Types of studies

We included both published and unpublished randomised controlled trials (RCTs). We assessed the method of randomisation to determine whether the studies were truly randomised. In the case of cross‐over trials, we only included them if pre‐cross‐over data were available. We incorporated trials that included a subset of participants with male subfertility if data were available for that subset.

Types of participants

Couples with male subfertility who had been trying to conceive for at least one year were eligible for inclusion. We included all couples with male factor subfertility, including oligo‐, terato‐, asthenospermia, or a combination of these, preferably measured by two separate semen samples.
Routine fertility evaluation should have consisted of confirmed ovulatory status (basal body temperature (BBT) chart, mid‐luteal progesterone or sonographic evidence of ovulation) and low risk for tubal pathology according to the medical history (Coppus 2007).

Types of interventions

We included RCTs with at least one of the following comparisons:

1. IUI versus TI or expectant management both in natural cycles;

2. IUI versus TI both in cycles with OH;

3. IUI in natural cycles versus TI in cycles with OH;

4. IUI in cycles with OH versus TI or expectant management in natural cycles;

5. IUI in natural cycles versus IUI in cycles with OH;

6. IVF versus TI or expectant management in natural cycles;

7. IVF versus TI in cycles with OH;

8. IVF versus IUI in natural cycles;

9. IVF versus IUI in cycles with OH;

10. ICSI versus TI or expectant management in natural cycles;

11. ICSI versus TI in cycles with OH;

12. ICSI versus IUI in natural cycles;

13. ICSI versus IUI in cycles with OH;

14. ICSI versus IVF.

We excluded trials comparing methods using insemination other than IUI, such as intracervical insemination (ICI), gamete intrafallopian transfer (GIFT) and fallopian tube sperm perfusion. In addition, we excluded trials comparing different types of ovarian stimulation protocol, as this is the subject of a different review (Cantineau 2007).

Types of outcome measures

Electronic searches

We searched for all published and unpublished RCTs that described (or might have described) treatments for male subfertility with no language restrictions. Marian Showell, Trials Search Co‐ordinator of the Cochrane Gynaecology and Fertility Group, performed a search of the following:

1. the Cochrane Menstrual Disorders and Subfertility Group Specialised Register of controlled trials (inception to April 2015) (Appendix 1);

2. the Cochrane Central Register of Controlled Trials (CENTRAL; inception to April 2015) (Appendix 2);

3. MEDLINE (inception to April 2015) (Appendix 3);

4. EMBASE (inception to April 2015) (Appendix 4);

5. PsycINFO (inception to April 2015) (Appendix 5);

6. CINAHL (inception to April 2015) (Appendix 6).

We placed no language restrictions.

We combined the MEDLINE search with the Cochrane highly sensitive search strategy for identifying RCTs, which appears in the Cochrane Handbook of Systematic Reviews of Interventions (Section 6.4.11) (Higgins 2011). We combined the EMBASE, PsycINFO and CINAHL searches with trial filters developed by the Scottish Intercollegiate Guidelines Network (SIGN) (www.sign.ac.uk/methodology/filters.html#random).

Other electronic sources of trials included the following:

1. trial registers for ongoing and registered trials:

   1. ClinicalTrials.gov, a service of the US National Institutes of Health (clinicaltrials.gov/ct2/home), and the WHO International Clinical Trials Registry Platform search portal (www.who.int/trialsearch/Default.aspx);

2. conference abstracts in the Web of Knowledge (wokinfo.com/);

3. OpenSIGLE database for grey literature from Europe (opensigle.inist.fr/);

4. LILACS database, as a source of trials from the Portuguese and Spanish speaking world (regional.bvsalud.org/php/index.php?lang=en) (choose 'LILACS' in 'all sources' drop‐down box);

5. PubMed (www.ncbi.nlm.nih.gov/pubmed/).

We searched the databases using the medical subject headings (MeSH terms) and keywords in Appendix 7.

Searching other resources

We checked the reference lists of all identified studies for relevant articles. We performed a handsearch of abstracts of the American Society for Reproductive Medicine (1999 to April 2015) and the European Society for Human Reproduction and Embryology (1997 to April 2015) meetings.

When important information was lacking from the original publications, we tried to contact the authors. We incorporated additional information in the review.

Selection of studies

After screening the titles and abstracts retrieved by the search, we obtained full texts of all potentially eligible studies. Two review authors (MC and MvW) independently selected the trials to be included according to the above‐mentioned criteria. We resolved disagreements by consensus or through arbitration by a third review author (AB). We documented the selection process with a PRISMA flow chart (Figure 1).

Figure 1

---

---

Study flow diagram.

