Effects of metformin in women with polycystic ovary syndrome treated with gonadotrophins for in vitro fertilisation and intracytoplasmic sperm injection cycles: a systematic review and meta-analysis of randomised controlled trials

Authors

  • S Palomba,

    Corresponding author
    • Obstetrics and Gynaecology Unit, Department of Obstetrics, Gynaecology and Paediatrics, Azienda Ospedaliera ASMN, Istituto di Ricovero e Cura a Carattere Scientifico, University of Modena and Reggio Emilia, Italy
    Search for more papers by this author
  • A Falbo,

    1. Obstetrics and Gynaecology Unit, Department of Obstetrics, Gynaecology and Paediatrics, Azienda Ospedaliera ASMN, Istituto di Ricovero e Cura a Carattere Scientifico, University of Modena and Reggio Emilia, Italy
    Search for more papers by this author
  • GB La Sala

    1. Obstetrics and Gynaecology Unit, Department of Obstetrics, Gynaecology and Paediatrics, Azienda Ospedaliera ASMN, Istituto di Ricovero e Cura a Carattere Scientifico, University of Modena and Reggio Emilia, Italy
    Search for more papers by this author

Correspondence: Dr S Palomba, Unit of Obstetrics and Gynecology, Arcispedale ‘Santa Maria Nuova’, Viale Risorgimento 80, 42123 Reggio Emilia, Italy. Email stefanopalomba@tin.it

Abstract

Background

Metformin is widely used for treating women with polycystic ovary syndrome (PCOS), and many patients with PCOS who are infertile receive gonadotrophins while being treated with metformin.

Objectives

To assess the effects of metformin administration in infertile patients with PCOS who receive gonadotrophins for in vitro fertilisation (IVF) and intracytoplasmic sperm injection (ICSI) cycles.

Search strategy

We searched international scientific databases, websites for the registration of trials, and bibliographies of retrieved articles, books, and review articles up to August 2012.

Selection criteria

Randomised controlled trials (RCTs).

Data collection and analysis

Authors independently reviewed and extracted the data.

Main results

Ten RCTs (with a total of 845 women with PCOS) were included in the final analysis. Metformin administration in IVF/ICSI cycles had no effect on the rates of pregnancy (OR 1.20, 95% CI 0.90–1.61) and live birth (OR 1.69, 95% CI 0.85–3.34). No effect of metformin dose, metformin pretreatment duration, and stopping time of metformin administration was observed on these reproductive end points. Metformin administration reduced the risk of ovarian hyperstimulation syndrome (OHSS; OR 0.27, 95% CI 0.16–0.46) and of miscarriage (OR 0.50, 95% CI 0.30–0.83), while increased that of implantation (OR 1.42, 95% CI 1.24–2.75).

Author's conclusions

In infertile patients with PCOS treated with gonadotrophins for IVF/ICSI cycles, metformin exerts no clinical effect on rates of pregnancy or live birth, but it reduces the risk of OHSS, and improves the rates of miscarriage and implantation. Further RCTs are needed to assess the reproductive effect of metformin in young well-selected patients with PCOS and specific phenotypes and features.

Introduction

Metformin is an insulin-sensitising drug that is commonly used for treating type–II diabetes mellitus.

During the last 15 years metformin has been widely studied in patients with polycystic ovary syndrome (PCOS),[1] wherein insulin resistance seems to play a pivotal role in the pathogenesis of the syndrome,[2] even if it exerts several biological actions not necessarily related to its insulin-sensitising properties.[1]

In patients with PCOS, metformin was reported to improve ovulatory menstrual cyclicity and clinical pregnancy rate,[3] to sensitise patients resistant to clomifene citrate (CC), and to optimise the efficacy of other drugs in inducing ovulation.[4]

The administration of gonadotrophins represents the final therapeutic step for infertility treatment in women with PCOS for the induction of controlled ovarian hyperstimulation during in vitro fertilisation (IVF) programmes,[5] and more and more patients with PCOS receive gonadotrophins under metformin therapy, as demonstrated by a recent online survey.[6]

A previous meta-analysis of randomised controlled trials (RCTs), updated to September 2008, failed to demonstrate any clinical effect of metformin on pregnancy and/or live birth rates in infertile patients with PCOS who received gonadotrophins for IVF and intracytoplasmic sperm injection (ICSI) cycles.[6] A significant benefit of metformin administration was observed only in terms of risk for ovarian hyperstimulation syndrome (OHSS).[7] However, these data were obtained from few studies and very small populations, thus limiting the power of the meta-analysis to exclude any possible treatment benefit. In fact, a total of six RCTs were included for an overall population of 459 subjects.[7] The authors highlighted the need for further large RCTs to definitively answer whether the use of metformin in women with PCOS undergoing IVF improves the rates of pregnancy and live birth.[7]

The aim of the present study was to systematically review the literature and to provide data synthesis by meta-analysis of RCTs in order to define whether metformin exerts clinical effects on reproductive end points in infertile patients with PCOS scheduled for IVF/ICSI programmes.

