Antioxidants for female subfertility

Summary of findings 1. Antioxidant(s) compared to placebo or no treatment/standard treatment for female subfertility

Outcomes	Relative effect (95% CI)	*Anticipated absolute effects^ (95% CI)**			Quality of the evidence (GRADE)	What happens
Antioxidant(s) compared to placebo or no treatment/standard treatment for female subfertility
Patient or population: women with subfertility Setting: Infertility clinics Intervention: Antioxidant(s) Comparison: placebo or no treatment/standard treatment
Outcomes	Relative effect (95% CI)	Without antioxidant(s)	With antioxidant(s)	Difference	Quality of the evidence (GRADE)	What happens
Live birth; antioxidants vs placebo or no treatment/standard treatment (natural conceptions and undergoing fertility treatments) № of participants: 1227 (13 RCTs)	OR 1.81 (1.36 to 2.43)	19.0%	29.8% (24.2 to 36.3)	10.8% more (5.2 more to 17.3 more)	⊕⊝⊝⊝ VERY LOW^a,b,c	We are uncertain whether antioxidants improve live birth rate compared with placebo or no treatment/standard treatment.
Clinical pregnancy; antioxidants vs placebo or no treatment/standard treatment (natural conceptions and undergoing fertility treatments) № of participants: 5165 (35 RCTs)	OR 1.65 (1.43 to 1.89)	18.8%	27.6% (24.8 to 30.4)	8.8% more (6.1 more to 11.6 more)	⊕⊕⊝⊝ LOW^a,d	Antioxidant(s) may improve clinical pregnancy rate, compared with placebo or no treatment/standard treatment (natural conceptions and undergoing fertility treatments).
Adverse events ‐ Miscarriage № of participants: 3229 (24 RCTs)	OR 1.13 (0.82 to 1.55)	4.8%	5.4% (4 to 7.3)	0.6% more (0.8 fewer to 2.5 more)	⊕⊕⊝⊝ LOW^a,c	Antioxidant(s) may result in little to no difference in adverse events ‐ Miscarriage
Adverse events ‐ Multiple pregnancy № of participants: 1886 (9 RCTs)	OR 1.00 (0.63 to 1.56)	4.3%	4.3% (2.7 to 6.5)	0.0% fewer (1.6 fewer to 2.2 more)	⊕⊕⊝⊝ LOW^a,c	Antioxidant(s) may result in little to no difference in adverse events ‐ Multiple pregnancy
Adverse events ‐ Gastrointestinal disturbances № of participants: 343 (3 RCTs)	OR 1.55 (0.47 to 5.10)	2.4%	3.7% (1.2 to 11.2)	1.3% more (1.2 fewer to 8.8 more)	⊕⊕⊝⊝ LOW^a,c	Antioxidant(s) may result in little to no difference in adverse events ‐ Gastrointestinal disturbances
Adverse events ‐ Ectopic pregnancy № of participants: 404 (4 RCTs)	OR 1.40 (0.27 to 7.20)	0.6%	0.9% (0.2 to 4.3)	0.3% more (0.4 fewer to 3.7 more)	⊕⊕⊝⊝ LOW^a,c	Antioxidant(s) may result in little to no difference in adverse events ‐ Ectopic pregnancy
*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: Confidence interval; OR: Odds ratio
GRADE Working Group grades of evidence High certainty: We are very confident that the true effect lies close to that of the estimate of the effect Moderate certainty: We are moderately confident in the effect estimate: The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different Low certainty: Our confidence in the effect estimate is limited: The true effect may be substantially different from the estimate of the effect Very low certainty: We have very little confidence in the effect estimate: The true effect is likely to be substantially different from the estimate of effect
^aDowngraded one level due to serious risk of bias. The no‐treatment group increases risk due to the lack of blinding. ^bDowngraded one level; overall the heterogeneity is low (0%), but in the placebo subgroup the heterogeneity statistic is 60% and some trials are showing potential benefit of the intervention while others are showing benefit of the placebo. ^cDowngraded one level as the event rate is low (< 400). ^dDowngraded one level as the heterogeneity statistic (63%) is considered to represent moderate to substantive heterogeneity.

Summary of findings 2. Head‐to‐head antioxidants for female subfertility

Outcomes	Relative effect (95% CI)	*Anticipated absolute effects^ (95% CI)**			Quality of the evidence (GRADE)	What happens
Head‐to‐head antioxidants for female subfertility
Patient or population: women with subfertility Setting: Infertility clinics Intervention: Head‐to‐head antioxidants Comparison: Other antioxidant
Outcomes	Relative effect (95% CI)	With one antioxidant	With another antioxidant	Difference	Quality of the evidence (GRADE)	What happens
Live birth; type of antioxidant (natural conceptions and undergoing fertility treatments) ‐ Melatonin lower dose versus melatonin higher dose № of participants: 140 (2 RCTs)	OR 0.94 (0.41 to 2.15)	24.0%	22.9% (11.5 to 40.4)	1.1% fewer (12.5 fewer to 16.4 more)	⊕⊕⊝⊝ LOW^a,b	There was no clear evidence of a difference between the lower and higher doses of melatonin
Clinical pregnancy; type of antioxidant (natural conceptions and undergoing fertility treatments) ‐ N‐acetylcysteine versus L‐carnitine № of participants: 164 (1 RCT)	OR 0.81 (0.33 to 2.00)	14.6%	12.2% (5.4 to 25.5)	2.4% fewer (9.2 fewer to 10.9 more)	⊕⊝⊝⊝ VERY LOW^c,d	There was no clear evidence of a difference between N‐acetylcysteine versus L‐carnitine
Clinical pregnancy; type of antioxidant (natural conceptions and undergoing fertility treatments) ‐ Melatonin lower dose versus melatonin higher dose № of participants: 140 (2 RCTs)	OR 0.94 (0.41 to 2.15)	24.0%	22.9% (11.5 to 40.4)	1.1% fewer (12.5 fewer to 16.4 more)	⊕⊕⊝⊝ LOW^a,b	There was no clear evidence of a difference between the lower and higher doses of melatonin
Adverse events ‐ Miscarriage № of participants: 304 (3 RCTs)	OR 1.54 (0.42 to 5.67)	3.0%	4.6% (1.3 to 15.1)	1.6 more (1.7 fewer to 12.1 more)	⊕⊕⊝⊝ LOW^a,b	There were no miscarriages in either melatonin study (140 women) There was no clear evidence of a difference between N‐acetylcysteine versus L‐carnitine (164 women)
Adverse events ‐ Multiple pregnancy	There were no trials reporting multiple pregnancy
Adverse events ‐ Gastrointestinal disturbances	There were no trials reporting gastrointestinal disturbances
Adverse events ‐ Ectopic pregnancy Melatonin lower dose versus melatonin higher dose № of participants: 120 (1 RCT)	Not estimable, there were no ectopic pregnancies in either group				⊕⊝⊝⊝ VERY LOW ^{3 4}	There was no clear evidence of a difference between the lower and higher doses of melatonin
*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: Confidence interval; OR: Odds ratio
GRADE Working Group grades of evidence High certainty: We are very confident that the true effect lies close to that of the estimate of the effect Moderate certainty: We are moderately confident in the effect estimate: The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different Low certainty: Our confidence in the effect estimate is limited: The true effect may be substantially different from the estimate of the effect Very low certainty: We have very little confidence in the effect estimate: The true effect is likely to be substantially different from the estimate of effect
^aDowngraded one level as there are only two trials in this analysis and one is small. ^bDowngraded one level as event rate is low (< 400). ^cDowngraded two levels as one study can not represent possible subfertile populations. ^dDowngraded two levels as only one study, event rate low and small number of participants

Background

Description of the condition

A couple that has tried to conceive for a year or longer without success is considered to be subfertile (Practice Committee of ARSM 2020) or less fertile than a typical couple. The World Health Organization (WHO) (Zegers‐Hochschild 2009) defines infertility as the “failure to achieve a clinical pregnancy after 12 months or more of regular unprotected sexual intercourse”. Levels of infertility in 2010 were similar to those in 1990 in most of the world, apart from declines in Sub‐Saharan Africa and in South Asia (Mascarenhas 2012). Thirty to forty per cent of cases of subfertility are due to causes in women (WHO). Influencing factors include ovulatory failure, tubal damage, endometriosis, poor egg quality and unexplained subfertility. It is suggested that up to 25% of couples who are planning a baby have difficulty (Boivin 2007; Hart 2003). Nine per cent of men and 11% of women of reproductive age are thought to experience infertility (Chandra 2013)

To overcome these fertility problems, many couples undergo assisted fertility techniques (assisted reproductive techniques (ART)). These include ovulation stimulation, intrauterine insemination (IUI), in vitro fertilisation (IVF) and intracytoplasmic sperm injection (ICSI).

Women use antioxidant supplements in preparation for ART or simultaneously with the treatment, or both, and some women use supplements alone with no ART in an attempt to improve their fertility.

Description of the intervention

Antioxidants are biological and chemical compounds that reduce oxidative damage, the imbalance between creation of reactive oxygen species and the body's ability to detoxify. They are a group of organic nutrients that include vitamins, minerals and polyunsaturated fatty acids (PUFAs). Some of the predominant antioxidants used in female subfertility are N‐acetylcysteine; melatonin; vitamins A, C and E; folic acid; myo‐inositol; zinc and selenium. They may be administered as a single antioxidant or as combined therapy.

PUFAs are classified into omega‐3, omega‐6 and omega‐9. Omega‐9 is synthesised by animals, but omega‐3 and ‐6 need to be supplemented in the diet. The main sources of omega‐6 are vegetable oils. Sources of omega‐3 are vegetable and fish oils. The ratio of omega‐6 to omega‐3 has risen in recent times (as a result of increased intake of vegetable oils) to the point where there is a reduced need for intake of omega‐6 and an increased need for intake of omega‐3 (Wathes 2007).

The amino acid L‐arginine also has antioxidant properties that aid in the inflammatory response and act against oxidative damage (Ko 2012).

When oxidative damage occurs, toxins are produced as a consequence of all cells using oxygen to survive. Toxic end‐products may include molecules that have unpaired electrons, which may lead to the formation of free radicals. Free radicals may cause further harmful reactions with lipids in membranes, amino acids in proteins and carbohydrates within nucleic acids. An antioxidant molecule is thought to be capable of slowing or preventing the oxidation of other molecules, and potentially of reducing the production of free radicals, which may cause this cellular damage.

Two major types of free radicals have been identified: reactive oxygen species (ROS) and reactive nitrogen species (RNS). Reactive oxygen species are products of normal cellular metabolism and consist of oxygen ions, free radicals and peroxides. The addition of one electron to oxygen forms the superoxide anion radical, which can then be converted to hydroxyl radical, peroxyl radical or hydrogen peroxide. Free radicals seek to participate in chemical reactions that relieve them of their unpaired electron, resulting in oxidation (Ruder 2008; Tremellen 2008). The presence of ROSs within the ovary and the endometrium may have physiological and pathological implications for women when they try to conceive. Oxidative stress (OS) is a result of an imbalance between the amount of ROS and the quantity of natural antioxidants present within the body, and results in overwhelming the body’s natural defence mechanism. ROS can attack lipids, proteins DNA and affect metabolic pathways (chemical transformations in the cells) (Agarwal 2012). Natural antioxidants present in the body include catalase, glutathione peroxidase, superoxide dismutase and glutathione reductase, vitamins C and E, ferritin and transferrin (Gupta 2007).

Indirect evidence from smoking and alcohol trials suggests that these factors have a negative impact on female fertility, potentially through the generation of excessive oxidative stress (Agarwal 2012; Ruder 2008). Other lifestyle factors such as diet, disease, pollution, stress and allergies also contribute to increased levels of free radicals (Agarwal 2012).

The global vitamin and supplement market has grown exponentially and has been reported in 2016 as being worth over USD 140 billion, and projected to reach USD 230 billion by 2027 (Global Supplement Report 2019; Global Supplement report 2016; Reportlinker.com 2010). In the UK there has been a 13.8% growth in vitamin and supplement manufacturing from 2015 to 2020. In 2009 sales of vitamins and dietary supplements in the UK "totalled £674.6 million, a growth of about 16% over the previous five years, with the two biggest‐selling areas being multivitamins (GBP 138.6 million) and fish oils (GBP 139.1 million)" (NHS News 2011). Vitamins and supplements are dispensed through various retail outlets, including health‐food shops, online retailers, health centres, fitness clubs, supermarkets and pharmacies.

In an effort to enhance fertility, couples are increasingly resorting to ART; however, these techniques do not cure the causes of subfertility, but rather overcome some of its barriers. Adjunctive measures, including courses of dietary supplements such as oral antioxidants, may be beneficial (Ebisch 2007). However, most antioxidant supplements are uncontrolled by regulation, and thus their effects may be unpredictable in the population.

How the intervention might work

Antioxidants are said to have an important role in the regulation of all processes involved in the birth of a healthy baby (Gupta 2007). The local development of oxidative stress will have significant adverse effects on these processes. Conditions with which the adverse effects of oxidative stress may be associated in subfertile women include endometriosis, hydrosalpinges (dilated fallopian tubes), polycystic ovary syndrome (PCOS) and potentially unexplained subfertility (Agarwal 2012; Ruder 2008; Zhao 2006).

At the time of conception, oxidative stress can lead to cell membrane lipid peroxidation, cellular protein oxidation and DNA damage, causing a negative effect upon the oocyte (immature egg cell), the embryo and implantation (Ruder 2008). Antioxidants would be expected to counteract the negative impact of oxygen‐free radicals by acting as free radical scavengers.

Supplementary antioxidants may have several methods of action. Fertility benefits of vitamin E include improvement in epithelial growth in blood vessels and in the endometrium (Ledee‐Bataille 2002). Higher vitamin D levels are associated with an increased likelihood of successful pregnancy and may particularly benefit women with PCOS in lowering hyperandrogenism (androgen excess) (Thomson 2012). Myo‐inositol helps ovarian function and decreases hyperandrogenism and insulin resistance (Nestler 1998); L‐arginine improves endometrial blood flow (Takasaki 2009); N‐acetylcysteine is needed for fertile cervical mucus and ovulation (Badawy 2007); and PUFAs influence prostaglandin (lipid compounds with hormone‐like effects) synthesis and steroidogenesis (creation of steroid hormones), and also play a role in the composition of cell membranes of the sperm and oocyte, which is important during fertilisation (Wathes 2007). Cohort studies have shown some evidence suggesting that in some instances taking a multivitamin tablet may increase fertility (Haggarty 2006) or even regulate ovulation (Charvarro 2008).

Why it is important to do this review

There is currently limited evidence on whether antioxidants improve fertility, and ongoing trials in this area show varied results. This review assesses the effectiveness of different antioxidants and different dosages. This is an update of a review first published in 2013 (Showell 2013) and updated in 2017 (Showell 2017).

Subfertile women are highly motivated to explore all avenues of treatment in their desire to have a healthy baby. Antioxidants are mostly unregulated and are readily available for purchase by consumers. Research has suggested that a significant number of women undergoing fertility treatment are taking oral supplements in the expectation that this will improve their chances of conception (O'Reilly 2014; Stankiewicz 2007). Consumer perception is that antioxidant therapy is not associated with harm and is associated only with benefit. It is important to establish whether or not this therapy does improve fertility and whether it is associated with any harm.

Objectives

To determine whether supplementary oral antioxidants compared with placebo, no treatment/standard treatment or another antioxidant improve fertility outcomes for subfertile women.

Methods

Criteria for considering studies for this review

Types of studies

Inclusion criteria

Randomised controlled trials (RCTs).
Cross‐over trials are included, but we used only first‐phase data in the analysis. Achieving outcomes such as pregnancy and live birth would preclude entry of couples into the next trial phase (Dias 2006).

Exclusion criteria

Any quasi‐randomised trials.

Types of participants

Inclusion criteria

Trials that included subfertile women who had been referred to a fertility clinic and might or might not be undergoing assisted reproductive techniques (ART) such as in vitro fertilisation (IVF), intrauterine insemination (IUI) or intracytoplasmic sperm injection (ICSI).

Exclusion criteria

Trials enrolling only fertile women attending a fertility clinic exclusively as the result of male partner infertility.
Trials enrolling women exclusively with any vitamin deficiency.

Types of interventions

Inclusion criteria

Any type of oral antioxidant supplementation versus control: placebo (plus or minus a co‐intervention) or no treatment/standard treatment (standard treatment includes folic acid < 1 mg);
Individual or combined oral antioxidants versus any antioxidant (head‐to‐head trials).

On clinical advice, we analysed trials that used folic acid (standard treatment) and those that included a co‐intervention (a fertility drug such as clomiphene citrate or metformin) in both arms in the antioxidant versus placebo or no treatment/standard treatment comparison, and not in the head‐to‐head comparison, as the controls were not considered to be active treatments.

