Trial characteristics, geographic distribution, and selected methodological issues of 1425 infertility trials published from 2012 to 2023: a systematic review.

Infertility is defined as an unsuccessful pregnancy after 6–12 months of unprotected sex. Globally, it affects one in every six couples on average, while the prevalence can be as high as a quarter of the population, e.g. in the Western Pacific ( World Health Organization, 2023 ). Infertility can be attributed to male or female factors, or both, and takes a profound psychological toll on both being affected by the condition. The inability to conceive naturally can evoke shame, guilt, frustration, and feelings of inadequacy, rendering individuals with infertility more vulnerable to suffering from discrimination and ostracism ( Cui, 2010 ; Wang et al. , 2022 ). Randomized controlled trials (RCTs) are a pivotal approach to address these challenges. They are the main source of evidence on the safety and effectiveness of interventions. For this to be possible, RCTs must pose questions that are pertinent to patients, be conducted with strong scientific rigour, and report findings in a transparent and accessible manner. Systematic reviews of a limited number of infertility trials revealed that under half of the trials had registration or specified primary outcomes at reporting ( Farquhar and Marjoribanks, 2018 ; Rimmer et al. , 2022 ). Among those specified outcomes, only half used pregnancy or live birth as primary outcomes ( Rimmer et al. , 2022 ). However, it remains unknown to what extent RCTs in infertility are conducted in such a manner, at a large scale. Understanding the strengths and limitations of RCTs from the recent past will help to pinpoint research gaps and prioritize the future research agenda. To obtain a holistic view of these issues in infertility, we herein provide an overview of RCTs in infertility published in the last decade for their characteristics, geographic distribution, and selected methodological issues.

This is a systematic review of RCTs in infertility published from 1 January 2012 to 30 August 2023. A research protocol was designed and uploaded to OSF ( https://osf.io/sgq6w ) and the study was registered at PROSPERO (CRD42024498624) before the study was conducted. We systematically searched Embase, Medline, and the Cochrane Central Register of Controlled Trials for RCTs in infertility published between 1 January 2012 and 30 August 2023 using a comprehensive search strategy ( Supplementary Table S1 ). Eligible RCTs were those involving infertile women and reporting either biochemical pregnancy, clinical pregnancy, ongoing pregnancy, or live birth as outcomes. Trials which did not report these outcomes were not identified in the literature search. Full texts of included trials were reviewed to confirm that at least one of the four above-specified pregnancy outcomes was reported. We excluded conference abstracts, interim analyses, secondary analyses, or follow-up studies. We did not limit our search based on the language of the articles. Article screening was conducted by W.L.L. and J.C. independently. To test whether the literature search and screening results were reliable and complete, we developed a validation set of infertility trials from Cochrane reviews. A detailed description of how the article set was constructed is provided in Supplementary File S1 . In short, we started by searching for Cochrane systematic reviews focusing on female infertility published between 2012 and 2023. Among 339 Cochrane reviews identified, we manually selected 78 reviews for those that had reported pregnancy or live births. We then applied our eligibility criteria against the infertility trials cited by these reviews, which boiled down to 84 trials included in our validation set. We downloaded articles in portable document format (PDF) via Endnote, a reference managing software, using the ‘find full-text’ function ( Clarivate, 2023 ). Articles not accessible through Endnote were manually obtained via university subscription or interlibrary loans. Our data extraction process involved a semi-automated workflow combining text mining, the process of extracting data from unstructured natural language by software, and manual data extraction conducted by the first author. The full articles were subjected to text mining first, and the manual data extraction came into play when automated data extraction could not pin down a fixed language expression in which the data are embedded, e.g. sample size. This hybrid approach for data extraction has been used widely by other authors of systematic reviews ( Gates et al. , 2021 ; Hill et al. , 2024 ). It serves as a sensitive, precise, and efficient alternative to data extraction solely by humans ( Lajeunesse, 2016 ; Pham et al. , 2021 ). To enable data extraction by computers, we parsed the articles in PDF format into the extensible markup language (XML), a readable and extractable format for computers. The parsing work