Metformin extended-release versus metformin immediate-release for adults with type 2 diabetes mellitus: a systematic review and meta-analysis of randomized controlled trials.

AIM
To compare the efficacy and tolerability metformin extended-release (MXR) and the conventional metformin immediate-release (MIR) in adults with type 2 diabetes mellitus (T2DM) METHODS: PubMed, the Cochrane Library, ClinicalTrials.gov and other sources were searched for randomized controlled trials (RCTs) that compared equal daily doses of MXR and MIR in adults with T2DM from platform inception to 19 March 2021. Random-effects model meta-analysis was performed to obtain, with 95% confidence intervals (CIs), pooled mean difference (MD) of change from baseline for continuous outcomes and risk ratio (RR) for dichotomous outcomes. Primary outcomes were HbA1c and key gastrointestinal (GI) symptoms (abdominal discomfort or pain, diarrhea, dyspepsia, and nausea & vomiting); fasting and post-prandial plasma glucose, other GI and serious adverse events (AEs), serum lipid control, and anthropometrics served as secondary outcomes.


RESULTS
Nine RCTs that randomized a total of 2609 adults revealed that MIR was statistically associated with better HbA1c lowering (MD 0.09% [0.01%, 0.17%]) and serum lipid control, and MXR only with reduced dyspepsia (RR 0.58 [0.34, 0.98]). MXR and MIR were similar in other considered outcomes.


CONCLUSIONS
MXR was associated with statistically worse but likely clinically insignificant HbA1c lowering, similar plasma glucose control, and minimal improvement of metformin intolerance versus MIR. Protocol Registration: PROSPERO (CRD42019148008).


Introduction
Diabetes mellitus (DM) is a chronic progressive metabolic disease and a global threat to health. It currently a icts more than 463 million people worldwide, at least 90% of which are type 2 DM (T2DM) cases [1]. DM is the 15 th leading cause of total years of life lost in 2017 [2] and its economic impact is approximately 2% of the global gross domestic product [3]. As the burden of this disease continues to worsen, developing and implementing integrative strategies for its prevention and control is thus a matter of urgency in all levels of care.
Comprehensive medical care for diabetes mellitus includes optimized and individualized glycemic goal and therapeutic plan [4]. As lifestyle modi cation (nutrition, physical activity, psychosocial care, and selfmanagement, education and support [4]) often alone fails to establish and sustain optimal glycemic control [5], pharmacotherapy forms an important component of management [6]. In this regard, with more than sixty years of clinical experience and evidentiary support of e cacy, safety and tolerability in numerous T2DM trials [7], the biguanide metformin has been recommended for use as a rst-line therapy, together with lifestyle modi cation, in managing the disease [8].
Unfortunately, a signi cant number of patients prescribed with metformin cannot tolerate the conventional immediate-release formulation at recommended doses because of gastrointestinal (GI) adverse events (AEs), leading to reduced treatment adherence [9,10]. Furthermore, this drug preparation notably demonstrates a lack of dose proportionality with increasing doses, resulting to decreased bioavailability [11]. As all these may contribute to untoward outcomes ranging from poor glycemic control to hospitalizations [12], intolerance to metformin immediate-release (MIR) is a pressing clinical concern. One supposed strategy to address MIR intolerance is alternatively prescribing using metformin extended-release (MXR). MXR slowly delivers the active component of the drug in the upper GI tract and allows for once-daily dosing and is thus designed to reduce GI AEs and to improve adherence [6,7,[9][10][11][12].
Furthermore, its bioavailability is not adversely affected either by variations in daily dosing frequency or meal intake [7,12]. Since its regulatory approval in Canada and in the United States 15 years ago [6], MXR has been marketed as an equally e cacious yet more GI-tolerable alternative to MIR.
Nevertheless, a systematic review has yet to be undertaken to validate assumptions regarding MXR prescription to adults with T2DM in avoiding or reducing metformin intolerance. The purpose of this study was thus to compare MXR and MIR in terms of e cacy and tolerability in the randomized controlled trial (RCT) setting involving adults with T2DM.

Protocol, Registration and Reporting
The methods of this systematic review and meta-analysis were pre-speci ed in a protocol that was registered in the International Prospective Register of Systematic Reviews (PROSPERO) with registration number CRD42019148008 (Supplement 1). Reporting was done in accordance to Preferred Reporting Items for Systematic Review and Meta-Analysis (PRISMA) guidelines. Activities and decisions at any point were independently conducted and made by two reviewers (A.A.A. and A.N.N.I.P.), with disagreements resolved through discussion or the senior author (C.A.J.). The deviations to the original protocol and justi cations are indicated in Supplement 2.