Exclusion criteria

Interventions that included antioxidants alone versus fertility drugs as controls. These fertility drugs included metformin and clomiphene citrate.

Types of outcome measures

Primary outcomes

Live birth rate per woman randomly assigned: if live birth data were unavailable and the trial reported ongoing pregnancy, we reported ongoing pregnancy as live birth (footnoted in the forest plot). We defined live birth as delivery of a live fetus after 20 completed weeks of gestation, and ongoing pregnancy as evidence of a gestational sac with fetal heart motion at 12 weeks, confirmed with ultrasound.

Secondary outcomes

Clinical pregnancy rate per woman (as confirmed by the identification of a gestational sac on ultrasound at seven or more weeks' gestation).
Any adverse effects reported by the trial. We subgrouped these events by the type of adverse event reported.

Search methods for identification of studies

We searched for all reports, published and unpublished, that described RCTs investigating oral antioxidant supplementation for subfertile women and its impact on live birth, pregnancy and adverse event rates. We used both indexed and free‐text terms, and applied no language or date restrictions.

Electronic searches

We searched the following databases:

The Cochrane Gynaecology and Fertility Group's (CGFG) specialised register of controlled trials; searched 12 September 2019, PROCITE platform (Appendix 1);
CENTRAL, via the Cochrane Register of Studies Online (CRSO); searched 12 September 2019, Web platform (Appendix 2);
MEDLINE; searched from 1946 to 12 September 2019, OVID platform (Appendix 3);
Embase; searched from 1980 to 12 September 2019, OVID platform (Appendix 4);
PsycINFO; searched from 1806 to 12 September 2019, OVID platform (Appendix 5);
AMED (Allied and Complementary Medicine); searched from 1985 to 12 September 2019, OVID platform (Appendix 6);
CINAHL; searched from 1961 to 12 September 2019, EBSCO platform (Appendix 7).

The MEDLINE search was limited by the Cochrane highly sensitive search strategy filter for identifying randomised trials, which appears in the Cochrane Handbook of Systematic Reviews of Interventions (Chapter 6, 6.4.11; Lefebvre 2011). We combined the Embase searches with trial filters developed by the Scottish Intercollegiate Guidelines Network (SIGN) (www.sign.ac.uk/what-we-do/methodology/search-filters/).

Searching other resources

(last searched September 2019):

International trial registers: the ClinicalTrials database, a service of the US National Institutes of Health (clinicaltrials.gov/ct2/home) and the World Health Organization International Trials Registry Platform search portal (www.who.int/trialsearch/Default.aspx);
Web of Knowledge for conference proceedings and published trials;
Google, using the keywords 'antioxidants female infertility' and 'antioxidants female subfertility';
Database for Abstracts of Reviews of Effects (DARE) for other reviews on this topic;
'Grey' literature (unpublished and unindexed), through the openGREY database (www.opengrey.eu/); (Appendix 8).

We also contacted known experts and personal contacts for information on any unpublished materials, and we checked the citation lists of appropriate papers for any relevant references.

Data collection and analysis

We conducted data collection and analysis in accordance with the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2019).

Selection of studies

Two review authors (MGS and RM‐P) independently reviewed titles and abstracts of trials for eligibility. We obtained the full texts of trials that we considered for inclusion. We sought further information from the authors of trials that did not contain sufficient information to make a decision about eligibility. We resolved any disagreements by reference to a third review author. We documented the selection process with a PRISMA flow chart.

Data extraction and management

Two review authors (MGS and RM‐P) independently extracted data from the included trials using a data extraction form. We compared the two sets of extracted data and resolved discrepancies by discussion. The review authors screened the trials to ensure that there were no duplicate publications.

We designed the data extraction forms to extract information on study characteristics and outcomes. We have included this information and present it in the Characteristics of included studies and the Characteristics of excluded studies tables, in keeping with the guidance provided by the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2019). If any information on trial methodology or any trial data were missing, we contacted the study authors by email and by post. The predominant questions for trial authors concerned live birth data, clinical pregnancy, methods of randomisation and allocation concealment.

Assessment of risk of bias in included studies

We assessed the included studies for risks of bias using the Cochrane 'Risk of bias' tool, to assess selection bias (sequence generation and allocation concealment); performance bias (blinding of participants and personnel); detection bias (blinding of outcome assessors); attrition bias (completeness of outcome data); reporting bias (selective outcome reporting); and other potential sources of bias. Two review authors (MGS and RM‐P) assessed the included studies according to these six criteria, resolving any disagreements by discussion with a third review author. We sought published protocols.

We took care to search for within‐study selective reporting, for example trials failing to report outcomes such as live birth or reporting them in insufficient detail to allow inclusion. Where protocols were available, we assessed studies for differences between study protocols and published results.

In cases where included studies failed to identify the primary outcome of live birth but did report pregnancy rates, we carried out an informal assessment to determine whether pregnancy rates were similar to those in studies that reported live birth.

Measures of treatment effect

We expressed the dichotomous data for live birth, pregnancy rate, miscarriage and adverse events as Mantel‐Haenszel odds ratios (ORs) with 95% confidence intervals (95% CIs).

Unit of analysis issues

We analysed the outcomes of live birth, pregnancy and adverse events per woman randomly assigned, counting multiple births as one live birth event.

Dealing with missing data

In cases where trial data were missing, we first sought information from the original trial investigators. Details of authors contacted and the questions asked of them are contained in Characteristics of included studies. In addition, and where possible, we performed analyses on all outcomes on an intention‐to‐treat basis, i.e. to include in the analyses all women randomly assigned to each group and to analyse all women in the group to which they were assigned, regardless of whether or not they received the allocated intervention.

Assessment of heterogeneity

We considered whether the clinical and methodological characteristics of included studies were sufficiently similar for meta‐analysis to provide a clinically meaningful summary. We assessed statistical heterogeneity according to the guidelines set out in the Cochrane Handbook for Systematic Reviews of Interventions (Deeks 2019). We examined heterogeneity between the results of different trials by visually examining the forest plots and the overlap of confidence intervals (poor overlap suggested heterogeneity), by considering the P value (a low P value or a large Chi² statistic relative to the degree of freedom suggests heterogeneity), and by identifying the I² statistic. If I² was 50% or higher, we assumed high heterogeneity, and conducted a sensitivity analysis. A high I² statistic suggests that variations in effect estimates were due to differences between trials rather than to chance alone.

Assessment of reporting biases

The search strategies covered multiple sources, without language or publication restrictions. We were alert to the possibility of duplication of data. We used a funnel plot to explore the possibility of small‐study effects in cases where estimates of intervention effect can be more beneficial in smaller studies (Page 2019).

Data synthesis

We conducted statistical analysis of the data using Review Manager 5 (RevMan 2014). We considered pregnancy outcomes to be positive, and higher numbers of pregnancy rates to be a benefit. We considered the outcomes of miscarriage and adverse events to be negative effects, and higher numbers harmful.

We combined data from primary studies using a fixed‐effect model in the following comparisons:

Antioxidants versus control (placebo or no treatment/standard treatment);
Antioxidants versus antioxidants, or head‐to‐head.

We displayed increases in the odds of a particular outcome, which may be beneficial (e.g. live birth) or detrimental (e.g. adverse events), graphically in meta‐analyses to the right of the centre line, and decreases in the odds of a particular outcome to the left of the centre line.

The aim was to define analyses that were comprehensive and mutually exclusive, so that we could slot all eligible study results into one stratum only. We specified comparisons so that any trials falling within each stratum could be pooled for meta‐analysis. Stratification allowed for consideration of effects within each stratum, as well as or instead of an overall estimate for comparison.

In trials with multiple arms, we pooled intervention groups versus the control group.

If individuals had been randomly re‐assigned after failed cycles, we did not pool the data in a meta‐analysis.

Subgroup analysis and investigation of heterogeneity

We conducted the following subgroup analyses:

Type of control, placebo or no treatment;
Type of antioxidant, whether individual or combined (three or more antioxidants combined);
Trials that enrolled women with different indications for infertility (i.e. PCOS, endometriosis, unexplained infertility or poor responders);
Trials that enrolled women who were also undergoing IVF or ICSI.

Sensitivity analysis

We conducted sensitivity analyses on the primary outcomes if we detected a high degree of heterogeneity (where the I² statistic was 50% or more), excluding studies:

with a high risk of bias, or
that used antioxidants plus a fertility drug (a co‐intervention) versus placebo plus a fertility drug

We also conducted a sensitivity analysis on the choice of using a fixed‐effect model by using a random‐effects model.

Overall quality of the body of evidence: 'Summary of findings' tables

We produced a 'Summary of findings' table, using GRADEpro GDT software (GRADEpro GDT 2015) and Cochrane methods (Schünemann 2019) for the main review comparison (Antioxidant(s) compared to placebo or no treatment/standard treatment). This table evaluates the overall quality of the body of evidence for the main review outcomes (live birth, clinical pregnancy and adverse events), using GRADE criteria (study limitations: risk of bias, consistency of effect, imprecision, indirectness and publication bias). We have included an additional 'Summary of findings' table for the main review outcomes for the head‐to‐head comparison, evaluating those trials that look at one antioxidant versus another antioxidant. Two review authors, working independently, made judgements about evidence quality ('high', 'moderate', 'low' or 'very low').

Results

Description of studies

Results of the search

2013 version of the review

The search retrieved 2127 abstracts and titles, which we screened to identify trials that met our inclusion criteria. We retrieved the full texts of 67 trials for appraisal. Only one study (Bonakdaran 2012) was not published in English, with the full text in Persian; however, the English abstract contained enough information to show that it did not meet the inclusion criteria, and we therefore excluded it. Of the 67 studies assessed, we included 28 and excluded 39. A repeat search in April 2013 revealed seven studies (Carlomagno 2012; Choi 2012; Mohammadbeigi 2012; Rosalbino 2012; Salehpour 2012; Schachter 2007; Salem 2012) that we placed into the 'Awaiting classification' section of the review. We found 12 ongoing trials in searches of the clinical trial registers.

2017 Update

We assessed 926 abstracts (after 222 duplicates were removed) for inclusion from the title and abstract found in a search dated from April 2013 to September 2016. We assessed 39 of these papers in full text. One study was published in Persian (Mohammadbeigi 2012) and required translation (see Acknowledgements). We excluded 15 articles (14 studies) of the 39, and included 24 (23 studies). Of the latter, six were from the seven trials placed in 'Awaiting classification' in the original review, while Salem 2012 was excluded due to inappropriate intervention and control. For the 2017 update, four of the 12 previously ongoing trials are now included (Bentov 2014; Mohammadbeigi 2012; Unfer 2011; Youssef 2015). The conference abstract of the included study Aboulfoutouh 2011 in the original review became a secondary reference of Youssef 2015 in the update, and Rezk 2004, formerly an excluded study, is now included as a secondary reference of Rizk 2005. Pourghassem 2010 was found to be the same trial as the excluded Ardabili 2012. We excluded Pasha 2011 due to an ineligible population. We added two trials (NCT03023514; NCT02058212) after the search in September 2016, so eight trials were now ongoing (Fernando 2014; NCT01019785; NCT03023514; NCT02058212; IRCT201112148408N1; CTRI/2012/08/002943; NCT01782911; NCT01267604).

We included 23 trials in the 2017 update: Battaglia 1999; Bentov 2014; Brusco 2013; Carlomagno 2012; Cheraghi 2016; Choi 2012; Colazingari 2013; Daneshbodi 2013; Deeba 2015; El Refaeey 2014; Ismail 2014; Keikha 2010; Lesoine 2016; Maged 2015; Mohammadbeigi 2012; Pacchiarotti 2016; Panti Abubakar 2015; Polak de Fried 2013; Razavi 2015; Rosalbino 2012; Salehpour 2012; Schachter 2007; Valeri 2015.

Fifty trials were included in this updated review and 50 have been excluded.

2020 Update

For the 2020 update we assessed 1268 abstracts (after removing 262 duplicates), checking titles and abstracts for eligibility criteria. The articles were found in a search dated from 1st January 2013 to 12th September 2019. We retrieved 39 full‐text papers for further eligibility criteria and from this we excluded six articles (four studies) and included 31 articles (25 studies); one of these (Schillaci 2012) was found through handsearching references (See the PRISMA flow chart (Figure 1). All papers were in English.

Figure 1

Study flow diagram.

For the current update one ongoing trial, Fernando 2014 became an included study (Fernando 2018). In addition to the seven ongoing trials from the 2017 update (NCT01019785; NCT03023514; NCT02058212; IRCT201112148408N1; CTRI/2012/08/002943; NCT01782911; NCT01267604), we added 26 new ongoing trials: ChiCTR1800019772; ChiCTR‐IPR‐15006369; EUCTR2015‐004233‐27‐IT; IRCT201009131760N9; IRCT201207156420N11; IRCT2012120311430N2; IRCT201306115942N2; IRCT20150831023831N2; IRCT201510266917N3; IRCT2016022821653N5; IRCT20160410027311N6; ISRCTN23488518; JPRN‐UMIN000016992; NCT01659788; NCT01665547; NCT01896492; NCT02239107; NCT02993588; NCT03085030; NCT03117725; NCT03306745; NCT03396380; NCT03476564; NCT04019899; PACTR201902584533870; TCTR20171109001.

We include 25 new studies in this review update: Al‐Alousi 2018; Behrouzi 2017; Caballero 2016; El Sharkwy 2019a; El Sharkwy 2019b; Espino 2019; Fernando 2018; Ghomian 2019; Hashemi 2017; Hefny 2018; Heidar 2019; Jahromi 2017; Jamilian 2018; Lu 2018; Mokhtari 2016; Mokhtari 2019; Mostajeran 2018; Rasekhjahromi 2018; Schillaci 2012; Sen Sharma 2017; Siavashani 2018; Taylor 2018; Tunon 2017; Xu 2018; Zadeh Modarres 2018.

Eight studies (Brusco 2013; Ciotta 2011; Colazingari 2013; Lesoine 2016; Pacchiarotti 2016; Papaleo 2009; Rosalbino 2012; Unfer 2011) were removed from the original review as they are now included in the Cochrane Review Inositol for subfertile women with polycystic ovary syndrome (Showell 2018).

Four pentoxifylline studies (Alborzi 2007; Aleyasin 2009; Balasch 1997; Creus 2008), were moved from the included category to excluded, as pentoxifylline is a prescription drug rather than an over‐the‐counter antioxidant supplement and therefore does not fit the inclusion criteria.

We now include 63 studies in this updated review (see Characteristics of included studies) and we exclude 58 (see Characteristics of excluded studies).

Included studies

Sixty‐three studies met the criteria for inclusion. Twenty were based in Iran (Behrouzi 2017; Cheraghi 2016; Daneshbodi 2013; Ghomian 2019; Hashemi 2017; Heidar 2019; Jahromi 2017; Jamilian 2018; Keikha 2010; Mohammadbeigi 2012; Mokhtari 2016; Mokhtari 2019; Mostajeran 2018; Rasekhjahromi 2018; Rashidi 2009; Razavi 2015; Salehpour 2009; Salehpour 2012; Siavashani 2018; Zadeh Modarres 2018), 10 in Egypt (Badawy 2006; El Refaeey 2014; El Sharkwy 2019a; El Sharkwy 2019b; Hefny 2018; Ismail 2014; Maged 2015; Rizk 2005; Nasr 2010; Youssef 2015). Eight were based in Italy (Battaglia 1999; Battaglia 2002; Carlomagno 2012; Gerli 2007; Lisi 2012; Rizzo 2010; Schillaci 2012; Valeri 2015), four in Turkey (Batioglu 2012; Cicek 2012; Eryilmaz 2011; Ozkaya 2011), three in Korea (Choi 2012; Kim 2006; Kim 2010), two in Spain (Espino 2019; Tunon 2017), in the USA (Taylor 2018; Westphal 2006), Argentina (Caballero 2016; Polak de Fried 2013) and China (Lu 2018; Xu 2018), and one each in the UK (Agrawal 2012), Hungary/Austria (Griesinger 2002), Mexico (Mier‐Cabrera 2008), Canada (Bentov 2014), Bangladesh (Deeba 2015), Nigeria (Panti Abubakar 2015), Israel (Schachter 2007), Australia (Fernando 2018), Iraq (Al‐Alousi 2018) and India (Sen Sharma 2017).

