Types of studies

Published and unpublished randomised controlled trials (RCTs) were eligible for inclusion. We excluded non‐randomised studies (e.g. studies with quasi‐randomisation, such as allocation based on alternate days or patient hospital numbers).

Crossover trials were eligible, but we would have only included data from the first phase in meta‐analyses, as the crossover is not a valid design in the context of fertility trials.

Types of participants

Subfertile women or couples who are trying to get pregnant either with sexual intercourse or intrauterine insemination (IUI), with or without ovulation induction (OI). We excluded women or couples undergoing assisted reproductive technology (ART) (e.g. in vitro fertilisation (IVF)) as this group of participants is the topic of another Cochrane review (Nastri 2015).

Types of interventions

Any intervention that caused intentional damage to the endometrium, performed with the objective of improving the reproductive outcomes of women desiring pregnancy. Intentional endometrial injury may be achieved by procedures such as endometrial pipelle biopsy or biopsy performed with a Novak curette. We excluded studies that evaluated interventions causing unintentional endometrial damage compared with control. Examples of unintentional endometrial injury are hysteroscopy, hysterosalpingography, endometrial sound, mock embryo transfer and cervical dilation.

Types of outcome measures

Electronic searches

We searched the following electronic databases, trial registers and websites from inception to 31 October 2015: the Cochrane Gynaecology and Fertility Specialised Register of Controlled Trials, the Cochrane Central Register of Controlled Trials, MEDLINE, EMBASE, PsycINFO and CINAHL. The MEDLINE search was combined with the Cochrane highly sensitive search strategy for identifying randomised trials which appears in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). The EMBASE, PsycINFO and CINAHL searches were combined with trial filters developed by the Scottish Intercollegiate Guidelines Network (SIGN) (http://www.sign.ac.uk/methodology/filters.html#random) (see Appendix 1; Appendix 2; Appendix 3; Appendix 4; Appendix 5; Appendix 6; Appendix 7; Appendix 8; Appendix 9).

Other electronic sources of trials included the following.

• Trial registers for ongoing and registered trials (with the search terms "endometrial injury", "endometrial scratching" and "endometrial biopsy"):

  • http://www.clinicaltrials.gov (a service of the US National Institutes of Health);

  • the World Health Organization International Trials Registry Platform (WHO ICTRP) search portal http://www.who.int/trialsearch/Default.aspx.

• DARE (Database of Abstracts of Reviews of Effects) on the Cochrane Library http://onlinelibrary.wiley.com/o/cochrane/cochrane_cldare_articles_fs.html (for reference lists from relevant non‐Cochrane reviews).

• The Web of Knowledge http://wokinfo.com/ (source of trials and conference abstracts).

• LILACS database http://regional.bvsalud.org/php/index.php?lang=en (for trials from the Portuguese‐ and Spanish‐speaking world).

Searching other resources

We handsearched reference lists of relevant articles retrieved by the search. We contacted experts in the field (e.g. authors of included studies) for information on additional trials, including unpublished or in‐progress trials.

Selection of studies

Firstly, two review authors (SL and MM or GT) independently screened the titles and abstracts of all articles retrieved from all searches according to the review inclusion criteria. The two review authors excluded any clearly irrelevant studies. We obtained the full‐text versions of all remaining potentially eligible studies, which two review authors then independently assessed for inclusion. We excluded articles that did not meet the review inclusion criteria. In instances where study eligibility was unclear, we contacted the study authors for clarification. The two review authors resolved any disagreements by discussion in the first instance, followed by consultation with a third review author (WPM) if required.

Data extraction and management

Two review authors (SL and MM or GT) independently extracted data from eligible studies using a data extraction form that we had pilot tested. They resolved any disagreements by discussion or by consultation with a third review author (WPM). Data extracted included study characteristics and outcome data. We corresponded with study investigators for further data on methods or results, or both, as required.

Assessment of risk of bias in included studies

Two review authors (SL and MM or GT) independently assessed the included studies for risk of bias using the Cochrane 'Risk of bias' assessment tool for the following bias domains: random sequence generation, allocation concealment, blinding of participants, personnel and outcome assessors, incomplete outcome data, selective reporting and other bias (see Appendix 10 for the rationale we used in assessing risk of bias). We resolved any disagreements by discussion or by consulting a third review author. We supported all judgements by excerpts from the study or comments from the review authors. We presented conclusions in 'Risk of bias' tables (see the 'Characteristics of included studies' section), which we incorporated into the interpretation of review findings by means of sensitivity analyses (see below). We took care to search for within‐trial selective reporting, such as trials that failed to report adverse outcomes. Where possible, we used published protocols or trial registration information for included studies to investigate selective reporting (i.e. a comparison of outcomes listed in the study protocol with outcomes reported in papers).

