Incorporation of the National Institute of Health (NIH) sex as a biological variable policy by R01 grant awardees

Abstract

Background

The National Institutes of Health (NIH) Sex as a Biological Variable (SABV) policy aims to improve the rigor and reproducibility of biomedical research by encouraging sex-inclusive study designs and sex-based analyses.

Methods

To evaluate policy implementation, we examine 574 funded publications (2017–2024) linked to grants from 21 NIH Institutes and Centers. We assess sex inclusion and sex-based reporting and analyses, while examining associations with author gender.

Results

Sixty-one percent of studies include both sexes, with human-subject research more likely to do so than non-human studies (p = 1.31×10^-23). Of the single-sex studies, 34% focus on sex-specific topics. Among studies to include both sexes, 83% report sample sizes by sex, yet only 44% conduct sex-based analyses. Sex-based analyses are more common in human-subject studies (p = 0.00002) and in articles with women first authors (50% vs. 39%, p = 0.036). Articles with women as first and last author dyads are significantly more likely to analyze data by sex (OR = 2.24, 95% CI: 1.33–3.79).

Conclusions

These findings suggest that while the SABV policy has encouraged sex inclusion, gaps remain in sex-based analyses. Author gender may influence these research practices, underscoring the need for continued efforts to fully integrate SABV into NIH-funded research studies.

Plain Language Summary

The U.S. National Institutes of Health (NIH) requires researchers to consider sex as a biological variable (SABV) to improve research quality and ensure findings benefit everyone. To assess how well this policy is being implemented, 574 NIH-funded publications from 2017–2024 were reviewed. Most studies (61%) included both sexes, particularly those involving humans. However, fewer than half (44%) analyzed results by sex, even when both sexes were included. Sex-based analyses were more common in human studies and in articles with women as first or senior authors. These findings show progress in sex inclusion, but many studies still fail to compare results by sex. Without these analyses, important differences may be missed. Stronger guidance and incentives are needed to ensure biomedical research fully considers sex and produces results that benefit all people.

Warden et al. analyze 574 NIH-funded publications (2017–2024) to assess implementation of the Sex as a Biological Variable policy. They find that most studies include both sexes, but fewer than half analyze results by sex; such analyses are more common in human studies and when women are lead authors, highlighting persistent gaps in how sex is considered in biomedical research.

Introduction

Sex is an important biological variable that shapes our understanding of human health and disease. It influences the incidence, presentation, progression, and treatment outcomes of numerous diseases and disorders. Despite this knowledge, biomedical research studies have historically favored male subjects, significantly limiting our understanding of sex differences and overlooking crucial insights into female biology^1–3. This knowledge gap has additional implications; failing to examine or report differences by sex may require additional studies to address questions that could have been resolved with more rigorous study designs and comprehensive reporting.

To address these challenges, the U.S. National Institutes of Health (NIH) introduced a landmark policy in 2016, requiring grantees to consider sex as a biological variable (SABV) in their experimental design, analyses, and reporting^4,5. This policy aims to enhance transparency, rigor, and reproducibility in biomedical research by encouraging investigators to systematically account for sex-based differences in their studies⁶. While investigators must address SABV in their grant submissions, it remains unclear how much weight this criterion carries in grant review and scoring, and whether its consideration influences downstream research practices and publication outcomes⁷.

In this study, we evaluate how the SABV policy is addressed in research supported by NIH R01 grants, the agency’s flagship mechanism for investigator-initiated research⁸. R01 awards provide sustained support for innovative research, culminating in peer-reviewed publications that reflect the quality and impact of the funded studies. By analyzing grants awarded in fiscal years 2017 and 2018, we link funding data to project outcomes to assess if and how SABV considerations are operationalized from funding to publication. Specifically, we examine sex inclusion and sex-based reporting and analyses in peer-reviewed publications and explore relationships between principal investigator (PI) and author gender, study subject type, and NIH funding Institute or Center.

