Metformin and gonadotropins for ovulation induction in patients with polycystic ovary syndrome: a systematic review with meta-analysis of randomized controlled trials

The current systematic review with meta-analysis of randomized controlled trials (RCTs) was aimed to evaluate the effects of metformin on reproductive outcomes in patients with polycystic ovary syndrome (PCOS) who receive gonadotropins for ovulation induction. After systematic review of electronic databases and websites for registration of RCTs, a total of 7 RCTs reporting data on 1023 cycles were included in the final analysis. Descriptive data showed an overall low studies’ quality due to unclear sequence generation and allocation concealment, lack of blinding procedure, incomplete outcome data and several biases and/or confounders. Data synthesis showed that metformin improved live-birth (odds ratio [OR] = 1.94, 95% confidence interval [CI] 1.10 to 3.44; P = 0.020) and pregnancy (OR = 2.25, 95% CI 1.50 to 3.38; P < 0.0001) rates, without significant heterogeneity across the studies (P = 0.230, estimation of inconsistency = 30%; and P = 0.710, estimation of inconsistency = 0%, respectively, for live-birth and pregnancy rates). A significant reduction of cancellation rate was observed after metformin administration (OR = 0.41, 95% CI 0.24 to 0.72, P = 0.002) without significant heterogeneity across the studies (P = 0.500, estimation of inconsistency = 0%). Metformin administration influenced or did not influence other secondary endpoints assessed with a significant heterogeneity. In conclusion, metformin administration increases the live-birth and pregnancy rate in PCOS patients who receive gonadotropins for ovulation induction. Further well designed, blinded, placebo-controlled, and adequately powered RCTs are need to confirm that metanalytic results.


Background
Metformin, an insulin sensitizer widely used for treating type-2 diabetes mellitus, is employed in patients with polycystic ovary syndrome (PCOS) in light of the scientific data showing the pivotal role of insulin resistance in the pathogenesis of the syndrome, and of its beneficial effects on metabolism and ovulatory function in PCOS women [1].
Experimental and translational data seem to suggest that metformin could influence the ovarian response to gonadotropins. In fact, it improves not only the systemic insulin sensitivity and serum androgen levels in PCOS patients [1,2] but also their ovarian morphology [3] and environment [2] by improving the intra-ovarian hyperandrogenism through local effect on ovarian steroidogenesis [4] and the intra-ovarian insulin-resistance [2,3] interfering with autocrine/paracrine insulin-related signaling [5,6]. These actions on the peripheral tissues are irrespective of systemic improvement in metabolism and ovulatory function [2]. In addition, scientific data suggest an improvement of the endometrial receptivity in PCOS patients under metformin treatment [7,8].
Gonadotropin administration represents a widely accepted therapeutic option to induce ovulation in PCOS patients with anovulatory infertility [9,10], despite its high direct and indirect costs and its high risk of side effects.
Based on these considerations, metformin theoretically could induce a normalization of the abnormal ovarian responsiveness to gonadotropins, which is characteristic in PCOS patients, as well as lead to an improvement of their endometrial receptivity with an overall beneficial effect in terms of pregnancies and live births. Furthermore, a previous meta-analysis [11] published on 2006 demonstrated that data regarding metformin administration during gonadotropin ovulation induction were inconclusive. In fact, the inclusion of only two randomized controlled trials (RCTs) with less than 50 subjects each and the lack of studies aimed to assess the pregnancy rates limited the power of the analysis to exclude a treatment benefit [11].
The current study was aimed to clarify the effects of metformin in infertile PCOS patients who receive gonadotropin for ovulation induction through a systematic review with meta-analysis of available RCTs.

Methods
The protocol design followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines for reporting systematic reviews and metaanalyses of RCTs [12].

Study selection
Criteria for inclusion and exclusion of studies were established prior to the literature search.
Only RCTs characterized by symmetric interventions between the two treatment arms, i.e., patients who received the same protocol for ovulation induction with gonadotropins and then randomized to metformin or to placebo/no treatment, were included. Crossover studies were also included, although only data from the precrossover phase were considered for meta-analysis.
Studies were excluded if non-randomized; if any followup data were either not available, not extractable, not documented, or if the authors did not respond; if data were inconsistent or suspected duplicate (corresponding author was contacted by email and asked for clarification, and, if no clarification was obtained, data sets were considered overlapping and only the wider ones were included); if they included subjects who received gonadotropins for in vitro fertilization (IVF) programs.
No limit was given for PCOS diagnosis, dose and protocol of intervention proposed, type of gonadotropin used, and/or stimulation protocol employed.