was performed using Grobid, a machine-learning tool developed for parsing raw documents from PDF to structured XML files, which was specifically designed for academic articles written in English ( Lopez, 2009 , 2023 ). It has been used for over a decade in research with satisfying reliability ( Miah et al. , 2022 ). The accuracy and precision consistently outperform other automated data extraction tools ( Lipinski et al. , 2013 ; Meuschke et al. , 2023 ). The table for recording extracted data was created before data extraction and piloted in 20 studies. It was then revised after discussions with all co-authors. For each type of data that was extracted with the aid of text-mining, we first constructed an algorithm designed to identify text patterns containing the desired data for extraction. We then manually extracted data from a dozen articles randomly selected to compare its results with the data extracted by the algorithm, after which the algorithm was adapted accordingly. This trial-and-error process was then repeated multiple times until the algorithm could accurately extract over 90% of the targeted data. For the articles where the algorithm identified the absence of the data (e.g. the blinding method was missing) or targeted data had no fixed pattern of expression (e.g. sample size or patient recruitment country), we complemented the text-mining approach by extracting data manually. For articles written in other languages, we first extracted available data from their English-language abstracts. If the data were not available in the abstract, we translated the full article into English using Google Translate ( Pedrana et al. , 2019 ) and extracted data manually or asked for help from a native speaker. To confirm the reliability of the data extracted using this method, we randomly selected 50 articles and manually checked the accuracy of each data extraction. To ascertain the retraction status of each article, we linked articles included in our database with Retraction Watch using the digital identifier of an object (DOI) ( Retraction Watch, 2024 ). Using the same method, we downloaded other metadata including conflicts of interest and funding statements from Dimensions, a database of abstracts, citations, and research grants, chosen for its robust and comprehensive metadata ( Hook et al. , 2018 ). The first author then manually ascertained whether the funding source was from the government or industry. To understand how efficient the public research investment was across different countries, we used the quantity of publication of trials, regardless of the quality of trials, divided by gross domestic product (GDP) per country in 2023 as an indicator of research efficiency. The GDP data per country in 2023 were downloaded from the website of the United Nations (2022) . Further, we used the number of trials per adjusted number of newborns (number of births divided by the total fertility rate of that country) as a proxy indicator to measure the adequacy of infertility trials conducted in each country. The number of newborns each year per country in 2023 was downloaded from Database Earth ( Database Earth, 2023 ). The total fertility rate of each country was downloaded from the United Nations (2023) . We used proportions and frequencies to describe the trend. All data extraction, statistical analyses, and graphs were performed using R software ( R Core Team, 2013 ).

From the detected 8757 records, we identified 1510 RCTs by screening their titles and abstracts ( Fig. 1 ). We further excluded 85 ineligible articles that were secondary analyses, conference abstracts, or non-RCTs after reading the full articles, resulting in 1425 RCTs included in our analysis. Flowchart of the study . We compared the identified 1425 RCTs against the validation set of infertility trials to assess the completeness of our literature search. Out of the 84 RCTs in the validation set, 77 (91.4%) were found in our 1425 RCTs. The reasons why our literature search missed the remaining seven RCTs are provided in Supplementary File S1 . On a manual checking of 50 articles for the accuracy of extracting trial characteristics, the false negatives (e.g. the trial received funding, but our extraction indicated it did not) and false positives (e.g. a trial did not register, but our extraction indicated it did) were all at or below 8% for each type of data extracted. The detailed breakdown of manual checking of results by data type can be found in Supplementary Table S2 . Of 1425 RCTs, the majority of the RCTs (93.3%, n = 1330) in our study were written in English ( Table 1 ). The median sample size was 140 and it ranged from 18 to 2772, with 33.3% (n = 474) of the trials having fewer than 100 participants, while 4.3% (n = 62) of RCTs could not be ascertained for their sample size as the author only provided the number of oocytes or embryos randomized. Furthermore, parallel arm studies (99.3%), single-centre studies (82.4%), and studies with controls