Efficacy of Various Antidiabetic Agents as Add-On Treatments to Metformin in Type 2 Diabetes Mellitus: Systematic Review and Meta-Analysis

Background and Aim. Diabetes mellitus is a chronic disease that has a great impact on patients and society. Metformin monotherapy is capable of maintaining a target glycemic control only for a short term. The aim of this study was to determine the efficacy of combination therapy of metformin with any antidiabetic agents in type 2 diabetes mellitus (T2DM) patients. Methods. Reports of randomized controlled trials (RCTs) of combination therapy of metformin with various antidiabetic agents in T2DM failing metformin alone were identified. Results. Eight studies were identified in our paper. Thiazolidinediones (TZDs) were as effective as dipeptidyl peptidase IV inhibitors (DPP IV inhs) in reducing HbA1c value (pooled mean difference −0.03%; 95% CI −0.16 to 0.10%). In comparison between TZDs and sulphonylureas (SUs), TZDs reduced fasting plasma insulin (FPI) more effectively than SUs (pool mean difference −5.72 μU/mL; 95% CI −8.21 to −3.22 μU/mL, P < 0.00001), but no significant differences were detected in the effects on HbA1c and fasting plasma glucose (FPG) (pooled mean difference −2.19 mg/dL; 95% CI −11.32 to 6.94 mg/dL, P = 0.64). Conclusions. Our study showed that TZDs reduced FPG better than did DPP IV inhs and decreased FPI more than did SUs.


Introduction
Diabetes mellitus is a chronic disease that has a great impact on patients and society. The estimate of worldwide prevalence by the International Diabetes Federation (IDF) in 2007 is about 246 million people, most of whom (85-95%) have type 2 diabetes mellitus (T2DM) [1]. The complications are the major causes of morbidity and mortality of people with diabetes [2][3][4]. Antidiabetic agents have an important role in normalizing plasma glucose levels [5,6]. Metformin is now recommended as the first agent for blood glucose lowering in type 2 diabetes patients [7][8][9][10][11]. Metformin has been proven to be efficacious in reducing cardiovascular risk [12] and is the only pharmacological treatment that could improve macrovascular outcomes in patients with diabetes [8]. However, metformin monotherapy is capable of maintaining a target glycemic control only for a short term [13]. Combination therapies of metformin with other oral antidiabetic agents are therefore necessary. There are many therapeutic options of adding second agents in metformintreated subjects which are recommended [7][8][9][10][11], such as insulin, insulin secretagogues, thiazolidinediones (TZDs), glucagon-like-peptide-1 (GLP-1) analogues, and dipeptidyl peptidase IV inhibitors (DPP IV inhs).
Our paper was aimed at determining the efficacy of combination therapy of metformin with any antidiabetic agents in type 2 diabetes inadequately controlled with metformin alone.

Search Strategy.
Reports of randomized controlled trials of combination therapy of metformin with various antidiabetic agents in type 2 diabetes failing metformin monotherapy were identified through a systematic literature search of MEDLINE (Pubmed), EMBASE, and The Cochrane Library. The following MeSH terms were used: diabetes mellitus type 2, metformin, sulfonylurea compounds, thiazolidinediones, dipeptidyl peptidase IV inhibitors, insulin, insulin NPH, and insulin long acting. This was followed by keyword search using as keywords glibenclamide, glyburide, gliclazide, glimepiride, glipizide, chlorpropamide, tolbutamide, meglitinide, nateglinide, repaglinide, pioglitazone, rosiglitazone, sitagliptin, vildagliptin, saxagliptin, alogliptin, glargine, lispro, aspart, glulisine, detemir, acarbose, voglibose, miglitol, exenatide, liraglutide, and pramlintide. Historical searches of reference lists of relevant randomized controlled trials, systematic and narrative reviews were also undertaken. No language restriction was imposed.

Study Selection.
Eligible studies were selected by two reviewers, and differences were resolved by agreement. The studies were included in this systematic review, if they (a) were randomized controlled trials in type 2 diabetes patients who had already been treated with metformin alone, (b) compared between two different antidiabetic drugs in combination with metformin, (c) included patients with baseline HbA1c ≥ 7%, (d) had no addition of a third oral antidiabetic agent or insulin, (e) lasted at least 12 weeks of treatment duration, and (f) reported outcome measure in terms of hemoglobin A1c (HbA1c). Abstract presentations were excluded.
We selected 12 weeks as the minimal study duration to assure the effect of medication on HbA1c level. HbA1c at ≥7% was chosen because the recent meta-analysis of the benefit of intensive glucose-lowering treatment reported overall HbA1c target level of <7%. Intensive glycemic control group (HbA1c < 7%) had no significant effects on stroke and all-cause mortality when compared with standard treatment where HbA1c goal was ≥7% [14]. Some guidelines recommend a target of HbA1c at ≤6.5% [7,8,11]. Achieving these goals may prove difficult. Only 60% of subjects have been reported to reach an HbA1c goal of ≤7.5% [15]. In addition, the ADA/EASD consensus treatment algorithm for the metabolic management of diabetes recommends a HbA1c goal at <7% [9, 10].

