Efficacy and Acceptability of Glycemic Control of Glucagon-Like Peptide-1 Receptor Agonists among Type 2 Diabetes: A Systematic Review and Network Meta-Analysis

Objective To synthesize current evidence of the impact of Glucagon-like peptide-1 receptor agonists (GLP-1 RAs) on hypoglycemia, treatment discontinuation and glycemic level in patients with type 2 diabetes. Design Systematic review and network meta-analysis. Data Sources Literature search (Medline, Embase, the Cochrane library), website of clinical trial, bibliographies of published systematic reviews. Eligibility Criteria Randomized controlled trials with available data comparing GLP-1 RAs with placebo or traditional anti-diabetic drugs in patients with type 2 diabetes. Data Synthesis Traditional pairwise meta-analyses within DerSimonian-Laird random effects model and network meta-analysis within a Bayesian framework were performed to calculate odds ratios for the incidence of hypoglycemia, treatment discontinuation, HbA1c<7.0% and HbA1c<6.5%. Ranking probabilities for all treatments were estimated to obtain a treatment hierarchy using the surface under the cumulative ranking curve (SUCRA) and mean ranks. Results 78 trials with 13 treatments were included. Overall, all GLP-1 RAs except for albiglutide increased the risk of hypoglycemia when compared to placebo. Reduction in the incidence of hypoglycemia was found for all GLP-1 RAs versus insulin (except for dulaglutide) and sulphonylureas. For the incidence of treatment discontinuation, increase was found for exenatide, liraglutide, lixisenatide and taspoglutide versus placebo, insulin and sitagliptin. For glycemic level, decrease was found for all GLP-1 RAs versus placebo. Dulaglutide, exenatide long-acting release (exe_lar), liraglutide and taspoglutide had significant lowering effect when compared with sitagliptin (HbA1c<7.0%) and insulin (HbA1c<6.5%). Finally, according to SUCRAs, placebo, thiazolidinediones and albiglutide had the best decrease effect on hypoglycemia; sulphanylureas, sitagliptin and insulin decrease the incidence of treatment discontinuation most; exe_lar and dulaglutide had the highest impact on glycemic level among 13 treatments. Conclusions Among 13 treatments, GLP-1 RAs had a significant reduction with glycemic level but a slight increase effect on hypoglycemia and treatment discontinuation. While albiglutide had the best decrease effect on hypoglycemia and treatment discontinuation among all GLP-1 RAs. However, further evidence is necessary for more conclusive inferences on mechanisms underlying the rise in hypoglycemia.

HbA1c<6.5%. Ranking probabilities for all treatments were estimated to obtain a treatment hierarchy using the surface under the cumulative ranking curve (SUCRA) and mean ranks.

Results
78 trials with 13 treatments were included. Overall, all GLP-1 RAs except for albiglutide increased the risk of hypoglycemia when compared to placebo. Reduction in the incidence of hypoglycemia was found for all GLP-1 RAs versus insulin (except for dulaglutide) and sulphonylureas. For the incidence of treatment discontinuation, increase was found for exenatide, liraglutide, lixisenatide and taspoglutide versus placebo, insulin and sitagliptin. For glycemic level, decrease was found for all GLP-1 RAs versus placebo. Dulaglutide, exenatide long-acting release (exe_lar), liraglutide and taspoglutide had significant lowering effect when compared with sitagliptin (HbA1c<7.0%) and insulin (HbA1c<6.5%). Finally, according to SUCRAs, placebo, thiazolidinediones and albiglutide had the best decrease effect on hypoglycemia; sulphanylureas, sitagliptin and insulin decrease the incidence of treatment discontinuation most; exe_lar and dulaglutide had the highest impact on glycemic level among 13 treatments.
Introduction low propensity for causing hypoglycemia and treatment discontinuation. Indeed, several clinical trials and meta-analyses [16][17][18][19][20][21] for GLP-1 RAs have demonstrated the lowering effect of glycemic levels as well as raised hypoglycemia and treatment discontinuation, although the mechanisms are not very clearly understood. However, since there are so much medicines to choose, which is better for clinical decision is still unknown. So there is a need to include all kinds of GLP-1 RAs simultaneously to assess the impact on hypoglycemia and treatment discontinuation between any two of them.
Therefore, we collected all randomized controlled trials (RCTs) of comparing GLP-1 RAs with placebo or traditional anti-diabetic drugs. A conventional pairwise meta-analysis was performed to summarize current evidence for the effect of GLP-1 RAs on hypoglycemia, treatment discontinuation and glycemic level in patients with T2DM. Additional network meta-analysis was conducted to assess the robustness of the pairwise meta-analysis, supplement missing evidence of head-to-head comparisons by combining both direct and indirect evidence and rank treatments in the evidence network.