Eligibility Criteria
RCTs that compared equal daily dosing of MXR and MIR on separate trial arms for at least 12 weeks which involved adults diagnosed with T2DM were considered. Studies with other interventions together with MXR and MIR were considered if the concomitants were provided equally in terms of dose, exposure or intensity in both arms. No restriction was set regarding publication status, year, and language. Studies that explicitly involved subjects with end-stage organ diseases were excluded.

Outcomes
The primary outcomes were glycemic control parameters, such as glycated hemoglobin or hemoglobin A1c (HbA1c), fasting plasma glucose (FPG) and post-prandial plasma glucose after 2 hours (PPG), and GI AEs such as abdominal distension, abdominal discomfort or pain, constipation, diarrhea, dyspepsia, atulence, and nausea and vomiting. Secondary outcomes were serum lipid control parameters, such as serum triglycerides (Tg), total cholesterol (TC), high-density lipoprotein-cholesterol (HDL-C) and lowdensity lipoprotein-cholesterol (LDL-C), anthropometric parameters such as weight and body mass index (BMI), and serious adverse events, such as hypoglycemia and lactic acidosis.

Search Strategy and Study Selection
PubMed, the Cochrane Library and ClinicalTrials.gov were searched for records of studies. A sensitive search strategy was implemented, including "metformin," "extended release," "gastric retention," "slow release," and "sustained release" (Supplement 3). Google searching was also performed for records not indexed in the speci ed databases. There was neither date, language nor study design restriction in the search. Reference lists of selected studies were also screened from eligible reports and relevant reviews. Unpublished results, print-only journals and conference proceedings were also sought by contacting the authors.
Items in the initial search were scanned through titles and abstracts to remove irrelevant papers and duplicates. The criteria for selection and exclusion of the remaining articles were implemented and documented. Figure 1 illustrates the outcome of the selection process through the PRISMA ow diagram.

Data Collection and Analysis
Data were independently extracted from included articles using a pre-piloted data collection form. Quality assessment was performed in each selected study through the Cochrane Collaboration risk of bias (RoB) tool [13]. A study was then deemed to have an overall low risk (low risk in all RoB domains), overall unclear risk (unclear risk in at least one domain and high risk in no domain) or overall high risk (high risk in at least one domain). Quantitative synthesis through meta-analysis with a Mantel-Haenszel random effects model was also performed with results reported, with 95% con dence interval (CI) and pvalue (statistical signi cance set at p < 0.05), as risk ratio (RR) for dichotomous outcomes and mean difference (MD) of change from baseline for continuous outcomes. Heterogeneity was declared if detected in at least one of the two tests, χ 2 (p < 0.1 indicating signi cant heterogeneity) and I 2 (at least 50% indicating substantial heterogeneity). Subgroup analyses were performed to explore possible sources of heterogeneity from differences in daily total metformin dose. Publication bias was ascertained through funnel plot using the outcome change from baseline HbA1c. Sensitivity analyses were conducted based on risk of bias (high risk, unclear risk, low risk) per particular domain and eliminating one included study at a time. Review Manager version 5.4 (Cochrane Collaboration) was utilized.

Imputations
In the absence of response from authors of included studies for queries on data, the following strategies were performed to address insu cient or missing continuous outcome values. The MD and standard deviation (SD) of change from baseline values were re-calculated using reported con dence intervals, means, and SDs or standard errors of baseline and endpoint measures. SDs of calculated change from baseline values were computed using the following formula (Equation 1): where Corr was set conservatively at 0.5. When the SD of interest was not reported or was reported but deemed unreasonably wide (with the particular study subsequently contributing too little weight in the analyses relative to its contribution to the pooled sample size) by author consensus, a validated summary statistic-level prognostic imputation method was performed utilizing undisputed SDs from included studies (Equation 2) [14]: where the numerator refers to the summation of undisputed SDs and the denominator refers to the number of undisputed SDs.

