Eligibility criteria

Studies will be selected according to the PECOS (ie, participants, exposures, comparator, outcomes, study designs) criteria (table 1). There will be no restrictions on the time frame or setting of the studies. Studies reported in English, French, Spanish, Portuguese and German languages will be considered. A list of possibly relevant titles in other languages will be provided as an appendix if relevant.

Information sources

Search strategies will be developed using text words related to the population, exposures and outcomes. Five electronic databases will be searched from their inception onwards: PubMed Central (includes MEDLINE), SPORTDiscus (via EBSCOhost interface), Scopus, the Cochrane Central Register of Controlled Trials and ProQuest Central: Health and Medical Collection; Nursing and Allied Health; Research Library: Health and Medicine. The electronic database search will be supplemented by searching for trial protocols through three registers: Clinical Trials (www.clinicaltrials.gov), EU Clinical Trials Register (www.clinicaltrialsregister.eu) and International Standard Randomised Controlled Trial Number (ISRCTN) (www.isrctn.com). To ensure literature saturation, the reference lists of included studies or relevant reviews identified, which may have been identified through the initial search strategy, will also be hand searched. All authors will search their personal files to make sure that all relevant material has been identified.

Search strategy

The PubMed Central search strategy will be developed with input from all authors using the Peer Review of Electronic Search Strategies standard.22 In addition, the search strategy will be peer-reviewed by a research librarian who has expertise in systematic review searching and is not otherwise associated with the project. A draft search strategy for PubMed Central is included in online supplemental file 1. Once the PubMed Central search strategy is finalised, the search strategy will be adapted to the syntax and subject headings of the other databases. The search will be updated toward the end of the review, prior to publication, to retrieve any articles published during the interim period.

Study record

All search results will be uploaded and stored in a systematic review management platform (Covidence systematic review software, Veritas Health Innovation, Melbourne, Australia), which will be accessible to all reviewers. Covidence will automatically remove duplicates by checking the following fields: titles, year, volume, authorship. Two reviewers will independently check the duplicates removed by Covidence and verify their accuracy.

Titles and abstracts will be independently screened by two reviewers, guided by the inclusion and exclusion criteria. Disagreements will be resolved with a consensus-based discussion, and, when in doubt, articles will be carried forward to full-text review. The full text of eligible papers, based on the titles and abstracts, will be downloaded and independently screened. If the reviewers are not in agreement, a third reviewer will be consulted and will provide recommendations. The reviewers will use the annotation facility on the decision dashboard to explain their decision and inform further discussions. If a study is reported in more than one publication, the multiple reports will be collated. When in doubt regarding the eligibility criteria of a study, the reviewers will contact the authors; with a maximum of three attempts, two emails and one phone call (if possible), over a 4-week period. Any ongoing trials, which have not yet been reported, will be recorded, so that they can be added to the ongoing studies table. A PRISMA flowchart detailing the search and selection process will be included (see online supplemental file 1 for a draft template); as well as a list of all full-text studies excluded, detailing the specific reason for exclusion.

A data extraction template will be created based on those used in similar meta-analyses.23 Data will be extracted by two reviewers. To ensure consistency across reviewers, calibration exercises will be conducted before starting the data collection process (ie, the data extraction form will be pilot-tested by each reviewer on five randomly selected studies). When outcome data are not reported in a usable format (ie, in a figure instead of a numerical format) specialist software will be used to extract the data from the figure (eg, WebPlotDigitizer Version 4.4).

In order to avoid double-counting data, records will be scrupulously compared, for example, juxtaposing author names, treatment comparisons, sample sizes and/or outcomes. If the same study data are reported in more than one publication, all publications will be treated as one dataset rather than multiple datasets. When we extract data, we will prioritise the following criteria: greatest number of participants, longest follow-up and primary reports where the primary outcome assessed is most relevant to our research questions. If the data differ across publications, it will be noted, investigated, and the authors contacted for more information; with a maximum of three attempts, two emails and one phone call, over a 4-week period.

Disagreements will be recorded and resolved with a consensus-based discussion between the two reviewers. Any disagreement that cannot be resolved will be referred to a third reviewer who will provide a recommendation. Study authors will be contacted if there are any doubts about the extracted data, with a maximum of three attempts, two emails and one phone call, over a 4-week period. If any disagreement cannot be resolved (ie, either through discussion between the reviewers or with the authors) the disagreement will be reported in the review.

Data items

Reviewers will extract data on the following: (i) study characteristics (ie, design, location, sources of funding, study aim), (ii) participant characteristics (ie, eligibility criteria, age, height, body mass, body mass index, training status, etc), (iii) exposure and comparison characteristics (ie, type, dosage, and duration of hormonal contraceptive use, menstrual cycle phase, type of menstrual irregularity, methods of determining participants’ ovarian hormonal status, etc), (iv) outcome characteristics for AK (ie, method of assessment, assessment characteristics, etc) and occurrence of ACL injuries (ie, method(s) used to confirm the injury, profile of the injury (injury mechanism, context of the injury, primary or recurrent injury, isolated ACL injury or other collateral structures injured), etc).

