Comparison of antidiabetic drugs added to sulfonylurea monotherapy in patients with type 2 diabetes mellitus: A network meta-analysis

Aims This study aimed to investigate the efficacy and safety of dual therapy comprising sulfonylurea (SU) plus antidiabetic drugs for the treatment of type 2 diabetes mellitus (T2DM). Methods We searched the PubMed, Cochrane library, and Embase databases for randomized clinical trials (≥24 weeks) published up to December 28, 2017. Subsequently, we conducted pairwise and network meta-analyses to calculate the odds ratios (ORs) and mean differences (MDs) with 95% confidence intervals (CIs) of the outcomes. Results The final analyses included 24 trials with a total of 10,032 patients. Compared with placebo, all treatment regimens were associated with a significantly higher risk of hypoglycemia, except the combinations of SU plus sodium-glucose co-transporter-2 inhibitor (SGLT-2i) [OR, 1.35 (95% CI: 0.81 to 2.25)] or alpha-glucosidase inhibitor (AGI) [OR, 1.16 (95% CI: 0.55 to 2.44)]. Notably, the combination of SU plus glucagon-like peptide-1 receptor agonist (GLP-1RA) was associated with the most significant increase in the risk of hypoglycemia. Furthermore, all SU-based combination regimens reduced the glycated hemoglobin (HbA1c) and fasting plasma glucose levels (FPG). However, only combinations containing SGLT-2i [MD, -1.00 kg (95% CI: -1.73 to -0.27)] and GLP-1RA [MD, -0.56 kg (95% CI: -1.10 to -0.02)] led to weight loss. Conclusions Our findings highlight the importance of considering the risk of hypoglycemia when selecting antidiabetic drugs to be administered concomitantly with SU. Although all classes of antidiabetic drugs improved glucose control when administered in combination with SU, SGLT-2i might be the best option with respect to factors such as hypoglycemia and body weight.


Introduction
According to the most recent data from the International Diabetes Federation (IDF), the number of adults affected by diabetes worldwide reached 425 million in 2017, and approximately 4 million deaths were attributed to this disease.Current estimates suggest that 629 million people worldwide will be affected by diabetes in by 2045 [1].Diabetes can be stratified into two types; of these, type 2 diabetes mellitus (T2DM) is a progressive disease characterized by initial insulin resistance and a subsequent decline in β cell function.Currently, metformin is the preferred therapeutic option for T2DM [2,3].However, metformin use is contraindicated for or not tolerated by some patients.In such cases, other hypoglycemic agents, including sodiumglucose co-transporter-2 inhibitors (SGLT-2is), dipeptidyl peptidase 4 inhibitors (DPP-4is), thiazolidinediones (TZDs), sulfonylureas (SUs), and alpha-glucosidase inhibitors (AGIs), may serve as first-line options [3].SUs improve blood glucose by stimulating the β cells to secrete insulin in a non-glucose-dependent manner.Given the low costs and favorable efficacy and safety profiles, these drugs are used extensively following an initial diagnosis of T2DM, despite the potential to increase the incidence of hypoglycemic events and weight gain [4,5].
As noted, however, β cell function declines over the course of T2DM, and most patients who initially used SUs will later require antidiabetic drug combination therapies to maintain glycemic control [6].Several such SU-based combination regimens including: SGLT-2is, DPP-4is, TZDs, AGIs, glucagon-like peptide-1 receptor agonists (GLP-1RAs), and insulin are currently administered to patients.The process of selecting the appropriate second-line antidiabetic drug is complicated by various factors; in addition, the glycemic effects of the drugs, such as hypoglycemia and weight gain, must be considered because these can impact the patient's adherence to treatment and quality of life [7,8].
With this work, we aimed to conduct a systematic review and network meta-analysis (NMA) of the most recently updated clinical trials to evaluate the efficacy and safety of antidiabetic drugs as add-on treatments for T2DM inadequately controlled with sulfonylurea monotherapy.Compared with the conventional pairwise meta-analysis method, the NMA enables us to calculate data from both direct and indirect comparisons of diverse regimens and to quantify and sort the efficacy and safety of each of these measures [9].Using this approach, we hope to provide evidence to assist clinicians and patients with decision-making.

