Effect of Metformin Intervention during Pregnancy on the Gestational Diabetes Mellitus in Women with Polycystic Ovary Syndrome: A Systematic Review and Meta-Analysis

Metformin is an effective insulin sensitizer treating type 2 diabetes mellitus. However, the functional consequences of metformin administration throughout pregnancy on gestational diabetes mellitus (GDM) with polycystic ovary syndrome (PCOS) have not been assessed. We therefore performed a meta-analysis and system review to determine the effect of metformin on GDM in PCOS. A meta-analysis was performed on the published studies before December, 2013. Meta-analysis examined whether metformin could reduce GDM occurrence in PCOS with a fixed effect model. The odds ratio (OR) with 95% confidence interval (95% CI) was calculated to estimate the strength of association. A total of 13 studies including 5 RCTs and 8 non-RCTs were enrolled. Ultimately, effectiveness analysis demonstrated that, in total, there was no significant availability of metformin on GDM in PCOS in contrast to placebo (OR = 1.07, 95% CI 0.60–1.92) in RCTs and significant availability of metformin on GDM (OR = 0.19, 95% CI 0.13–0.27) was indicated in non-RCTs. In summary, according to the results of our meta-analysis, strictly, metformin did not significantly effect on GDM with PCOS, though more multicenters RCTs still need to be investigated.


Introduction
Polycystic ovarian syndrome (PCOS), which is one of the common endocrine disorders, is one of the main causes of ovulatory infertility, affecting 5-10% of women of reproductive age [1,2]. PCOS is characterized by the presence of typical ultrasound features of polycystic ovaries, oligomenorrhea, and clinical and/or biochemical hyperandrogenism and, commonly, by insulin resistance, hyperinsulinemia, morbid obesity, and infertility [3][4][5][6][7][8][9]. Insulin resistance appears in both obese and nonobese women with PCOS [10]. Among these women, insulin promotes intraovarian steroidogenesis by interacting with luteinizing hormone (LH) leading to inappropriate advancement of granulose cell differentiation and arrest of follicle growth. The outcomes with hyperinsulinemia may directly enhance ovarian secretion and abnormal follicular development, which ultimately lead to ovarian dysfunction [11,12]. Moreover, hyperinsulinemia has been suggested as pathogenic factors in pregnancy complications [6]. Insulin resistance or obesity with PCOS leads to a significant increase in gestational outcomes and difficulties during labor.
During recent years, metformin, which is an effective oral biguanide insulin sensitizer, has been widely used for treating type 2 diabetes mellitus (T2DM) as an antihyperglycemic agent [13], by improving tissue sensitivity to insulin while inhibiting hepatic glucose production, enhancing peripheral glucose uptake, and decreasing insulin levels [14,15], and approved by the United States Food and Drug Administration (FDA) [16]. Therefore, metformin has become an ideal first line therapy for individuals with T2DM. When treated in women with PCOS, especially these patients with hyperinsulinemia, metformin corrects hyperinsulinemia and also reduces ovarian androgens, LH and sex hormone binding globulins. Metformin has been increasingly regarded to be effective and safe medicine for the metabolic and endocrine 2 Journal of Diabetes Research abnormalities in PCOS [17,18]. Its use as a drug for ovulation induction in PCOS has been extensively investigated and has been found to increase the likelihood of ovulation and decrease miscarriage rates, particularly in patients who have clomiphene resistance before. As we all know, metformin is a category B drug for use in pregnancy (absence of teratogenic effects based on animal data) and its characteristics of effectiveness and safety and, as a result, its use in pregnancy have become increasingly popular worldwide, although there are no guidelines for its continuous use in pregnancy and there is debating on potential adverse effects on both the mother and the fetus because of its crossing the placenta [19,20].
Moreover, metformin has been considered as a potentially effective agent during pregnancy to treat gestational diabetes mellitus [21]. On the other hand, several randomized controlled trials (RCTs) evaluating the reproductive effect of metformin administration and reporting the occurrence of GDM are not available in literatures. Experimental and clinical studies seem to suggest that metformin does not have any effect on the incidence of GDM in women with PCOS. Furthermore, unclear and nonpowered data support the use of metformin for the prevention of GDM, preeclampsia, and other gestational complications in PCOS patients [22].
Based on these considerations, we conducted a systematic review and meta-analysis of trials for a more objective appraisal of evidence regarding the effect of pregestational metformin administration on gestational complications, especially the gestational diabetes in women with PCOS.