We tried to contact authors of all the included studies to obtain further details and clarification, but we could not obtain data for meta‐analysis from 24 trials (Al‐Alousi 2018; Caballero 2016; Carlomagno 2012; Choi 2012; Daneshbodi 2013; Deeba 2015; Ghomian 2019; Hashemi 2017; Hefny 2018; Heidar 2019; Jamilian 2018; Keikha 2010; Kim 2006; Kim 2010; Mohammadbeigi 2012; Mokhtari 2016; Ozkaya 2011; Rasekhjahromi 2018; Razavi 2015; Schillaci 2012; Siavashani 2018; Taylor 2018; Valeri 2015; Zadeh Modarres 2018), and one did not report on the outcomes included in this review (Salehpour 2009). In one trial (Gerli 2007) (see Table 1), only half of the participants declared that they wanted to become pregnant before the study began; we have therefore included this trial, but have not used the data in the meta‐analysis (see Characteristics of included studies).

Table 1. Gerli 2007‐ data not included in meta‐analysis

Outcome	Data	Notes
Clinical pregnancy rate; myo‐inositol + folic acid	4/23	Only 42 of the 92 women enrolled in this trial declared a desire to become pregnant
Clinical pregnancy rate; folic acid + placebo	1/19	‐
Miscarriage rate; myo‐inositol + folic acid	Miscarriage reported, but unknown whether from treatment or control	1 miscarriage occurred in the first trimester, but it is unknown from which group
Miscarriage rate; folic acid + placebo	Unknown	‐

Duration of treatment ranged from 10 to 12 days (Battaglia 2002) to 12 months (Nasr 2010). Nine trials (Eryilmaz 2011; Ghomian 2019; Hefny 2018; Ismail 2014; Maged 2015; Mostajeran 2018; Rizk 2005; Salehpour 2012; Sen Sharma 2017) gave treatment for four to five days during the menstrual cycle and the treatment was repeated per unsuccessful cycle.

One trial (Bentov 2014) was terminated before the end due to the publication of a paper (Levin 2012) describing the negative effects of polar body biopsy, an adjunctive treatment in this trial, on the development of the embryo. The trial began in 2010 and ran until 2012, enrolling 39 women. This study was included in the meta‐analysis but was rated at high risk of bias in two domains; 'incomplete outcome reporting' and in 'other bias'.

Participants

The trials randomly assigned 7760 subfertile women who were attending a fertility clinic and might or might not be undergoing ART procedures such as IVF, IUI or ICSI. The age range of randomly‐assigned participants was 18 to 45 years; at the upper age range Battaglia 1999 enrolled women who were between 37 and 44 years, and Fernando 2018 enrolled women as young as 18 years old.

Twenty‐seven trials (Behrouzi 2017; Cheraghi 2016; Choi 2012; Daneshbodi 2013; El Refaeey 2014; El Sharkwy 2019a; El Sharkwy 2019b; Ghomian 2019; Hefny 2018; Heidar 2019; Ismail 2014; Jamilian 2018; Keikha 2010; Maged 2015; Mohammadbeigi 2012; Mokhtari 2016; Nasr 2010; Mostajeran 2018; Panti Abubakar 2015; Rasekhjahromi 2018; Razavi 2015; Rizk 2005; Salehpour 2012; Schachter 2007; Sen Sharma 2017; Siavashani 2018; Zadeh Modarres 2018) included women with PCOS. Other participants in the trials were enrolled for endometriosis, ovulation failure, tubal blockages, recurrent implantation failure, poor ovarian reserve and unexplained subfertility. One trial included women aged 35 to 42 years with poor oocyte quality and poor response (Rizzo 2010). Schillaci 2012 looked at the use of myo‐inositol for two different groups of women: those with PCOS and those with poor response. Only those women with poor ovarian response are included in this review, and the group of women with PCOS will be included in the update of Showell 2018 (Inositol for subfertile women with polycystic ovary syndrome). Nine trials (Agrawal 2012; Al‐Alousi 2018; Batioglu 2012; Battaglia 1999; Fernando 2018; Griesinger 2002; Taylor 2018; Tunon 2017; Westphal 2006) included women with more than one fertility problem: these reasons included a percentage of male‐partner subfertility, unexplained subfertility, ovulatory problems, poor responders, PCOS, tubal blockages and endometriosis. One trial included a small percentage of women whose subfertility was caused by the male partner (Griesinger 2002).

One trial enrolled only women who were aged over 40 (Valeri 2015), and Taylor 2018 enrolled women of advanced maternal age (36 to 42 years). One trial (Gerli 2007) included participants in whom "infertility was an ailment in only half of the participants in each group". The author of this trial states that there was "no difference in the proportions of infertile women in the groups".

Thirty‐three studies included women undergoing IVF/ICSI (Al‐Alousi 2018; Batioglu 2012; Battaglia 1999; Battaglia 2002; Bentov 2014; Caballero 2016; Carlomagno 2012; Cheraghi 2016; Choi 2012; Eryilmaz 2011; Espino 2019; Fernando 2018; Griesinger 2002; Heidar 2019; Jahromi 2017; Jamilian 2018; Kim 2006; Kim 2010; Lisi 2012; Lu 2018; Mokhtari 2016; Ozkaya 2011; Polak de Fried 2013; Rizzo 2010; Salehpour 2009; Schillaci 2012; Siavashani 2018; Taylor 2018; Tunon 2017; Valeri 2015; Xu 2018; Youssef 2015; Zadeh Modarres 2018). Twenty studies included women undergoing natural intercourse or ovulation induction with timed intercourse or IUI (Agrawal 2012; Badawy 2006; Behrouzi 2017; Cicek 2012; Deeba 2015; El Refaeey 2014; El Sharkwy 2019a; El Sharkwy 2019b; Ghomian 2019; Hefny 2018; Ismail 2014; Maged 2015; Mohammadbeigi 2012; Mokhtari 2019; Mostajeran 2018; Panti Abubakar 2015; Rasekhjahromi 2018; Rizk 2005; Salehpour 2012; Sen Sharma 2017). The remaining 10 studies enrolled women who were either having, no adjunctive treatment, or each trial included a number of differing treatments, i.e. some women having IVF while others were having IUI, and only one trial enrolled women undergoing laparoscopic ovarian drilling (Nasr 2010).

Further details of inclusion and exclusion criteria are available in the Characteristics of included studies table.

Interventions

A variety of antioxidants were used in the included trials. Comparisons covered antioxidants versus placebo, no treatment or standard treatment (folic acid < 1 mg), and head‐to‐head comparisons (antioxidant versus antioxidant).

Comparison of antioxidants versus placebo, no treatment and standard treatment included the following: combinations of antioxidants; L‐arginine, vitamin E, myo‐inositol, D‐chiro‐inositol, carnitine, selenium, vitamin B complex, vitamin C, vitamin D+calcium, CoQ10, melatonin, folic acid and omega‐3 polyunsaturated fatty acids. Combined antioxidants were labelled as Octatron® (Youssef 2015), multiple micronutrients (Agrawal 2012; Deeba 2015; Ozkaya 2011; Panti Abubakar 2015), Fertility Blend (Westphal 2006) and Seidivid (Tunon 2017). The time that women received treatment or control in these trials ranged from 2½ menstrual cycles to six months. Four of these trials (Agrawal 2012; Deeba 2015; Panti Abubakar 2015; Westphal 2006) enrolled women undergoing ovulation induction with timed intercourse, and three (Ozkaya 2011; Tunon 2017; Youssef 2015) included women undergoing IVF/ICSI. More details of these combination antioxidants are given in the Characteristics of included studies. The remaining 56 trials gave single antioxidants. The duration of treatment in these trials ranged from 10 to 12 days to one year with a one‐year follow‐up.

The comparison 'antioxidants versus antioxidants' included only four trials (El Sharkwy 2019a; Espino 2019; Fernando 2018; Keikha 2010). El Sharkwy 2019a studied the effects of N‐acetylcysteine (NAC) versus L‐carnitine, while Espino 2019 and Fernando 2018 looked at different doses of melatonin and were also included in the placebo comparison. Keikha 2010 looked at NAC versus vitamin C. Only El Sharkwy 2019a, Espino 2019 and Fernando 2018 were used in the meta‐analysis, as Keikha 2010 did not report on live birth, clinical pregnancy or adverse events. The head‐to‐head comparisons were included in an attempt to assess whether one antioxidant was more effective than another.

In summary:

33 included trials compared antioxidants versus placebo: Al‐Alousi 2018; Badawy 2006; Battaglia 2002; Bentov 2014; Cheraghi 2016; Choi 2012; Daneshbodi 2013; El Sharkwy 2019b; Fernando 2018; Griesinger 2002; Hashemi 2017; Hefny 2018; Heidar 2019; Ismail 2014; Jahromi 2017; Jamilian 2018; Kim 2006; Mier‐Cabrera 2008; Mohammadbeigi 2012; Mokhtari 2016; Mokhtari 2019; Mostajeran 2018; Nasr 2010; Ozkaya 2011; Panti Abubakar 2015; Polak de Fried 2013; Rizk 2005; Salehpour 2009; Salehpour 2012; Siavashani 2018; Taylor 2018; Westphal 2006; Zadeh Modarres 2018;
26 trials compared antioxidants with 'no treatment' or standard treatment: Agrawal 2012; Batioglu 2012; Battaglia 1999; Behrouzi 2017; Caballero 2016; Carlomagno 2012; Cicek 2012; Deeba 2015; El Refaeey 2014; Eryilmaz 2011; Espino 2019; Gerli 2007; Ghomian 2019; Lisi 2012; Lu 2018; Maged 2015; Rasekhjahromi 2018; Rashidi 2009; Razavi 2015; Rizzo 2010; Schachter 2007; Schillaci 2012; Sen Sharma 2017; Valeri 2015; Xu 2018; Youssef 2015;
four trials compared one antioxidant with another antioxidant (head‐to‐head comparisons): El Sharkwy 2019a; Espino 2019; Fernando 2018; Keikha 2010;
18 trials compared antioxidants plus a co‐intervention with a placebo or no treatment plus a co‐intervention at the same dosage: Badawy 2006; Behrouzi 2017; Cheraghi 2016; El Refaeey 2014; El Sharkwy 2019a; El Sharkwy 2019b; Ghomian 2019; Hefny 2018; Maged 2015; Mostajeran 2018; Rasekhjahromi 2018; Rashidi 2009; Razavi 2015; Rizk 2005; Rizzo 2010; Salehpour 2012; Schachter 2007; Sen Sharma 2017. The co‐interventions used were clomiphene citrate, letrozole and metformin;
in two trials (Kim 2010;Tunon 2017), the control was unspecified, and we tried unsuccessfully to contact these authors by email and by post.

Seven trials (Cheraghi 2016; Espino 2019; Fernando 2018; Griesinger 2002; Maged 2015; Rashidi 2009; Schachter 2007) were multi‐arm and fitted into more than one of the above categories. In one trial (Cheraghi 2016) all women were prescribed the oral contraceptive pill as a pretreatment to ICSI.

Outcomes

Live birth

The primary outcome for this review was live birth. Thirteen studies reported on live birth: Agrawal 2012; Battaglia 2002; Bentov 2014; Cicek 2012; Espino 2019; Fernando 2018; Jahromi 2017; Nasr 2010; Panti Abubakar 2015; Polak de Fried 2013; Schachter 2007; Tunon 2017; Xu 2018. We sent emails and letters to authors of all other included trials to ask whether they had any data on live birth. We received live birth data from Battaglia 2002, Panti Abubakar 2015, and Polak de Fried 2013 by email. Agrawal 2012, Cicek 2012 and Schachter 2007 reported on ongoing pregnancy, which we used as a surrogate for live birth. Caballero 2016 reports on live birth but numbers per treatment and control groups are not available, despite our attempts to contact these authors.

Clinical pregnancy

Forty‐two trials reported on clinical pregnancy rates in the text of the trial reports or through direct communication with the authors: Agrawal 2012; Badawy 2006; Batioglu 2012; Battaglia 1999; Battaglia 2002; Behrouzi 2017; Bentov 2014; Caballero 2016; Carlomagno 2012; Cheraghi 2016; Choi 2012; Cicek 2012; Deeba 2015; El Refaeey 2014; El Sharkwy 2019a; El Sharkwy 2019b; Eryilmaz 2011; Espino 2019; Fernando 2018; Gerli 2007; Griesinger 2002; Ismail 2014; Jahromi 2017; Kim 2010; Lisi 2012; Lu 2018; Maged 2015; Mokhtari 2019; Mostajeran 2018; Nasr 2010; Panti Abubakar 2015; Polak de Fried 2013; Rashidi 2009; Rizk 2005; Rizzo 2010; Salehpour 2012; Schachter 2007; Sen Sharma 2017; Tunon 2017; Westphal 2006; Xu 2018 Youssef 2015. Two trials reported only biochemical pregnancy or conception (Al‐Alousi 2018; Ghomian 2019) and another six trials reported only 'pregnancy rates' (Heidar 2019; Mier‐Cabrera 2008; Mohammadbeigi 2012; Razavi 2015; Schillaci 2012; Siavashani 2018) (see data in Table 2). Rasekhjahromi 2018 provides pregnancy data, but we were unable to use it in the meta‐analysis, as the conference abstract only provided an overall pregnancy rate, with no definition of pregnancy, and with no breakdown into the different groups. Hefny 2018 reports on pregnancy but provides no data. Eleven trials did not report any pregnancy outcomes (Daneshbodi 2013; Hashemi 2017; Jamilian 2018; Keikha 2010; Kim 2006; Mokhtari 2016; Ozkaya 2011; Salehpour 2009; Taylor 2018; Valeri 2015; Zadeh Modarres 2018). We tried to contact authors of all the trials that did not report clinical pregnancy rates.

See: Summary of findings 1 Antioxidant(s) compared to placebo or no treatment/standard treatment for female subfertility; Summary of findings 2 Head‐to‐head antioxidants for female subfertility

Table 2. 'Biochemical' and 'pregnancy' data for those trials that did not specifically report 'clinical pregnancy'

Trial	Pregnancy in antioxidant group	Pregnancy in control group
Mier‐Cabrera 2008	0/16 (vitamins C + E), at follow‐up over 9 months 3/16	0/18 (placebo), at follow‐up over 9 months 2/18
Mohammadbeigi 2012	9/22 (vitamin D)	7/22 (placebo)
Razavi 2015	6/32 (selenium)	1/32 (placebo)
Al‐Alousi 2018	20/60 (omega)	15/58 (placebo)
Ghomian 2019	7/33 (NAC + CC)	5/33 (CC)
Heidar 2019	6/20 (selenium)	5/20 (placebo)
Siavashani 2018	5/20 (chromium)	4/20 (placebo)
Schillaci 2012	0/6 (myo‐inositol + 200 µg folic acid twice a day)	0/6 (400 µg folic acid once a day)

CC: clomiphene citrate; NAC: N‐acetylcysteine

Adverse events

Twenty eight trials, in both the antioxidant versus placebo/no treatment and the head‐to‐head comparisons reported on adverse events.

The following adverse events were reported:

Miscarriage: 27 trials either reported on miscarriage, or we calculated the numbers from the differences between live birth and clinical pregnancy rates (Agrawal 2012; Badawy 2006; Battaglia 1999; Battaglia 2002; Behrouzi 2017; Bentov 2014; Choi 2012; Cicek 2012; El Refaeey 2014; El Sharkwy 2019a; El Sharkwy 2019b; Eryilmaz 2011; Espino 2019; Fernando 2018; Ismail 2014; Jahromi 2017; Nasr 2010; Panti Abubakar 2015; Polak de Fried 2013; Rizzo 2010; Rizk 2005; Schachter 2007; Sen Sharma 2017; Tunon 2017; Westphal 2006; Xu 2018; Youssef 2015). We did not include the data from Rizk 2005 in the meta‐analysis for miscarriage, as no pregnancies were reported in the control group, and adding these miscarriage data would have skewed the analysis. Choi 2012 stated that miscarriage rates were similar for each group but there were no data reported in the abstract. There were six early miscarriages reported by Fernando 2018 that occurred in the biochemical to the clinical pregnancy stage, four from 120 women in the combined treatment arms and two of 40 women in the placebo group, with no miscarriages from the clinical pregnancy stage to live birth. Nasr 2010 also stated that all miscarriages occurred in the biochemical stage.
Multiple pregnancy: Nine trials reported on multiple pregnancy (Badawy 2006; Behrouzi 2017; El Refaeey 2014; Ismail 2014; Nasr 2010; Polak de Fried 2013; Rizk 2005; Salehpour 2012; Youssef 2015). We did not include Rizk 2005 in the meta‐analysis for multiple pregnancy, as no pregnancies occurred in the control group, and adding these data would have skewed the analysis.
Gastrointestinal disturbances: Three trials reported on nausea (Cicek 2012; Maged 2015; Westphal 2006). No cases of gastrointestinal disturbances were reported in treatment or control groups in Cicek 2012;
Ectopic pregnancy: Four trials reported ectopic pregnancies (Agrawal 2012; Behrouzi 2017; Fernando 2018; Jahromi 2017);
Ovarian hyperstimulation syndrome (OHSS): two trials reported on OHSS (Kim 2006; Rizk 2005).There were no cases of OHSS in treatment or control groups in Rizk 2005, and Kim 2006 did not provide data for OHSS;
Preterm birth: two trials (Fernando 2018; Nasr 2010) reported on preterm birth. The births reported by Fernando 2018 were between 34 and 37 weeks, and Nasr 2010 did not define the gestation of the preterm births.