composed of routine care or no treatment (91.3%) were the most common study designs. Of all the trials, 46.8% (n = 668) did not provide information on whether and how the blinding was performed, while the most common type of blinding method was open-label studies which made up 19.8% of all trials. Baseline characteristics of infertility trials from 2012 to 2023. Trials focusing on IVF made up 69.6% of all trials (n = 992). Ovulation induction and intrauterine insemination contributed to 12.4% (n = 177) and 10.6% (n = 151) of all trials, respectively, while the remaining 7.4% of trials included surgery, tubal flushing, lifestyle interventions, etc. Among trials reporting a primary outcome (n = 629), pregnancy of any stage as the primary outcome accounted for 47.2% (n = 297), while oocyte or embryonic parameters accounted for 19.7% (n = 124) and live birth accounted for 16.4% (n = 103) of the trials. The majority of the trials declared that they did not receive funding from any source (76.9%), while the proportion of studies funded by government and industry was 19.2% and 3.9%, respectively. Among the top five countries producing the highest number of infertility trials, the proportions of funded trials were as follows: China (31.9%), Iran (27.1%), Egypt (4.3%), Turkey (0%), and India (8.5%). Sweden (72.7%) had the highest proportion of funded trials, followed by New Zealand (71.4%) and Denmark (66.7%) ( Supplementary Fig. S1 ). Unregistered trials made up 47.8% of all trials. Over time, the proportion of studies that were unregistered halved from 70.0% in 2012 to 34.6% in 2022. The proportion of trials adopting placebo and single-centre designs remained almost unchanged ( Supplementary Fig. S2 ). There were 536 trials (37.6%) that did not specify what the primary outcome was ( Table 1 ). The proportion of trials specifying primary outcomes rose steadily from 49.5% in 2012 to 61.4% in 2022. A total of 260 (18.3%) of the trials had more than one primary outcome, which was often composed of a mix of various outcomes on effectiveness (e.g. number of oocytes retrieved, number of high-quality embryos, and clinical pregnancy rate, all as a primary outcome in a single RCT; data not shown). European Union (EU) countries were first grouped together, and then the trend was analysed across a breakdown of individual EU countries. Regarding the number of trials produced by each region ( Fig. 2A ), China was the top contributor (n = 310), followed by Iran (n = 284), Egypt (n = 138), Turkey (n = 82), India (n = 71), and the USA (n = 69). The number of trials published by the EU was 343, with Italy being the most prolific country within the EU (n = 55), closely followed by Spain (n = 52) and Belgium (n = 50) ( Fig. 2B ). Internationally collaborative trials, defined as trials recruited in different countries, were most common in the EU countries (31%), while this ratio was below 5% among the other top three most prolific countries ( Supplementary Fig. S3 ). The geographic distribution of trials in infertility published from 2012 to 2023. ( A ) The overall trend in the number of trials, with European Union (EU) countries grouped collectively. ( B ) The trends within individual EU countries showing the contribution of each member state to the overall regional trend. Among the top seven most prolific regions, including the EU, China, Iran, Egypt, Turkey, India, and the USA, China and Iran saw the most rapid growth in the number of trials published each year, overtaking the EU as the top contributor in this field in 2018 ( Fig. 3A ). Egypt saw a mild decrease in the number of trials produced each year, while the number of trials published by the remaining regions remained stable over the years. Regarding EU countries, Italy held the top spot before 2017 but slumped since then, overtaken by Spain as the top contributor in 2022 ( Fig. 3B ). The annual number of trials in infertility published by each country or region in the top seven most prolific countries (or regions). ( A ) The number of trials in infertility by each country or region when European Union (EU) countries are grouped. ( B ) The number of trials in infertility by each country or region within EU countries. The lines represent the rolling average of three consecutive years. Among 1425 RCTs, there were 16 retractions and four expressions of concern, of which 13 articles were from Egypt and three were from China, while the remaining four were from India, Iran, Italy, and Turkey. The list of retracted articles and their reasons for retraction are provided in Supplementary Table S3 . Regarding the number of publications in RCTs of infertility per trillion GDP, Greece ranked as the top among all regions at 4.6 ( Fig. 4B ); Iraq was the second, at 3.9 publications of RCTs per trillion GDP, followed by Iran and Denmark, both at 2.5 ( Fig. 4A and B ). While the USA and China had the highest GDP among all regions, both were at the bottom of this ranking, at 0.04 and 0.06 RCTs per trillion