Types of studies

We will include randomised controlled trials (RCTs) irrespective of language of publication, publication status, year of publication, or sample size. We will exclude quasi‐randomised trials (e.g. studies with evidence of inadequate sequence generation such as use of alternate days or patient numbers) and non‐randomised studies as they tend to have high risk of bias. We will include cross‐over trials and cluster‐randomised trials (if any are available), as these designs are valid in this context. However, we will use only first‐phase data of cross‐over trials in our analyses to avoid a carryover effect.

Types of participants

We will include studies of women of reproductive age who met medically defined diagnostic criteria for PMS (including PMDD, which is a severe form of PMS). Diagnosis of PMS requires that symptoms are confirmed by prospective recording for at least two menstrual cycles. Diagnosis must have been made by healthcare professionals before inclusion of women in the study. We will exclude studies that were based solely on self‐diagnosis.

Types of interventions

We will include trials comparing the following interventions.

• Vitamins or minerals or both in any dose and through any route of administration versus placebo or no treatment.

• Vitamins or minerals or both in any dose and through any route of administration versus other treatment.

• Vitamins or minerals or both in any dose and through any route of administration combined with other treatment versus that other treatment alone or that other treatment plus placebo.

• Vitamins or minerals or both in any dose and through any route of administration versus other vitamins or minerals or both in any dose and through any route of administration.

Types of outcome measures

Electronic searches

We will search the following electronic databases, trial registers, and websites from their inception to the present.

• Cochrane Gynaecology and Fertility Group (CGFG) Specialised Register of Controlled Trials (Procite platform) (Appendix 2).

• Cochrane Central Register of Studies Online (CRSO) (web platform) (Appendix 3).

• MEDLINE (Ovid platform) (Appendix 4).

• Embase (Ovid platform) (Appendix 5).

• PsycINFO (Ovid platform) (Appendix 6).

• Allied and Complementary Medicine Database (AMED) (Ovid platform) (Appendix 7).

• Cumulative Index to Nursing and Allied Health Literature (CINAHL) (Ebsco platform) (Appendix 8).

We will combine the MEDLINE search with the Cochrane Highly Sensitive Search Strategy for identifying randomised trials, which appears in the Cochrane Handbook for Systematic Reviews of Interventions (Version 5.0.2, Chapter 6, 6.4.11). We will combine Embase, PsycINFO, and CINAHL searches using trial filters developed by the Scottish Intercollegiate Guidelines Network (SIGN) ‐ http://www.sign.ac.uk/methodology/filters.html#random.

Other electronic sources of trials will include the following.

• Trial registries for ongoing and registered trials ‐ ClinicalTrials.gov, a service of the US National Institutes of Health, at http://www.clinicaltrials.gov; and the World Health Organization International Trials Registry platform search portal, at http://www.who.int/trialsearch/Default.aspx.

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

• ProQuest Dissertations & Theses (PQDT), at http://www.proquest.com/libraries/academic/dissertations‐theses/ (for unpublished dissertations and theses).

• Conference abstracts and other trials on the Web of Science, at http://wokinfo.com/.

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

• OpenGrey, at http://www.opengrey.eu/ (for unpublished literature from Europe).

• Latin American Caribbean Health Sciences Literature (LILACS) database, at http://regional.bvsalud.org/php/index.php?lang=en (for trials from the Portuguese and Spanish speaking world).

• PubMed and Google Scholar (for recent trials not yet indexed in the major databases).

Searching other resources

We will handsearch reference lists of reviews and articles retrieved by the search. We will also handsearch relevant journals and conference abstracts that are not included in the CGFG Register, in liaison with the CGFG Information Specialist.

Selection of studies

We will download all titles and abstracts retrieved through the electronic search to a reference management database (EndNote). After removing duplicates, we will transfer these data to Covidence (Covidence). Two review authors (SK and CK) will independently examine the remaining references. We will exclude studies that clearly do not meet the inclusion criteria. We will obtain copies of full texts of potentially relevant references. Two review authors (SK and CK) will independently assess the eligibility of retrieved reports/publications. We will resolve disagreements through discussion, or, if required, we will consult a third review author (PL). We will identify and exclude duplicates and will collate multiple reports of the same study, so that each study rather than each report is the unit of interest in the review. We will use details regarding the selection process to complete a PRISMA flow diagram and 'Characteristics of excluded studies' tables (Liberati 2009).

Data extraction and management

Two review authors (SK and CK) will independently extract study characteristics and outcome data from included studies using Covidence. We will note any outcome data not reported in a usable way in the 'Characteristics of included studies' tables. We will resolve disagreements by reaching consensus or by involving a third review author (PL). We will correspond with study investigators to request further data on methods and/or results, as required. We will estimate data values from graphs when necessary.

Assessment of risk of bias in included studies

Two review authors (SK and CK) will independently assess risk of bias in included studies by using the criteria available in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011); we will resolve differences by discussion or by appeal to a third review author (PL). We will assess each study on the basis of six categories: selection bias (sequence generation and allocation concealment); performance bias (blinding of participants and personnel); detection bias (blinding of outcome assessment); attrition bias (incomplete outcome data); reporting bias (selective outcome reporting); and other biases (any 'other issues' also considered for assessment). We will judge each item as being at high, low, or unclear risk of bias, as set out in the criteria displayed in Appendix 9 (Higgins 2011). However, we anticipate that the studies included in this review will be highly susceptible to detection and attrition bias owing to the combination of self‐reported outcome measures and the likelihood that dropout might be associated with the individual's condition. We will acknowledge studies without effective blinding of participants, personnel, and outcome assessors as having high risk of performance and detection bias, respectively. If at least 80% of randomised women were included in the analysis, we will rated the study as having low risk of attrition bias.

