Types of studies

Randomised controlled trials (RCTs) that compared IVM with conventional ART (IVF or ICSI) in women with PCOS were considered for inclusion in this review. If a factorial design was identified (for example, trials with four groups: IVM alone, IVF alone, IVM + drug, IVF + drug) data would be pooled where IVM followed by IVF or ICSI was compared with conventional IVF or ICSI.

Quasi‐randomised trials were not included. Cross‐over trials were not eligible for inclusion unless pre‐cross‐over data were available, in which case only these data would be used for the purposes of the review.

Types of participants

Types of interventions

Conventional IVF or ICSI after controlled ovarian hyperstimulation (COH) versus IVM followed by IVF or ICSI. Embryo transfers of both fresh and frozen, but not mixed embryos were included.

Trials evaluating alternative interventions in which the option of IVM or conventional ART was not randomised (for example ovarian drilling, use of the oral contraceptive pill, metformin, aromatase inhibitors, steroids, progestins, growth hormone, L‐arginine) were excluded.

Studies using different protocols for IVM, for example human chorionic gonadotrophin (HCG) or gonadotrophin priming for the maturation of oocytes before IVM, were eligible as long as they compared women with PCOS undergoing COH programmes.

All ovarian stimulation protocols were eligible (use of gonadotrophin‐releasing hormone agonists (GnRHa), GnRH antagonists, gonadotrophins alone (recombinant or urinary preparations), or clomiphene with gonadotrophins). In all the above cases, pooling data was our intended purpose.

Types of outcome measures

Electronic searches

(1) The Menstrual Disorders and Subfertility Group Specialised Register of controlled trials was searched for any relevant trials from the title, abstract, or keyword sections

(2) The following electronic databases were also searched:

MEDLINE and PubMed (http://www.ncbi.nlm.nih.gov/pubmed/) (1966 to May 2013) (Appendix 1), CENTRAL (the Menstrual Disorders and Subfertility Group’s Central register of Controlled Trials) (1966 to May 2013) (Appendix 2), CINHAL (1966 to May 2013) (Appendix 3), EMBASE (1966 to May 2013) (Appendix 4) and the Trial registers for ongoing and registered trials: 'ClinicalTrials.gov', a service of the US National Institutes of Health (http://clinicaltrials.gov/ct2/home) and The World Health Organization International Trials Registry Platform search portal (http://www.who.int/trialsearch/Default.aspx) (Appendix 5) and OpenSIGLE database for grey literature from Europe (http://opensigle.inist.fr/).

Searching other resources

Further searches were made in the National Institute of Clinical Excellence (NICE) fertility assessment and treatment guidelines (NICE 2004). Reference lists of relevant systematic reviews and RCTs were handsearched.

Selection of studies

Three review authors (CS, MK and NV) independently checked the titles and abstracts retrieved by the search and examined them for compliance with the above criteria. It was our intention that additional information would be sought from the authors of trials which appeared to meet the eligibility criteria but had either aspects of methodology that were unclear or data in an unsuitable form for meta‐analysis. Differences of opinion were to be resolved after correspondence with AM and SB.

Data extraction and management

Our intention was to extract the following data from the studies included in the review.

Assessment of risk of bias in included studies

It was our intention that data on trial quality would be extracted in duplicate. This information would have been presented in the 'Characteristics of included studies' table to provide a context for discussing the reliability of results.

Measures of treatment effect

For dichotomous data (e.g. live birth rates), our aim was to use the numbers of events in the control and intervention groups of each study to calculate Mantel‐Haenszel odds ratios (ORs) with 95% confidence intervals (CIs) for all outcomes. Where data to calculate ORs were not available, we would have utilised the most detailed numerical data available that might facilitate similar analyses of included studies.

Unit of analysis issues

Our intention was that the primary analysis would be per woman randomised; per pregnancy data might also be included for some outcomes (e.g. miscarriage). Data that would not allow valid analysis (e.g. "per cycle" data) would be briefly summarised in an additional table but would not be meta‐analysed. Multiple live births (e.g. twins or triplets) were to be counted as one live birth event. Only first‐phase data from cross‐over trials were to be included. If studies reported only “per cycle” data, we would have contacted authors to request “per woman” data.

Dealing with missing data

Attempts were made to obtain missing data from the original trialists of the three ongoing studies by sending the corresponding authors two e‐mails. No data were obtained.

Our initial intention was to analyse the data on an intention‐to‐treat basis. Where these were unobtainable, imputation of individual values would have been undertaken for the primary outcomes only. Live births would be assumed not to have occurred in participants without a reported outcome. For other outcomes, only the available data would have been analysed. Any imputation undertaken would have been subjected to sensitivity analysis (see below).

Assessment of heterogeneity

It was our intention to consider whether the clinical and methodological characteristics of the included studies were sufficiently similar for meta‐analysis to provide a clinically meaningful summary. We would have assessed statistical heterogeneity by the measure of the I^2. An I² measurement greater than 50% would be taken to indicate substantial heterogeneity (Higgins 2011).

Assessment of reporting biases

In order to avoid or at least minimise reporting biases (e.g. publication bias, multiple publication bias, language bias etc), we performed a comprehensive search for eligible studies and were alert for duplication of data. If there were 10 or more studies in an analysis, we planned to use a funnel plot to explore the possibility of small‐study effects.

Data synthesis

It was our intention that if the studies were sufficiently similar, we would combine the data using a fixed‐effect model

An increase in the odds of a particular outcome, which might be beneficial (e.g. live birth) or detrimental (e.g. adverse effects), would be displayed graphically in the meta‐analyses to the right of the centre‐line and a decrease in the odds of an outcome to the left of the centre‐line. A fixed‐effect analysis would be used if the underlying effect size was the same for all the trials in the analysis. If not, a random‐effects analysis would be implemented.

Subgroup analysis and investigation of heterogeneity

It was our intention that where data were available, we would conduct subgroup analyses to determine the separate evidence within the following subgroups:

1. average age for women of less than and over 35 years;

2. various types of subfertility;

3. different protocols for COH;

4. parity of women;

5. previous COH for IVF attempts.

If we had detected substantial heterogeneity, we would have explored possible explanations in sensitivity analyses. We would have taken any statistical heterogeneity into account when interpreting the results, especially if there were any variation in the direction of effect.

Sensitivity analysis

It was our intention to conduct sensitivity analyses for the primary outcomes to determine whether the conclusions were robust to arbitrary decisions made regarding the eligibility and analysis. These analyses would have included consideration of whether the review conclusions would have differed if:

1. eligibility were restricted to studies without high risk of bias;

2. a random‐effects model had been adopted;

3. alternative imputation strategies had been implemented;

4. the summary effect measure was risk ratio rather than odds ratio.