Materials and Methods
A systematic review was conducted with all pertinent studies that were found in the electronic databases MEDLINE, EMBASE, and the Cochrane Central Register of Controlled Trials (CENTRAL) that examined metformin and pregnancy outcome on the women with PCOS from 1966 to December, 2013. The search strategy included the terms pregnancy, pregnant complication, GDM, gestational diabetes mellitus, diabetes, metformin, biguanide, insulin-sensitizing drugs, insulinsensitizers, polycystic ovary syndrome, and PCOS. The search was concluded by (1) the perusal of the reference sections of all relevant studies in all languages and (2) a manual search of the key journals and abstracts from the major annual meetings in the fields of endocrinology and obstetrics and gynecology. Articles were excluded from the analysis if they did not have either adequate disease-matched control groups or data on the outcome of the pregnancy with respect to major gestational diabetes and exposure to metformin throughout pregnancy or in at least the first trimester. The control groups consisted of women with PCOS who were not treated with metformin or placebo.
The inclusion and exclusion criteria were as follows.
Inclusion criteria: (1) PCOS was exposed to metformin throughout the pregnant period, or at least during first trimester; (2) studies could be a prospective, retrospective, or a case-control study; however, there had to be a control group; (3) studies were written in any language.
Exclusion criteria: (1) if type of articles was review or letter to the editor; (2) animal studies; (3) studies with no control group or with an inappropriate control group; (4) women with exposures to other known pathogenic factors or other maternal disorders that might affect the outcome.
Titles and, especially, abstracts were screened, and the potential relevant researches were identified. Two independent reviewers (ZHUO and YU), who were not blinded to the names of investigators or sources of publications, identified and selected the articles that met the inclusion and exclusion criteria below. The reviewers worked independently and in duplicate. Disagreements between two of them were resolved by consensus or arbitration between each other. Meta-analysis was conducted using Review Manager 5.2 software (Review Manager (RevMan) (computer program) (Version 5.2 for Windows), Oxford, England: The Cochrane Collaboration, 2012). We included trials evaluating the reproductive effects of metformin administration on the gestational complications especially GDM in the women with PCOS in the current meta-analysis. We considered all studies in which the diagnosis of PCOS was based on the normal criteria such as Rotterdam ESHRE/ASRM consensus criteria, NIH criteria, or AES criteria eligible.
The quality of the studies included was evaluated by the Cochrane guidelines [23]. Specifically, allocation concealment, blinding, intention-to-treat (ITT) analysis, and followup were assessed for each trial. Allocation concealment was graded as adequate (A), unclear (B), or inadequate (C) according to the criteria provided by the Cochrane Group [24] particularly. Blinding was reported as yes, no, or not reported for patients, outcome assessors, investigators, or data analysts. The use of ITT analysis was recorded and indicated as yes or no.
The -test of Cochran RevMan was used to measure the heterogeneity between articles and was calculated as the weighted sum of squared differences between individual study effects. A Cochran -test ⩽ 0.05 represents a statistical homogeneity. For data with statistical homogeneity, the Mantel-Haenszel method was used to calculate the weighted summary RR under the fixed effects model. On the other hand, the random-effect model of meta-analysis was used in the presence of unexplained statistical heterogeneity.
⩽ 0.05 of 95% CI not containing 1.0 RR was considered statistically significant. Potentially appropriate studies to be included in the meta-analysis (n = 32) Studies included in the meta-analysis (n = 13)

Flowchart of Study Selection.
The flowchart of the study selection according to the inclusion and exclusion criteria above is shown in Figure 1. A total of 13 trials potentially resulted for inclusion in the meta-analysis [22,[25][26][27][28][29][30][31][32][33][34][35][36]. In particular, we excluded the trials as follows: the data on the occurrence of GDM were not available from papers and could not be obtained from the investigators by e-mail contact; it was not possible to contact the corresponding investigators and the occurrence rate of gestational diabetes has not been evaluated. Table 1 summarizes the quality of the trials included in the analysis. The main characteristics of the populations, the interventions received, and the outcomes obtained in the trials are summarized in Tables 2  and 4, respectively. A total of five RCTs were also included in the final analysis as Table 3. However, three RCTs including Fougner et al. [30], Vanky et al. [32], and Salvesen et al. [35] are duplicate, and Vanky et al. [22] reported an epianalysis of two randomized controlled trials as Vanky et al. [32] and Salvesen et al. [35]. Finally, we reviewed two RCTs [32,35] in our meta-analysis. No research included in the system review and meta-analysis had occurrence of gestational diabetes as a primary end point, and none was powered to detect differences in GDM incidence. In addition, other potential gestational complications, such as miscarriage, preeclampsia, and preterm delivery, including the infants' characteristics, were investigated in these studies. The populations studied in the different trials were heterogeneous for several demographic and biochemical/metabolic characteristics.