Methods
The methods and results of this NMA have been reported in accordance with the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) [10] recommendations and checklist (S1 Table ).

Study selection and data extraction
Included studies were required to meet the following criteria: (1) RCT design; (2) study duration !24 weeks; (3) inclusion of adult (age: !18 years) patients with T2DM and inadequate glycemic control with sulfonylurea monotherapy; (4) drugs in the SGLT-2i, DPP-4i, GLP-1RA, TZD, AGI, metformin, and insulin classes; and (5) assessment of at least one of the following continuous outcomes-glycated hemoglobin (HbA1c), fasting plasma glucose (FPG), body weight-and dichotomous outcomes-hypoglycemia, serious adverse events (SAEs).SAEs were defined as fatal or life-threatening events, events requiring or extending inpatient hospitalization, those resulting in ongoing or significant incapacity or interfering substantially with normal life functions, and/or those that caused a congenital anomaly or birth defect.The primary outcome was hypoglycemia.Studies that met the following criteria were excluded: (1) patients with serious cardiovascular disease or severe renal impairment; (2) pregnant patients; (3) sample size <100; (4) non-randomized trials; and (5) publication as a conference report, letter, or abstract.
Two authors (D.Q. and T.Z.) independently assessed the titles and abstracts identified in the initial search and reviewed the full texts of all identified studies that met the inclusion criteria.The following details were recorded using a pre-defined spreadsheet: first author (publication year), study duration, interventions (types and doses), sample size, and baseline participant information (HbA1c, body mass index, body weight, age, sex).For studies with different follow-up durations, the longest reported duration was recorded.Any conflicts were resolved by the third author (P.Z.).

Risk of bias
The Cochrane risk-of-bias tool [11] was used to assess the quality of the included RCTs.This tool assesses 7 domains, namely random allocation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, incomplete outcome data, selective reporting, and other biases.For each domain, the risk of bias was determined to be low, unclear, or high.

Statistical analysis
In this study, the results of comparisons of dichotomous (hypoglycemia, SAEs) and continuous variables (HbA1c, FPG, and body weight) are reported as odds ratios (ORs) and mean differences (MDs), respectively, together with the corresponding 95% confidence intervals (CI).
A DerSimonian-Laird random effects model was used for the conventional pairwise metaanalysis [12].Heterogeneity was evaluated using the I 2 statistic, with an I 2 of 25%, 50%, or 75% indicating low, moderate, or high heterogeneity, respectively [13].For the NMA, a frequentist random-effects model was used with the assumption of a common between-study covariance structure across the treatment arms; this is often referred to as a homogenous variance assumption [14,15].In such a model, we included only 2 arms from trials including 3 overlapping arms (e.g., only A versus B from a trial with A versus B versus A + B).In addition, if the same drug was evaluated at different doses in the trials, we combined different doses into a single dose.For this analysis, we used the mvmeta and network commands in Stata software (Stata Corp, College Station, TX, USA) and the programmed Stata routines [16,17].To rank the probabilities of each intervention in terms of efficacy and safety outcomes, we used surface under the cumulative ranking (SUCRA) curves and mean ranks.Higher SUCRA values indicate better efficacy or safety [18].A 0.5 zero-cell correction was applied when studies reported zero events [19].The heterogeneity variance (tau [τ]) was estimated using a restricted maximum likelihood method and employed to estimate heterogeneity [20].For the purposes of this analysis, we assumed that common methods of comparing glucose-lowering strategies were used similarly when reported by different trials and that participants included in those trials could be randomly allocated to any of the compared treatments.
To check for possible inconsistency, a loop-specific approach was used to evaluate differences between the direct and indirect estimates in each closed triangular or quadrangular loop [21].Additionally, a "design-by-treatment" interaction model was applied to evaluate global heterogeneity within networks [22].The small-study effects of active treatments versus placebo were assessed using comparison-adjusted funnel plots [23].
All P-values were 2-tailed, and P values of <0.05 were considered statistically significant.All analyses were performed using Stata 14.1 software.