Methods

The protocol design followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines for reporting systematic reviews and meta-analyses of RCTs.[8]

Criteria for inclusion and exclusion of studies were established prior to the literature search.

All RCTs evaluating the effect of metformin in women with PCOS who received gonadotrophin treatment for IVF cycles were included in the current systematic review and meta-analysis. In particular, only RCT characterised by symmetric interventions between the two treatment arms were included, that is patients who received the same protocol for controlled ovarian hyperstimulation with gonadotrophins for IVF/ICSI programmes and were then randomised to treatment with metformin or to no treatment/placebo. Crossover studies were included, even if only data from the pre-crossover phase were considered for meta-analysis.

Studies were excluded if non-randomised, if any follow-up data were either not available, not extractable, not documented, or if the authors did not respond. Additional exclusion was made if studies were inconsistent or were suspected to be duplicates. In this last case, the corresponding author was contacted by email and asked for clarification: if no clarification could be obtained, data sets were considering to overlap and only the wider data sets were included.

No limit was given for PCOS diagnosis, dose, and protocol of intervention proposed, the type of gonadotrophin used, and the stimulation protocol employed.

The bibliographic search for the identification of RCTs (articles and/or abstracts) was conducted every month until August 2012, without limits for the English language.

A combination of the following medical subject headings or keywords was included: ‘ART’, ‘assisted reproduction technique’, ‘controlled ovarian hyperstimulation’, ‘controlled ovarian stimulation’, ‘gonadotropins’, ‘gonadotrophins’, ‘ICSI’, ‘infertility’, ‘insulin-sensitizer’, ‘insulin-sensitizier’, ‘insulin-sensitizing drugs’, ‘intra-cytoplasmic sperm injection’, ‘in vitro fertilisation’, ‘IVF’, ‘live birth’, ‘metformin’, ‘OHSS’, ‘ovarian hyperstimulation syndrome’, ‘ovulation induction’, ‘PCOS’, ‘polycystic ovarian syndrome’, ‘polycystic ovary disease’, ‘polycystic ovary syndrome’, ‘pregnancy’, ‘RCTs’, ‘randomised controlled trials’, ‘randomized controlled trials’, ‘sterility’.

We searched MEDLINE through PubMed (from 1966 to August 2012), EMBASE (from 1966 to August 2012), CINAHL (from 1981 to August 2012), Cochrane Library (from 1970 to August 2012), Clinical Evidence, UpToDate, and DARE for relevant studies. The Institute for Scientific Information (ISI), Web of Science,[9] Scopus,[10] Google Scholar,[11] and websites for the registration of controlled trials were also consulted for relevant clinical trials until August 2012.[12-15]

The bibliographies of retrieved articles, books, and expert opinion review articles were manually searched and reviewed. There was no systematic attempt to search the grey literature.

The titles and abstracts were initially screened, and potentially relevant articles were identified and reviewed for inclusion/exclusion criteria. Successively, the protocols and results of the studies were examined according to specific inclusion criteria. Finally, only studies that met the inclusion criteria were considered for the final analysis.

Two independent reviewers (A.F. and S.P.) not blinded at any point to the authors or sources of publication reviewed, in parallel, the full articles of all citations that possibly matched the predefined selection criteria. Final inclusion or exclusion decisions were made on an examination of the articles in full. Disagreements about inclusion between the reviewers were discussed and solved by consensus or arbitration after consultation with an independent third author (G.B.L.S.).

The primary end points were the rates of pregnancy and live birth, defined as number of pregnancies and live births per woman randomised.[16] The secondary end points included the rates of miscarriages, cancelled cycles, OHSS, and implantation, the stimulation length, the gonadotrophin dose, the serum estradiol (E2) levels at human chorionic gonadotrophin (hCG) injection, and the number of oocytes retrieved.

The process of data abstraction examined the methodological and procedural characteristics of each study and a wide range of variables, including the 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.

To obtain data for each end point from all of the RCTs, whenever it was possible, the collaboration of all corresponding authors was requested to obtain data missing from the papers included in the study. The collaboration of all investigators was also requested for unpublished (grey literature) and preliminary data.

Statistical analysis

For dichotomous data we presented the results as odds ratios (ORs) with 95% confidence intervals (95% CIs), and combined these for a meta-analysis to calculate a pooled estimate of the treatment effect for each outcome across the studies. On the other hand, in case of continuous data, we used the weighted mean difference (WMD) if the outcomes were measured the same way among trials, whereas we used the standardised mean difference (SMD) to combine trials that measured the same outcome but used different methods.