We find that while most R01‑associated publications include both sexes, fewer than half analyze data by sex, revealing a persistent gap between sex inclusion and the research practices intended by the SABV policy. We also observe that studies using animal models are markedly less likely than human subject studies to include both sexes or to perform sex‑based analyses, underscoring a key area where SABV implementation remains limited. We observe notable variation as to how the policy is implemented across NIH Institutes and Centers and identify associations between author gender and the likelihood of performing sex‑based analyses. Our findings bridge a critical gap in understanding how funding agency policies translate into measurable scientific outputs. By highlighting trends, challenges, and opportunities for improvement, this work contributes to ongoing efforts to promote rigor in biomedical research.

Methods

Data collection & coding

Data for this project were obtained from the NIH RePORTER tool (https://reporter.nih.gov/), a publicly accessible database providing comprehensive information on NIH-funded research projects, including details about awarded grants, project descriptions, and associated outcomes such as publications. Using this tool, we searched for R01 awards funded in FY2017 and FY2018 that reported project outcomes, yielding a total of 7671 projects (Fig. 1). The R01 mechanism, the NIH’s oldest and most frequently utilized grant program, supports independent research projects and serves as a primary indicator of investigator-driven research. To ensure the inclusion of awards with documented research progress, only those with available project outcomes were selected. This criterion limited our analysis to awards with an interim or final Research Performance Progress Report (RPPR) submitted. Data were exported on June 27th, 2023, into a CSV file containing key information such as PI name, institution, total award amount, funding institute or center (IC), and project title.

Fig. 1. Study flow diagram of the identification and screening of eligible R01 Project records and resulting publications retrieved from NIH RePORTER.

Adapted from: Page MJ, McKenzie JE, Bossuyt PM, Boutron I, Hoffmann TC, Mulrow CD, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ 2021;372:n71. doi: 10.1136/bmj.n71

Projects were assigned a random number using Microsoft Excel’s randomizer function and sorted numerically. The first 1000 awards were then selected for additional screening. For each selected award, we identified the most recent peer-reviewed article associated with the project and manually extracted data, including the PubMed ID (PMID), journal name, publication year, first author’s first name, and last author’s first name. If the most recent publication was a preprint, the next most recent peer-reviewed article was selected. This method provided a standardized approach across awards and allowed for a sample of articles from a variety of journals.

Each identified article was reviewed for the inclusion of human or vertebrate subjects. Articles were included if the study reported the use of any human or vertebrate subjects in any part of the experimental methods, including methods describing the generation of primary cell culture. Review articles, method reports, and articles restricted to fetal organisms or immortal cell lines were excluded from further analyses. This resulted in a final sample of 574 articles.

Articles containing human or vertebrate subjects were further categorized by subject type: human, non-human, or both. Each article was coded for the reported sex of subjects as male, female, both sexes, or unspecified. Studies were categorized as using both sexes even when sex was reported for only one subject group (e.g., human participants) but not for another (e.g., mice). Articles that included both sexes were also evaluated for a description of the sample size (n) by sex and if they included sex-based data analyses (including sex as a covariate or subgroup, or if data were analyzed separately by sex).

For all reviewed articles, we also noted whether the authors provided a rationale for the use of a single sex or for the absence or presence of sex-based analyses. If studies utilized only a single sex, we additionally coded for sex-specific conditions, such as pregnancy, when applicable.

Name to gender assignment

The first (given) names of the PIs and authors were used to assign gender using the GenderAPI tool (https://gender-api.com/). GenderAPI was selected due to its low rates of misclassification and non-classification^9,10. The tool assigns gender as male, female, or unknown.

For articles where the gender of both the first and last authors could be determined, the author dyads were categorized into one of the following groups: Man-Man (MM), Man-Woman (MW), Woman-Man (WM), or Woman-Woman (WW). This categorization allowed for an analysis of gender representation and collaboration dynamics in the authorship of the reviewed articles.

The terms male and female typically refer to biological sex. As such, we use the terms man/men or woman/women to describe the gender of the authors or PIs to provide greater distinction and clarity. Gender is a dynamic, social construct and may differ from assumptions based on name. This is a known limitation of binary name-to-gender assignment tools.

Statistical analyses

Data were analyzed using GraphPad Prism version 11.0.0. Descriptive statistics were provided for all variables of interest. Categorical variables were summarized with frequencies and percentages. Chi-square tests were used to compare author or PI gender, subject type, and the inclusion of both sexes, sex-based reporting, and sex-based analyses. P values <0.05 were considered significant.