Search Results
A total of 1272 database records (415 from PubMed, 738 from the Cochrane Library and 119 from ClinicalTrials.gov) were obtained through the latest systematic search performed on 15 October 2020, and another one was found through the Google search engine. After removing 471 records as duplicates, the rest were subjected through title and abstract screening wherein 790 were excluded. Nine full-text articles of RCTs, that directly compared MXR and MIR, satis ed the eligibility criteria (Table 1) and were included in both the qualitative and quantitative syntheses. On the other hand, 1 unpublished clinical trial [15] and 2 full-text articles [16,17] were excluded because of unavailable study results, metformin daily doses were adjusted based on gastrointestinal tolerance on a per-patient basis (and thus were virtually not equal between the two arms), and concomitant non-metformin anti-T2DM medications were not controlled, respectively (Figure 1).
There was substantial variation in the total daily MXR and MIR dose given in the studies such as 500 mg/day (combined with the sulfonylurea glimepiride in equivalent dosing in both arms) in 1 study [20], 1000 mg/day in 2 studies [19,21], 1500 mg/day in 2 studies [18,24] and 2000 mg/day in 4 studies [22,23,25,26]. Two studies had an additional arm (MXR daily dose 500 mg greater than the two equivalent arms [19] and MXR daily dose 1000 mg less than the two equivalent arms [26]). Notably, 1 study had 2 MXR arms with total metformin daily dose equal to each other (MXR was provided once daily in one arm and twice daily in the other) and to the MIR arm [18]; the 2 MXR arms were thus pooled for quantitative analyses. Six studies had dose build-up weeks within the treatment period, which lasted from one week to one month [18,[21][22][23]25,26]. The formal duration of treatment also varied across studies: 12 weeks in 4 trials [19,22,24,26], 16 weeks in 2 trials [20,25], 18 weeks in 1 trial [21] and 24 weeks in 2 trials [18,23]. The frequency of follow-up for assessment of outcomes ranged from 1 to 4 weeks in seven trials [18,19,21,22,[24][25][26], while two trials performed the same only at baseline and at endpoint (16 weeks [20] and 24 weeks [23], respectively).

Imputations
Four studies [18,19,24,26] only reported separate baseline and endpoint means + SDs; thus, Equation 1 was applied to generate MD + SD change from baseline values. Equation 2 was used to generate SDs for one study [18] that reported values that were unanimously considered by the authors as too wide (resulting to a relative weight in the analyses that was too small compared to the contributed sample size, see Supplement 4) and for another [21] where the SDs for mean change from baseline HbA1c could not be obtained.

Risk of Bias and Publication Bias
All included trials were judged to have unclear risk of selection bias (random sequence generation and allocation concealment) as none of them presented adequate relevant details for assessment. Two studies [23,26] were considered to have an overall unclear risk of bias. An overall high risk of bias was noted for the remaining 7 studies [18][19][20][21][22]24,25], 2 of which were explicitly unblinded [24,25] (Figure 2 and Supplement 5).
The generated funnel plot ( Figure S6) using the primary outcome of change from baseline HbA1c showed funnel plot asymmetry. However, since this plot involves less than 10 studies, other sources of asymmetry such as selective outcome reporting, poor methodological quality, true heterogeneity, artefactual elements and chance cannot be reliably ruled out [27].

Secondary Outcomes
In terms of change from baseline in serum lipid control parameters, MIR was associated with improved mean Tg lowering (0. 18 (Tables S7.11 to S7.13). In subgroup analyses, statistically signi cant change was found only for HDL-C elevation at metformin 2000 mg/day in favor of MIR. Sensitivity analyses show that the statistical signi cance of these ndings were not robust when (1) studies with unclear or high risk for performance and/or reporting bias were removed, (2) at least one of the studies, except the smallest [21], were eliminated and (3) elimination of studies with unclear or high risk for performance, detection, attrition or reporting bias was performed, respectively (Tables S8.9 to S8.11). Mean lowering of LDL-C (-0.03 mmol/L [95% CI, -0.12 mmol/L, 0.06 mmol/L], 5 studies [18,19,21,23,24]) was not signi cantly different between the two arms ( Figure S7.14), with neither signi cant nor substantial heterogeneity (χ 2 p = 0.41, I 2 = 0%). The nding was robust on sensitivity analysis (Table S8.