Data Extraction and Quality
Assessment. Data extraction and study quality assessment were performed independently by two investigators using a standardized form. Disagreements were resolved by a third investigator. The data extracted were publication year, study country, study design, study duration, outcome parameters, type and dosage of interventions, patient characteristics, and number of participants.
The methodological quality of each trial was assessed using the scale developed by Jadad et al. [16]. The Jadad's scale is divided into three dimensions: randomization, blinding and reasons for withdrawals and dropouts. The possible maximum score is 5 points, the studies with the score of 2 points or less are of low quality, while those with 3 points or more are of high quality.

Statistical Analysis.
Primary outcome was HbA1c, secondary outcomes were fasting plasma glucose (FPG) and fasting plasma insulin (FPI). Efficacy was reported as mean change value from baseline to final assessment. When the variations of these changes were not reported, we estimated them by using the following equation: [17] SD 1 (C)= SD 1 (B) 2 where SD 1 (C) is the standard deviation of change, SD 1 (B) and SD 1 (F) are the standard deviations of baseline and final values: respectively, and R is the correlation coefficient and was assumed to be 0.5. In addition, when the variations were not reported at all, the pooled SD calculated from the studies data that reported SD was used. Treatment effect was estimated with mean difference in the change value between the treatment group and the control group. The inverse variance-weighted method was used for the pooling of mean difference and the estimation of 95% confidence interval [18]. Random effects model was used to combine the results of individual studies when Q-statistic test was significant at the level of 0.1 [19], otherwise the fixed effects model was used [18]. I-squared statistic which is the percentage of total variation across studies was used to quantify the level of heterogeneity [20]. Sensitivity analysis was performed to examine the causes of heterogeneity. Publication bias was assessed by the method of Egger et al. [21]. The statistical analysis was undertaken with RevMan 5.0 (Cochrane collaboration). The statistical significance was set at P < 0.05.

Study Characteristics.
The initial search identified 121 potentially relevant randomized studies of additional therapy to metformin in type 2 diabetes mellitus. All were published in English. One hundred and thirteen trials were excluded for the following reasons. Seven trials were excluded as they evaluated agents already withdraw from the market, that is, inhaled insulin, muraglitazar, and tesaglitazar. Fortyfive trials that included subjects previously receiving various antidiabetic regimens including metformin and did not report data separately for each antidiabetic agent therefore were also excluded. Thirty-three studies were excluded since they were placebo-controlled or non-treatment-controlled. Nine trials with entry HbA1c < 7% were excluded. We excluded one trial which added a third oral glucoselowering agent or insulin. In addition, this particular trial was previously reported in three preliminary publications and therefore was excluded. One trial was excluded as the duration of study was less than 12 weeks. Seven trials were further excluded because they were duplication or interim analysis. One trial evaluated vildagliptin 100 mg once daily which is currently not a recommended dose (the recommended dose is now 50 mg twice daily), thus it was excluded. Six studies were abstract presentation and were then excluded. The remaining eight trials met our inclusion criteria and were included in the meta-analysis [22][23][24][25][26][27][28][29].