Regression analysis

In the sex-based analyses model, the dependent variable is a binary variable that indicates if an article analyzed data by sex. Logistic regression models were fitted for author gender using the author dyad categories of Man-Man, Man-Woman, Woman-Man, and Woman-Woman, with Man-Man as the reference. Odds ratios (OR) and 95% confidence intervals were used to summarize the logistic regression model findings.

Ethics

The publicly accessible administrative data examined here do not constitute human subjects research and are not subject to Institutional Review Board review or approval.

Results

This sample included 574 articles supported by funding from 21 NIH Institutes and Centers, along with one FDA R01 administered by the NIH, published between 2017 and 2024 (Table 1). The articles were published in 403 journals, with a median impact factor of 4.5. The majority of PIs were men (64%, n = 369). Among the articles reviewed, the gender distribution of first authors was nearly balanced: 50% were men (n = 289), 47% were women (n = 267), and gender could not be determined for 3% (n = 18). In contrast, the majority of last authors were men (64%, n = 365). The gender identity of both the first and last authors, referred hereinafter as the author dyad, was available for 543 articles. The most frequent author dyad was man-first and man-last authors (35%, n = 199), whereas man-first and woman-last author dyads were the least frequent (14%, n = 83).

Table 1.

Summary characteristics of R01 projects and associated articles

Summary characteristics	N	Percent
Total articles	574	100
PI gender
Man	369	64
Woman	190	33
Unknown	15	3
First author gender
Man	289	50
Woman	267	47
Group Author	0	0
Unknown	18	3
Last author gender
Man	365	64
Woman	195	34
Group author	1	0
Unknown or not applicable	13	2
Author dyads
Man first - man last	199	35
Man first - woman last	83	14
Woman first - man last	155	27
Woman first - woman last	106	18
Missing	31	5
Subject type
Human subjects	283	49
Both human & non-human vertebrate subjects	25	4
Non-human vertebrates subjects	266	46
NIH institute or center
NHLBI	99	17
NCI	74	13
NIMH	54	9
NIDDK	53	9
NINDS	44	8
NEI	35	6
NIDA	32	6
NICHD	31	5
NIAID	26	5
NIA	25	4
Other	101	18

Human subjects research accounted for 49% of the articles (n = 283), while non-human subjects research represented 46% (n = 266; Table 1). A smaller proportion of articles (4%, n = 25) included both human and non-human vertebrate subjects within their experimental design and analyses. For subsequent analyses, these articles were grouped with the human subjects sample, resulting in a total of 308 articles.

Sex inclusion

The majority of articles included both sexes (61%, n = 350; Supplemental Table 1). Articles involving human subjects were significantly more likely to include both sexes compared to articles focused solely on non-human subjects [χ² (3, n = 574) = 109.6, p = 1.31 × 10⁻²³]. Of the 152 articles that included only a single sex, 34% (n = 51) focused on a sex-specific research topic (e.g., pregnancy, ovarian cancer, prostate cancer, etc.). Figure 2A summarizes the proportion of articles using both sexes, male-only subjects, or female-only subjects across the top 10 NIH Institutes and Centers most represented in the sample. Data for all NIH Institutes and Centers included in the sample are provided in Supplemental Table 2. A description of the research subject type (human vs. non-human subjects) by NIH Institute and Center can be found in Supplemental Table 3. Within the top 10 NIH Institutes and Centers represented in the sample, the proportion of articles which did not indicate the sex of the subjects ranged between 6% (NICHD, n = 2) and 23% (NIAID, n = 6). No significant differences in sex inclusion were observed based on the gender of the PI or first author (Supplemental Table 4). However, articles with women as last authors were significantly more likely to include both sexes [χ²(1, n = 574) = 3.953, p = 0.047] as well as those authored by man-first and woman-last author dyads [χ²(3, n = 574) = 10.09, p = 0.018].

Fig. 2. Sex inclusion and sex-based analyses across NIH Institutes and Centers.

A Proportion of articles that included both sexes, male-only subjects, female-only subjects, or those that did not specify sex across the top 10 NIH Institues and Centers represented in the sample. B Proportion of articles that conducted sex-based analyses across the same Institutes and Centers.