To measure heterogeneity (the variation in the outcomes among the studies), we used Cochran's Q-test and P > 0.05 to represent statistical homogeneity. Additionally, the potential heterogeneity of the treatment effects of each trial was examined by testing for interactions between the source trial and the treatment effects and estimation of inconsistency (I2).[17]

In case of statistical homogeneity, the fixed-effects model was applied, whereas in case of statistical heterogeneity the random-effects model of meta-analysis was used. Specifically, dichotomous data were combined for meta-analysis using the Mantel–Haenszel method,[18] and the DerSimonian–Laird method,[19] when the fixed-effects model and the random-effects model were applied, respectively. On the other hand, continuous data were combined for meta-analysis using inverse variance method and the DerSimonian–Laird method,[19] when the fixed effects model and the random effects model were applied, respectively.

Data synthesis was performed, firstly, on the overall population included, and subsequently after classifying the selected RCTs according to metformin dose, metformin pretreatment duration, and stopping time of metformin administration.

A P value of lower than 0.05 was considered statistically significant. A statistical trend was arbitrarily established for P values that ranged between 0.05 and 0.07. StatsDirect (CamCode, Ashwell, UK) was used for the statistical analysis.

Results

Systematic review

Figure S1 shows the flow diagram of the study selection according to the PRISMA guidelines.[8] No unpublished study or study protocol in progress was identified.

The characteristics of the excluded studies are detailed in Table S1.

After checking for the inclusion and exclusion criteria, we included a total of ten published RCTs in the final analysis,[20-29] involving an overall population of 900 infertile patients with PCOS enrolled and 845 assessed, and reporting data on 426 subjects treated with metformin and 419 controls who received either placebo or no treatment.

The quality of the selected studies is detailed in Figure S2.

The main characteristics of the included RCTs are summarised in Table S2.

A wide variability was found among studies in the characteristics of participants, interventions performed, and outcomes measured. The diagnosis of PCOS was made according to the European Society of Human Reproduction and Embryology (ESHRE)/American Society of Reproductive Medicine (ASRM) criteria,[30] or to the National Institutes of Health (NIH)[31] in seven[20, 21, 23, 26-29] and one[24] RCTs, respectively. Conversely, in two RCTs non-standardised criteria were used for PCOS diagnosis (Table S2).[22, 25] Resistance to CC was a peculiar inclusion criterion in one study,[26] whereas in another RCT only patients with CC–resistance or CC–failure were included.[24] The presence of insulin resistance was an inclusion criterion in only one RCT,[22] whereas none of the other RCTs reported/specified the presence of insulin resistance or glucose metabolism impairment.

Metformin was used as a pretreatment and co-administered in nine studies,[20, 22-29] whereas it was just co-administered with gonadotrophins in only one RCT.[21] The daily dose of metformin employed was 1000,[21, 22] 1500,[20, 24, 27-29] 1700,[25, 26] 2000,[23] or 2550 mg.[24] Metformin was given until oocyte retrieval, hCG administration, or embryo transfer in six RCTs,[20-22, 24, 25, 29] whereas it was administrated until pregnancy test and 12 weeks of gestation, respectively, in three and one RCTs.[23, 26-28]

Metformin treatment was controlled for placebo or no treatment in seven and three RCTs, respectively.[20-29]

In almost all of the studies a long-protocol gonadotrophin releasing hormone (GnRH) agonist suppression was used,[20-27] whereas a GnRH agonist flare-up protocol was used in just one RCT,[28] and a flexible GnRH antagonist protocol was adopted in another study.[29] Gonadotrophins were administered using step-up,[20] step-down,[23-25, 27, 28] fixed,[22, 29] or personalised protocols.[21, 26]

The type of luteal phase support varied among studies, and included vaginal progesterone capsules,[22] vaginal progesterone gel,[29] vaginal progesterone pessaries,[25, 26] and intramuscular progesterone.[20, 27, 28] In three RCTs the data on the drugs used for the luteal phase support were specified.[21, 23, 24]

Finally, the definition of clinical pregnancy also varied among the studies included.[20-29] In particular, clinical pregnancy was not defined in three RCTs.[20, 21, 24] It was defined as the presence of an intrauterine gestational sac by ultrasound performed at 7 weeks of gestation in two RCTs,[22, 23] or as the presence of one or more gestational sacs diagnosed by ultrasound or definitive clinical signs of pregnancy in two other RCTs,[27, 28] whereas in only one RCT it was defined as the presence of a viable pregnancy beyond 12 weeks of gestation.[25]

Meta-analysis

Primary and secondary outcomes are depicted in Figure S3.

Live birth rate

Data on the live birth rate were available in seven RCTs.[22-28] By a meta-analysis of those results, metformin administration did not significantly affect the live birth rate (OR 1.69, 95% CI 0.85–3.34, = 0.132), with significant heterogeneity across the studies (Cochran's Q-test, = 0.001, I2 = 72.4%; Figure S3).

Pregnancy rate

By the meta-analysis of nine RCTs,[20-28] no significant effect of metformin was observed on the pregnancy rate (OR 1.20, 95% CI 0.90–1.61, = 0.253), without significant heterogeneity across the studies (Cochran's Q-test, = 0.253, I2 = 21.4%; Figure S3).