Meta-Analysis.
13 studies met the inclusion criteria, including 5 random control trial studies in the calculation of overall GDM occurrences. Exposure to metformin throughout the pregnancy was not associated with a decreased rate of GDM occurrence. The OR for GDM in RCTs with 289 subjects was 1.07 (95% CI 0.60-1.92, = 0.89, and 2 = 0%) ( Figure 2) (Fougner et al. [30] and Selvesen et al. 's studies [35] were eliminated but Vanky et al. 's study [32] was enrolled because of the same object of three RCTs; meanwhile, Vanky et al. 's study [22] was eliminated because of the fact that it is an epianalysis of Vanky et al. [26] and Vanky et al. [32]); inversely, the OR including the studies with disease-matched controls or patient self as control was 0.19 (95% confidence interval (CI), 0.13-0.27, = 0.48, 2 = 0%) ( Figure 3). Subanalysis was conducted to separate studies into those with PCOS and those with patients' selves with PCOS subjects as controls (Figures 4 and 5, resp.). With 457 subjects in the treatment group, the OR for the nondiabetes pregnancy subjects group was 0.17 (95% CI, 0.11-0.27, = 0.40, 2 = 2%). With 118 subjects in the treatment group for PCOS with self as control, the OR was 0.24 (95% CI, 0.11-0.54, = 0.29, and 2 = 11%).

Discussion
In our mind, this is the first systematic review and metaanalysis of trials including RCTs performed to establish the     potential effect of metformin invention during the gestational period on gestational diabetes occurrence risk in patients with polycystic ovary syndrome.
Gestational diabetes mellitus (GDM) represents an important medical and social problem in the general pregnancy population, and it becomes harmful to the health of mother and fetus as well as neonate. Boomsma et al. [37] investigated that women with PCOS demonstrated a significantly higher risk of developing GDM (OR = 2.94), pregnancy-induced hypertension (OR = 3.67), and preeclampsia (OR = 3.47), and their babies had a significantly higher risk of admission to a neonatal intensive care unit (OR = 2.31) and a higher perinatal mortality (OR = 3.07). Thiebaugeorges et al. [38] has included that the traditional therapy in pregnancy for GDM includes the use of diet and exercise and/or insulin injection. Meanwhile, several researchers have already demonstrated that the administration of metformin throughout pregnancy may decrease miscarriage and improve pregnancy outcomes including GDM, preeclampsia, and preterm delivery [34,[39][40][41][42][43][44].
Although a population-based cohort research, a system review, and meta-analysis recently identified that the patients with PCOS are at higher risk for pregnancy complications [45,46], moreover, unless metformin was recommended for treatment in pregnancy complications especially DM because of its characteristics of safety, effectiveness, and nonteratogenicity, they could not reach an agreement on its use in prevention or treatment of GDM, hyperinsulinemia, and hyperandrogenemia throughout the pregnancy in women with PCOS. In our knowledge, during the past several decades, the retrospective and nonrandomized studies have confirmed beneficial effects of metformin on pregnancy miscarriage and pregnancy complications, in particular GDM in women with PCOS, whereas the potential effect of metformin to pregnancy complications in women with PCOS has been explored in five randomized, placebo-controlled trials so far [26,32]. The studies suggested no statistical difference in the prevalence of gestational diabetes, preeclampsia, and preterm delivery between the metformin-and the placebo-treated groups. As a result, meta-analytical summaries of existing studies therefore are of importance to GDM in PCOS.
Although it yields a reassuring trend, this analysis exemplifies the need for more research on metformin to DM during pregnancy. It was our initial aim to identify all possible adverse pregnancy outcomes including miscarriage, minor anomalies, major malformations, intrauterine growth retardation, GDM, preterm delivery, live birth, caesarean section rate, or the characteristics of infants in particular the birthheight, birth-weight, APGAR in 1 or 5 or 10 minutes, and head circumference. However, we were limited to focusing mainly on the pregnancy complications on GDM because of the paucity of data and the paradoxical effect of metformin on GDM in PCOS.
Notwithstanding our data, metformin administered alone throughout the pregnancy, or at least during first trimester, or in combination with other infertility treatments for inducing ovulation in patients with PCOS seems to have significant effect on GDM prevention. In fact, our non-RCTs  Test for overall effect: Z = 0.23 (P = 0.82) data demonstrated reduction of GDM occurrence risk in patients who received metformin in comparison with those who did not, for an overall relative risk for GDM of 0.19 (95% CI, 0.13-0.27). These findings were also confirmed by categorizing studies according to the objects control.
Nevertheless, on the basis of RCT studies available now, metformin does not appear to be unavailable for use during pregnancy with respect to diabetes mellitus, for an overall relative risk for GDM of 1.07 (95% CI, 0.60-1.92). As with placebo use in pregnancy, it is challenging to establish the Test for overall effect: Z = 7.81 (P < 0.00001) Figure 4: The meta-analysis of the odds ratio for gestational diabetes in the studies with PCOS as control.
necessity of metformin use in pregnancy considering the background incidence of diabetes in the confounder of the underlying diseases with PCOS that have been linked to increased rates of diabetes. However, it is not always true that RCTs and non-RCTs arrive at the same conclusions and, consequently, it is important to ask what should be done when they do not agree. First of all, the evidence of data derived from multiple randomized clinical trials or meta-analyses is stronger than data derived from a single randomized trial or unrandomized studies [47]. Thus, if a RCT study differs from several non-RCTs, then there is a pressing need to identify the studies much more deeply. We examine the possible reasons why two types of studies arrived at different conclusions and the potential threats to validity of observational outcomes and RCTs. Our analysis and review indicate that several methodological issues are relevant because they may affect the interpretation of some of the studies published on GDM. Most of the studies were retrospective or prospective of cohorts or case-control studies or observation researches and conducted at single center or multiple centers, and many studies included a small number of women, or else, the five RCTs, including the three studies of the same research object and one study epianalyzing the other two trials, took place in the same country as Norway and the study group of patients mainly recruited from the near clinic center imposed the restrictions on the outcomes, and thus the current situation limited the validity of the results. The second factors of different conclusion may be selection bias. Perhaps the most serious advantages of RCT are blindness and randomization because there may be large observed and unobserved differences in patients between the treatment and control groups [48]. These differences may explain why our meta-analysis concludes different results of RCTs and non-RCTs as RCTs were developed for the purpose of eliminating this bias.
During pregnancy, extra care must be taken to ensure the safety of the fetus or neonates. Our meta-analytical summary and review of inclusive studies therefore have indicated monitoring metformin safety. Although all included Heterogeneity: 2 = 1.12, df = 1 (P = 0.29); I 2 = 11% Test for overall effect: Z = 3.47 (P < 0.0005) Figure 5: The meta-analysis of the odds ratio for gestational diabetes in the studies with patients' selves as control.
studies are very heterogeneous for protocols and doses of the drug administered and for characteristics of the studied populations, the detected observation of the results, such as lactic acidosis, birth defects, and maternal or neonatal hypoglycemia, is just a case-report and has no statistical analytical significance.
Our systematic review and meta-analysis of RCTs studies demonstrates no statistically significant benefit of metformin administration on the DM risk in PCOS patients who received the drug as monotherapy. However, we examine that non-RCTs have a characteristic of a lower GDM occurrence rate in the metformin group compared with the PCOSmatched control group. Thus, at the moment, there is no clinical evidence to suggest that metformin administration reduces the risk of diabetes in women with PCOS.