Characteristics of the included studies
The PRISMA flow chart of this study is shown in Fig 1 .A total of 38,655 articles were obtained from the searched databases, from which 427 potentially relevant articles were identified after de-duplication and abstract screening.Trials not involving add-on sulfonylurea therapy (245), those with a non-RCT design (34) or a study duration of <24 weeks (41), meta-analyses and pooled analyses (64), conference abstracts (28), and studies with no available data (30) were excluded.Finally, 24 eligible studies satisfied the inclusion criteria and were included in this NMA.
The included studies were published from 2000 to 2017 and had enrolled 10,032 (range, 105-1041) participants with T2DM.All 24 RCTs  reported data for follow-ups ranging from 24 to 52 weeks' duration, while 2 [46,47] additionally reported data between 52and 104 weeks.The mean [range] baseline HbA1c was 8.5% [7.6-9.9%], the mean ages ranged from 52 to 75 years, the mean [range] baseline BMI was 28.6 [23.8-32.2]kg/m 2 , and the mean [range] baseline weight was 78.1 [62.6-99] kg.Table 1 summarizes the detailed information from the included studies.The baseline characteristics of participants in these studies were deemed sufficiently similar in terms of age, sex, HbA1c, body weight, and body mass index (BMI) to permit network comparison (S1 Fig).There were also no specific clinical reasons (based on the inclusion and exclusion criteria of every trial in the network) to suggest that the type of participants under one comparison would be different from the type of participants in other comparisons.The S3-S6 Tables detail the numbers of participants included in the efficacy and safety outcome analyses by study and drug class.
The Cochrane system bias evaluation is shown in S7 Table and S2 Fig.
All studies were randomized and double blinded, and the risks of bias for random sequence generation, concealment of treatment allocation, and blinding of participants and personnel were low or unclear.Three studies had a high risk of reporting bias, of which 1 presented a high risk of detection bias.One study had incomplete outcome data.

Network consistency
The networks of eligible comparisons of efficacy and safety outcomes are graphically displayed in Fig 2 .There were no loop inconsistencies between the evidence derived from direct and indirect comparisons for the 95% CIs of the IF values including zero values (S8 Table ).In addition, the design-by-treatment model did not detect global inconsistency within any network (p for all >0.05,S9 Table ).The contribution of each study to NMA is shown in S10 Table.

Other safety outcome: Serious adverse events
SAE data were available for 6335 participants in 17 RCTs (S3 Table ).The NMA indicated no significant differences in the ORs of SAE for any agent when added to SU.Excluding placebo,   basal insulin and AGI received the highest (76.8%) and lowest (36.2%)SUCRA values, respectively (Table 3).

Publication bias
A comparison-adjusted funnel plot of all outcomes is displayed in S2 Fig.This analysis indicated no evidence of small-study effects with respect to active treatments versus placebo in the network.