We will provide in the 'Risk of bias' table a quote from the study report and/or a statement as justification for judgement for each item. We will summarise results in both a 'Risk of bias graph' and a 'Risk of bias summary', if feasible. When interpreting treatment effects and meta‐analyses, we will take into account the risk of bias of included studies that contribute to that outcome. When information on risk of bias relates to unpublished data or correspondence with a trial author, we will note this in the 'Risk of bias' table.

Measures of treatment effect

For dichotomous data (e.g. occurrence of adverse events), we will use numbers of events in control and intervention groups of each study to calculate Mantel‐Haenszel odds ratios (ORs). For continuous data (i.e. end scores and change scores for PMS symptoms), if all studies report exactly the same outcomes, we will calculate mean differences (MDs) between treatment groups. If similar outcomes are reported on different scales, we will calculate standardised mean differences (SMDs). We will reverse the direction of effects of individual studies, if required, to ensure consistency across trials. We will treat ordinal data (i.e. scores obtained from visual analogue scales (VASs) and quality of life scores) as continuous data. We will extract end scores in preference to change scores, when available, and will include them in the 'Summary of findings' table. We will present 95% confidence intervals (CIs) for all outcomes. When data needed to calculate ORs or MDs are not available, we will utilise the most detailed numerical data available that may facilitate similar analyses of included studies (e.g. test statistics, P values). We will compare the magnitude and direction of effects reported by studies with how they are presented in the review, while taking account of legitimate differences.

Unit of analysis issues

The unit of analysis is per woman randomised. In a study with multiple groups, we will combine all relevant experimental intervention groups into a single group to create a single pair‐wise comparison (Higgins 2011). We will include only data from the first phase in cases of cross‐over trials to minimise a carryover effect.

Dealing with missing data

In so far as it is possible, we will carry out analyses on an intention‐to‐treat basis for all outcomes. This means that we will attempt to include in analyses all participants randomised to each group, and we will analyse all participants in the group to which they were allocated, regardless of whether or not they received the allocated intervention. We will attempt to contact study authors to obtain missing data. We anticipate finding data missing from trial reports because of loss to follow‐up; we will attempt to impute missing data for primary outcomes using the 'informative missingness (IM)' approach suggested by Higgins 2008. For continuous outcomes, we will apply the informative missing difference of means (IMDOM), and for dichotomous outcomes, we will calculate the informative missing odds ratio (IMOR) using 'metamiss' or 'metamiss2' in STATA 14 (StataCorp 2015). For secondary outcomes, if we are unable to obtain missing data, we will include only available data.

Assessment of heterogeneity

We will consider whether clinical and methodological characteristics of included studies are sufficiently similar for meta‐analysis to provide a clinically meaningful summary. We will assess statistical heterogeneity using the I² measurement. We will consider I² greater than 50% to indicate substantial heterogeneity (Higgins 2003).

Assessment of reporting biases

In view of the difficulty of detecting and correcting for publication bias and other reporting biases, review authors will aim to minimise the potential impact of biases by ensuring a comprehensive search for eligible studies, and by staying alert for duplication of data. If we include 10 or more studies in an analysis, we will use a funnel plot to explore the possibility of small‐study effects (i.e. the tendency for smaller studies to estimate the intervention effect as more beneficial than reported by larger studies).

Data synthesis

We will carry out statistical analyses with Review Manager 5 software (RevMan 2014), using the Mantel‐Haenszel fixed‐effect model for combining data when included trials examine the same intervention, and when trial populations and methods are acknowledged as sufficiently similar (DerSimonian 1986). We will convert continuous data/standardised mean differences to an interpretable scale/natural units. We will analyse end scores and change scores separately and will not pool these data. We will report separately the results of each comparison listed under Types of interventions. If we identify substantial heterogeneity between included studies, we will summarise findings narratively.

• Vitamins and/or minerals versus placebo or no treatment (stratified by type of comparator and pooled).

• Vitamins and/or minerals versus other treatment (separate comparisons for different comparators).

• Vitamins and/or minerals plus other treatment versus other treatment (± placebo) (separate comparisons for different comparators).

• Vitamins and/or minerals versus other vitamins and/or minerals (separate comparisons for different comparators).

Subgroup analysis and investigation of heterogeneity

When data are available, we will perform subgroup analyses using the RevMan 2014 test for subgroup differences to assess the influence of the following issues on effect size.

• Subgroups by severity of PMS symptoms (with reference to the trial inclusion criteria): mild to moderate PMS versus severe PMS. Severity of PMS will depend on self‐rated scores from rating tools, and PMDD will be defined as severe PMS.

• Subgroups by dosage of vitamins or minerals or both.

If we detect substantial heterogeneity, we will explore possible explanations by performing subgroup analyses and/or sensitivity analyses. We will take any statistical heterogeneity into account when interpreting trial results, especially if we note any variation in the direction of effect.

Sensitivity analysis

We will perform sensitivity analysis to determine the effects of the following factors.

• Repeating the analysis while excluding unpublished studies (if any).

• Repeating the analysis while excluding studies judged to be at 'high' or 'unclear' risk of bias for allocation concealment.

• Repeating the analysis using a random‐effects model.

• As a consequence of the imputation method (informative missingness (IM)) for missing outcome data, we will perform a series of sensitivity analyses to examine the impact of attrition on treatment effect estimates by using a different assumption regarding missing data, an available case analysis (ACA), and imputed case analyses (ICAs), as suggested by Higgins 2008.