Method
Systematic review registration PROSPERO register, CRD42014015328

Search strategy
In consultation with a medical librarian, a search strategy for MEDLINE, EMBASE and the Cochrane library (from inception to June 1st, 2014) was established. The following search strategy for Ovid-MEDLINE was adapted for other databases: 1. exp glucagon-like peptide-1 agonists/ 2. (glucagon like peptide Ã or GLP-1).tw.
3. (exenatide or liraglutide or albiglutide or taspoglutide or lixisenatide or LY2189265).tw. 4. randomized controlled trial.pt. 5. (randomized or randomised).tw. 6. (1 or 2 or 3) and (4 or 5) In addition, completed but unpublished trials were identified from www.clinicaltrials.gov website using the similar search strategy. The bibliographies of published systematic reviews were also searched. All relevant authors and principal manufacturers were contacted to supplement incomplete reports of the original papers or to provide new data for unpublished studies.

Study selection
All the studies included are in English and they are eligible for inclusion only if they were RCTs involving GLP-1 RAs, active anti-diabetic drugs or placebo with complete data on hypoglycemia, treatment discontinuation or glycemic level. Trials are excluded if only they meet one of the following: (1) trials are not RCT (e.g., review, expert comment, editor opinion, new agent introduction, single case report, or case series); (2) if several studies included the same clinical trial, we only include the one which had the longest follow-up time and excluded the other early studies; (3) experimentation on animals or in vitro; (4) not conducted in T2DM; (5) pharmacokinetics research; (6) trials underway, unfinished, or suspended; (7) economical evaluation research; (8) other unrelated researches. These studies were approved by the local ethics committees and written informed consent was obtained from all the patients. The eligibility of studies for inclusion criteria was assessed independently by four reviewers (ZXL, YZ, XCQ and ZRY) in duplicate.

Data extraction and quality evaluation
Data were extracted using ADDIS software [22] with respect to trial information (author, publication year, sample size, trial duration, types of intervention and control), population characteristics (background therapy, diabetes duration, age, baseline level of HbA1c), reported outcomes (Number of hypoglycemia, treatment discontinuation, HbA1c<7.0% and HbA1c<6.5% events in each group) and information on methodology. Four investigators (ZXL, ZRY, XCQ and XTZ) extracted data independently, in duplicate. Any discrepancies were resolved by consensus between the two independent reviewers or by a senior investigator (FS).
Quality of studies was assessed according to JADAD scale[23], including adequate method for randomization, appropriate blinding procedures, and detailed report of withdrawals. The JADAD score was not used as a selection criterion, but only for descriptive purpose.

Data analysis
Methods for direct treatment comparisons. Traditional pairwise meta-analyses was performed using DerSimonian-Laird random effects model [24]. Odds ratio (OR) for hypoglycemia, treatment discontinuation, HbA1c<7.0% and HbA1c<6.5% with 95% confidence interval (CI) were calculated as effect measures. For studies that did not report intention-to-treat, we analyzed outcomes as all-patients randomized. The I 2 -statistic was calculated as a measure of the proportion of the overall variation that is attributable to between-study heterogeneity [25].
Methods for indirect and mixed comparisons. A random-effects network meta-analysis within a Bayesian framework[26] was performed to evaluate the relative effectiveness of each kind of GLP-1 RAs on hypoglycemia, HbA1c<7.0%, HbA1c<6.5% and the relative acceptability on treatment discontinuation. Bayesian network meta-analysis is a generalization of traditional meta-analysis that allows all evidence to be taken into account simultaneously (both direct and indirect). It can be applied whenever a connected network of evidence is available [26]. ORs for hypoglycemia, treatment discontinuation, HbA1c<7.0% and HbA1c<6.5% with 95% credible interval (CrI) were summarized. The posterior densities for all unknown parameters were estimated using MCMC (Markov chain Monte Carlo) for each model. Each chain used 40 000 iterations with a burn-in of 20 000.
Network meta-analyses enable estimation of the probability that each intervention is the best for each outcome. Probabilities for each treatment taking each possible rank were plotted in absolute rankograms or cumulative rankograms. Besides, the surface under the cumulative ranking curve (SUCRA)[27] were used to estimate the ranking probabilities for all treatments in order to obtain a treatment hierarchy. SUCRA is a percentage interpreted as the percentage of efficacy of a treatment on the outcome that would be ranked first without uncertainty, which is equal to 1 when the treatment is certain to be the best and 0 when it is certain to be the worst [27].
An absolute measure of fit D res , was considered to formally check the model's overall fit. D res is the posterior mean of the residual deviance (the deviance for the fitted model minus the deviance for the saturated model). Ideally, each data point should contribute about one to the posterior mean deviance so that it can be compared to the number of data points for the purpose of checking model fit [28].
Loop-specific approach was used to evaluate the presence of inconsistency locally in network meta-analysis models, that is, if the information of both sources of evidence is similar enough to be combined [29]. This method evaluates the consistency assumption in each closed loop of the network separately. Difference (inconsistency factor) with 95% CIs between direct and indirect estimations for a specific comparison was calculated to assess the presence of inconsistency in each loop. Inconsistency was defined as disagreement between direct and indirect evidence with a 95% CI excluding 0.