Fernando 2018 also reported on headache, congenital abnormality (a missing kidney), low birth weight, placenta previa, pre‐eclampsia and fatigue.

We tried to contact authors of all the trials that did not report adverse events. We could not assume that there were no adverse events in trials where these were not reported.

Design

All 63 included trials were of parallel‐group design. Three trials (Fernando 2018; Griesinger 2002; Schachter 2007) were four‐armed, which used different dosages of melatonin, vitamin C versus placebo and doses of vitamin B complex versus no treatment respectively, and four trials were three‐armed (Cheraghi 2016; Espino 2019; Maged 2015; Rashidi 2009).

The sample size of the included trials ranged from 12 participants (Schillaci 2012) to 804 participants (Badawy 2006). The 12 participants from the Schillaci 2012 trial are a subgroup of poor responders using inositol, with the other population being women with PCOS (n = 17) who will be included in Showell 2018. Taylor 2018 is the second smallest trial with 21 participants. Nineteen trials included in the meta‐analysis (Agrawal 2012; Battaglia 2002; Behrouzi 2017; Bentov 2014; Cicek 2012; El Refaeey 2014; El Sharkwy 2019a; El Sharkwy 2019b; Eryilmaz 2011; Fernando 2018; Ismail 2014; Jahromi 2017; Lisi 2012; Lu 2018; Mokhtari 2019; Mostajeran 2018; Nasr 2010; Salehpour 2012; Xu 2018) reported carrying out a power calculation.

Funding

Funding sources were reported by only 27 of the 63 included trials. Three studies (Bentov 2014; Espino 2019; Taylor 2018) reported the support of Ferring Pharmaceuticals. Bentov 2014 also reported that one of the authors had a consultancy agreement with Fertility Neutraceuticals, responsible for manufacturing and distribution of the CoQ10 product, and is also on the Science Advisory Board for Ferring. Taylor 2018 was also supported by Theralogix Science, a manufacturer of vitamins and supplements. Espino 2019 was supported by FundeSalud, jointly financed by Ferring and the Government. Valeri 2015 reported funding by a pharmaceutical company, and Carlomagno 2012 included an author who was an employee of a pharmaceutical company. Schachter 2007 and Tunon 2017 were supported by the companies that were producing the supplements that were used in the trials. One trial reports self‐funding (Agrawal 2012), and 17 reported gaining funding from their institutions (Behrouzi 2017; Carlomagno 2012; Cheraghi 2016; Fernando 2018; Ghomian 2019; Hashemi 2017; Heidar 2019; Jahromi 2017; Jamilian 2018; Lu 2018; Mier‐Cabrera 2008; Razavi 2015; Salehpour 2009; Siavashani 2018; Westphal 2006; Xu 2018; Zadeh Modarres 2018) Two trials (Mokhtari 2019; Mostajeran 2018) reported that they had no financial support. See details in Characteristics of included studies.

Excluded studies

We retrieved the full text of trials that we identified as potentially eligible for inclusion (see Figure 1). We excluded 58 trials; 34 of these were because the population did not meet criteria for inclusion in this review: Aflatoonian 2014; Ardabili 2012; Baillargeon 2004; Benelli 2016; Bonakdaran 2012; Cheang 2008; Ciotta 2012; Costantino 2009; Dastorani 2018; Elgindy 2010; Fatemi 2017; Firouzabadi 2012; Genazzani 2008; Hebisha 2016; Hernández‐Yero 2012; Iuorno 2002; Jamilian 2016a; Jamilian 2016b; Kamencic 2008; Kilicdag 2005; Le Donne 2012; Li 2013; Mokhtari 2016a; Moosavifar 2010; Nestler 1999; Nestler 2001; Nordio 2012; Oner 2011; Pasha 2011; Pizzo 2014; Santanam 2003; Taheri 2015; Vargas 2011; Yoon 2010. Many of these trials recruited women with PCOS who were not attending a subfertility clinic and whose main concern was not pregnancy but rather ways to control their symptoms of PCOS. Seven were quasi‐controlled trials and therefore were not randomised: Aksoy 2010; Al‐Omari 2003; Crha 2003; Henmi 2003; Nazzaro 2011; Papaleo 2007; Tamura 2008. Ten had inappropriate treatment or control for inclusion: Asadi 2014; Elnashar 2007; Farzadi 2006; Hashim 2010; Immediata 2014; Kermack 2017; Papaleo 2008; Raffone 2010; Salem 2012; Twigt 2011. Four trials (Alborzi 2007; Aleyasin 2009; Balasch 1997; Creus 2008) were excluded as they were using pentoxifylline, a treatment that would have been included in the review prior to this update. Two trials (Elnashar 2005 and Siavashani 2016) were conference abstracts of other excluded trials (Elnashar 2007 and Jamilian 2016a respectively). Two were secondary analyses (Pal 2016; Ruder 2014). One was a duplicate study (Ghotbi 2007) of the excluded study Alborzi 2007, and we excluded Nichols 2010 after the lead investigator confirmed that this trial had been abandoned before recruitment because of lack of funding. One trial (Rezk 2004), previously excluded, was now added as a sub‐study of the included study Rizk 2005.

Ongoing trials

In the 2017 update four ongoing trials became included trials (Agrawal 2012; Bentov 2014; Mohammadbeigi 2012; Youssef 2015); two became excluded trials (Ardabili 2012 (formerly known as Pourghassem 2010), and Pasha 2011). One (Unfer 2011) became an included trial in the review Showell 2018, so that five of the original 12 trials remained ongoing (NCT01019785; IRCT201112148408N1; CTRI/2012/08/002943; NCT01782911; NCT01267604). Three further ongoing trials (Fernando 2014; NCT03023514; NCT02058212) were added in the 2017 update (Fernando 2014 became the included trial Fernando 2018 in the latest update of this review).

We include 33 ongoing trials in this review update. In addition to the seven ongoing trials from previous versions of this review, we identified a further 26 ongoing trials: ChiCTR1800019772; ChiCTR‐IPR‐15006369; EUCTR2015‐004233‐27‐IT; IRCT201009131760N9; IRCT201207156420N11; IRCT2012120311430N2; IRCT201306115942N2; IRCT20150831023831N2; IRCT201510266917N3; IRCT2016022821653N5; IRCT20160410027311N6; ISRCTN23488518; JPRN‐UMIN000016992; NCT01659788; NCT01665547; NCT01896492; NCT02239107; NCT02993588; NCT03085030; NCT03117725; NCT03306745; NCT03396380; NCT03476564; NCT04019899; PACTR201902584533870; TCTR20171109001.

Risk of bias in included studies

See Figure 2 for a summary of risks of bias in individual trials, and Figure 3 for a summary of each 'Risk of bias' item across all included trials.

Figure 2

Methodological risk of bias summary: review authors' judgements about each methodological bias item for each included study.

Figure 3

Methodological risk of bias graph: review authors' judgements about each methodological bias item presented as percentages across all included trials.

Allocation

Sequence Generation

All of the 63 included trials were randomised with a parallel design. Thirty‐seven trials described their methods of sequence generation, which typically were computer‐generated or used a random‐number table: Agrawal 2012; Batioglu 2012; Battaglia 2002; Battaglia 1999; Bentov 2014; Cicek 2012; El Refaeey 2014; El Sharkwy 2019a; El Sharkwy 2019b; Eryilmaz 2011; Espino 2019; Fernando 2018; Gerli 2007; Ghomian 2019; Hashemi 2017;Heidar 2019; Ismail 2014; Jahromi 2017; Jamilian 2018; Lisi 2012; Maged 2015; Mokhtari 2019; Mier‐Cabrera 2008; Mohammadbeigi 2012; Nasr 2010; Ozkaya 2011; Polak de Fried 2013; Rashidi 2009; Razavi 2015; Rizzo 2010; Schachter 2007; Siavashani 2018; Taylor 2018; Valeri 2015; Xu 2018; Youssef 2015; Zadeh Modarres 2018. One trial (Panti Abubakar 2015) used a coin toss. Twenty‐six trials simply reported the trial as randomised with no description of method, and were judged to be at unclear risk of bias for sequence generation: Al‐Alousi 2018; Badawy 2006; Behrouzi 2017; Caballero 2016; Carlomagno 2012; Cheraghi 2016; Choi 2012; Daneshbodi 2013; Deeba 2015; Griesinger 2002; Hefny 2018; Keikha 2010; Kim 2006; Kim 2010; Lu 2018; Mier‐Cabrera 2008; Mokhtari 2016; Mostajeran 2018; Rasekhjahromi 2018; Rizk 2005; Salehpour 2009; Salehpour 2012; Schillaci 2012; Sen Sharma 2017; Tunon 2017; Westphal 2006. There were no trials that we judged as high risk for this domain. We conducted a sensitivity analysis on the exclusion of trials that we considered to be at high risk in any of the 'Risk of bias' domains.

Allocation concealment

We judged 24 trials to be at low risk for allocation concealment: Agrawal 2012; Badawy 2006; Battaglia 1999; Battaglia 2002; Bentov 2014; El Sharkwy 2019a; El Sharkwy 2019b; El Refaeey 2014; Fernando 2018; Griesinger 2002; Ismail 2014; Jahromi 2017; Jamilian 2018; Lisi 2012; Maged 2015; Mokhtari 2019; Razavi 2015; Rizk 2005; Schachter 2007; Siavashani 2018; Taylor 2018; Xu 2018; Youssef 2015; Zadeh Modarres 2018. One trial (Eryilmaz 2011) replied through email correspondence that no allocation concealment was used and so was deemed to be at high risk. The remainder either did not describe any methods of allocation concealment or the description was not clear. We tried unsuccessfully to contact these authors about allocation concealment techniques.

Blinding

We considered that the blinding status of participants could influence findings for the outcomes of live birth, pregnancy and adverse effects, as antioxidants are easily available and it would be possible for participants to self‐medicate. If the participants were not blinded or the trial was not placebo‐controlled, or both, we therefore considered the trial to be at high risk. Forty‐one of the 63 included trials described some form of blinding of participants or investigators, or both.

One trial (Taylor 2018) was quadruple‐blinded. Six were triple‐blinded, with participants, clinicians/investigators and outcome assessors blinded: Agrawal 2012; Badawy 2006; Battaglia 2002; El Sharkwy 2019b; Fernando 2018; Mier‐Cabrera 2008. Ten were double‐blinded with blinding of participants and clinicians: Al‐Alousi 2018; Bentov 2014; El Sharkwy 2019a; Griesinger 2002; Hashemi 2017; Jamilian 2018; Razavi 2015; Rizk 2005; Salehpour 2009; Westphal 2006. Eighteen stated that they were double‐blinded but did not declare who was blinded: Cheraghi 2016; Carlomagno 2012; Daneshbodi 2013; Gerli 2007; Ghomian 2019; Hefny 2018; Heidar 2019; Ismail 2014; Jahromi 2017; Keikha 2010; Mokhtari 2016; Mokhtari 2019; Mostajeran 2018; Polak de Fried 2013; Siavashani 2018; Tunon 2017; Valeri 2015; Zadeh Modarres 2018. Tunon 2017 states that the trial is double‐blinded but there is no description of what type of control is used in the study, so we considered it to be at high risk for this domain, as it might be a 'no‐treatment trial'. Ghomian 2019 and Valeri 2015 were also no‐treatment trials but reported being double‐blinded.

Six trials were single‐blinded: the participants were blinded in Panti Abubakar 2015 and Salehpour 2012; the embryologists were blinded in Espino 2019; and the outcome assessors were blinded in Lisi 2012, El Refaeey 2014 and Mohammadbeigi 2012.

The remaining 22 trials did not report any blinding; however, 13 of these used 'no treatment' as the control, making blinding for these trials problematic. We therefore considered these trials to be at high risk for this domain: Battaglia 1999; Batioglu 2012; Behrouzi 2017; Caballero 2016; Carlomagno 2012; Cicek 2012; Eryilmaz 2011; Lu 2018; Maged 2015; Rasekhjahromi 2018; Sen Sharma 2017; Xu 2018; Youssef 2015. Only Xu 2018 stated that it was an open study. Nine trials did not report on blinding: Choi 2012; Deeba 2015; Kim 2006; Kim 2010; Ozkaya 2011; Rashidi 2009; Rizzo 2010; Schachter 2007; Schillaci 2012.

We rated Espino 2019 at high risk of bias for the blinding domain. In the treatment‐versus‐control arm of the trial, the control is 'no treatment' so blinding was not possible, although the paper states that; "Embryo quality was graded by blinded embryologists" (page 2/11). However, we consider it to be at low risk in the head‐to‐head comparison; "Melatonin treatments comprised immediate‐release melatonin formula (Guinama, Valencia, Spain) that was encapsulated in identical two‐piece gelatine capsules (containing 3 mg or 6 mg melatonin) and dispensed in identical 50‐capsule containers". We were unable to use two different judgements in the 'Risk of bias' table, so we used the 'high risk' judgement in the 'Risk of bias' table and covered the 'low risk' judgement for the head‐to‐head comparison in the text here, in the Effects of interventions section and in the footnotes of the appropriate forest plot, in order to ensure that the study was removed in the sensitivity analysis.

Incomplete outcome data

Fourteen trials included in the meta‐analysis had no losses to follow‐up: Badawy 2006; Batioglu 2012; Battaglia 1999; Espino 2019; Lisi 2012; Maged 2015; Nasr 2010; Polak de Fried 2013; Rashidi 2009; Rizk 2005; Rizzo 2010; Schachter 2007; Sen Sharma 2017; Westphal 2006. Four trials reported losses but used intention‐to‐treat (ITT) analysis: Agrawal 2012; Fernando 2018; Ismail 2014; Youssef 2015. Fourteen trials had losses and described from which groups they were lost, but did not use ITT in the reporting of trials; however, we used ITT for them in the meta‐analysis: Battaglia 2002; Behrouzi 2017; El Refaeey 2014; El Sharkwy 2019a; El Sharkwy 2019b; Jahromi 2017; Lu 2018; Mier‐Cabrera 2008; Mokhtari 2019; Mostajeran 2018; Panti Abubakar 2015; Salehpour 2012; Tunon 2017; Xu 2018.

Bentov 2014 had explained loss to follow‐up but reported data as percentages, so it is unclear if ITT was used. This trial was also terminated before finishing enrolment, and we therefore rated it at high risk for this domain. Cheraghi 2016 explained the losses but was considered at high risk for attrition, as the losses were over 25% of the randomised women.

Three trials (Cicek 2012, Eryilmaz 2011; Griesinger 2002;) had losses to follow‐up with no explanation of which groups were affected, but we took data from these trials as totals were given after dropouts, and we assumed that the groups were equal on allocation.

The data from the following 26/63 included trials could not be added to the meta‐analysis. Ten of these trials had different reported outcomes from those of the review, Jamilian 2018 and Zadeh Modarres 2018 had no attrition, and two trials (Ghomian 2019; Siavashani 2018) both used intention‐to‐treat. Four trials (Al‐Alousi 2018; Hashemi 2017; Heidar 2019; Salehpour 2009) explained their losses to follow‐up. We judged Schillaci 2012 to be at high risk for the attrition domain as the paper provided preliminary results only, the numbers given in the text are different from numbers in the baseline characteristics table and there appear to be two dropouts from the intervention group which go unexplained. Mokhtari 2016 was judged as unclear as no mention of attrition was given.

Gerli 2007 had more than 30% dropouts from the treatment group, and data were unavailable for the 15 other trials: Caballero 2016; Carlomagno 2012; Choi 2012; Daneshbodi 2013; Deeba 2015; Hefny 2018; Keikha 2010; Kim 2006; Kim 2010; Mohammadbeigi 2012; Ozkaya 2011; Rasekhjahromi 2018; Razavi 2015; Taylor 2018; Valeri 2015. We tried to contact authors when the data were unavailable.

Selective reporting

We considered a trial to be at low risk for selective reporting if a trial registration number or protocol was provided and the clinical outcomes of live birth or clinical pregnancy or both were reported. Thirteen trials (Bentov 2014; Cheraghi 2016; El Refaeey 2014; El Sharkwy 2019a; Fernando 2018; Hashemi 2017; Heidar 2019; Jahromi 2017; Lisi 2012; Mokhtari 2019; Siavashani 2018; Xu 2018; Youssef 2015) were classified as low risk.