Search strategy
The bibliographic search for identification of articles, abstracts, and study protocols was conducted monthly up to October 2013, with no language restriction.
The The bibliographies of retrieved articles, books and expert opinion review articles were manually searched and reviewed. No systematic attempt to search the grey literature, defined as information produced on all levels of government, academics, and/or business and industry in electronic and print formats not controlled by commercial publishing (International Conferences on grey literature, New York 2004), was made.
First, the titles and abstracts were screened and potentially relevant articles were identified and reviewed for inclusion/exclusion criteria. Then, the protocols and results of the studies were examined according to specific inclusion criteria. Lastly, only studies that met the inclusion criteria were considered for the final analysis.
Two independent reviewers (A.F., S.P.) not blinded at any point to the authors or sources of publication simultaneously reviewed the full manuscripts of all citations that possibly matched the predefined selection criteria. Final inclusion or exclusion decisions were made on examination of the full manuscripts. Disagreements between the reviewers on inclusion were discussed and solved by consensus or arbitration after consultation with an independent third author (G.B.L.S.).

Data extraction
The primary endpoints were the live-birth rate, defined as the number of deliveries that resulted in at least one live born baby for initiated cycle, and the pregnancy rate, defined as number of pregnancies per initiated cycles [13].
The secondary endpoints included the rates of miscarriages, multiple pregnancy, cancelled cycles for either poor-or hyper-response, ovarian hyperstimulation syndrome (OHSS), the stimulation length, the gonadotropin dose, and the serum estradiol (E 2 ) levels at human chorionic gonadotropin (hCG) injection [13].
The process of data abstraction examined the methodological and procedural characteristics of each study as well as a wide range of variables, including demographic, hormonal and metabolic characteristics of the study population, definition of PCOS, and treatments received with particular regard for type, protocols, and doses. These data were all extracted and tabulated.
The collaboration of all corresponding authors was requested, whenever possible, to obtain data missing from the papers included in the study, as well as unpublished and preliminary data.

Quantitative data analysis and synthesis
Statistical analyses were performed according to the statistical guidelines for review authors developed by The Cochrane Collaboration and published in the Cochrane Handbook for Systematic Reviews of Interventions [14]. All the statistical analyses were performed by using Review Manager Version 5 [15], provided by the Cochrane Menstrual Disorders and Subfertility Group.
The analysis of the treatment effect was performed according to the intention-to-treat (ITT) principle considering dropouts and missing data as treatment failures. In consideration of the potential effect of metformin in a pretreatment phase on pregnancy and live-birth, when possible, the analysis was also performed also according to the per-protocol method, considering the results only from the patients who really received the infertility treatment (gonadotropins with and/or without metformin).
Odds ratio (OR), with 95% confidence interval, was used as a valid way of describing an intervention effect for each dichotomous outcome using the Mantel-Haenszel method [16]. In particular, OR describes the multiplication of the odds of the outcome that occur with use of the intervention [16]. Continuous outcome differences between the two groups were presented as mean difference (MD) with 95% confidence interval.
A fixed-effect model was initially employed in the analysis, unless a significant heterogeneity occurred; a random effects model analysis was used in order to account for the extra uncertainty due to heterogeneity.
Potential heterogeneity of the treatment effects of each trial was examined by testing for interactions between source trial and treatment effects and estimation of inconsistency (I 2 ) [14,17]. Specifically, I 2 represents an estimate of the degree of inconsistency among studies; I 2 scores from 0% to 40% might not be important; from 30% to 60% may represent moderate heterogeneity; from 50% to 90% may represent substantial heterogeneity; and from 75% to 100% considerable heterogeneity [14].
A P value lower than 0.05 or 95% CI that did not contain unity was considered statistically significant. A statistical trend was arbitrarily established for P values that ranged between 0.05 and 0.09.
The number needed to treat (NNT) was calculated only for outcomes which were statistically significantly different between metformin and control/placebo groups, i.e. the expected number of people who need to receive the experimental (metformin) rather than the comparator intervention (no metformin) for one additional person to either incur or avoid an event in a given time frame.