Model fit and inconsistence check
Statistical inconsistency between direct and indirect comparisons was generally low for four outcomes. Most loops (networks of three or four comparisons that arise when collating studies involving different selections of competing treatments) were consistent, since their 95% CIs included 0 according to the forest plots, which meant the direct estimation of the summary effect did not differentiate from the indirect estimation. The summary estimations of network meta-analysis are relatively robust.
The model fit was evaluated using the posterior mean of the residual deviance D res . The values of the D res for hypoglycemia, treatment discontinuation, HbA1c<7.0% and HbA1c<6.5% were 121.98, 83.34, 131.68 and 86.80 respectively, which were close to corresponding 152, 104,145 and 104 of the number of data points for four outcomes, meaning that model's overall fit is relatively satisfactory.

Discussion
Aside from adequate glycemic control, increasing attention is being paid to the hypoglycemia and treatment discontinuation effect of GLP-1 RAs recently [14, 15]. Our network meta-analysis suggested that all GLP-1 RAs significantly increase the risk of hypoglycemia compared with placebo (except for albiglutide), and reduce the risk of hypoglycemia compared with insulin (except for dulaglutide) and SU. In terms of the increasing incidence of treatment discontinuation, exenatide, liraglutide, lixisenatide and taspoglutide had significant effect when compared with either placebo, insulin, SU or sitagliptin, and exe_lar only increased the incidence of treatment discontinuation significantly when compared with SU. This was accompanied by taspoglutide in comparison to TZD. Besides, all GLP-1 RAs decreased glycemic level compared with placebo, and dulaglutide, exe_lar, liraglutide and taspoglutide had significant lowering effect when compared with sitagliptin (HbA1c<7.0%) and insulin(HbA1c<6.5%). Regarding to HbA1c <6.5%, there was also a significant lowing effect for exe_lar and liraglutide in comparison to SU and sitagliptin, dulaglutide in comparison to sitagliptin, exe_lar in comparison to Met.

Effect on hypoglycemia
Hypoglycemia is a common complication of intensive diabetes therapy, which could cause fall, seizure, coma, and even death [109]. The UK Prospective Diabetes Study (UKPDS) reported that maintenance of tight glycemic control in T2DM with insulin treated led to a significant increase in the incidence of hypoglycemia [110]. Our network meta-analysis showed that the significant increasing in the incidence of hypoglycemia was associated with all GLP-1 RAs except for albiglutide, which was consistent with Riddle's study [90]. Riddle's results showed that the incidence of symptomatic hypoglycemia was 28% for lixisenatide and 22% for placebo, and 1.2% subjects had severe hypoglycemia with lixisenatide vs. 0.0% with placebo. While, the beneficial hypoglycemia lowering effect of all GLP-1 RAs was observed when compared with insulin (except for dulaglutide) and SU, which was consistent with previous reviews [10,111] and clinical trials [112]. Besides, studies also reported that the incidence of hypoglycemia was similar across GLP-1 RA treatment groups, and most of patients with hypoglycemia had the history of treating with concomitant SU therapy [111,113].
To date, the mechanism of hypoglycemia for T2DM has not been clearly identified. It may involve complex regulation, but it has been shown that β-cell failure precede defects of α-cell response to lowering glucagon levels in T2DM, indicating that the counter-regulatory effect of glucagon to hypoglycemia is impaired in T2DM [114,115]. Fukuda's [116] study reported that the degree of α-cell dysfunction is related with the lack of β-cell function in diabetes. Commonly, insulin represses glucagon secretion as a pulsatile manner, but this coordination is disrupted in patients with T2DM and it could potentially contribute to glucagon dysregulation [117]. So finally, the defect of an increment in glucagon secretion during hypoglycemia is the result of β-cell failure in advanced T2DM [118].