Sex-based reporting & sex-based analyses

Of the 350 articles that included both sexes, 83% (n = 290) provided a description of the sample size by sex and 44% (n = 155) analyzed data by sex (Supplemental Table 5). Articles involving human subjects were significantly more likely to report and analyze data by sex, respectively [χ² (1, n = 350) = 117.0, p = 2.90 × 10⁻²⁷; χ² (1, n = 350) = 18.03, p = 0.00002). Figure 2B summarizes the proportion of articles which analyzed data by sex across the top 10 NIH Institutes and Centers most represented in the sample. Data for all NIH Institutes and Centers included in the sample are provided in Supplemental Table 6. There was variability in sex-based reporting and sex-based analyses by NIH Institutes and Centers. Within the top 10 NIH Institutes and Centers represented in the sample, the proportion of articles that provided a description of the sample size by sex ranged between 55% (NINDS, n = 16) and 100% (NICHD n = 20). The proportion of articles which analyzed data by sex ranged between 17% (NEI, n = 3 and NIAID, n = 2) to 75% (NICHD, n = 15).

While there were no differences in sex-based reporting by PI or last author gender, or author dyad, articles with women first authors were significantly more likely to analyze data by sex (50%, n = 87 vs. 39%, n = 66; p = 0.036, Supplemental Table 5. In an unadjusted model, having women first and last authors was positively associated with sex-based analyses with an OR of 2.24 (95% CI 1.33−3.79; Fig. 3).

Fig. 3. Association between author gender and sex-based analyses.

Unadjusted odds ratios for the likelihood of conducting sex-based analyses by author dyad gender. Articles with women as both first and last authors had significantly higher odds of conducting sex-based analyses compared to other author gender combinations (OR = 2.24, 95% CI 1.33–3.79).

Rationale

All 574 articles were screened to assess whether they provided a rationale or justification for conducting single-sex studies and whether they included sex-based analyses. Only 4% (n = 24) of the articles explicitly justified their approach to sex inclusion or the presence/absence of sex-based analyses (Table 2). Among the studies that provided rationale for conducting single-sex studies, the most commonly cited reasons included experimental model considerations (n = 4), where researchers selected male or female subjects based on a particular animal disease model, and experimental design constraints (n = 2) where single-sex studies were conducted due to logistical feasibility. Other justifications included hormonal variation (n = 2) with researchers opting for a single sex to avoid confounding variables, and previous examination of sex differences (n = 2) where prior research suggested no significant sex-based variation in a particular disease characteristic. Additionally, some studies justified the use of a single sex due to the disease being predominant in one sex (n = 2).

Table 2.

Rationale for single-sex studies or decisions regarding sex-based analyses

Rationale	N	Representative example
Single-sex studies
Experimental design	2	“Given the large number of experiments required…we elected to perform experiments in female rats in a subsequent study.”
Experimental model	4	“Male mice were preferred for these studies because of the high incidence of [disease condition] as described.”
Hormonal variation	2	“Analyses were limited to male mice to provide the most accurate data comparisons across current and historical experimental cohorts without the introduction of potential confounding variables from hypothetical physiological and hormonal variations in a mixed gender sampling.”
Knowledge of sex differences	2	“While this study did not focus on the effect of sex on [disease state], prior studies have suggested…no difference in [disease characteristic] in male mice compared to female mice.”
Predominant in one sex	2	“This study was performed using female mice, given that publications demonstrate higher mortality rates in women…”
Sex-based analyses
Sample size	1	“Due to a small sample, models were not adjusted for potential measured confounders (e.g., age, sex, comorbidities)…”
Experimental design	2	“The present study was not designed to determine the impact of sex and/or circulating hormone levels…thus we did not control for menstrual cycle.” “To consider sex as an important biological variable, we first examined the effect of [drug exposure] in male and female animals.”
No sex differences identified	7	“Both males and females were used and showed no obvious differences; therefore, data for both were pooled.”
Sex differences found	2	“Exploratory analyses revealed nuanced effects of sex within the older adult group.”