Results
Search data Figure 1 shows the flow diagram of the study selection according to the PRISMA statement [12].
No unpublished study or in-progress study protocol was identified.

Studies description
The included studies [18][19][20][21][22][23][24] reported data on an overall population of 334 PCOS subjects (167 and 167 for metformin and control arm, respectively). A total of 1023 cycles were analyzed (438 and 585 cycles under metformin and no metformin, respectively). Study quality is detailed in Figure 2. An overall low studies' quality due to unclear sequence generation and allocation concealment, lack of blinding procedure, incomplete outcome data and several biases and/or confounders were observed. Table 1 summarizes the main characteristics of the RCTs included in the final analysis.
A wide variability was found across studies in the characteristics of participants, interventions performed and outcomes measured.
In 2 [23,24] and 3 [18,19,21] RCTs, the diagnosis of PCOS was made according to the European Society of Human Reproduction and Embryology (ESHRE)/American Society of Reproductive Medicine (ASRM) [25] or to the National Institute of Health (NIH) [26] criteria, respectively, whereas in 2 RCTs [20,22] non-standardized criteria were used. In particular, PCOS was diagnosed by the presence of oligomenorrhea, clinical and biochemical signs of hyperandrogenism, polycystic ovaries (PCO) and follicle stimulating hormone (FSH)/luteinizing hormone (LH) level higher than 2 in 1 RCT [20], whereas it was diagnosed by the presence of oligomenorrhea (an interval of at least 56 days between menses) or amenorrhea (an interval of at least 6 months between menses) in the other one RCT [22].
PCOS phenotype of the studied population was defined in no RCT.
Two RCTs [19,23] included only women with normal glucose tolerance. On the other hand, in 2 other RCTs [21,22], the presence of insulin resistance, univocally defined as a glucose-to-insulin ratio <4.5 mg/10-4, was a specific inclusion criterion.
-Days of treatment.
No data on live-births.
-Abnormal renal and liver tests.
-Use of drugs with possible effect on endogenous sex hormones.
-Any infertility factor other than PCOS (by semen analysis, hysterosalpingography, and/or laparoscopy).
Only insulin-resistant women were included.
70 subjects -Serum E 2 levels. IUI was performed in ovulating women who failed to conceive.
Exclusion: -N. dominant follicles. TI was performed in non-ovulating women.
-Days of treatment. Cycle was cancelled in presence of more than 3 follicles ≥14 mm, or in absence of ovarian response after 35 days of treatment. -BMI >30 and <18 kg/m 2 .
-Neoplastic, metabolic (including glucose intolerance), hepatic, and cardiovascular disorder or other concurrent medical illness.
-Current or previous (within 6 mths) use of oral contraceptives, glucocorticoids, antiandrogens, antidiabetic, and anti-obesity and hormonal drugs.
-Tubal or male factor infertility (by hysterosalpingogram and semen analysis).
-Intention to start a diet or a specific program of physical activity.
Only insulin-resistant women were included. -Rate of mono-ovulatory cycles.
-Signs of liver or kidney insufficiency and heart or vascular disease.
The study was divided into two phases. Only the 2 nd phase was considered in the analysis. -Multiple pregnancy rate. -Abortion.
No data on live-births.
Timed intercourses were performed in 5 RCTs [18][19][20]22,23], except in 1 RCT [21], in which women who previously failed to ovulate underwent timed intercourse whereas women who ovulated in the previous cycles but did not achieve a pregnancy underwent intrauterine insemination. In 1 study [24], the strategy used was not specified.
No significant effect of metformin on other secondary endpoints analyzed was detected (data not shown).