Conclusion
Metformin is an effective insulin sensitizer treating type 2 diabetes mellitus. However, the functional consequences of metformin administration throughout pregnancy on gestational diabetes mellitus (GDM) have not been assessed. Metformin has been increasingly regarded to be effective and safe medicine for the metabolic and endocrine abnormalities in PCOS. In our knowledge, this is the first systematic review and meta-analysis of trials including RCTs performed to establish the potential effect of metformin intervention during the gestational period on gestational diabetes occurrence risk in patients with polycystic ovary syndrome. We therefore performed a meta-analysis and system review to determine the effect of metformin on GDM in patients with PCOS. A meta-analysis was performed on the published studies before December, 2013. Meta-analysis examined whether metformin could reduce GDM occurrence in PCOS with a fixed effect model. The odds ratio (OR) with 95% confidence interval (95% CI) was calculated to estimate the strength of association. A total of 13 studies including five RCTs and eight non-RCTs were enrolled. Ultimately, effectiveness analysis demonstrated that, in total, there was no significant availability of metformin on GDM in PCOS in contrast to placebo (OR = 1.07, 95% CI 0.60-1.92) in RCTs and significant availability of metformin on GDM (OR = 0.19, 95% CI 0.13-0.27) was indicated in non-RCTs. Subanalysis was conducted to separate studies into those with PCOS and those with patients' selves with PCOS subjects as controls. With 457 subjects in the treatment group, the OR for the non-diabetes pregnancy subjects group was 0.17 (95% CI, 0.11-0.28). With 118 subjects in the treatment group for PCOS with self as control, the OR was 0.24 (95% CI, 0.11-0.54). In summary, according to the results of our meta-analysis, strictly, metformin did not significantly have effect on GDM with PCOS, though more multicenters RCTs still need to be investigated. Thus, at the moment, there is no clinical evidence to suggest that metformin administration reduces the risk of diabetes in women with PCOS.