Summary, Interpretation and Context of Results
MXR was associated with statistically inferior HbA1c lowering and with reduced cumulative incidence of dyspepsia compared to MIR. Both interventions were found to be equivalent in terms of shorter-term glycemic control (FPG and PPG lowering), cumulative incidence of all other GI AEs, anthropometrics, and LDL-C lowering. Interestingly, MIR led to better Tg and TC lowering as well as HDL-C elevation than MXR.
MXR rst received regulatory approval for use as a therapeutic agent for adult-onset T2DM in 2005, with accumulated evidence purporting similar glycemic control and improved GI tolerability compared to the conventional MIR formulation [6]. While subsequent RCTs comparing MXR and MIR in equal daily doses individually reported general glycemic control equivalence [20][21][22][23][24][25][26], a portion of these studies that reported cumulative incidence of AEs demonstrated similarity in GI tolerability pro le as well [21,[23][24][25]. Moreover, the measures of dispersion reported in   [18] -whose results were integral to the regulatory approval of MXR a year prior to its publication -were uncharacteristically wide in comparison to those reported in other included studies. When used directly in the meta-analysis, these measures decreased considerably the weight of the trial relative to its contribution in the pooled sample size. This compelled us to use imputation Equation 2 [14] for all continuous outcomes in this trial where it provided information (HbA1c, FPG, Tg, TC, LDL-C and HDL-C). The status of statistical signi cance in the ndings between the non-imputed and imputed scenarios was the same for FPG, Tg, LDL-C and HDL-C, while the difference in HbA1c and TC lowering became statistically signi cant in favor of MIR when the imputation was performed (Supplement 4).
The results suggest that most included trials were likely to be individually underpowered in ascertaining the difference in glycemic control between MXR and MIR. On the other hand, it remains to be seen whether the statistically better HbA1c lowering noted for MIR in this study translates to tangible difference in morbidity or mortality outcomes compared to MXR, as there exists no study that compares "hard" outcomes between the two. Nevertheless, the current body of evidence suggests that this particular nding is unlikely to be clinically signi cant [28,29]. While comparing cumulative incidences of AEs with MXR and MIR suggests that the former only minimally improved metformin intolerance relative to the latter, it is possible that comparing cumulative incidences are not sensitive enough to gauge the difference in metformin tolerability between the two formulations. Incidence rates and incidence rate ratios, rather than cumulative incidences and RRs, may be able to tease out clinically relevant ndings in GI AE burden between these two formulations, if the difference in tolerability truly exists. As the frequency and severity of GI AEs has been suggested to subside as a function of time that metformin is taken [10], future RCTs comparing MXR and MIR should aim to comprehensively describe and report AE occurrence associated with these interventions beyond cumulative incidence. In terms of signi cant secondary outcomes, the nding that MIR lowered Tg by 5.99 mg/dL to 26.05 mg/dL and TC by 0.06 mg/dL to 7.92 mg/dL compared to MXR can be considered clinically relevant for at least some cases (especially those with Tg and/or TC level at or near the risk factor thresholds for these parameters). The clinical signi cance of improved HDL-C increase with MIR compared to MXR, however, remains uncertain, as it is yet to be convincingly shown that such elevation by pharmacologic agents, among patients without genetic perturbations (e.g., de ciency or excess in serum levels) in the metabolism of the said lipoprotein, improves clinical outcomes [30].

Limitations
While the included trials were conducted in 14 different countries, majority of the participants were Caucasian in origin. The pooled age estimate of the participants (mean [SD] of 54.3 [11.5] years) suggests that the ndings of this meta-analysis would likely not be applicable in relatively younger and older patients with T2DM, both growing subsets of the population a icted with the disease. Majority of the included studies are considered to have an overall high risk of bias and the reporting quality of one e cacy (PPG) and several tolerability parameters of primary interest is arguably super cial (cumulative incidence rather than incidence rate without a measure of severity in all studies) and incomplete (only 3 studies reporting PPG, and relatively few or no GI AEs presented in many included studies). The pooled estimates also re ect ndings across 4 daily metformin doses. The impact of comorbidities and concomitant intake of non-T2DM medications in the ndings could also not be ascertained as these data were not speci ed in most included trials. Findings from one reportedly completed study that would have been eligible for inclusion in the systematic review were not included due to unavailability [15], hence not all of the existing evidence was analyzed. Imputation strategies were also implemented which may not adequately estimate the actual aggregate measures observed from the involved trials. Finally, individual patient data meta-analysis was not and could not be performed which would otherwise allow consistent de nition of exposures and outcomes as well as proper adjustments for confounders [31].

Conclusions
In this meta-analysis of RCTs comparing MXR and MIR, MXR was statistically associated with worse but likely clinically insigni cant HbA1c lowering and with reduced cumulative incidence of dyspepsia. On the other hand, intake of MIR led to improved Tg and TC lowering as well as HDL-C elevation compared to MXR. In terms of FPG and PPG reduction, cumulative incidence of GI and serious AEs other than dyspepsia, LDL-C lowering, and anthropometrics, the difference between the two formulations is not statistically signi cant. While the ndings of this study remain to be validated and the de nitive clinical implications yet to be elucidated by future trials, these data may be utilized by physicians in discussing metformin-related treatment options with their adult patients with T2DM especially in the setting of signi cant GI intolerance. Additional head-to-head high-quality RCTs, with protocols informed by the study-level aws and quality assessments mentioned in this study, that will also provide individual patient-level data are needed. In particular, the use of incidence rates rather than cumulative incidence especially for safety and tolerability related outcomes should be considered. Investigations that aim to mechanistically prove or disprove the ndings of this study are of interest, including the noted difference of both formulations in serum lipid control parameters. Difference in cost-effectiveness and adherence rates between these two formulations may also be explored.

Declarations FUNDING
Funding was neither sought nor received in making this study.
AUTHOR CONTRIBUTIONS