Discussion
The present systematic review and meta-analysis demonstrated that metformin improves the pregnancy rate of more than two-fold with the NNT of 5.7 benefits without heterogeneity among studies [18][19][20][21][22][23][24]. This beneficial effect seems to translate into a significantly better rate of live births. In fact, metformin administration increased significantly also the live birth rate of about two-fold. Moreover, only two RCTs [21,24] evaluated the live birth rate as study endpoint on a total of 298 and 363 cycles performed under metformin or no treatment/placebo, respectively.
The benefit of metformin treatment in improving live birth and pregnancy rates was confirmed also after data synthesis performed according to per protocol principle. Although this kind of analysis is generally used for safety-related outcomes, we would explore the specific effect of metformin on gonadotropin stimulation, excluding the patients who had a pregnancy under metformin during the pre-treatment phase [27].
Current data showed a reduced rate of cancelled cycles in PCOS patients who received metformin. In particular, less cycles resulted cancelled both for excessive and poor response under metformin. Of note, in 3 RCTs [21][22][23] was demonstrated a higher proportion of PCOS patients who achieved monofollicular cycles under metformin, and mono-ovulation should be considered the best result for treating anovulatory infertility.
The reduction in cancellation rate is very important from a clinical point of view, and could be crucial for  the reproductive benefit of metformin. However, at the moment, it is not possible to define the real impact of the reduction in cancellation rate on the increased pregnancy/live birth rate. To this regard, a specific data synthesis of the noncancelled cycles would be useful. Unfortunately that analysis was not possible for the lack of this outcome in many papers. Thus, we can hypothesize that the beneficial effect of metformin on reproduction can be explained not only with the fewer cancelled cycles, but also with a potential effect on oocyte quality and/or endometrial competence that cannot be formally excluded.
Regarding the endometrial receptivity, similarly to what occurred in IVF cycles [8], the current meta-analysis showed that serum E 2 levels were lower in PCOS patients who received metformin. This figure could affect the endometrial receptivity. However, experimental data [33,34] demonstrated a direct effect of metformin on the endometrium of PCOS patients. Moreover, no effect on the miscarriage rate was detected under metformin, confirming our previous meta-analysis on the lack of any effect of metformin on the risk of miscarriage in infertile PCOS patients under treatment [35]. On the other hand, a reduced risk of miscarriage and of implantation failure under metformin therapy was recently observed in PCOS patients undergoing IVF cycles [8].
Current data demonstrated no beneficial effect of metformin on OHSS risk in patients who received gonadotropins for ovulation induction, whereas only a trend was found for multiple pregnancies. These findings can be explained by the very low risk for and incidence of OHSS and multiple pregnancy observed in many of the included studies due to use of safe gonadotropin protocols and strict criteria for ovulation triggering.
The main strengths of the current systematic review and meta-analysis regard the use of strict inclusion and exclusion criteria, the lack of heterogeneity in all primary reproductive outcomes assessed, and the absence of further clinical trials or RCTs in progress or only designed that make current data conclusive for the next few years.
However, a definitive interpretation of the data and general applicability of the findings to make clinical recommendations seemed to be premature. In fact, the present meta-analysis has several limitations not regarding only the overall study quality (Figure 2).
First of all, it included a limited number of studies and of subjects. In fact, after a careful bibliographic search, only 7 RCTs were included in the final analysis, for an overall population of about 300 subjects. The publication bias was not tested in our analysis in consideration of the low power due to small number of studies included. In addition, many of the included studies had an open design and lacked adequate power analysis.
Secondly, infertile PCOS populations with heterogeneous characteristics were studied. Only 5 out of 7 included RCTs utilized conventional criteria for PCOS diagnosis, no standardization or sub-analysis according to phenotype was possible, and there was no distinction between CC-resistant patients and CC responders. In addition, factors such as insulin resistance and/or body mass index (BMI) may have biased our results. Unfortunately, no sub-analysis according to BMI was possible since almost no included study selected its participants for BMI and only few for insulin resistance. In addition, several confounders related to other patient characteristics, including previous parity, additional infertility diagnosis, duration of infertility, age, and so on may impact our findings.
Finally, protocols for metformin administration were heterogeneous. Specifically, metformin was given with a daily dose of 1500 mg or 1700 mg. These differences in metformin doses could again bias our findings, although recent data [35] demonstrated the lack of significant difference between metformin doses and treatment effectiveness even at lower dosages. On the other hand, metformin was generally administrated as gonadotropin pretreatment and coadministration, and only in 1 RCT [23] it was given as pretreatment. Of note, no study explored the effect of metformin pretreatment alone.

Conclusions
The current systematic review with meta-analysis demonstrated that metformin administration significantly increases the live birth and pregnancy rates of about two-fold, and reduces the cancellation rate of about 60% in PCOS patients who receive gonadotropins for ovulation induction.
This beneficial effect on reproduction can be explained with an effect on the reduced cancellation rate, even if an effect on oocyte and endometrial quality cannot be excluded.
In consideration of the suboptimal quality of the studies included, further well designed, blinded, placebocontrolled, and adequately powered RCTs are need to confirm current results. Unfortunately, no clinical trial on this issue is currently underway.