Miscarriage rate

Data on rates of miscarriage were available in eight RCTs.[20, 22-28] After the meta-analysis, a significantly lower miscarriage rate after metformin administration (OR 0.50, 95% CI 0.30–0.83, = 0.010) was observed, without significant heterogeneity across the studies (Cochran's Q-test, = 0.269, I2 = 20.3%; Figure S3).

Cancellation rate

By meta-analysis of eight RCTs reporting data on the cancellation rate, no significant effect of metformin on that parameter (OR 0.71, 95% CI 0.42–1.22, = 0.262) was observed, without significant heterogeneity across the studies (Cochran's Q-test, = 0.060, I2 = 48.3%; Figure S3).[21-23, 25-29]

OHSS rate

Data on the OHSS rate were available in nine RCTs.[21-29] After meta-analysis, a significantly lower OHSS rate after metformin administration (OR 0.27, 95% CI 0.16–0.46, P < 0.0001) was observed, without significant heterogeneity across the studies (Cochran's Q-test, = 0.307, I2 = 15.6%; Figure S3).

Stimulation length

By meta-analysis of all RCTs included, no significant effect of metformin was observed on the stimulation length (WMD –0.070, 95% CI –0.52 to 0.66, = 0.816), with significant heterogeneity across the studies (Cochran's Q-test, = 0.008, I2 = 59.8%; Figure S3).[20-29]

Gonadotrophin doses

After combining the data from all of the RCTs included, no significant effect of metformin was observed on gonadotrophin doses (WMD –118.66, 95% CI –368.56 to 131.24, = 0.352), with significant heterogeneity across the studies (Cochran's Q-test, < 0.0001, I2 = 79.7%; Figure S3).[20-29]

Serum E2 levels

Data on the serum E2 levels were available in seven RCTs.[21, 22, 24, 26-29] After meta-analysis, significantly lower serum E2 levels after metformin administration was observed (WMD –581.89, 95% CI –999.58 to –164.20, = 0.006), with significant heterogeneity across the studies (Cochran's Q-test, < 0.0001, I2 = 87.2%; Figure S3).

Oocytes retrieved

After combining the data from all of the RCTs included, a lower number of oocytes retrieved was observed after metformin administration (WMD –1.11, 95% CI –1.86 to –0.36, = 0.004), without significant heterogeneity across the studies (Cochran's Q-test, = 0.199, I2 = 26.6%; Figure S3).[20-29].

Implantation rate

Data on the rate of implantation were obtained from six RCTs.[20, 23, 25-28] After meta-analysis, a significant effect of metformin in increasing the implantation rate (OR 1.42, 95% CI 1.24–2.75, P = 0.040) was observed, without significant heterogeneity across the studies (Cochran's Q-test, = 0.201, I2 = 31.3%; Figure S3).

Subanalysis

The data synthesis according to metformin doses, metformin pretreatment duration, and stopping time of metformin administration are shown in Tables 12, and 3, respectively. In particular, the selected RCTs were categorised into low (≤1000 mg daily) and high (> 1000 mg daily) doses, in long-term (> 3 weeks), short-term (≤ 3 weeks), or no metformin pretreatment, and in stopping time of metformin administration, that is until oocyte retrieval, hCG administration, or embryo transfer (ET).

Table 1. Data synthesis performed according to metformin doses
OutcomeHigh doses (>1000 mg daily)Low doses (</=1000 mg daily)
  1. —, * – Not evaluable.

Live birth rate

OR 1.83, 95% CI 0.83 to 4.06, = 0.135

Cochran's Q-test, = 0.0006, I2 = 76.8%

Six RCTs included[23-28]

OR 1.21, 95% CI 0.43 to 3.37, = 0.925

Cochran's Q-test, = —, I² = —

One RCT included[22]

Pregnancy rate

OR 1.13, 95% CI 0.81 to 1.58, = 0.508

Cochran's Q-test, = 0.159, I2 = 35.2%

Seven RCTs included[20, 23-28]

OR 1.50, 95% CI 0.78 to 2.87, = 0.292

Cochran's Q-test, = 0.571, I= *%

Two RCTs included[21, 22]

Miscarriage rate

OR 0.50, 95% CI 0.25 to 1.02, = 0.056

Cochran's Q-test, = 0.228, I2 = 26.3%

Eight RCTs included[20, 23-29]

OR 1.04, 95% CI 0.19 to 5.57, = 0.700

Cochran's Q-test, P= —, = —

One RCT included[22]

Cancellation rate

OR 0.65, 95% CI 0.20 to 2.05, = 0.460

Cochran's Q-test, = 0.028, I2 = 60.2%

Six RCTs included[23, 25-29]

OR 0.40, 95% CI 0.10 to 1.63, = 0.330

Cochran's Q-test, = 0.764, I2=*%

Two RCTs included[21, 22]