Four trials were considered to be at high risk of bias for various reasons. These include higher clinical pregnancy numbers reported than biochemical pregnancy numbers (Batioglu 2012); in Gerli 2007 only half the population declared wanting to become pregnant, with miscarriages reported but with no information on which groups they occurred in. Mier‐Cabrera 2008 stated that they would collect live births but these was not reported and there was no trial registration number. Westphal 2006 combined the miscarriage and side‐effect data from their trial with an extra three months of data from a non‐randomised source.

We assigned unclear risks of bias to the remaining 46 trials (see Characteristics of included studies). The trials were classified as 'unclear' firstly if they had a trial registration number but no report of clinical outcomes in a trial where you would expect these to be reported, and secondly if no trial registration number was provided but clinical outcomes such as live birth or clinical pregnancy were reported.

Failure to report live birth in subfertility trials is common, and is a major source of bias (Clarke 2010); it should be the default primary outcome in fertility trials. Only 13 trials reported live birth: Agrawal 2012; Battaglia 2002; Bentov 2014; Caballero 2016; Cicek 2012; Caballero 2016; Fernando 2018; Jahromi 2017; Nasr 2010; Panti Abubakar 2015; Polak de Fried 2013; Tunon 2017; Xu 2018; this represents an increase of only six trials from the 2017 version of this review (three of the original trials (Aleyasin 2009a; Ciotta 2011; Unfer 2011) were removed as they were now included in Showell 2018).

Two trials (Agrawal 2012; Schachter 2007) reported ongoing pregnancy and Espino 2019 reported full‐term pregnancies, which we took to be live birth in the analysis. Mier‐Cabrera 2008 stated that they would report live birth, but reported only pregnancy. Caballero 2016 was not included in the meta‐analysis as the numbers per group were not available. Tunon 2017 did not provide clarification of the control used, so for the purposes of this meta‐analysis we consider the trial to be a no‐treatment control. Adverse events were not well reported in most studies. We attempted to contact all authors about live birth and adverse outcomes.

A funnel plot for clinical pregnancy (Figure 4) was symmetrical, except for an absence of studies in the lower left of the pyramid. This suggests a small‐study effect, indicating the potential for publication bias whereby small unpublished studies with negative results were not represented. Estimates of the intervention effect tend to be more beneficial in smaller studies and thus introduce the potential for selective reporting and publication bias.

Figure 4

Funnel plot of comparison: 1 Antioxidant(s) versus placebo or no treatment/standard treatment, outcome: 1.5 Clinical pregnancy; antioxidants vs placebo or no treatment/standard treatment (natural conceptions and undergoing fertility treatments).

Other potential sources of bias

We rated Bentov 2014 at high risk in this domain, for women receiving varying adjunctive treatments and early termination of the study, respectively. See details in Characteristics of included studies.

Reasons for studies with data included within the review but not in the analysis

Gerli 2007 (see Table 1) was not incorporated into the analysis, as only half the women randomly assigned reported a desire to become pregnant. Ninety‐two women were randomly assigned, 45 to the treatment group and 47 to the control group. Twenty‐three from the treatment group and 19 from the control group wished to conceive; four from the treatment group and one from the control group became pregnant. This trial also had more than 30% dropouts from the treatment group.

Effects of interventions

1. Antioxidant supplement versus placebo, no treatment/standard treatment

Primary outcome: Live birth

1.1 Live birth; antioxidants versus placebo or no treatment/standard treatment

See Analysis 1.1.

Due to the very low‐quality evidence we are uncertain whether antioxidants improve live birth rate compared with placebo or no treatment/standard treatment (odds ratio (OR) 1.81, 95% confidence interval (CI) 1.36 to 2.43; P < 0.001, I² = 29%; 13 RCTs, 1227 women; very low‐quality evidence; Figure 5). This suggests that among subfertile women with an expected live birth rate of 19%, the rate among women using antioxidants would be between 24% and 36% (summary of findings Table 1).

Figure 5

Forest plot of comparison: 1 Antioxidant(s) versus placebo or no treatment/standard treatment, outcome: 1.1 Live birth; antioxidants vs placebo or no treatment/standard treatment (natural conceptions and undergoing fertility treatments).

In the 13 trials that reported live birth (Agrawal 2012; Battaglia 2002; Bentov 2014; Cicek 2012; Espino 2019; Fernando 2018; Jahromi 2017; Nasr 2010; Panti Abubakar 2015; Polak de Fried 2013; Schachter 2007; Tunon 2017; Xu 2018), the OR for live birth was 1.81 and for clinical pregnancy was 1.80. When we pooled all 35 studies that reported clinical pregnancy, the OR for clinical pregnancy was lower, at 1.65. This suggests that the clinical pregnancy rate in the 13 trials that reported live birth may have been a small overestimation of the effect of the antioxidants, and hence that the live birth rate in these trials may also be a small overestimate (summary of findings Table 1).

The test for subgroup differences showed no evidence of a difference between the placebo and no‐treatment subgroups (Chi² = 0.05, df = 1, P = 0.83, I² = 0%).

1.2 Live birth; type of antioxidant

See Analysis 1.2.

We considered each type of antioxidant separately. Only four comparisons contained more than one trial, including Schachter 2007, a four arm trial.

1.2.1Nasr 2010; compared N‐acetylcysteine with placebo (OR 3.00, 95% CI 1.05 to 8.60; P = 0.04, 60 women).

1.2.2Battaglia 2002; compared L‐arginine with placebo (OR 0.43, 95% CI 0.09 to 2.09; P = 0.30, 37 women).

1.2.3Bentov 2014; compared CoQ10 with placebo and Xu 2018 compared CoQ10 with no treatment (OR 1.50, 95% CI 0.78 to 2.88; P= 0.23, I² = 0%; 2 RCTs, 225 women). Antioxidants were not associated with an increased live birth rate compared with placebo or with no treatment in women taking CoQ10.

1.2.4Polak de Fried 2013; compared Vitamin D with placebo (OR 0.79, 95% CI 0.21 to 3.02; P = 0.73; 52 women).

1.2.5Schachter 2007, a four‐armed trial with two arms comparing a Vitamin B complex with no treatment and Vitamin B complex plus metformin versus metformin (also considered to be 'no treatment'), showing no association with increased live birth rate compared to no treatment (OR 2.07, 95% CI 0.93 to 4.57; P = 0.07, I² = 0%; 102 women).

1.2.6Agrawal 2012 and Tunon 2017 compared combined antioxidants with no treatment, and Panti Abubakar 2015 compared combined antioxidants with placebo. Combined antioxidants were associated with an increased live birth rate compared with placebo or no treatment (OR 2.59, 95% CI 1.52 to 4.40; P < 0.001, I² = 78%; 3 RCTs, 378 women).

1.2.7Cicek 2012 compared Vitamin E to no treatment (OR 1.43, 95% CI 0.50 to 4.10; P = 0.51; 103 women).

1.2.8Fernando 2018 and Jahromi 2017 compared melatonin with placebo and Espino 2019 compared it with no treatment (OR 1.60, 95% CI 0.68 to 3.76; P = 0.28, I² = 0%; 3 RCTs, 270 women). Antioxidants were not associated with an increased live birth rate compared with placebo or no treatment in women taking melatonin.

The test for subgroup differences showed no evidence of a difference between the subgroups of antioxidant type (Chi² = 7.96, df = 7, P = 0.34, I² = 12%).

1.3 Live birth rate; indications for subfertility

See Analysis 1.3.

1.3.1Polycystic ovary syndrome

Three trials reported on women with PCOS: Panti Abubakar 2015; Nasr 2010; and Schachter 2007 (a four‐armed trial, which contributed to two comparisons in this analysis). Antioxidants were associated with an increased live birth rate compared with placebo or no treatment in women with PCOS (OR 3.34, 95% CI 1.90 to 5.86; P < 0.001, I² = 28%; 3 RCTs, 362 women). Each trial included different antioxidants: N‐acetylcysteine, combined antioxidants and Vitamin B complex.

1.3.2 Tubal subfertility

One trial (Battaglia 2002) enrolled women with tubal subfertility undergoing IVF (OR 0.43, 95% CI 0.09 to 2.09; P = 0.30; 37 women).

1.3.3 Varying indications

Three trials (Agrawal 2012; Fernando 2018; Tunon 2017) enrolled women with various causes of subfertility (OR 1.70, 95% CI 1.02 to 2.83; P = 0.04, I² = 50%; 3 RCTs, 338 women). Antioxidants were associated with an increased live birth rate compared with placebo or no treatment in women with varying indications for subfertility.

1.3.4 Unexplained subfertility

Two trials (Cicek 2012; Espino 2019) enrolled women with unexplained subfertility (OR 1.50, 95% CI 0.60 to 3.72: P = 0.38, I² = 0%: 2 RCTs, 133 women).

1.3.5 Poor ovarian reserve

Two trials (Jahromi 2017; Xu 2018) enrolled women with poor ovarian reserve, but Jahromi 2017 reported no live births in either the treatment or control groups (OR 1.75, 95% CI 0.83 to 3.67; P = 0.14; 2 RCTs, 266 women).

1.4 Live birth; IVF/ICSI

See Analysis 1.4.

Nine trials (Battaglia 2002; Bentov 2014; Espino 2019; Fernando 2018; Jahromi 2017; Polak de Fried 2013; Schachter 2007; Tunon 2017; Xu 2018) compared antioxidants with placebo or no treatment in women having IVF/ICSI treatment and reporting live birth. Antioxidants were not associated with an increased live birth rate compared with placebo or no treatment in women undergoing IVF/ICSI (OR 1.36, 95% CI 0.96 to 1.93; P = 0.08, I² = 0%; 9 RCTs, 806 women). Jahromi 2017 reported no live births in either the treatment or control groups.

Secondary outcome: Clinical pregnancy

Only 35 of the 63 included trials presented or provided data that could be used in this meta‐analysis. We could not use the data for the remaining 28 trials in the meta‐analysis, as they provided either only 'pregnancy' or biochemical pregnancy data (see Table 2), only bio‐markers or embryo/oocyte numbers, or insufficient information in the reports, which were mainly conference abstracts. We tried to contact these authors to obtain the clinical pregnancy data; some responded saying that they did not have the data, while others did not respond at all.

1.5 Clinical pregnancy; antioxidants versus placebo or no treatment/standard treatment

See Analysis 1.5.

Antioxidants may improve the clinical pregnancy rate compared with placebo or no treatment (OR 1.65, 95% CI 1.43 to 1.89; P < 0.001, I² = 63%; 35 RCTs, 5165 women; low‐quality evidence; Figure 6). This suggests that among subfertile women with an expected clinical pregnancy rate of about 19%, the rate among women using antioxidants would be between 25% and 30% (summary of findings Table 1). Heterogeneity was moderately high.

Figure 6

Forest plot of comparison: 1 Antioxidant(s) versus placebo or no treatment/standard treatment, outcome: 1.5 Clinical pregnancy; antioxidants vs placebo or no treatment/standard treatment (natural conceptions and undergoing fertility treatments).

The test for subgroup differences showed no evidence of a difference between the placebo and no‐treatment subgroups (Chi² = 0.31, df = 1, P = 0.58, I² = 0%).

Sensitivity analyses

Using a random‐effects model did not change the direction of the results, and the I² remained at 63%.

1. We conducted a sensitivity analysis, excluding trials with a high risk of bias in any domain.

Sixteen trials (Batioglu 2012; Behrouzi 2017; Bentov 2014; Cheraghi 2016; Cicek 2012; El Refaeey 2014; Eryilmaz 2011; Espino 2019; Lisi 2012; Lu 2018; Maged 2015; Sen Sharma 2017; Tunon 2017; Xu 2018; Youssef 2015; Westphal 2006) had a rating of high risk in any one or more of the 'Risk of bias' domains (mostly in the domain of blinding in the no‐treatment trials) (see Characteristics of included studies). When these trials were removed in a sensitivity analysis, there remained an association between antioxidants and an increased clinical pregnancy rate when compared to placebo (OR 1.74, 95% CI 1.45 to 2.08; P < 0.001, I² = 78%; 19 RCTs, 3449 women). Heterogeneity was high.

2. We conducted a sensitivity analysis, excluding studies that used a fertility drug (metformin, clomiphene or letrozole) as a control plus a placebo or no treatment (these agents were in both the intervention and control arms, with an antioxidant in addition in the intervention arm). When these 13 trials were removed from the analysis (Badawy 2006; Behrouzi 2017; Cheraghi 2016; El Refaeey 2014; El Sharkwy 2019b; Espino 2019; Ismail 2014; Maged 2015; Mostajeran 2018; Rizk 2005; Salehpour 2012; Schachter 2007; Sen Sharma 2017) there remained an association between antioxidants and an increased clinical pregnancy rate compared no treatment (OR 1.40, 95% CI 1.17 to 1.67; P < 0.001, I² = 31%; 24 RCTs, 2968 women). Two trials (Cheraghi 2016; Schachter 2007) were multi‐armed, but only those arms with a fertility drug plus placebo/no treatment were removed in this analysis.

1.6 Clinical pregnancy; type of antioxidant

See Analysis 1.6.

We considered each type of antioxidant separately.

1.6.1 N‐acetylcysteine was associated with an increased clinical pregnancy rate when compared with placebo, no treatment or standard treatment (OR 1.36, 95% CI 1.05 to 1.77; P = 0.02, , I² = 71%; 8 RCTs, 1590 women). Heterogeneity was very high, perhaps as a result of the high risk of bias for blinding in Behrouzi 2017; Cheraghi 2016; Maged 2015, and the unclear risk of bias for sequence generation in Badawy 2006; Behrouzi 2017; Cheraghi 2016 Mostajeran 2018; Rizk 2005; Salehpour 2012, or the additional treatment of laparoscopic drilling that women received in Nasr 2010.

1.6.2 Combined antioxidants (similar antioxidants were combined in each trial) were associated with an increased clinical pregnancy rate when compared to placebo or no treatment (OR 1.90, 95% CI 1.33 to 2.70; P < 0.001, I² = 70%; 5 RCTs, 689 women). Heterogeneity was high, with two of the trials enrolling small numbers of women.

1.6.3 Melatonin was associated with an increased clinical pregnancy rate when compared with placebo, no treatment or standard treatment (OR 1.66, 95% CI 1.12 to 2.47; P = 0.01, I² = 0%; 7 RCTs, 678 women).

1.6.4 There was no clear evidence of a difference in clinical pregnancy rates between Vitamin E and no treatment (OR 1.43, 95% CI 0.50 to 4.10; P = 0.51; 103 women).

1.6.5 There was no clear evidence of a difference in clinical pregnancy rates between ascorbic acid and placebo (OR 0.91, 95% CI 0.66 to 1.25; P = 0.55, I² = 46%; 2 RCTs, 899 women).

1.6.6 There was no clear evidence of a difference in clinical pregnancy rates between L‐arginine and placebo or no treatment (OR 1.05, 95% CI 0.32 to 3.46; P = 0.94, I² = 67%; 2 RCTs, 71 women).

1.6.7 There was no clear evidence of a difference in clinical pregnancy rates between myo‐inositol plus folic acid and placebo or no treatment (OR 1.24, 95% CI 0.50 to 3.06: P = 0.64; 94 women).

1.6.8 CoQ10 was associated with an increased clinical pregnancy rate when compared to placebo or no treatment (OR 2.49, 95% CI 1.50 to 4.13; P < 0.001, I² = 47%; 4 RCTs, 397 women).

1.6.9 L‐carnitine was associated with an increased clinical pregnancy rate when compared to placebo (OR 11.14, 95% CI 5.70 to 21.81; P < 0.001, I² = 85%; 2 RCTs, 450 women). The high heterogeneity may be due to the very high numbers of clinical pregnancy in the treatment group (42/85) when compared to the low numbers in the control group (1/85) in Ismail 2014.

1.6.10 There was no clear evidence of a difference in clinical pregnancy rates between vitamin D and placebo (OR 0.83, 95% CI 0.25 to 2.76; P = 0.76; 2 RCTs, 92 women). Rashidi 2009 reported no clinical pregnancies in either treatment or control group.

1.6.11 There was no clear evidence of a difference in clinical pregnancy rates between vitamin B complex in the two arms of Schachter 2007 and placebo or no treatment (OR 1.94, 95% CI 0.82 to 4.58; P = 0.13, I² = 0%; 1 RCT, 102 women).

The test for subgroup differences showed that there were differences between the type of antioxidant subgroups (Chi² = 51.55, df = 10, P < 0.001, I² = 80.6%).

1.7 Clinical pregnancy rate; indications for subfertility

See Analysis 1.7.

1.7.1 Polycystic ovary syndrome

Antioxidants were associated with an increased clinical pregnancy rate when compared with placebo, no treatment or standard treatment in women with PCOS (OR 4.24, 95% CI 3.23 to 5.56; P < 0.001, I² = 51%; 15 RCTs, 1908 women).