Treatment discontinuation increasing effect
Our network meta-analysis showed that exenatide, liraglutide, lixisenatide and taspoglutide had significant increasing effect on the incidence of treatment discontinuation when compared with either placebo, insulin, SU or sitagliptin. Exe_lar only increased the incidence of treatment discontinuation when compared with SU. This was accompanied by taspoglutide in comparison to TZD. Similar results were indicated in relevant clinical trials [113].
Several reasons may be account for this. Firstly, All GLP-1 RAs are injected subcutaneously, and cannot be administered orally. The incidence of treatment discontinuation among patients who had injection site adverse events was high [119]. Secondly, the adverse events of GLP-1 RAs like nausea, diarrhea, and vomiting, also account for the incidence of treatment discontinuation [120]. Especially for the most commonly occurred nausea, which usually lasts a long time, is a tough experience for T2DM to bear.

Glycemic level lowering effect
The beneficial glycemic level lowering effect of all GLP-1 RAs in our analysis was consistent with previous studies [10,121]. Scheen's [122] study reported that the HbA1c lowering potential for GLP-1 RAs is approximately at 1%-1.5% on average. A review of 8 head-to-head phase III clinical programs showed that the primary efficacy endpoint in all of the GLP-1 RAs was change in HbA1c from baseline with a noninferiority margin of 0.4% [111]. Similar results were indicated in relevant clinical trials. The significant glycemic level lowering effect of EXQW was observed in series of DURATION trials, with mean reductions of -0.9% to -1.63% [18,44,[123][124][125][126]. Liraglutide was found to lower HbA1c by -0.9 to -1.1% [127].

Strengths
A major strength of our study is the inclusion of a substantially greater number of trials of GLP-1 RAs than earlier meta-analysis[16, 20, 21], thus it is the largest completed evaluation of GLP-1 RAs' effect on hypoglycemia, treatment discontinuation and glycemic level to date. Furthermore, the network meta-analysis based on Bayesian model makes indirect comparison among multiple treatments available, especially when there are few trials for direct comparison between different anti-diabetic drugs, such as comparisons between dulaglutide and insulin in our study. Network meta-analysis has been proved to be the most appropriate method for multiple treatments comparison to date [26,129]. In addition, the network technique enables the estimation of the probability that one intervention is the best for one outcome. Thus it can provide an explicit ranking when many treatments are competing for one outcome. Our study provided the ranks of GLP-1 RAs and traditional anti-diabetic drugs on hypoglycemia, treatment discontinuation and glycemic level for the first time.

Limitations
Several limitations are worthy to be mentioned. First, only trials only in English were included, and our literature search was from inception to June 1st, 2014, and didn't include literatures published after June 1st, 2014, which may lead to potential publication bias and selection bias. Secondly, none of the trials included was specially designed to evaluate the effect of GLP-1 RAs on hypoglycemia, treatment discontinuation and glycemic level. Thirdly, the different duration of years of T2DM in 78 trials may cause heterogeneous, which may influence the different response to therapy and increase the possibility of hypoglycemia. Thus the results of our study should be considered as hypothesis generation, and any conclusions should be drawn with caution.

Conclusion
In conclusion, our network meta-analysis presents the associations amongGLP-1 RAs, traditional anti-diabetic drugs and placebo on hypoglycemia, treatment discontinuation and glycemic level. GLP-1 RAs had the lowering effect on glycemic level, increasing effect on hypoglycemia and treatment discontinuation. While, GLP-1 RAs were associated with lower incidence of hypoglycemia when compared with active comparators. However, further evidence is necessary for more conclusive inferences on mechanisms underlying the increasing in hypoglycemia.