OHSS rate

OR 0.40, 95% CI 0.20 to 0.80, = 0.010

Cochran's Q-test, = 0.386, I2 = 4.8%

Seven RCTs included[23-29]

OR 0.15, 95% CI 0.06 to 0.38, < 0.0001

Cochran's Q-test, P = 0.691, I2 = *%

Two RCTs included[21, 22]

Stimulation length

WMD 0.19, 95% CI –0.56 to 0.93, = 0.620

Cochran's Q-test, = 0.005, I2 = 65.1%

Eight RCTs included[20, 23-29]

WMD -0.42, 95% CI −1.10 to 0.25, P = 0.222

Cochran's Q-test, P = 0.522, I2 = *%

Two RCTs included[21, 22]

Gonadotrophin doses

WMD –52.37, 95% CI –375.56 to 270.82, = 0.751

Cochran's Q-test, P < 0.0001, I2 = 82.7%

Eight RCTs included[20, 23-29]

WMD -326.84, 95% CI −505.99 to −147.69, P = 0.0003

Cochran's Q-test, P = 0.542, I2 = *%

Two RCTs included[21, 22]

Serum E2 levels

WMD –581.89, 95% CI –999.58 to –164.20, = 0.006

Cochran's Q-test, < 0.0001, I2 = 87.2%

Five RCTs included[24, 26-29]

WMD -1460.56, 95% CI −3929.06 to 1007.93, P = 0.246

Cochran's Q-test, P < 0.0001, I2 = *%

Two RCTs included[21, 22]

N. oocytes retrieved

WMD –1.16, 95% CI –1.96 to –0.37, = 0.004

Cochran's Q-test, P = 0.134, I2 = 37.0%

Eight RCTs included[20, 23-29]

WMD -0.668, 95% CI −2.885 to 1.549, P = 0.555

Cochran's Q-test, P = 0.323, I2 = *%

Two RCTs included[21, 22]

Implantation rate

OR 1.42, 95% CI 1.24 to 2.75, P = 0.040

Cochran's Q-test, P = 0.201, I2 = 31.3%

Six RCTs ncluded[20, 23, 25-28]

No available RCT
Table 2. Data synthesis performed according to metformin pretreatment length
OutcomeLong term (>3 weeks)Short term (≤ 3 weeks)No pretreatment
  1. —, * – Not evaluable.

Live birth rate

OR 1.71, 95% CI 0.65 to 4.49, P = 0.276

Cochran's Q-test, P = 0.001, I2 = 78.4%

Five RCTs included[22-24, 26, 28]

OR 1.68, 95% CI 0.95 to 2.99, P = 0.103

Cochran's Q-test, P = 0.082, I2 = *%

Two RCTs included[25, 27]

No available RCT
Pregnancy rate

OR 0.95, 95% CI 0.65 to 1.39, P = 0.858

Cochran's Q-test, P = 0.588, I2 = 0%

Six RCTs included[20, 22-24, 26, 28]

OR 1.71, 95% CI 0.97 to 2.99, P = 0.085

Cochran's Q-test, P = 0.090, I2 = *%

Two RCTs included[25, 27]

OR 1.76, 95% CI 0.74 to 4.19, P = 0.283

Cochran's Q-test, P = —, I2 = —%

One RCT included[21]

Miscarriage rate

OR 0.41, 95% CI 0.21 to 0.78, P = 0.0086

Cochran's Q-test, P = 0.162, I2 = 36.7%

Six RCTs included[20, 22-24, 26, 28]

OR 0.71, 95% CI 0.30 to 1.68, P = 0.579

Cochran's Q-test, P = 0.636, I2 = *%

Two RCTs included[25, 27]

No available RCT
Cancellation rate

OR 0.88, 95% CI 0.44 to 1.76, P = 0.857

Cochran's Q-test, P = 0.104, I2 = 48%

Five RCTs included[22, 23, 26, 28, 29]

OR 0.72, 95% CI 0.07 to 7.32, P = 0.783

Cochran's Q-test, P = 0.031, I2 = *%

Two RCTs included[25, 27]

OR 0.31, 95% CI 0.03 to 3.05, P = 0.583

Cochran's Q-test, P = —, I2 = —%

One RCT included[21]

OHSS rate

OR 0.55, 95% CI 0.25 to 1.20, P = 0.185

Cochran's Q-test, P = 0.640, I2 = 0%

Six RCTs included[22-24, 26, 28, 29]

OR 0.20, 95% CI 0.07 to 0.54, P = 0.014

Cochran's Q-test, P = 0.699, I2 = *%

Two RCTs included[25, 27]

OR 0.14, 95% CI 0.05 to 0.38, < 0.001

Cochran's Q-test, P = —, I2 = —%

One RCT included[21]