Outcomes and prioritisation

The primary outcomes are the physical assessment of AK laxity and the secondary outcomes are the occurrence of non-contact ACL injuries.

Risk of bias assessment

Risk of bias will be initially evaluated at the individual study level, using the Newcastle Ottawa Quality Assessment Scale for Cohort or Case–Control Studies.24 The Newcastle Ottawa Quality Assessment Scale is a domain-based risk of bias tool that comprises eight items within three categories to assess the key bias domains: (i) selection; (ii) comparability and (iii) outcome/exposure. We have developed coding systems, that are very similar to formerly published work in our research area,25 according to our outcomes (ie, anterior knee laxity and ACL injury occurrence) and ensured that the assessment is specific to each outcome. We have opted for using the Newcastle Ottawa Quality Assessment Scale without the star-rating system, as the PRISMA explanation and elaboration26 states that presenting assessments for each domain in the tool is preferable to reporting an overall ‘quality score’ because it enables users to understand the specific domains that are at risk of bias in each study. Accordingly, we will separate the key bias domains covered by the Newcastle Ottawa Quality Assessment Scale when assessing bias and we will present the results in a table.

Certainty in cumulative evidence

Certainty will be assessed by two independent reviewers using a strategy based on the recommendations of the Grading of Recommendations Assessment Development and Evaluation working group.27 Any differences between reviewers will be resolved by discussion and, if needed, in consultation with a third reviewer. Certainty in cumulative evidence will be based on consideration of five domains, namely risk of bias (assessed using the NOS as described above), indirectness, inconsistency, imprecision or evidence of publication bias. Directness will be ascertained based on the methods used to identify and confirm menstrual cycle phase, along with injury confirmation. This information is considered essential, given that if unconfirmed, any result observed cannot be directly attributed to the phase under investigation. This will be evaluated based on the response to two questions:

(Q1) Was the ovarian hormone profile confirmed?

If the authors provide a definition for the sampled population and report using blood samples to confirm ovarian hormone status, the a priori rating will be maintained, if not the study will be downgraded a level (eg, a study that is classified as ‘high’ quality, would be downgraded to ‘moderate’ quality).

(Q2) Was the injury medically diagnosed either as part of the study or prior to the study?

These questions are based on methodological conclusions made in previous studies.18 23

Consistency will be ascertained using the meta-analysis results and will be based on visual inspection of effect size and variance estimates across the different levels (eg, within study variation, between study variation and between outcome variation). Precision will be judged based on the number of outcomes available (with outcomes based on <3 data points downgraded) and on interpretation of width of the credible intervals (CrIs). Small-study effects (ie, publication bias) will be visually inspected with funnel plots and quantified with a multi-level extension of Egger’s regression-intercept test.28 Collectively, these procedures will result in a final level of certainty for each outcome (table 2): namely of ‘high’, ‘moderate’, ‘low’ or ‘very low’. This certainty appraisal strategy will not be used to exclude any study.

Data synthesis

Data will be presented in summary tables, which will describe the study characteristics and outcomes. A Bayesian framework was chosen over a frequentist approach as it provides a more flexible modelling approach that will enable results to be interpreted intuitively through reporting of subjective probabilities rather than null hypothesis tests or frequentist CIs.29 For the primary outcomes comprising assessment of menstrual cycle phase, menstrual irregularities and hormonal contraceptive use on AK laxity, both repeated measures data and independent group data will be used to create standardised mean difference effect sizes. Standard distributional assumptions will be used to estimate within study sampling error.30 For repeated measures data where a correlation value is required, a standard value of 0.7 will be used to generate an informative prior with variance included to account for correlations ranging from 0.5 to 0.9. For the secondary outcome comprising the occurrence of ACL injuries, count data will be used to calculate ORs and within study sampling error. All meta-analyses will comprise a three-level hierarchical model to account for random variation across studies and covariance between multiple outcomes reported from the same study. Inferences from all analyses will be performed on posterior samples generated by Hamiltonian Markov Chain Monte Carlo with Bayesian 95% CrIs. Interpretations will be based on visual inspection of the posterior sample, the median pooled effect size value (ES_0.5: 0.5-quantile) and 95% CrIs for location parameters and 75% CrIs for variance parameters. Heterogeneity will be quantified using the posterior distribution of the between study variance parameter. Where possible, meta-regressions will be used to explore sources of variance including the type of hormonal contraceptive (eg, OCPs, implants, injections, etc). Meta-regression will be performed when there is sufficient data including a minimum of four data points per category level or 10 data points for continuous variables.31 Sensitivity analyses will also be conducted to assess the influence of research quality (inclusion of ‘moderate’ and ‘high’ quality studies only) on overall conclusions and effect size magnitudes. Results of meta-analyses will be presented in tables and visually through forest and funnel plots. Where quantitative pooling is not possible due to insufficient data, narrative synthesis will be conducted.