) additional absolute HbA1c reduction with a third therapy added to MET/SU compared with MET/SU [2] and in the present study we found an additional 0.56% (6.2 mmol/mol; DPP-4) to 0.94% (10.3 mmol/mol; TZD) HbA1c reduction for a third therapy added to an existing dual therapy.This reduction in HbA1c is clinically relevant.
The UK Prospective Diabetes Study reported a 37% reduction in the risk of microvascular complications and 14% reduction in the risk of myocardial infarction for each 1% (11 mmol/mol) reduction in HbA1c [48].For a difference of 0.8% (9 mmol/mol) in HbA1c, the Action in Diabetes and Vascular Disease: Preterax and Diamicron Modified Release Controlled Evaluation (ADVANCE) trial reported a 14% reduction in the risk of major microvascular events [49].A meta-analysis on the effects of glucose lowering and cardiovascular disease, which included ADVANCE, found that a difference of 0.88% (9.7 mmol/mol) in HbA1c was associated with a 9% reduction in the risk of a major cardiovascular event [50].
The specific question addressed by our study was whether there were clinically relevant differences between the various currently available triple therapy combinations.We found no statistical difference in the reduction of HbA1c between six of the drug classes added as a third therapy to existing dual therapy when compared with dual therapy.However, differences were observed between DPP-4 and GLP-1, INS, and TZD when triple therapies were compared.
These changes in HbA1c should be considered in the context of other clinically relevant effects.
Estimates of all pairwise comparisons of triple therapies for HbA1c, body weight and hypoglycaemia provided in Supplementary tables 3 to 5 can potentially be used to assist in the selection of drug class as add on to existing dual therapy.For example, compared with DPP-4 (a common first choice added to existing dual therapy), GLP-1 is associated with a 0.25% (2.8 mmol/mol) significantly greater absolute reduction in HbA1c and a 1.9 kg significantly more favourable weight change.In contrast, compared with DPP-4, INS is associated with a 0.35% (3.9 mmol/mol) significantly greater reduction in HbA1c but also a significant 2.3 kg weight gain.
Furthermore, there are also clinically relevant differences in risk of hypoglycaemia between the various therapeutic agents [1,51].
The major strength of this study was that we have compared the relative effects of nine drug classes as add-ons to existing dual therapy on the reduction of HbA1c.However, we were unable to conduct in depth analyses of potential confounders due to the limited number of studies available for each pairwise comparison.For instance, the differences in baseline mean age, body mass index, body weight, HbA1c and duration of diabetes among the 40 trials may have influenced the estimates but we were unable to conduct network meta-regression to explore the effects that these differences may have had on our results.Nevertheless, the baseline mean HbA1c weighted by sample size were similar between drug classes, which ranged from 8.3% (67 mmol/mol) to 8.5% (69 mmol/mol) except for MEG (7.5% (59 mmol/mol)), SGLT2 (8.1% (65 mmol/mol)) and AGI (8.8% (73 mmol/mol)).The lack of direct evidence between drug classes also meant that most comparisons were estimated through direct evidence from one or two studies and indirect evidence.We tested for the overall inconsistency in the network using a global method, yet we were unable to test loop specific inconsistency since most pairwise comparisons only have one study [52].Furthermore, we have combined study arms that had the same drug class combinations but the drugs used within the same drug class differed among studies, although differences between drugs within the same class are mostly related to adverse effects, rather than in their efficacy in reducing HbA1c.Therefore, it is clinically plausible to combine studies that used different drugs within the same drug class.We have also combined all existing dual therapy combinations since, in clinical practice, a third therapy is only added when existing dual therapy is no longer effective in controlling blood glucose.The effect of a drug class such as DPP-4 added to any existing dual therapy combinations should result in similar mean difference in HbA1c.We were unable to estimate the odds of severe hypoglycaemia associated with these triple therapy combinations as few studies reported these severe events.Moreover, hypoglycaemia data were the least robust as non-standardised definitions and grading were used.
Clinically relevant additional reduction in HbA1c can be achieved with triple therapy.As described above, the estimates provided in this study may be used to guide clinical practice since it is unlikely that a randomised trial will compare all triple therapy combinations included here.Nevertheless, readers should be aware of the limitations in this study and treat these estimates as a general guide rather than precise evidence.
Astra Zenica, Bristol-Myers Squibb, Glaxo Smith Kline, Janssen-Cilag, Merck Sharp & Dohme, Medtronics, Novartis, Novo Nordisk, Sanofi-aventis, Servier, and Takeda.The number of studies available per direct comparison is provided in the network.The size of the node reflects the number of studies available for the therapy combination.

Described but inappropriate=-1, Described and appropriate=0
Was the study described as double blind?

Yes=1, No=0
Was the method of double blinding described and appropriate (identical placebo, active placebo, dummy, etc.)?

Yes=1, No=0
Deduct one point if the study was described as double blind but the method of blinding was inappropriate (e.g.comparison of tablet vs. injection with no double dummy).

Described but inappropriate=-1, described and appropriate=0
Was there a description of withdrawals and dropouts?

Table 1 :
Characteristics of included studies