Stimulation length

WMD 0.07, 95% CI −0.87 to 1.01, P = 0.889

Cochran's Q-test, P = 0.006, I2 = 66.9%

Seven RCTs included[20, 22-24, 26, 28, 29]

WMD 0.42, 95% CI −0.31 to 1.16, P = 0.258

Cochran's Q-test, P = 0.299, I2 = *%

Two RCTs included[25, 27]

Not calculable

One RCT included[21]

Gonadotrophin doses

WMD -166.24, 95% CI −469.95 to 137.48, P = 0.283

Cochran's Q-test, P < 0.0001, I2 = 80.6%

Eight RCTs included[20-24, 26, 28, 29]

WMD 74.46, 95% CI −129.05 to 277.98, P = 0.473

Cochran's Q-test, P = 0.998, I2 = *%

Two RCTs included[25, 27]

Not calculable

One RCT included[21]

Serum E2 levels

WMD -618.42, 95% CI −1088.82 to −148.01, P = 0.01

Cochran's Q-test, < 0.0001, I2 = 89.3%

Four RCTs included[24, 26, 28, 29]

Not calculable

One RCT[27]

No available RCT
N. oocytes retrieved

WMD -1.45, 95% CI −2.37 to −0.53, P = 0.002

Cochran's Q-test, P = 0.177, I2 = 32.9%

Seven RCTs included[20, 22-24, 26, 28, 29]

WMD −0.14, 95% CI −1.60 to 1.32, P = 0.851

Cochran's Q-test, P = 0.311, I2 = *%

Two RCTs included[25, 27]

Not calculable

One RCT included[21]

Implantation rate

OR 0.28, 95% CI 0.12 to 0.62, P = 0.002

Cochran's Q-test, P = 0.169, I2 = 40.5%

Four RCTs included[20, 23, 26, 28]

OR 0.71, 95% CI 0.27 to 1.83, P = 0.517

Cochran's Q-test, P = 0.636, I2 = *%

Two RCTs included[25, 27]

No available RCT
Table 3. Data synthesis performed according to stopping time of metformin administration
OutcomeUntil oocyte retrieval/human chorionic gonadotrophins triggering/embryo transferUntil pregnancy testUntil 12 weeks of gestation
  1. —, * – Not evaluable.

Live birth rate

OR 1.31, 95% CI 0.46 to 3.71, P = 0.618

Cochran's Q-test, P = 0.034, I2 = 70.4%

Three RCTs included[22, 24, 25]

OR 1.38, 95% CI 0.90 to 2.11, P = 0.177

Cochran's Q-test, P = 0.310, I2 = 14.6%

Three RCTs included[23, 27, 28]

OR = 75.6, 95% CI 8.03 to 711.5, < 0.001

Cochran's Q-test, P = —, I2 = —%

One RCT included[26]

Pregnancy rate

OR 1.38, 95% CI 0.91 to 2.11, P = 0.161

Cochran's Q-test, P = 0.141, I2 = 42.1%

Five RCTs included[20-22, 24, 25]

OR 0.92, 95% CI 0.59 to 1.44, P = 0.819

Cochran's Q-test, P = 0.730, I2 = 0%

Three RCTs included[23, 27, 28]

OR 2.02, 95% CI 0.72 to 5.63, P = 0.278

Cochran's Q-test, P = —, I2 = —%

One RCT included[26]

Miscarriage rate

OR 0.84, 95% CI 0.41 to 1.74, P = 0.777

Cochran's Q-test, P = 0.834, I2 = 0%

Four RCTs included[20, 22, 24, 25]

OR 0.52, 95% CI 0.20 to 1.33, P = 0.249

Cochran's Q-test, P = 0.628, I2 = 0%

RCTs included:[23, 27, 28]

OR 0.08, 95% CI 0.02 to 0.39, P = 0.001

Cochran's Q-test, P = —, I2 = —%

One RCT included[26]

Cancellation rate

OR 0.72, 95% CI 0.29 to 1.76, P = 0.620

Cochran's Q-test, P = 0.341, I2 = 10.2%

Four RCTs included[21, 22, 25, 29]

OR 0.65, 95% CI 0.11 to 3.83, P = 0.631

Cochran's Q-test, P = 0.014, I2 = 76.7%

Three RCTs included[23, 27, 28]

OR 0.12, 95% CI 0.01 to 2.46, P = 0.278

Cochran's Q-test, P = —, I2 = —%

One RCT included[26]

OHSS rate

OR 0.22, 95% CI 0.11 to 0.42, < 0.0001

Cochran's Q-test, P = 0.430, I2 = 0%

Five RCTs included[2, 21, 22, 25, 29]

OR 0.47, 95% CI 0.19 to 1.13, P = 0.133

Cochran's Q-test, P = 0.126, I2 = 0%

Three RCTs included[23, 27, 28]

OR 0.12, 95% CI 0.01 to 2.46, P = 0.278

Cochran's Q-test, P = —, I2 = —%

One RCT included[26]