1.7.2 Unexplained subfertility

There was no clear evidence of a difference in clinical pregnancy rates when antioxidants were compared with placebo, no treatment or standard treatment in women with unexplained subfertility (OR 0.84, 95% CI 0.61 to 1.16; P = 0.29, I² = 0%; 4 RCTs, 997 women).

1.7.3 Tubal subfertility

There was no clear evidence of a difference in clinical pregnancy rates when antioxidants were compared with placebo, no treatment or standard treatment in women with tubal subfertility (OR 1.05, 95% CI 0.32 to 3.46; P = 0.94, I² = 67%; 2 RCTs, 71 women).

1.7.4 Varying indications

There was no clear evidence of a difference in clinical pregnancy rates when antioxidants were compared with placebo, no treatment or standard treatment in women with varying indications (OR 1.14, 95% CI 0.85 to 1.52; P = 0.38, I² = 54%; 6 RCTs, 1135 women).

1.7.5 Poor responders

There was no clear evidence of a difference in clinical pregnancy rates when antioxidants were compared with placebo, no treatment or standard treatment in women who were poor responders (OR 1.88, 95% CI 0.64 to 5.47; P = 0.25; 1 RCT, 65 women).

1.7.6 Poor ovarian reserve

There was no clear evidence of a difference in clinical pregnancy rates when antioxidants were compared with placebo, no treatment or standard treatment in women with poor ovarian reserve (OR 1.70, 95% CI 0.86 to 3.37; P = 0.13, I² = 0%; 2 RCTs, 266 women).

1.7.7 Endometriosis

There was no clear evidence of a difference in clinical pregnancy rates when antioxidants were compared with placebo, no treatment or standard treatment in women with endometriosis (OR 1.19, 95% CI 0.71 to 1.98; P = 0.51; 1 RCT, 280 women).

1.8 Clinical pregnancy rate; IVF/ICSI

See Analysis 1.8.

There was no clear evidence of a difference in clinical pregnancy rates when antioxidants were compared with placebo, no treatment or standard treatment in women undergoing IVF/ICSI (OR 1.15, 95% CI 0.95 to 1.40; P = 0.15, I² = 0%; 18 RCTs, 2341 women).

Secondary outcome: Adverse events

1.9 Adverse events

See Analysis 1.9; Figure 7

Figure 7

Forest plot of comparison: 1 Antioxidant(s) versus placebo or no treatment/standard treatment, outcome: 1.9 Adverse events.

We subgrouped adverse event data according to the types of events that occurred, as reported by the trials. These included miscarriage, multiple pregnancy, gastrointestinal disturbances, ectopic pregnancy and headache, congenital (missing kidney), low birth weight, preterm birth, placenta previa, pre‐eclampsia, fatigue and OHSS. There was no evidence to suggest an association between antioxidants and adverse events, but data were limited, with 24 trials reporting on miscarriage, nine trials reporting on multiple pregnancy, three reporting on gastrointestinal upsets, four reporting ectopic pregnancy, two reporting headache and preterm birth, and one reporting on congenital abnormality (missing kidney), low birth weight, placenta previa, pre‐eclampsia, fatigue and OHSS.

1.9.1 Miscarriage

There was no difference in miscarriage rates when antioxidants were compared with placebo or no treatment (OR 1.13, 95% CI 0.82 to 1.55; P = 0.46, I² = 0%; 24 RCTs, 3229 women; low‐quality evidence). This means that given the rate of 5% miscarriages in the control population, the use of antioxidants would be expected to result in a miscarriage rate of between 4% and 7% (summary of findings Table 1). Most of the trials in this analysis were small, although one trial (Badawy 2006) enrolled 804 women. There were no events in three of the studies (Battaglia 2002; Espino 2019; Fernando 2018).

1.9.2 Multiple pregnancy

There was no difference in multiple pregnancy rates when antioxidants were compared with placebo or no treatment (OR 1.00, 95% CI 0.63 to 1.56; P = 0.99, I² = 0%; 9 RCTs, 1886 women; low‐quality evidence; Figure 7). This means that if the multiple pregnancy rate in the control population is 4%, use of antioxidants instead would be expected to result in a multiple pregnancy rate between 3% and 7% (summary of findings Table 1). There were no events reported in one of the studies (Nasr 2010).

1.9.3 Gastrointestinal disturbances

Three trials reported on gastrointestinal disturbances (Cicek 2012; Maged 2015; Westphal 2006). There was no difference in gastrointestinal disturbances when antioxidants were compared with placebo or no treatment (OR 1.55, 95% CI 0.47 to 5.10; P = 0.47, I² = 0%; 3 RCTs, 343 women; low‐quality evidence; Figure 7). This means that with a rate of 2% gastrointestinal disturbances in the control population, use of antioxidants would be expected to result in a rate of between 1% and 11% (summary of findings Table 1).

1.9.4 Ectopic pregnancy

Four trials (Agrawal 2012; Behrouzi 2017; Fernando 2018; Jahromi 2017) reported on ectopic pregnancy. There was no difference between the groups (OR 1.40, 95% CI 0.27 to 7.20; P = 0.69, I² = 0%; 4 RCTs, 404 women, low‐quality evidence). This means that with a rate of 0.6% ectopic pregnancy in the control population, use of antioxidants would be expected to result in an ectopic pregnancy rate between 0.2% and 4% (summary of findings Table 1).

1.9.5 Headache

Two trials (Fernando 2018; Ismail 2014) reported on headache. There was no difference between the groups (OR 0.89, 95% CI 0.45 to 1.75; P = 0.73, I² = 0%; 2 RCTs, 330 women; moderate‐quality evidence).This means that with a rate of 17% headache in the control population, use of antioxidants would be expected to result in a headache rate between 8% and 26%.

1.9.6 Congenital abnormality (missing kidney)

Fernando 2018 reported on a congenital abnormality. There was no clear evidence of a difference between the groups (OR 1.02, 95% CI 0.04 to 25.46; P = 0.99; 160 women; very low‐quality evidence).

1.9.7 Low birth weight < 2.500 g

Fernando 2018 reported on a low birth weight. There was no clear evidence of a difference between the groups (OR 0.11, 95% CI 0.00 to 2.74; P = 0.18; 160 women; very low‐quality evidence).

1.9.8 Preterm birth

Two trials (Fernando 2018; Nasr 2010) reported on a preterm birth. There was no clear evidence of a difference between the groups (OR 1.31, 95% CI 0.17 to 9.93; P = 0.80, I² = 0%; 2 RCTs, 220 women; moderate‐quality evidence).This means that with a rate of 1% preterm birth in the control population, use of antioxidants would be expected to result in a preterm birth rate between 0.2% and 13%.

1.9.9 Placenta previa

Fernando 2018 reported on placenta previa. There was no clear evidence of a difference between the groups (OR 1.02, 95% CI 0.04 to 25.46; P = 0.99; 160 women; very low‐quality evidence).

1.9.10 Pre‐eclampsia

Fernando 2018 reported on pre‐eclampsia. There was no clear evidence of a difference between the groups (OR 1.71, 95% CI 0.08 to 36.35; P = 0.73; 160 women; very low‐quality evidence).

1.9.11 Fatigue

Fernando 2018 reported on fatigue. There was no clear evidence of a difference between the groups (OR 1.86, 95% CI 0.75 to 4.62; P = 0.18; 160 women; very low‐quality evidence).

1.9.12 Ovarian hyperstimulation syndrome

Rizk 2005 reported on OHSS but there were no events in either the antioxidant or placebo group.

2. Head‐to‐head antioxidants

Three trials (El Sharkwy 2019a; Espino 2019; Fernando 2018) were included in the head‐to‐head comparison. El Sharkwy 2019a enrolled women with PCOS undergoing ovulation induction, Espino 2019 included women with unexplained subfertility undergoing IVF, and Fernando 2018 enrolled women with varying indications who were also undergoing IVF.

Primary outcome: Live birth

2.1 Live birth; type of antioxidant

See Analysis 2.1; Figure 8

Figure 8

Forest plot of comparison: 2 Head‐to‐head antioxidants, outcome: 2.1 Live birth; type of antioxidant (natural conceptions and undergoing fertility treatments).

We considered each type of antioxidant separately.

2.1.1 Two trials (Espino 2019; Fernando 2018) reported on live birth. There was no difference between the lower and higher dose of melatonin (OR 0.94, 95% CI 0.41 to 2.15; P = 0.89, I² = 0%; 2 RCTs, 140 women; low‐quality evidence). This suggests that among subfertile women with an expected live birth rate of 24%, the rate among women using a lower dose of melatonin compared to a higher dose would be between 12% and 40% (summary of findings Table 2).

Sensitivity analysis

We were unable to perform a sensitivity analysis in any of the head‐to‐head analyses as there were no trials with a high risk of bias in any domain. Espino 2019 appears to have a high risk of bias for blinding in the 'Risk of bias' table (see Characteristics of included studies), but this is only the case for the antioxidant versus placebo/no treatment comparison. In the head‐to‐head comparison Espino 2019 explains the blinding clearly."Melatonin treatments comprised immediate‐release melatonin formula (Guinama, Valencia, Spain) that was encapsulated in identical two‐piece gelatine capsules (containing 3 mg or 6 mg melatonin) and dispensed in identical 50‐capsule containers". However in the treatment versus control comparison, the control is 'no treatment', so blinding not possible, although; "Embryo quality was graded by blinded embryologists"

2.2 Live birth; indications for subfertility

See Analysis 2.2

2.2.1 Unexplained subfertility

Espino 2019 enrolled women with unexplained subfertility. There was no clear evidence of a difference between the groups (OR 1.00, 95% CI 0.15 to 6.77; P = 1.00; 20 women).

2.2.2 Varying indications

Fernando 2018 enrolled women with varying indications. There was no clear evidence of a difference between the groups (OR 0.93, 95% CI 0.37 to 2.32; P = 0.88; 120 women).

2.3 Live Birth; IVF/ICSI

See Analysis 2.3

Two trials (Espino 2019; Fernando 2018) enrolled women who were undergoing IVF.There was no clear evidence of a difference in live birth rates in women undergoing IVF when lower versus higher doses of melatonin were used (OR 0.94, 95% CI 0.41 to 2.15; P = 0.89, I² = 0%; 2 RCTs, 140 women).

Secondary outcome: Clinical pregnancy

Three trials (El Sharkwy 2019a; Espino 2019; Fernando 2018) reported on clinical pregnancy.

2.4 Clinical pregnancy; type of antioxidant

See Analysis 2.4; Figure 9.

Figure 9

Forest plot of comparison: 2 Head‐to‐head antioxidants, outcome: 2.4 Clinical pregnancy; type of antioxidant (natural conceptions and undergoing fertility treatments).

2.4.1El Sharkwy 2019a reported on N‐acetylcysteine versus L‐carnitine. There was no clear evidence of a difference between these two antioxidants (OR 0.81, 95% CI 0.33 to 2.00; P = 0.65; 164 women).

2.4.2Espino 2019 and Fernando 2018 reported on different doses of melatonin. There was no difference in rates of clinical pregnancy between a lower and higher dose of melatonin (OR 0.94, 95% CI 0.41 to 2.15; P = 0.89, I² = 0%; 140 women; low‐quality evidence). This suggests that among subfertile women with an expected clinical pregnancy rate of 24%, the rate among women using a lower dose of melatonin compared to a higher dose would be between 12% and 40% (summary of findings Table 2).

2.5 Clinical pregnancy; indications for subfertility

See Analysis 2.5

2.5.1 Polycystic ovary syndrome

El Sharkwy 2019a enrolled women with PCOS. There was no clear evidence of a difference between the groups (OR 0.81, 95% CI 0.33 to 2.00; P = 0.65; 164 women).

2.5.2 Unexplained subfertility

Espino 2019 enrolled women with unexplained subfertility. There was no clear evidence of a difference between the groups (OR 1.00, 95% CI 0.15 to 6.77; P = 1.00; 20 women).

2.5.3 Varying indications

Fernando 2018 enrolled women with various reasons for their subfertility. There was no clear evidence of a difference between the groups (OR 0.93, 95% CI 0.37 to 2.32; P = 0.88; 120 women).

2.6 Clinical pregnancy; IVF/ICSI

Two trials(Espino 2019; Fernando 2018) reported on women who were undergoing IVF/ICSI. There was no clear evidence of a difference between the groups (OR 0.94, 95% CI 0.41 to 2.15; P = 0.89, I² = 0%; 140 women).

Secondary outcome: Adverse events

2.7 Adverse events

See Analysis 2.7; Figure 10

Figure 10

Forest plot of comparison: 2 Head‐to‐head antioxidants, outcome: 2.7 Adverse events.

We subgrouped adverse event data according to the type of events that occurred, as reported by the trials. These included miscarriage, ectopic pregnancy and headache, congenital (missing kidney), low birth weight, birth between 34 and 37 weeks placenta previa and pre‐eclampsia. There was no evidence to suggest an association between antioxidants and adverse events, but data were very limited, with only three trials reporting on miscarriage, and only one trial reporting on the remaining adverse events.

2.7.1 Miscarriage

Three trials (El Sharkwy 2019a; Espino 2019; Fernando 2018) report on miscarriage, but there were no events in Espino 2019 or Fernando 2018. There was no clear evidence of a difference between NAC and L‐carnitine in El Sharkwy 2019a (OR 1.54, 95% CI 0.42 to 5.67; P = 0.52; 3 RCTs, 304 women; low‐quality evidence). This suggests that among subfertile women with an expected miscarriage rate of 3.0%, the rate among women using a NAC versus L‐carnitine would be between 1.3% and 15% (summary of findings Table 2).

Fernando 2018 reported on ectopic pregnancy and headache, congenital (missing kidney), low birth weight, birth between 34 and 37 weeks, placenta previa and pre‐eclampsia.

2.7.2 Ectopic pregnancy

There were no ectopic pregnancies in either the lower‐ or higher‐dose melatonin.

.2.7.3 Congenital (missing kidney)

There was no clear evidence of a difference between the lower‐ or higher‐dose melatonin (OR 1.53, 95% CI 0.06 to 38.36; P = 0.80; 120 women).

2.7.4 Low birth weight < 2.500 g

There were no babies born with low birth weight in either the lower‐ or higher‐dose melatonin groups.

2.7.5 Birth between 34 and 37 weeks

There was no clear evidence of a difference between the lower‐ or higher‐dose melatonin (OR 0.49, 95% CI 0.03 to 8.10; P = 0.62; 120 women).

2.7.6 Placenta previa

There was no clear evidence of a difference between the lower‐ or higher‐dose melatonin (OR 1.53, 95% CI 0.06 to 38.36; P = 0.80; 120 women).

2.7.7 Pre‐eclampsia

There was no clear evidence of a difference between the lower‐ or higher‐dose melatonin (OR 0.49, 95% CI 0.03 to 8.10; P = 0.62; 120 women).

Discussion

Summary of main results

Effectiveness of antioxidants versus placebo or no treatment

Very low‐quality evidence means that we are uncertain whether antioxidants improve the live birth rate compared with placebo or no treatment/standard treatment. Thirteen trials with a total of 1227 women reported on live birth (summary of findings Table 1). The differences between the trials (heterogeneity) were low (I² = 29% with a fixed‐effect model).

We conducted subgroup analyses, in accordance with our protocol, by type of comparison and type of antioxidant. The association between antioxidants and an increased live birth rate persisted. There was an association between the use of combination antioxidants and increased live birth, but heterogeneity was high. When we considered specific indications for subfertility, there was an association between the use of antioxidants and increased live birth among women with polycystic ovary syndrome (PCOS) and those trials that enrolled women with varying indications for subfertility.

We found no difference between antioxidants and an increased live birth rate among women undergoing IVF or ICSI.

We performed a sensitivity analysis excluding trials at high risk of bias in any domain, and those that used folic acid or a fertility drug as a control (these were in both the intervention and control arms with an antioxidant in addition in the intervention, and classified as no treatment). When these trials were removed from the analysis there remained an association between antioxidants and an increased live birth rate, with heterogeneity moderately low.

Antioxidants may improve clinical pregnancy rate when compared with either placebo or no treatment, although the quality of this evidence was assessed as low (summary of findings Table 1). Heterogeneity was moderate, but there was no evidence that the effects differed by type of control (placebo or no treatment). We conducted sensitivity analyses excluding trials at high risk of bias and those using a standard or co‐intervention agent as their control. There remained an association between increased clinical pregnancy rates and antioxidants in the analysis when these trials were removed.

When we considered individual antioxidant interventions separately, N‐acetylcysteine, 'combined antioxidants', melatonin, CoQ10 and L‐carnitine showed an association between antioxidant and an increased clinical pregnancy rate, although heterogeneity in the N‐acetylcysteine, 'combined antioxidants and L‐carnitine groups was high. We found no difference between ascorbic acid, L‐arginine, vitamin D or vitamin B complex and clinical pregnancy rate, although these subgroups contained only three or fewer trials.