GDP, respectively. The number of trials in infertility published per trillion gross domestic product (GDP) among the top 17 most prolific countries (or regions). ( A ) The number of trials in infertility published per trillion GDP when European Union (EU) countries are grouped. ( B ) The number of trials in infertility published per trillion GDP within EU countries. Within EU countries, the number of publications per million newborns, adjusted by total fertility rate, was highest in Belgium (n = 655), with Denmark (n = 644) and Greece (n = 396) following ( Fig. 5B ). Outside the EU, Iran (n = 417) had the highest number of publications per adjusted annual newborns in millions ( Fig. 5A ). While India, Nigeria, and Pakistan had the highest fertility rates, their trials per adjusted million newborns were at the bottom of the ranking, with each having less than seven publications per adjusted million annual newborns. The number of trials in infertility published per newborn each year in millions, adjusted by total fertility rate, in all countries (regions) . ( A ) The number of trials in infertility published per newborn each year in millions when European Union (EU) countries are grouped. ( B ) The number of trials in infertility published per newborn each year in millions within EU countries.

In a comprehensive overview of RCTs in infertility, we found a dominance of trials focusing on IVF, highlighting a need for reconsideration of other topics to be studied and a realignment of established research priorities. Trials from countries with potentially the largest number of infertile couples, such as India, were under-represented, raising a question about the equity of healthcare resources and generalizability of the results of trials. Inadequate trial registration rates and insufficient reporting of primary outcomes highlight the importance of improving trial design and reporting quality in this field. On a positive note, the consistent improvement in trial registrations suggests that the enforcement of trial registrations from the journals is effective and hence should be continued. These findings provide insights for the optimization of study design and prioritization of the research agenda.

In this overview of literature comprising 1425 infertility trials, we observed that nearly two-fifths of the trials used primary outcomes that were neither pregnancy nor live birth, while nearly 70% of all trials focused on IVF. Trials from China, Egypt, and Iran made up half of the global infertility trials, while trials from regions with the fastest-growing populations, such as India and Nigeria, were under-represented. In terms of methodological issues, half of the trials were not registered, and a third of the trials did not specify a primary outcome. Encouragingly, there has been a consistent rise in the trial registration rate and the number of trials specifying primary outcomes over time. To our knowledge, this is the largest comprehensive overview of infertility trials published in the last decade. We provided a holistic overview of the current state and evolving trends of infertility trials globally, laying the groundwork for future-focused research questions on specific methodological issues, such as temporal changes in trial registration and the proportion of trials conducting interim analyses. Mapping such a landscape represents an important stepping stone towards addressing these questions and optimizing resources used for future research in infertility. Prospective trial registration is a crucial measure for fostering research transparency and accountability. However, the registration rate, including prospective and retrospective trial registration, was not optimal in infertility trials. While we did not distinguish between prospective and retrospective registration in the trials, the low overall registration rate suggests that fewer than half were likely registered prospectively. This inadequate compliance with trial registration was also noted by others within and beyond the fertility field ( Farquhar et al. , 2017a ). The reasons for prospective registration deficiencies are multifactorial. They might be attributed to poor awareness among trialists, lack of organization and supervision of the trials, or inadequate enforcement by journals to which the manuscript is submitted. Lack of prospective registration warrants attention as unregistered trials have a higher risk of bias for random sequence generation, allocation concealment, and selective reporting, all of which undermine the reliability and validity of the trials’ results ( Farquhar et al. , 2017b ). Moreover, unregistered trials are less likely to be identified and included in secondary analyses, hence contributing to publication bias. Encouragingly, we saw the proportion of registered trials increased over time, thanks to the mandatory trial registrations required by the International Committee of Medical Journal Editors and enforced by journal editors ( Zarin et al. , 2017 ). Other measures, such as the mandatory requirement for National Institutes of Health-funded trials to register and post-results at ClinicalTrials.gov, might also play a role in boosting prospective trial registration ( Riley et al. , 2018 ). However, we cannot rule out that the rising trial registration was due to retrospective trial registrations, which offer no safeguard against selective reporting. To further improve prospective trial registrations, it is critical that there is concerted enforcement of prospective registration throughout the entire lifecycle of the trials made by national legislation, funding bodies, ethics committees, journal editors, and universities through self-regulation ( Viergever and Li, 2015 ). Our study showed an insufficient reporting of primary outcome, consistent with the observations in the fields of endometriosis and male infertility ( Hirsch et al. , 2016 ; Rimmer et al. , 2022 ). Omissions of primary outcomes can hinder the interpretability, replicability, and synthesis of RCTs, which are crucial for informing clinical guidelines and decision-making. As demonstrated in our findings, there has been a mild but consistent increase in the proportion of trials specifying primary outcomes over time. This could be attributed to the Consolidated Standards of Reporting Trials (CONSORT) checklist, which requires authors to pre-specify the outcomes explored in the study and has been widely endorsed by journal editors. To improve the primary outcome reporting, a continued effort should be made to increase the endorsement and compliance of CONSORT. With regard to the primary outcomes used, trials often set primary outcomes that are peripheral to pregnancy and live birth, including hormonal levels, and oocyte or embryonic parameters, in line with the findings by Rimmer et al. , 2022 . The motives for choosing these intermediate outcomes are that they take a shorter time to observe and are easier to measure. Furthermore, selective reporting could be another cause, in which authors choose to report positive results over negative ones. In addition, using upstream endpoints other than live birth may be more suitable in certain scenarios. For example, when an intervention has a clear, proposed mechanism, testing that mechanism can be more efficient, cost-effective, and ethical before assessing its impact on live birth in large, expensive trials ( Wilkinson and Stocking, 2021 ). While focusing on these upstream outcomes enables researchers to understand the underlying mechanisms behind infertility treatments, they might not always correlate directly with the ultimate goal of achieving a successful pregnancy or live birth. Core outcome sets in infertility have been developed and endorsed by the reproductive health community to encourage reporting of outcomes relevant to patients with infertility, even when these are not the primary outcome ( Duffy et al. , 2020b , 2021 ). Future studies are needed to assess the barriers preventing the uptake of the core outcome set by trialists and researchers. Moreover, studies with multiple primary outcomes were also observed in our analysis. Having multiple primary outcomes can threaten a study’s validity if not handled properly, as it increases the risk of Type I errors and may lead to selective reporting. While we did not assess whether statistical methods were used to control Type I error inflation in these cases, it is unlikely that such adjustments were frequently applied. We observed that infertility trials with sample sizes of fewer than 100 were common. Trials of smaller sample sizes have the advantages of shorter time to complete and lower cost to conduct, and sometimes they are the only option when the prevalence of the diseases is rare. While trials of smaller sample sizes can provide evidence conclusive enough for changing clinical practices, for most research questions on infertility, a large sample size is often needed when using pregnancy or live birth as a primary outcome, as a significant portion of infertility couples can conceive naturally in the absence of any treatment ( Wang et al. , 2020 ). While we did not measure whether the sample size calculation was based on robust statistical reasoning, a review of sample sizes in infertility trials by Stocking et al. found that even the largest infertility trials lacked the power to detect plausible improvements in live birth rates ( Stocking et al. , 2019 ). Compared to large trials, trials of smaller sizes are susceptible to publication bias, and carry a higher risk of having false-positive results and inflated treatment effects ( Dechartres et al. , 2013 ). While the CONSORT and its extension have outlined specific items related to sample size calculations for the reporting of RCTs, future research is needed to assess to what extent trials have complied with the checklist and to identify the potential hurdles for compliance. EU countries produced the highest number of trials, likely