Stimulation length

WMD -0.03, 95% CI −0.37 to 0.32, P = 0.878

Cochran's Q-test, P = 0.435, I2 = 0%

Six RCTs included[20-22, 24, 25, 29]

WMD 0.85, 95% CI 0.02 to 1.68, P = 0.045

Cochran's Q-test, P = 0.299, I2 = 17.2%

Three RCTs included[23, 27, 28]

Not calculable

One RCT included[26]

Gonadotrophin doses

WMD -192.75, 95% CI −419.80 to 34.30, P = 0.096

Cochran's Q-test, P = 0.029, I2 = 60%

Six RCTs included[20-22, 24, 25, 29]

WMD 341.01, 95% CI −218.54 to 900.57, P = 0.232

Cochran's Q-test, P = 0.010, I2 = 78.2%

Three RCTs included[23, 27, 28]

Not calculable

One RCT included[26]

Serum E2 levels

WMD -948.17, 95% CI −1944.73 to 48.40, P = 0.062

Cochran's Q-test, P < 0.0001, I2 = 89.9%

Two RCTs included[24, 29]

WMD -279.73, 95% CI −548.21 to −11.25, P = 0.041

Cochran's Q-test, P = 0.688, I2 = *%

Two RCTs included[27, 28]

Not calculable

One RCT included[26]

N. oocytes retrieved

WMD 0.14, 95% CI −1.14 to 1.43, P = 0.826

Cochran's Q-test, P = 0.646, I2 = 0%

Six RCTs included[20-22, 24, 25, 29]

WMD -1.32, 95% CI −2.40 to −0.23, P = 0.018

Cochran's Q-test, P = 0.557, I2 = 0%

Three RCTs included[23, 27, 28]

Not calculable

One RCT included[26]

Implantation rate

OR 0.74, 95% CI 0.30 to 1.82, P = 0.672

Cochran's Q-test, P = 0.467, I2 = *%

Two RCTs included[20, 25]

OR 0.52, 95% CI 0.20 to 1.33, P = 0.249

Cochran's Q-test, P = 0.628, I2 = 0%

Three RCTs included[23, 27, 28]

OR 1.5, 95% CI 0.31 to 8, P = 0.812

Cochran's Q-test, P = —, I2 = —%

One RCT included[26]

No significant effect of metformin doses and metformin pretreatment duration was observed on the rates of pregnancy and live birth. A significant effect of metformin was observed only when it was administrated until 12 weeks of pregnancy. However, these data were obtained from only one RCT.[26].

Miscarriage rate was positively affected by higher doses, a longer duration of pretreatment, and delaying the stopping time of metformin administration, whereas OHSS rate and serum E2 levels were influenced, respectively, by longer metformin pretreatment duration and higher metformin doses. The number of oocytes retrieved was reduced for higher doses and after long-term treatment. Finally, the implantation rate was enhanced after long-term pretreatment.

Discussion

The present systematic review and meta-analysis serves to update the previously published data,[7] and to clarify definitively the clinical effect of metformin on reproductive outcomes when administered in infertile patients with PCOS treated with gonadotrophins for IVF/ICSI programmes.

First of all, our data confirmed that metformin administration gives no clinical benefit in terms of rates of pregnancy and live birth.[7] However, the inclusion of ten RCTs reporting data for an overall population of 900 subjects,[21-30] about double that included in the Cochrane review,[7] and the lack of registered clinical trials actually in progress make our data the best clinical evidence available on this issue at the moment and, probably, for the next few years.

The findings obtained on our secondary end points are novel and surprising. In fact, reduced risks of miscarriage and implantation failure under metformin therapy were observed. These findings were obtained after the synthesis of data from eight and six RCTs, respectively, and although the luteal phase support varied widely among studies, they were homogeneous for both end points.[20, 22-28].

Women with PCOS seem to be at increased risk of miscarriage,[32-34] even if no univocal explanation can be given at the moment. Several factors, both systemic and local, could be involved alone or in combination in the pathogenesis of the incidence of miscarriage in PCOS subjects, and several experimental and clinical data have suggested that metformin could act on these same factors, reducing the overall risk for miscarriage in PCOS.[1] Furthermore, a recent meta-analysis failed to demonstrate that metformin, administered alone or in combination with other infertility treatments for inducing ovulation in patients with PCOS, exerts a significant effect on abortion prevention both in the overall population and in populations categorised according to the treatment received.[35] However, in contrast to the current meta-analysis, all of the studies included were more heterogeneous for protocols, doses of metformin administered, and for characteristics of the populations studied.[26].