When we considered specific 'indications for subfertility', we found an association between antioxidants and increased clinical pregnancy in women with PCOS. Heterogeneity here was moderate, which is probably due to the varying antioxidants, as shown by a significant result in the test for subgroup differences. We found no difference between antioxidants and clinical pregnancy rates in women with unexplained subfertility, with tubal subfertility, with varying indications, or in trials that enrolled women with poor ovarian reserve.

There was no association between antioxidants and clinical pregnancy rates in women undergoing IVF or ICSI.

There was insufficient evidence to draw any conclusions about adverse events such as miscarriage, multiple pregnancy, gastrointestinal disturbances, ectopic pregnancy, headache or preterm birth when comparing antioxidants with placebo or no treatment/standard treatment. We rated the quality of evidence for miscarriage, multiple pregnancy and gastrointestinal disturbances as moderate to very low (summary of findings Table 1). The outcomes of congenital abnormality, low birth weight, placenta previa, pre‐eclampsia, fatigue and ovarian hyperstimulation syndrome were reported by only one trial.

Effectiveness of antioxidants versus antioxidants－head‐to‐head

Low‐quality evidence indicates that there was no difference between lower‐ and higher‐dose melatonin in live birth rates. Two trials with a total of 140 women reported on live birth (summary of findings Table 2). The differences between the trials (heterogeneity) were low (I² = 0% with a fixed‐effect model).

One trial enrolled women with unexplained subfertility and the other enrolled women with varying indications for subfertility, so we were unable to make any assumptions about the use of different doses of melatonin for different indications of subfertility. We were also unable to perform a sensitivity analysis, as neither trial was rated at high risk of bias in any domain.

We found no clear difference between different doses of melatonin and an increased live birth rate among women undergoing IVF or ICSI.

Three trials with a total of 304 women reported on clinical pregnancy. There was no clear evidence of a difference between lower and higher doses of melatonin for increased clinical pregnancy rates, with the quality of this evidence assessed as low (summary of findings Table 2). Heterogeneity was low. A single trial studied the effect of N‐acetylcysteine versus L‐carnitine on clinical pregnancy.

The three trials in this analysis all enrolled women with differing indications for subfertility.

There was no difference with lower or higher doses of melatonin and clinical pregnancy in women undergoing IVF or ICSI. We were also unable to perform a sensitivity analysis as neither trial was considered to be at low or unclear risk of bias in any domain.

Overall completeness and applicability of evidence

Of the 63 trials included in this review, 42 reported on clinical pregnancy but only 13 trials reported on live birth. Miscarriage, harmful events and costs of the included trials generally were not well reported. Twenty‐five trials reported on miscarriage, nine reported on multiple pregnancy, three trials discussed gastrointestinal disturbances, four ectopic pregnancy, two ovarian hyperstimulation syndrome, two preterm birth, and one for headache, congenital abnormality (a missing kidney), low birth weight, placenta previa, pre‐eclampsia and fatigue. The trials were generally quite small, and heterogeneity between them was considered low overall, with the exception of the clinical pregnancy analysis.

The antioxidants melatonin and CoQ10 may have had beneficial effects on the outcomes of this review, and although this was also the case for combination antioxidants and N‐acetylcysteine, these analyses showed large differences between the trials so we could not be sure about this result. Similarly, the indications for subfertility within the trials were representative of the general subfertile population, but apart from trials on PCOS (with 16 trials across all comparisons), there were very few trials specific to one indication for subfertility (six for varying indications, four for unexplained subfertility, two for tubal subfertility, two for poor ovarian reserve, one for endometriosis, and one for poor responders), and when pooling was possible within these indications, we had to take into account that the women were also receiving different types of antioxidants and differing adjunctive interventions such as laparoscopic ovarian drilling, timed intercourse or IVF/ICSI. Apart from PCOS, it was therefore difficult to show any benefit or harm from antioxidants for a particular indication of subfertility.

Only three trials were included in the head‐to‐head analysis, and only two of them used the same antioxidants, which we grouped as lower and higher doses of melatonin; we found no differences in live birth, clinical pregnancy or adverse events between the different dosages.

Quality of the evidence

The quality of the evidence according to the 'Summary of findings' tables (summary of findings Table 1; summary of findings Table 2) was considered to be low to very low for all outcomes in the antioxidant versus placebo/no treatment and in the head‐to‐head comparisons. Heterogeneity for the live birth outcome in the antioxidant versus placebo/no treatment comparison was 29%, and 63% for clinical pregnancy.

The overall quality of evidence was limited by serious risks of bias associated with poor reporting of methods, imprecision and inconsistency, leading to a downgrading of the evidence. The risk of bias within the evidence (because of methodological limitations) was moderately high (see Figure 2; Figure 3; and Characteristics of included studies). Not all trials described their sequence generation or allocation concealment methods, and most trials randomly assigned only small numbers of women.

The funnel plot for clinical pregnancy (Figure 4) was not symmetrical, which suggests that the high number of small studies may have had an excessively positive effect on the overall results. This high risk of bias in the included trials is also described in other antioxidant reviews (Lu 2012; Showell 2011) and seems to be common in this area of complementary medicine.

Potential biases in the review process

There may have been some potential for bias in the review process, as there were some changes to the protocol. These included additions and deletions to inclusion/exclusion criteria and to the subgroup analyses (see Differences between protocol and review). None of these changes were made as a result of the findings of included studies, but rather to improve the structure of the review.

Agreements and disagreements with other studies or reviews

The results of our review are in agreement with those of other published reviews. Sekhon 2010 and Grajecki 2012 concluded that, despite numerous advances made in this area and possible positive effects of antioxidants, there is a need for further investigation using better‐quality randomised controlled trials within a larger population to determine the efficacy and safety of these supplements. A Cochrane Review, Antioxidants for male subfertility (Smits 2019), found a small significant effect in favour of antioxidants for pregnancy and live birth and no apparent association with any reported adverse events, but there were too few similar trials to provide conclusive evidence. Another Cochrane Review (Showell 2018) showed uncertainty in the use of myo‐inositol for women with PCOS.

Similar to the results of our review, Arhin 2017 states that "within the limits of this review micronutrients appear to positively influence the outcomes of pregnancy rate and live birth in couples undergoing IVF and calls for larger clinical trials to strengthen the evidence". However this review includes both women and men and also includes non‐random studies.

Zhang 2020, looks at the use of CoQ10 for poor responders and discusses one of our included studies (Xu 2018), it concludes that CoQ10 has good prospects for women who were poor responders but results need to be confirmed with further studies. This is in line with the conclusions of CoQ10 in our review. Also in agreement with our review, the Pundir 2019 overview says there is low‐ or very low‐quality evidence to suggest that supplementation with NAC can improve ovulation and pregnancy rates in women with PCOS, but these need to be further evaluated by adequately‐powered and well‐conducted randomised controlled trials. Thakker 2015 also says, albeit with more positivity, that "NAC showed significant improvement in pregnancy and ovulation rate as compared to placebo. The findings need further confirmation in well‐designed randomised controlled trials to examine clinical outcomes such as live birth rate in longer follow‐up periods".

Similarly to the conclusions of our review, Lagana 2018 found that the use of myo‐inositol in women without PCOS made little difference to any other outcomes except for the reduction in the amount of gonadotropins used in IVF. A systematic review by Pacis 2015 did not find any evidence to support the use of vitamin D in women undergoing ART. Another three systematic reviews (Fang 2017; Irani 2014; Thomson 2012) looked at vitamin D for subfertile women with PCOS. They reported that there is some evidence for the beneficial effects of vitamin D supplementation on menstrual dysfunction, but the current evidence is limited and additional randomised controlled trials are required.

Two Cochrane Reviews (Bjelakovic 2008; Bjelakovic 2012) reported an increased risk of mortality associated with the use of supplemental antioxidants. Bjelakovic 2012 found this association with beta‐carotene and possibly vitamin E and vitamin A, but not with vitamin C or selenium. The review included healthy participants and participants with various stable diseases. Bjelakovic 2008 reported on the use of antioxidants (beta‐carotene, vitamin A, vitamin C and vitamin E) to prevent gastrointestinal cancers and found that there may be an increased risk of mortality for participants taking these antioxidants. The review authors found that selenium may have preventative effects on gastrointestinal cancers. Neither review supports the use of antioxidants as a preventative measure, and they call for tighter regulations. Bjelakovic 2008 suggests that antioxidants should be regulated as drugs.

A review of systematic reviews, Elnashar 2019 agreed with our review, in that they found an association with antioxidants and an increase in clinical pregnancy rates, which was also the case for women with PCOS. However, unlike our review there was no association found with live birth and the use of antioxidants. The overview summarises that there is a need for further randomised trials within larger populations to determine efficacy and safety.

Figure 1

Study flow diagram.

Figure 2

Methodological risk of bias summary: review authors' judgements about each methodological bias item for each included study.

Figure 3

Methodological risk of bias graph: review authors' judgements about each methodological bias item presented as percentages across all included trials.

Figure 4

Figure 5

Figure 6

Figure 7

Forest plot of comparison: 1 Antioxidant(s) versus placebo or no treatment/standard treatment, outcome: 1.9 Adverse events.

Figure 8

Forest plot of comparison: 2 Head‐to‐head antioxidants, outcome: 2.1 Live birth; type of antioxidant (natural conceptions and undergoing fertility treatments).

Figure 9

Forest plot of comparison: 2 Head‐to‐head antioxidants, outcome: 2.4 Clinical pregnancy; type of antioxidant (natural conceptions and undergoing fertility treatments).

Figure 10

Forest plot of comparison: 2 Head‐to‐head antioxidants, outcome: 2.7 Adverse events.

Analysis 1.1

Comparison 1: Antioxidant(s) versus placebo or no treatment/standard treatment, Outcome 1: Live birth; antioxidants vs placebo or no treatment/standard treatment (natural conceptions and undergoing fertility treatments)

Analysis 1.2

Comparison 1: Antioxidant(s) versus placebo or no treatment/standard treatment, Outcome 2: Live birth; type of antioxidant

Analysis 1.3

Comparison 1: Antioxidant(s) versus placebo or no treatment/standard treatment, Outcome 3: Live birth; indications for subfertility

Analysis 1.4

Comparison 1: Antioxidant(s) versus placebo or no treatment/standard treatment, Outcome 4: Live birth; IVF/ICSI

Analysis 1.5

Comparison 1: Antioxidant(s) versus placebo or no treatment/standard treatment, Outcome 5: Clinical pregnancy; antioxidants vs placebo or no treatment/standard treatment (natural conceptions and undergoing fertility treatments)

Analysis 1.6

Comparison 1: Antioxidant(s) versus placebo or no treatment/standard treatment, Outcome 6: Clinical pregnancy; type of antioxidant

Analysis 1.7

Comparison 1: Antioxidant(s) versus placebo or no treatment/standard treatment, Outcome 7: Clinical pregnancy; indications for subfertility

Analysis 1.8

Comparison 1: Antioxidant(s) versus placebo or no treatment/standard treatment, Outcome 8: Clinical pregnancy; IVF/ICSI

Analysis 1.9

Comparison 1: Antioxidant(s) versus placebo or no treatment/standard treatment, Outcome 9: Adverse events

Analysis 2.1

Comparison 2: Head‐to‐head antioxidants, Outcome 1: Live birth; type of antioxidant (natural conceptions and undergoing fertility treatments)

Analysis 2.2

Comparison 2: Head‐to‐head antioxidants, Outcome 2: Live Birth; indications for subfertility

Analysis 2.3

Comparison 2: Head‐to‐head antioxidants, Outcome 3: Live Birth; IVF/ICSI

Analysis 2.4

Comparison 2: Head‐to‐head antioxidants, Outcome 4: Clinical pregnancy; type of antioxidant (natural conceptions and undergoing fertility treatments)

Analysis 2.5

Comparison 2: Head‐to‐head antioxidants, Outcome 5: Clinical pregnancy; indications for subfertility

Analysis 2.6

Comparison 2: Head‐to‐head antioxidants, Outcome 6: Clinical pregnancy; IVF/ICSI

Analysis 2.7

Comparison 2: Head‐to‐head antioxidants, Outcome 7: Adverse events

Summary of findings 1. Antioxidant(s) compared to placebo or no treatment/standard treatment for female subfertility

Outcomes	Relative effect (95% CI)	*Anticipated absolute effects^ (95% CI)**			Quality of the evidence (GRADE)	What happens
Antioxidant(s) compared to placebo or no treatment/standard treatment for female subfertility
Patient or population: women with subfertility Setting: Infertility clinics Intervention: Antioxidant(s) Comparison: placebo or no treatment/standard treatment
Outcomes	Relative effect (95% CI)	Without antioxidant(s)	With antioxidant(s)	Difference	Quality of the evidence (GRADE)	What happens
Live birth; antioxidants vs placebo or no treatment/standard treatment (natural conceptions and undergoing fertility treatments) № of participants: 1227 (13 RCTs)	OR 1.81 (1.36 to 2.43)	19.0%	29.8% (24.2 to 36.3)	10.8% more (5.2 more to 17.3 more)	⊕⊝⊝⊝ VERY LOW^a,b,c	We are uncertain whether antioxidants improve live birth rate compared with placebo or no treatment/standard treatment.
Clinical pregnancy; antioxidants vs placebo or no treatment/standard treatment (natural conceptions and undergoing fertility treatments) № of participants: 5165 (35 RCTs)	OR 1.65 (1.43 to 1.89)	18.8%	27.6% (24.8 to 30.4)	8.8% more (6.1 more to 11.6 more)	⊕⊕⊝⊝ LOW^a,d	Antioxidant(s) may improve clinical pregnancy rate, compared with placebo or no treatment/standard treatment (natural conceptions and undergoing fertility treatments).
Adverse events ‐ Miscarriage № of participants: 3229 (24 RCTs)	OR 1.13 (0.82 to 1.55)	4.8%	5.4% (4 to 7.3)	0.6% more (0.8 fewer to 2.5 more)	⊕⊕⊝⊝ LOW^a,c	Antioxidant(s) may result in little to no difference in adverse events ‐ Miscarriage
Adverse events ‐ Multiple pregnancy № of participants: 1886 (9 RCTs)	OR 1.00 (0.63 to 1.56)	4.3%	4.3% (2.7 to 6.5)	0.0% fewer (1.6 fewer to 2.2 more)	⊕⊕⊝⊝ LOW^a,c	Antioxidant(s) may result in little to no difference in adverse events ‐ Multiple pregnancy
Adverse events ‐ Gastrointestinal disturbances № of participants: 343 (3 RCTs)	OR 1.55 (0.47 to 5.10)	2.4%	3.7% (1.2 to 11.2)	1.3% more (1.2 fewer to 8.8 more)	⊕⊕⊝⊝ LOW^a,c	Antioxidant(s) may result in little to no difference in adverse events ‐ Gastrointestinal disturbances
Adverse events ‐ Ectopic pregnancy № of participants: 404 (4 RCTs)	OR 1.40 (0.27 to 7.20)	0.6%	0.9% (0.2 to 4.3)	0.3% more (0.4 fewer to 3.7 more)	⊕⊕⊝⊝ LOW^a,c	Antioxidant(s) may result in little to no difference in adverse events ‐ Ectopic pregnancy
*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: Confidence interval; OR: Odds ratio
GRADE Working Group grades of evidence High certainty: We are very confident that the true effect lies close to that of the estimate of the effect Moderate certainty: We are moderately confident in the effect estimate: The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different Low certainty: Our confidence in the effect estimate is limited: The true effect may be substantially different from the estimate of the effect Very low certainty: We have very little confidence in the effect estimate: The true effect is likely to be substantially different from the estimate of effect
^aDowngraded one level due to serious risk of bias. The no‐treatment group increases risk due to the lack of blinding. ^bDowngraded one level; overall the heterogeneity is low (0%), but in the placebo subgroup the heterogeneity statistic is 60% and some trials are showing potential benefit of the intervention while others are showing benefit of the placebo. ^cDowngraded one level as the event rate is low (< 400). ^dDowngraded one level as the heterogeneity statistic (63%) is considered to represent moderate to substantive heterogeneity.