Data extraction and management

The same two review authors independently used a data extraction form to extract data from published reports. We resolved disagreements by consensus or through arbitration by a third review author (AB). This data extraction form included information on the type of study, quality of the selected studies, types of participants, types of interventions and the types of outcome measures (Appendix 8). An analysis of agreement between the two review authors on assessment of the method of randomisation and study design resulted in 100% agreement.

Assessment of risk of bias in included studies

As part of the data collection process, two review authors (MC and MvW) independently extracted data for trial characteristics that have been recognised as potential sources of bias, such as the method used in generating the allocation sequence, how allocation was concealed and differences in drop‐out rates between study arms. We used the criteria outlined in the Cochrane Handbook for Systematic Reviews of Interventions Version 5.1.0 (Higgins 2011). Where there was uncertainty, we contacted the authors to clarify aspects of study design. We resolved disagreements by consensus or through arbitration by a third review author (AB).

Two review authors independently assessed the included studies for risk of bias using the Cochrane 'Risk of bias' assessment tool outlined in the Cochrane Handbook for Systematic Reviews of Interventions Version 5.1.0 using the following domains (Higgins 2011):

1. selection bias (random sequence generation and allocation concealment);

2. performance bias (blinding of participants and personnel);

3. detection bias (blinding of outcome assessors);

4. attrition bias (incomplete outcome data);

5. reporting bias (selective reporting);

6. other bias.

These domains were assessed to have:

1. high risk of bias;

2. unclear risk of bias;

3. low risk of bias.

We resolved disagreements by discussion or by consulting a third review author. We described all judgements fully and presented the conclusions in the 'Risk of bias' table, which was incorporated into the interpretation of review findings by means of sensitivity analyses.

We judged that blinding of the researcher, the personnel or the participants could not influence any of the outcomes. Therefore, we assessed all included trials at low risk of bias for blinding.

Measures of treatment effect

We performed statistical analyses in accordance with the guidelines for statistical analysis developed by Cochrane, outlined in the Cochrane Handbook for Systematic Reviews of Interventions Version 5.1.0 (Higgins 2011).
All outcomes were binary. We expressed results for each included study as Mantel‐Haenszel odds ratios (ORs) with 95% confidence intervals (CI).

Unit of analysis issues

The primary analysis was per couple randomised. If an included study only reported per cycle data, we contacted the author for additional information. We planned to include studies that could not provide us with per couple data in the review but not in the meta‐analysis, and describe them separately. We included both parallel group and cross‐over trials in the analysis. For cross‐over trials, we used only pre‐cross‐over data. Furthermore, we counted multiple live births (e.g. twins or triplets) as one live birth event.

Dealing with missing data

For missing data, we attempted to contact the authors. When we could not obtain the missing data from the authors, we explained the assumptions we made in the extraction and analysis of the data. We analysed only the available data.

Assessment of heterogeneity

We noted statistical heterogeneity between the results of different studies by visually inspecting the scatter in the data points on the graphs and the overlap in their CIs and using the I² statistics. Following the Cochrane Handbook for Systematic Reviews of Interventions, we judged an I² value greater than 50% to indicate substantial heterogeneity. We used a random‐effects model for sensitivity analysis and explored the original trials for clinical and methodological heterogeneity.

Assessment of reporting biases

Publication bias might influence the interpretation of the pooled results. To detect publication bias, we used a funnel plot, plotting sample size versus effect size, if there were sufficient studies. This plot is only relevant when five or more studies per comparison are included. The graph is symmetrical when bias is absent.

Data synthesis

If appropriate, we combined the data in a meta‐analysis using Review Manager 5 (RevMan 2014), using a fixed‐effect model and presenting OR with 95% CI. For reporting purposes, we translated primary outcomes to absolute risks.
We considered live birth rate and pregnancy outcomes as a positive consequence of treatment. For adverse outcomes such as OHSS, multiple pregnancy rate, miscarriage rate and total fertilisation failure, which are negative consequences, higher numbers were considered to be detrimental (increased odds signify relative harm). This needs to be take into consideration when interpreting the analyses.

Subgroup analysis and investigation of heterogeneity

A priori, we planned to perform separate subgroup analyses if there were more than two studies in each subgroup, for trials with different ovarian stimulation protocols (oral ovulation induction agents (anti‐oestrogens) versus gonadotrophins (FSH, human menopausal gonadotropin (hMG)).

Sensitivity analysis

We conducted sensitivity analyses for the primary outcomes, to examine stability regarding the pooled outcomes.

1. Restriction to studies without high risk of bias.

2. Use of a random‐effects model.

3. Use of risk ratio (RR) rather than OR.