Our findings confirmed that metformin reduces the incidence of OHSS by about 70%.[7] This effect can be explained both with the reduced ovarian response to gonadotrophins under metformin therapy and with the pleiotropic effects of the drug. However, a direct effect of metformin on vascular mediators of OHSS cannot be formally excluded. In contrast with Tso et al.,[7] we detected a significant reduction of serum E2 levels in patients with PCOS who received metformin without any difference in gonadotrophin dose. A lower number of oocytes retrieved was also observed. Moreover, metformin acts on several PCOS features and PCOS-related characteristics involved in the pathogenesis of the ovarian hyper-response that is typically increased in PCOS patients.[1] A direct effect of metformin on the ovarian morphology and environment has been demonstrated.[36, 37] In particular, metformin could act to improve intraovarian hyperandrogenism through local actions on ovarian steroidogenesis and through intraovarian insulin resistance interfering with autocrine/paracrine insulin-related signalling.[38-45] The current meta-analysis also showed that serum E2 levels were lower in patients with PCOS who received metformin. This could affect endometrial receptivity, also explaining the improved miscarriage and implantation rates under metformin therapy.

The strengths of this meta-analysis include the analysis of an overall population of 900 subjects, the use of strict selection criteria (i.e. RCTs with symmetric interventions between the two treatment arms), and the search of clinical trials in progress on the main website registers.

The current meta-analysis has also several limitations.

Firstly, outcome measures were incomplete for several RCTs, even after correspondence with the corresponding authors. Secondly, there were many confounding variables related to the patients, to the treatments received, and to the definition of clinical outcomes. In fact, a lack of uniformity in patients' characteristics (i.e. age, body mass index, insulin resistance, and ovarian reserve) and selection for infertility treatment was noted.

In particular, the inclusion of one RCT on women with PCOS and with reduced ovarian reserve requires consideration, even if the data from that study did not substantially affect the results of the meta-analysis.[28] In fact, after the exclusion of these data, the odds ratios for rates of pregnancy and live birth changed from 1.20 (95% CI 0.90–1.61, = 0.253; Cochran's Q-test, = 0.253, I2 = 21.4%) to 1.28 (95% CI 0.93–1.74, = 0.146; Cochran's Q-test, = 0.26, I² = 21.3%), and from 1.69 (95% CI 0.85–3.34, = 0.132; Cochran's Q-test, = 0.001, I2 = 72.4%) to 1.93 (95% CI 0.88–4.26, = 0.102; Cochran's Q-test, = 0.001, I² = 75.6%), respectively. Moreover, a number of subjects studied in all of the RCTs included in the current analysis were older women with potentially poor reproductive prognosis.[20-27, 29].

In addition, even if the PCOS diagnosis was performed using well-recognised criteria in eight of the ten studies included, the same PCOS criteria used for the initial selection of patients introduce a wide heterogeneity arising from the PCOS phenotypes. In fact, in no study did the authors take into account the different PCOS phenotypes as potential predictors of a metformin effect.

Heterogeneous protocols for gonadotrophins and, in particular, for metformin administrations were used. In particular, the effect of different protocols for metformin administration was not explored in any RCTs included in the current meta-analysis. In this respect, a further data synthesis of the selected RCTs grouped according to metformin doses, pretreatment duration, and length of treatment administration was performed in order to clarify the effect, if any, of a specific protocol of metformin administration.

No relevant clinical effect of our subanalysis was detected on the primary end points, that is rates of pregnancy and live birth. However, metformin seemed to exert benefits on the rates of live birth when administrated until 12 weeks of gestation. However, these data were obtained from only one RCT.[26].

Our subanalysis also showed that higher doses, a longer duration of metformin pretreatment, and delaying the stopping time of metformin treatment positively affected the miscarriage rate. Similarly, the rate of implantation improved only with long-term metformin pretreatment.

Conclusions

Here, we confirm and consolidate the conclusion of the Cochrane review performed by Tso et al. that there is a lack of evidence for metformin treatment before or during IVF/ICSI cycles in unselected patients with PCOS improving the rates of pregnancy or live birth.[7] Metformin reduces the risk of OHSS, and should be considered as a preventive strategy for patients with PCOS at high-risk of OHSS, as well as for young, non-obese, hyperandrogenic women with PCOS and with a normal ovarian reserve. The administration of metformin before and/or during gonadotrophin ovarian hyperstimulation in infertile patients with PCOS scheduled for IVF/ICSI cycles reduces the risk of miscarriage and of implantation failure, and these beneficial effects seem to be influenced by higher metformin doses and a longer duration of metformin pretreatment/co-treatment. Further RCTs are needed to assess the reproductive effect of metformin in young well-selected patients with PCOS and with specific phenotypes and features.

Disclosure of interest

The authors declare the absence of any potential conflicts of interest, whether of a financial or other nature.

Contribution to authorship

All authors made a substantial contribution to the conception and design of the review, to the acquisition of data, and to the analysis and interpretation of data. All authors drafted the article and revised it critically, and gave final approval of the version to be published.

Details of ethics approval

No ethical approval was required.

Funding

No funding for the systematic review and other support was received.

Acknowledgements

The authors would like to thank all of the authors contacted who provided extra data and permitted the optimisation of the current meta-analysis.

Ancillary