due to their high uptake of assisted reproductive technology ( Hummel and Kettel, 1997 ; Berg Brigham et al. , 2013 ; Wyns et al. , 2021 ). This widespread acceptance of infertility treatments drives interest and demand for infertility research through clinical trials. Additionally, these countries benefit from well-established funding agencies and research infrastructure that actively support reproductive health studies. While the landscape of infertility trials should be applauded for its geographic diversity, we also observed a geographical disparity: a small number of infertility trials were conducted in countries with potentially a large number of infertile patients. For example, India represents almost 20% of the world’s population, yet only 5% of global infertility trials are conducted there ( Worldometer, 2024b ). Africa accounts for 18% of the global population, while the number of trials in Africa outside Egypt was little to none ( Worldometer, 2024a ). One rationale for promoting more clinical trials in these populous countries is that the disease characteristics and research agenda in these countries are different from others. For example, over three-quarters of infertility cases in sub-Saharan women are attributed to infection, while this rate in women worldwide is only 30% ( Ombelet et al ., 2008 ). On the other hand, while 70% of infertility trials are dedicated to IVF treatment, IVF in countries with a high total fertility rate is either woefully inadequate or prohibitively expensive ( Chiware et al. , 2021 ). Nigeria has a fertility rate of 5.4, one of the fastest expanding populations in the world with a potentially large number of infertile couples, but the number of IVF clinics per million people was 0.46 ( Ombelet and Onofre, 2019 ; TC Health 2024 ). This disparity underscores the need to align research resources with the pressing healthcare demands of rapidly growing populations ( Duffy et al. , 2020a ). In addition, there is a pressing need for more inclusive research efforts that address the specific needs of these under-represented regions, ensuring that the benefits of infertility research and treatments are globally accessible. The number of trials per GDP can serve as a proxy for research prioritization, indicating which countries allocate more effort to infertility research irrespective of wealth. While high values can suggest that a country focuses more on infertility research relative to its economic capacity, this alone does not fully explain the patterns observed in our analysis. Greece, Iraq, Iran, and Egypt are among the top countries for infertility trials published per trillion GDP, yet they have unusually low proportions of funded trials. The reasons behind this paradox are unclear. There could be fewer organizations or governmental programs dedicated specifically to infertility research funding in these regions. There could also be a lower awareness of declaring funding resources in these countries. Our study has limitations. The first is that while we optimized the algorithm for text-mining iteratively, there have been inaccuracies in data extraction. However, our manual checking revealed that the false-positive and false-negative rates were all at or below 8%. Using text mining for data extraction has been widely adopted by other authors of systematic reviews, particularly when the volume of studies makes manual extraction impractical ( Thomas et al. , 2011 ; Pereira et al. , 2012 ; Walker et al. , 2022 ). Manually checking data extracted through text mining, as done in this study, is routinely performed to ensure the validity of the extracted data. Given the lack of a standardized framework for data extraction, a separate focused study could help further validate our approach. Secondly, while we extracted characteristics of the trials, some of the information was not granular enough to provide detailed insights into specific aspects of the research. For example, while we showed the proportion of trials that have used a placebo as a comparator, we did not ascertain whether the use of a placebo was plausible in the clinical context where the drug was under investigation. While trial registration status was ascertained, distinguishing between prospective and retrospective registrations was unfeasible due to the large volume of studies requiring manual extraction. Future research delving deeper into this question might provide more useful insights. Thirdly, we restricted our scope to trials reporting pregnancy outcomes, excluding those that reported only earlier endpoints. However, while trials reporting pregnancy outcomes are of primary interest to patients and carry significant clinical implications, early-stage trials also play an important role and should be considered when evaluating the full spectrum of infertility trials. Finally, the landscape is derived from trials that are published, suggesting the conclusions cannot be extrapolated to unpublished infertility trials.