Discussion
The results of this analysis suggest that TZDs were as effective as DPP IV inhs in reducing HbA1c value in type 2 diabetes patients who had been treated with metformin alone, however, FPG better improved with TZDs than with DPP IV inhs. From its mechanism of actions, TZDs may reduce FPI more than does DPP IV inhs. In addition, there are some issues that should be concerned. First, patients in both trials received fixed-dose TZDs (rosiglitazone 8 mg/day [22], pioglitazone 30 mg/day [23]) and there was no retitration. Indeed, dose of TZDs in combination therapy with metformin should be titrated from initial dose to the maximum dose (rosiglitazone 4-8 mg/day, pioglitazone 15-45 mg/day) based on efficacy in reducing hyperglycaemia and tolerability such as side effects. Second, maximum dose of TZD was not used (pioglitazone 30 mg OD was used) whiles maximum recommended dose of the DPP IV inh (vildagliptin 50 mg bid) was used in one trial and patients compliance was not determined [23]. That may affect the comparability between the comparators. TZDs reduced FPI more effectively than SUs, but no significant differences were detected in the effects on HbA1c and FPG. All subjects in the included studies had hyperinsulinemia (Table 4). Hyperinsulinemia is recognized as a risk factor that affects the development and progression of atherosclerosis and cardiovascular disease which are the major causes of morbidity and mortality in diabetes patients [30]. TZDs improve glycaemic control by promoting the local effect of insulin. TZDs also decrease the gluconeogenesis from hepatic tissues, resulting in reduced insulin resistance, which leads to improved glycemic control with no enhancement in the insulin secretion. [31], while SUs improve blood glucose level by triggering insulin release from the pancreatic β-cell [32]. Thus, TZDs and SUs may have similar effect in controlling blood glucose level but have different effect on FPI level.    Test for overall effect: Z = 3.34(P = 0.0008) Figure 3: Effects of various antidiabetic agents on FPI as an add-on treatment to metformin in T2DM. Evidence of interstudy heterogeneity was discovered in the comparison of TZDs against SUs based on HbA1c (χ 2 = 8.66, P = 0.03; I 2 = 65%, Figure 1). Similar results were found with FPG outcome (χ 2 = 12.42, P = 0.006; I 2 = 76%) and FPI parameter (χ 2 = 16.23, P = 0.001; I 2 = 82%). Sensitivity analysis was performed by excluding Garber's study [25] which was different from other studies in various aspects. Firstly, it lasted only 24 weeks, the shortest treatment duration among the trials that compared TZDs against SUs. Hence, TZD may not demonstrate maximum effectiveness in reducing hyperglycemia that was supported by ADOPT trial in which TZDs showed the maximum benefit in decreasing HbA1c at 48 week [33]. Secondly, the dose of metformin could be up-and downtitrated, while other trials used metformin at baseline dose and maintained the dose throughout the course of studies. Finally, the medication could be titrated at every visit (4 week thereafter for 24 weeks) depending on mean daily glucose or fructosamine levels, while other trials employed restriction of forced-titration period (approximately 1 to 16 weeks). When this study was excluded, no significant heterogeneity was observed with regard to HbA1c (χ 2 = 2.74, P = 0.25; I 2 = 27%) and FPG results (χ 2 = 3.70, P = 0.16; I 2 = 46%). The effect estimates remained unchanged. The pooled mean differences were 0.01% (95% CI −0.11 to 0.14%) and −6.06 mg/dL (95% CI −13.08 to 0.95 mg/dL), respectively for HbA1c and FPG values. Interstudy heterogeneity in the pooling of FPI may be There were variations and limitations among the included studies. Several treatment strategies including diet control, exercise training, and antidiabetic agents are useful tools to improve glycaemic control in diabetes. Only two studies provided details about dietary advice [24,25]. There were variations in criteria for dose titration. Garber et al. [25] titrated study medication according to mean daily glucose levels measured by self-monitoring of blood glucose (SMBG) method or fructosamine levels. In addition, downward titration was permitted only in glibenclamidearm [25]. While in other trials, the study medications were titrated to higher or lower doses based on tolerability, such as actual hypoglycemia or enhanced risk of hypoglycemia [24]. Umpierrez et al. [26] administered active agent with forced titration to a maximum dose or adjusted dose based on fasting SMBG. However, only pioglitazone treatment was uptitrated based on HbA1c levels [26]. Down-and uptitrating the dose were also performed according to mean daily glucose from SMBG results [27]. Furthermore, the duration allowed for the dose titration also varied among the included trials. Most trials had limited period for dose titration and maintained that dose until the end of study except the study by Garber et al. in which the dose could be increased or decreased every 4 weeks throughout 24-week study duration [25]. In the other trial, duration for dose titration differed between the intervention and the control groups [26]. The methodological quality of the included studies was assessed by Jadad's score. Studies scoring 3 or greater were rated as high quality trials [16]. Of the eight trials, seven [22][23][24][25][27][28][29] were rated as high quality. One study was not double-blind study and did not report the method to generate allocation and the reasons for dropout [26].

ISRN Endocrinology
A statistical pooling of pioglitazone versus rosiglitazones and biphasic insulin aspart 30 versus glibenclamide was not possible. However, in both trials, there were no significant differences between the treatment and the control groups in all outcomes. This may be explained by the fact that pioglitazone and rosiglitazone are categorized in the same glucose-lowering agent group. In comparison between biphasic insulin aspart 30 and glibenclamide, the initial dose of biphasic insulin aspart 30 was 0.2 U/kg/day and the dose was titrated every one to seven days in step of two to four units per injection. Thus, patients whose blood glucose levels were uncontrolled with suggested dose and required additional dose were absolutely not able to achieve target glycaemic goals.

Conclusions
TZDs, DPP IV inhs, and SUs may have similar effect in reducing HbA1c in type 2 diabetes patients whose blood glucose levels were not adequately controlled with metformin alone. However, TZDs decreased FPG better than did DPP IV inhs and reduced FPI more than did SUs. Given the limitations of the published data, large sample size, high quality, randomized controlled trials of combination treatment with metformin, and other agents are warranted.