Measures of treatment effect

For dichotomous data (e.g. live birth), we used the numbers of events in the control and intervention groups of each study to calculate the Mantel‐Haenszel risk ratios (RRs). For continuous outcomes (e.g. pain) we calculated the mean difference. We presented 95% confidence intervals (CIs) for all outcomes.

Unit of analysis issues

We used the number of randomised women as the denominator for live birth, clinical pregnancy, pain and bleeding. We used the number of clinical pregnancies as the denominator for miscarriage, multiple pregnancy and ectopic pregnancy as these outcomes can only occur in women who achieve a clinical pregnancy.

Dealing with missing data

We analysed the data on an intention‐to‐treat basis as far as possible and attempted to obtain missing data from the study investigators. When we were unable to obtain missing data, we performed imputation of individual values as described below.

• We assumed that live births and pregnancies had not occurred in participants without a reported outcome.

For other outcomes, we only analysed the available data. We subjected any imputation undertaken to sensitivity analysis.

Assessment of heterogeneity

We considered whether the clinical and methodological characteristics of the included studies were sufficiently similar for meta‐analysis to provide a clinically meaningful summary. We assessed statistical heterogeneity using the I² statistic, where we took an I² statistic value greater than 50% to indicate substantial heterogeneity. We planned to investigate the causes of any observed heterogeneity with prespecified subgroup analyses.

Assessment of reporting biases

In view of the difficulty of detecting and correcting for publication bias and other reporting biases, we aimed to minimise their potential impact by ensuring a comprehensive search for eligible studies, including in trial registries, and by being alert to data duplication. If there were 10 or more included studies in an analysis, we planned to use a funnel plot to explore the possibility of small study effects (a tendency for estimates of the intervention effect to be more beneficial in smaller studies).

Data synthesis

One review author entered the data and performed the statistical analysis in Review Manager (RevMan) (RevMan 2014). When a study reported ongoing pregnancy but did not report live birth, we pooled ongoing pregnancy data with live birth data from other included studies. Where this occurred, we also performed sensitivity analyses. We discussed data that we could not pool in a narrative format in the text. When we could confidently rule out significant clinical and statistical heterogeneity, we combined data from primary studies in a meta‐analysis with Revman (RevMan 2014). We used the Mantel‐Haenzel random‐effects model for the following comparisons.

• Intentional endometrial injury vs. either no intervention or a sham procedure.

• Higher vs. lower degree of intentional endometrial injury (e.g. two interventions vs. one intervention; Novak curette vs. pipelle).

• Different timing of intentional endometrial injury (e.g. follicular phase vs. luteal phase).

We combined data using a random‐effects model, as we considered that the method and instruments used to cause endometrial injury were likely to differ across the trials in each analysis, and that most participants had unexplained infertility which is thought to be a heterogeneous condition. We displayed an increase in the risk of a particular outcome, which may be beneficial (e.g. live birth) or detrimental (e.g. miscarriage), graphically in the meta‐analyses to the right of the centre‐line and displayed a decrease in the risk of an outcome to the left of the centre‐line.

Subgroup analysis and investigation of heterogeneity

We planned to perform the following subgroup analyses only if substantial heterogeneity existed (I² statistic value of greater than 50%) and enough data were available.

• Type of conception (e.g. IUI, OI, timed intercourse, regular intercourse); as benefit from endometrial injury may vary depending on the type of conception.

• Cause of subfertility (e.g. unexplained subfertility, polycystic ovarian syndrome, endometriosis etc.); as benefit from endometrial injury may vary depending on the cause of subfertility.

• Timing of endometrial injury (e.g. follicular phase, luteal phase); as benefit from endometrial injury may vary depending on the phase of the menstrual cycle in which the injury is performed.

• Length of study period (e.g. only one attempted conception cycle, between one to three cycles, more than three cycles); to account for a higher probability of pregnancy with longer study durations, and to investigate the potential duration of benefit following endometrial injury.

• Severity of injury (e.g. two interventions vs. one intervention; Novak curette vs. pipelle).

Sensitivity analysis

We conducted sensitivity analyses on the primary outcomes to determine whether the conclusions were robust to arbitrary decisions we made regarding the eligibility and analysis. These analyses included consideration of whether the review conclusions would have differed if the following occurred.

• We restricted eligibility to studies considered to be at low risk of bias.

• We did not perform any imputation for live birth.

• We did not pool ongoing pregnancy data with live birth data.

• We had used a fixed‐effect model.

• The summary effect measure was odds ratio rather than relative risk.