Summary of findings 1. Antioxidant(s) compared to placebo or no treatment/standard treatment for female subfertility

Summary of findings 2. Head‐to‐head antioxidants for female subfertility

Outcomes	Relative effect (95% CI)	*Anticipated absolute effects^ (95% CI)**			Quality of the evidence (GRADE)	What happens
Head‐to‐head antioxidants for female subfertility
Patient or population: women with subfertility Setting: Infertility clinics Intervention: Head‐to‐head antioxidants Comparison: Other antioxidant
Outcomes	Relative effect (95% CI)	With one antioxidant	With another antioxidant	Difference	Quality of the evidence (GRADE)	What happens
Live birth; type of antioxidant (natural conceptions and undergoing fertility treatments) ‐ Melatonin lower dose versus melatonin higher dose № of participants: 140 (2 RCTs)	OR 0.94 (0.41 to 2.15)	24.0%	22.9% (11.5 to 40.4)	1.1% fewer (12.5 fewer to 16.4 more)	⊕⊕⊝⊝ LOW^a,b	There was no clear evidence of a difference between the lower and higher doses of melatonin
Clinical pregnancy; type of antioxidant (natural conceptions and undergoing fertility treatments) ‐ N‐acetylcysteine versus L‐carnitine № of participants: 164 (1 RCT)	OR 0.81 (0.33 to 2.00)	14.6%	12.2% (5.4 to 25.5)	2.4% fewer (9.2 fewer to 10.9 more)	⊕⊝⊝⊝ VERY LOW^c,d	There was no clear evidence of a difference between N‐acetylcysteine versus L‐carnitine
Clinical pregnancy; type of antioxidant (natural conceptions and undergoing fertility treatments) ‐ Melatonin lower dose versus melatonin higher dose № of participants: 140 (2 RCTs)	OR 0.94 (0.41 to 2.15)	24.0%	22.9% (11.5 to 40.4)	1.1% fewer (12.5 fewer to 16.4 more)	⊕⊕⊝⊝ LOW^a,b	There was no clear evidence of a difference between the lower and higher doses of melatonin
Adverse events ‐ Miscarriage № of participants: 304 (3 RCTs)	OR 1.54 (0.42 to 5.67)	3.0%	4.6% (1.3 to 15.1)	1.6 more (1.7 fewer to 12.1 more)	⊕⊕⊝⊝ LOW^a,b	There were no miscarriages in either melatonin study (140 women) There was no clear evidence of a difference between N‐acetylcysteine versus L‐carnitine (164 women)
Adverse events ‐ Multiple pregnancy	There were no trials reporting multiple pregnancy
Adverse events ‐ Gastrointestinal disturbances	There were no trials reporting gastrointestinal disturbances
Adverse events ‐ Ectopic pregnancy Melatonin lower dose versus melatonin higher dose № of participants: 120 (1 RCT)	Not estimable, there were no ectopic pregnancies in either group				⊕⊝⊝⊝ VERY LOW ^{3 4}	There was no clear evidence of a difference between the lower and higher doses of melatonin
*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: Confidence interval; OR: Odds ratio
GRADE Working Group grades of evidence High certainty: We are very confident that the true effect lies close to that of the estimate of the effect Moderate certainty: We are moderately confident in the effect estimate: The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different Low certainty: Our confidence in the effect estimate is limited: The true effect may be substantially different from the estimate of the effect Very low certainty: We have very little confidence in the effect estimate: The true effect is likely to be substantially different from the estimate of effect
^aDowngraded one level as there are only two trials in this analysis and one is small. ^bDowngraded one level as event rate is low (< 400). ^cDowngraded two levels as one study can not represent possible subfertile populations. ^dDowngraded two levels as only one study, event rate low and small number of participants

Summary of findings 2. Head‐to‐head antioxidants for female subfertility

Table 1. Gerli 2007‐ data not included in meta‐analysis

Outcome	Data	Notes
Clinical pregnancy rate; myo‐inositol + folic acid	4/23	Only 42 of the 92 women enrolled in this trial declared a desire to become pregnant
Clinical pregnancy rate; folic acid + placebo	1/19	‐
Miscarriage rate; myo‐inositol + folic acid	Miscarriage reported, but unknown whether from treatment or control	1 miscarriage occurred in the first trimester, but it is unknown from which group
Miscarriage rate; folic acid + placebo	Unknown	‐

Table 1. Gerli 2007‐ data not included in meta‐analysis

Table 2. 'Biochemical' and 'pregnancy' data for those trials that did not specifically report 'clinical pregnancy'

Trial	Pregnancy in antioxidant group	Pregnancy in control group
Mier‐Cabrera 2008	0/16 (vitamins C + E), at follow‐up over 9 months 3/16	0/18 (placebo), at follow‐up over 9 months 2/18
Mohammadbeigi 2012	9/22 (vitamin D)	7/22 (placebo)
Razavi 2015	6/32 (selenium)	1/32 (placebo)
Al‐Alousi 2018	20/60 (omega)	15/58 (placebo)
Ghomian 2019	7/33 (NAC + CC)	5/33 (CC)
Heidar 2019	6/20 (selenium)	5/20 (placebo)
Siavashani 2018	5/20 (chromium)	4/20 (placebo)
Schillaci 2012	0/6 (myo‐inositol + 200 µg folic acid twice a day)	0/6 (400 µg folic acid once a day)
CC: clomiphene citrate; NAC: N‐acetylcysteine

Table 2. 'Biochemical' and 'pregnancy' data for those trials that did not specifically report 'clinical pregnancy'

Comparison 1. Antioxidant(s) versus placebo or no treatment/standard treatment

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1.1 Live birth; antioxidants vs placebo or no treatment/standard treatment (natural conceptions and undergoing fertility treatments) Show forest plot	13	1227	Odds Ratio (M‐H, Fixed, 95% CI)	1.81 [1.36, 2.43]

1.1.1 Placebo	7	628	Odds Ratio (M‐H, Fixed, 95% CI)	1.89 [1.18, 3.03]
1.1.2 No treatment	6	599	Odds Ratio (M‐H, Fixed, 95% CI)	1.77 [1.22, 2.56]
1.2 Live birth; type of antioxidant Show forest plot	13	1227	Odds Ratio (M‐H, Fixed, 95% CI)	1.81 [1.36, 2.43]

1.2.1 N‐acetyl‐cysteine	1	60	Odds Ratio (M‐H, Fixed, 95% CI)	3.00 [1.05, 8.60]
1.2.2 L‐arginine	1	37	Odds Ratio (M‐H, Fixed, 95% CI)	0.43 [0.09, 2.09]
1.2.3 CoQ10	2	225	Odds Ratio (M‐H, Fixed, 95% CI)	1.50 [0.78, 2.88]
1.2.4 Vitamin D	1	52	Odds Ratio (M‐H, Fixed, 95% CI)	0.79 [0.21, 3.02]
1.2.5 Vitamin B complex	1	102	Odds Ratio (M‐H, Fixed, 95% CI)	2.07 [0.93, 4.57]
1.2.6 Combined antioxidants	3	378	Odds Ratio (M‐H, Fixed, 95% CI)	2.59 [1.52, 4.40]
1.2.7 Vitamin E	1	103	Odds Ratio (M‐H, Fixed, 95% CI)	1.43 [0.50, 4.10]
1.2.8 Melatonin	3	270	Odds Ratio (M‐H, Fixed, 95% CI)	1.60 [0.68, 3.76]
1.3 Live birth; indications for subfertility Show forest plot	11		Odds Ratio (M‐H, Fixed, 95% CI)	Subtotals only

1.3.1 Polycystic ovary syndrome	3	362	Odds Ratio (M‐H, Fixed, 95% CI)	3.34 [1.90, 5.86]
1.3.2 Tubal subfertility	1	37	Odds Ratio (M‐H, Fixed, 95% CI)	0.43 [0.09, 2.09]
1.3.3 Varying indications	3	338	Odds Ratio (M‐H, Fixed, 95% CI)	1.70 [1.02, 2.83]
1.3.4 Unexplained subfertility	2	133	Odds Ratio (M‐H, Fixed, 95% CI)	1.50 [0.60, 3.72]
1.3.5 Poor ovarian reserve	2	266	Odds Ratio (M‐H, Fixed, 95% CI)	1.75 [0.83, 3.67]
1.4 Live birth; IVF/ICSI Show forest plot	9	806	Odds Ratio (M‐H, Fixed, 95% CI)	1.36 [0.96, 1.93]

1.5 Clinical pregnancy; antioxidants vs placebo or no treatment/standard treatment (natural conceptions and undergoing fertility treatments) Show forest plot	35	5165	Odds Ratio (M‐H, Fixed, 95% CI)	1.65 [1.43, 1.89]

1.5.1 Placebo	17	3292	Odds Ratio (M‐H, Fixed, 95% CI)	1.70 [1.42, 2.05]
1.5.2 No treatment/standard treatment	19	1873	Odds Ratio (M‐H, Fixed, 95% CI)	1.57 [1.28, 1.94]
1.6 Clinical pregnancy; type of antioxidant Show forest plot	35	5165	Odds Ratio (M‐H, Fixed, 95% CI)	1.65 [1.43, 1.89]

1.6.1 N‐acetylcysteine	8	1590	Odds Ratio (M‐H, Fixed, 95% CI)	1.36 [1.05, 1.77]
1.6.2 Combined antioxidants	5	689	Odds Ratio (M‐H, Fixed, 95% CI)	1.90 [1.33, 2.70]
1.6.3 Melatonin	7	678	Odds Ratio (M‐H, Fixed, 95% CI)	1.66 [1.12, 2.47]
1.6.4 Vitamin E	1	103	Odds Ratio (M‐H, Fixed, 95% CI)	1.43 [0.50, 4.10]
1.6.5 Ascorbic acid	2	899	Odds Ratio (M‐H, Fixed, 95% CI)	0.91 [0.66, 1.25]
1.6.6 L‐arginine	2	71	Odds Ratio (M‐H, Fixed, 95% CI)	1.05 [0.32, 3.46]
1.6.7 Myo‐inositol plus folic acid	1	94	Odds Ratio (M‐H, Fixed, 95% CI)	1.24 [0.50, 3.06]
1.6.8 CoQ10	4	397	Odds Ratio (M‐H, Fixed, 95% CI)	2.49 [1.50, 4.13]
1.6.9 L‐carnitine	2	450	Odds Ratio (M‐H, Fixed, 95% CI)	11.14 [5.70, 21.81]
1.6.10 Vitamin D	2	92	Odds Ratio (M‐H, Fixed, 95% CI)	0.83 [0.25, 2.76]
1.6.11 Vitamin B complex	1	102	Odds Ratio (M‐H, Fixed, 95% CI)	1.94 [0.82, 4.58]
1.7 Clinical pregnancy; indications for subfertility Show forest plot	31		Odds Ratio (M‐H, Fixed, 95% CI)	Subtotals only

1.7.1 Polycystic ovary syndrome	15	1908	Odds Ratio (M‐H, Fixed, 95% CI)	4.24 [3.23, 5.56]
1.7.2 Unexplained	4	997	Odds Ratio (M‐H, Fixed, 95% CI)	0.84 [0.61, 1.16]
1.7.3 Tubal subfertility	2	71	Odds Ratio (M‐H, Fixed, 95% CI)	1.05 [0.32, 3.46]
1.7.4 Varying indications	6	1135	Odds Ratio (M‐H, Fixed, 95% CI)	1.14 [0.85, 1.52]
1.7.5 Poor responders	1	65	Odds Ratio (M‐H, Fixed, 95% CI)	1.88 [0.64, 5.47]
1.7.6 Poor ovarian reserve	2	266	Odds Ratio (M‐H, Fixed, 95% CI)	1.70 [0.86, 3.37]
1.7.7 Endometriosis	1	280	Odds Ratio (M‐H, Fixed, 95% CI)	1.19 [0.71, 1.98]
1.8 Clinical pregnancy; IVF/ICSI Show forest plot	18	2341	Odds Ratio (M‐H, Fixed, 95% CI)	1.15 [0.95, 1.40]

1.9 Adverse events Show forest plot	27		Odds Ratio (M‐H, Fixed, 95% CI)	Subtotals only

1.9.1 Miscarriage	24	3229	Odds Ratio (M‐H, Fixed, 95% CI)	1.13 [0.82, 1.55]
1.9.2 Multiple pregnancy	9	1886	Odds Ratio (M‐H, Fixed, 95% CI)	1.00 [0.63, 1.56]
1.9.3 Gastrointestinal disturbances	3	343	Odds Ratio (M‐H, Fixed, 95% CI)	1.55 [0.47, 5.10]
1.9.4 Ectopic pregnancy	4	404	Odds Ratio (M‐H, Fixed, 95% CI)	1.40 [0.27, 7.20]
1.9.5 Headache	2	330	Odds Ratio (M‐H, Fixed, 95% CI)	0.89 [0.45, 1.75]
1.9.6 Congenital (missing kidney)	1	160	Odds Ratio (M‐H, Fixed, 95% CI)	1.02 [0.04, 25.46]
1.9.7 Low birth weight < 2.500 g	1	160	Odds Ratio (M‐H, Fixed, 95% CI)	0.11 [0.00, 2.74]
1.9.8 Preterm birth	2	220	Odds Ratio (M‐H, Fixed, 95% CI)	1.31 [0.17, 9.93]
1.9.9 Placenta praevia	1	160	Odds Ratio (M‐H, Fixed, 95% CI)	1.02 [0.04, 25.46]
1.9.10 Pre‐eclampsia	1	160	Odds Ratio (M‐H, Fixed, 95% CI)	1.71 [0.08, 36.35]
1.9.11 Fatigue	1	160	Odds Ratio (M‐H, Fixed, 95% CI)	1.86 [0.75, 4.62]
1.9.12 Ovarian hyperstimulation syndrome (OHSS)	1	150	Odds Ratio (M‐H, Fixed, 95% CI)	Not estimable

Comparison 1. Antioxidant(s) versus placebo or no treatment/standard treatment

Comparison 2. Head‐to‐head antioxidants

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
2.1 Live birth; type of antioxidant (natural conceptions and undergoing fertility treatments) Show forest plot	2	140	Odds Ratio (M‐H, Fixed, 95% CI)	0.94 [0.41, 2.15]

2.1.1 Melatonin lower dose versus melatonin higher dose	2	140	Odds Ratio (M‐H, Fixed, 95% CI)	0.94 [0.41, 2.15]
2.2 Live Birth; indications for subfertility Show forest plot	2		Odds Ratio (M‐H, Fixed, 95% CI)	Subtotals only

2.2.1 Unexplained subfertility	1	20	Odds Ratio (M‐H, Fixed, 95% CI)	1.00 [0.15, 6.77]
2.2.2 Varying Indications	1	120	Odds Ratio (M‐H, Fixed, 95% CI)	0.93 [0.37, 2.32]
2.3 Live Birth; IVF/ICSI Show forest plot	2	140	Odds Ratio (M‐H, Fixed, 95% CI)	0.94 [0.41, 2.15]

2.4 Clinical pregnancy; type of antioxidant (natural conceptions and undergoing fertility treatments) Show forest plot	3		Odds Ratio (M‐H, Fixed, 95% CI)	Subtotals only

2.4.1 N‐acetylcysteine versus L‐carnitine	1	164	Odds Ratio (M‐H, Fixed, 95% CI)	0.81 [0.33, 2.00]
2.4.2 Melatonin lower dose versus melatonin higher dose	2	140	Odds Ratio (M‐H, Fixed, 95% CI)	0.94 [0.41, 2.15]
2.5 Clinical pregnancy; indications for subfertility Show forest plot	3		Odds Ratio (M‐H, Fixed, 95% CI)	Subtotals only

2.5.1 Polycystic ovary syndrome	1	164	Odds Ratio (M‐H, Fixed, 95% CI)	0.81 [0.33, 2.00]
2.5.2 Unexplained subfertility	1	20	Odds Ratio (M‐H, Fixed, 95% CI)	1.00 [0.15, 6.77]
2.5.3 Varying indications	1	120	Odds Ratio (M‐H, Fixed, 95% CI)	0.93 [0.37, 2.32]
2.6 Clinical pregnancy; IVF/ICSI Show forest plot	2	140	Odds Ratio (M‐H, Fixed, 95% CI)	0.94 [0.41, 2.15]

2.7 Adverse events Show forest plot	3		Odds Ratio (M‐H, Fixed, 95% CI)	Subtotals only

2.7.1 Miscarriage	3	304	Odds Ratio (M‐H, Fixed, 95% CI)	1.54 [0.42, 5.67]
2.7.2 Ectopic pregnancy	1	120	Odds Ratio (M‐H, Fixed, 95% CI)	Not estimable
2.7.3 Congenital (missing kidney)	1	120	Odds Ratio (M‐H, Fixed, 95% CI)	1.53 [0.06, 38.36]
2.7.4 Low birth weight < 2.500 g	1	120	Odds Ratio (M‐H, Fixed, 95% CI)	Not estimable
2.7.5 Birth between 34 and 37 weeks	1	120	Odds Ratio (M‐H, Fixed, 95% CI)	0.49 [0.03, 8.10]
2.7.6 Placenta praevia	1	120	Odds Ratio (M‐H, Fixed, 95% CI)	1.53 [0.06, 38.36]
2.7.7 Pre‐eclampsia	1	120	Odds Ratio (M‐H, Fixed, 95% CI)	0.49 [0.03, 8.10]

Comparison 2. Head‐to‐head antioxidants