Discussion
Only 1 previous meta-analysis [48] assessed the risk of hypoglycemia of DPP-4i versus placebo when added to SU, and no head-to-head trials have estimated the relative effects of other antidiabetic drugs (especially SGLT-2i and GLP-1RA).To address this paucity of research, we performed the current NMA to combine high-quality data from the most updated trials and thus comprehensively compare the effects of SGLT-2i, DPP-4i, GLP-1RA, TZD, Met, AGI and basal insulin in patients with T2DM that was inadequately controlled by SU monotherapy.We
Hypoglycemia is a serious clinical event that cannot be neglected during T2DM treatment.Related symptoms such as weakness, nervousness, trembling, and palpitations negatively affect the patient's quality of life and are closely associated with the risks of cardiovascular disease and hospital admission [49][50][51].Additionally, even mild hypoglycemia can cause mental distress [52].In our NMA, we found that when compared with placebo, all agents except SGLT-2i  and AGI were associated with an increased risk of hypoglycemia when added to SU.Of these agents, GLP-1RA most significantly increased the risk of hypoglycemia relative to all other agents except Met and basal insulin and thus received the worst ranking.Our findings were consistent with the previous meta-analysis conducted by Salvo et al. [48], in which the combination of DPP-4i and SU was associated with a 50% increased risk of hypoglycemia versus placebo plus SU.However, we did not detect significant differences in the risks of hypoglycemia associated with SGLT-2i, DPP-4i, TZD, Met, or AGI plus SU.Generally, SGLT-2i reduces the risk of hypoglycemia, whereas SU drugs increase this risk.Although our results indicate that SGLT-2i and AGI did not cause statistically significant increases in the risk of hypoglycemia when added to SU, the SU dosage may need to be adjusted in such regimens to mitigate this risk [48].Beta-cell dysfunction plays a vital role at all stages in the pathogenesis of T2DM, and these effects are compounded by insulin resistance [53,54].Our study demonstrated that all classes of antidiabetic drugs improved glucose control relative to the placebo (0.59-1.12% decrease in HbA1c and 0.68-2.37mmol/l decrease in FPG) when combined with SU.Of these drugs, GLP-1RA was associated with the greatest reduction in HbA1c levels when added to SU monotherapy.Furthermore, no single agent other than GLP-1RA could significantly reduce HbA1c to a greater extent than any other when added to SU.However, only 2 RCTs included in our study reported durations >52 weeks [46,47].As HbA1c is a relatively stable variable used to reflect long-term glucose control, future research may be needed to estimate the longterm efficacies of these agents when combined with SU.
Weight gain has been suggested to correlate with an increased risk of diabetes, suggesting that weight control would be favorable for blood glucose control [55,56].Our NMA showed that compared to placebo, only SGLT-2i and GLP-1RA were associated with significant decreases in body weight when added to SU. Notably, SGLT-2i yielded significant improvements in this parameter when compared with all other agents except GLP-1RA and AGI and was therefore ranked the best.A previous NMA by Kay et al. [57] demonstrated that all DPP-4is were associated with mean body weight gains relative to placebo when added to metformin and SU.Although that study focused on triple combination therapy (DPP-4i + Met + SU), these findings were consistent with our study results.
To our knowledge, ours is the first NMA to compare the efficacy and safety profiles of all available agents in patients with T2DM inadequately controlled with SU monotherapy.The previous meta-analysis conducted by Salvo and his colleague [48] evaluated the effects of DPP-4i on the risk of hypoglycemia relative to placebo, rather than to other active agents.Therefore, we aimed to fill the gaps left by that study.We comprehensively estimated the relative effects of all available agents administered in combination with SU to provide additional information that would assist clinicians with decision-making.However, some potential limitations of our study should be considered.First, the majority of the RCTs had study durations of 24-52 weeks, while only 2 had follow-up durations >52 weeks.Obviously, therefore, our conclusions should be applied cautiously when evaluating the long-term effects of these antidiabetic drugs when combined with SU.Second, although we evaluated the effects of each class of antidiabetic drugs as a whole, some within-class differences were observed.Further studies might focus on the effects of different doses of these drugs when combined with SUs for T2DM.Third, the varying definitions of hypoglycemic events in the included studies may have contributed to clinical heterogeneity.Fourth, the quality of our analysis might have been compromised by inter-study differences in patient discontinuation rates.Fifth, we did not assess the baseline age, sex, HbA1c, body weight, BMI, diabetes duration, or duration of treatment, as effect modifiers when estimating efficacy and safety outcomes; the focus of further studies should be on evaluating the effects of these variables.Finally, only 1 RCT included data for basal insulin added to SU, which had a relatively wide confidence interval.This factor might have had some effect on our conclusions.
In conclusion, all classes of antidiabetic drugs improved glucose control when added to SU.However, SGLT-2i exhibited superior effects in terms of weight loss and did not increase the risk of hypoglycemia, suggesting that it might be the best option.Clinicians should particularly consider the risk of hypoglycemia when selecting antidiabetic drugs for administration together with SU.