For sex-based analyses, several studies indicated that they did not identify any preliminary sex differences (n = 7) and data from both sexes were pooled. Other justifications included small sample size (n = 1) or experimental design limitations (n = 1), where the study was not designed to statistically evaluate sex differences. However, preliminary sex differences were identified in three studies (n = 3), and the authors noted why they chose to conduct additional sex-based analyses.

Discussion

Sex is a fundamental biological variable that influences disease mechanisms, treatment responses, and health outcomes. Yet, gaps remain in how sex inclusion and sex-based analyses are implemented across different types of studies and by scientific discipline or medical speciality. This study builds on prior work by not only assessing sex inclusion, sex-based reporting, and sex-based analyses in peer-reviewed publications but also by linking research practices back to their parent funding awards and respective NIH Institute or Center.

The importance of sex-inclusive research has been recognized for decades. The NIH Revitalization Act of 1993 mandated the inclusion of women in NIH-funded clinical research to ensure that findings would be applicable to both men and women. Our findings align with previous reports indicating that women account for approximately 41% of participants in clinical research studies, though this varies by clinical trial stage and disease area^11,12. Similar to findings by Geller and colleagues¹³, our analyses show that despite the routine inclusion of women in research, sex-based analyses remain infrequent. A recent large-scale meta-analysis of oncology clinical trials found that only 0.5% reported outcome differences by sex. Among those that did, significant differences in survival, therapeutic response, and adverse events were observed between male and female participants, illustrating how clinically meaningful patterns can remain undetected when sex-specific analyses are not performed¹⁴.

In contrast, preclinical research using animal models continues to lag in these areas, despite the SABV policy being designed to address such disparities. Our results suggest that its impact has been inconsistent. Many preclinical studies still fail to include both sexes, report sample size by sex, or conduct sex-based analyses. Although our analyses focused on human and animal studies, we note that similar sex-reporting gaps exist in cell-based research, particularly in studies relying on immortalized cell lines where the sex of the cells is frequently unreported^15,16. These persistent gaps highlight the need for more rigorous enforcement and education to promote sex-inclusive preclinical research practices.

Among the articles screened, only 24 (4%) provided a rationale or justification in their methods regarding their decisions to study a single sex or conduct (or not conduct) sex-based analyses. Greater transparency in methodological decisions would enhance reproducibility, particularly if authors explicitly justified their approach. Most rationales provided were justifiable and increased clarity; however, a few instances still cited ‘hormonal variability’ as a reason to exclude female subjects—an argument that has been debunked in animal models and is widely recognized as an inadequate justification¹⁷. Addressing these misconceptions through education and training is crucial to ensuring robust and unbiased research practices.

Our study defined ‘sex-based analyses’ broadly, but there are different levels at which sex can be included in study design and statistical analyses. The first step—including both sexes in a study—only helps ensure that findings are more generalizable. The next step—estimating effects for males and females separately—generally requires larger sample sizes of both groups to achieve power for subgroup analyses. A final step—estimating whether effects are significantly different between sexes- will generally require even larger sample sizes than those for subgroup analyses. Depending on the study design and available population, powering a study for subgroup analyses by sex or to detect differences by sex may be considered feasible where there is evidence for sex differences and the knowledge would justify the effort and resources, and could result in different scientific or clinical implications.

Recently, Kastrati and colleagues evaluated 216 meta-analyses of randomized controlled trials and identified 115 claims of sex differences in treatment effects¹⁸. However, only a minority of these claims were supported by formal interaction tests, and many included misinterpretations of subgroup analyses (e.g., assuming a statistically significant effect in one sex but not the other without a significant sex-by-treatment interaction). Reports of statistically robust and clinically actionable sex-treatment interactions were uncommon. These findings highlight that sex-disaggregated analyses must be accompanied by rigorous statistical methods and cautious interpretation.

To achieve this rigor, statistical analysis plans should explicitly state how sex will be addressed, consistent with study design and power considerations. Pre-specifying subgroup analyses, even if not fully powered, can provide relevant, contextualized results that can be used in planning future studies. Of note, without sufficient power, findings should be interpreted as largely exploratory. Including sex as a covariate in regression models may help reduce variability in an outcome, possibly address confounding, and result in more precise estimates of the main effect of interest. In this case, the interpretation of the main effect should be as the effect one would observe when comparing two persons or animals of the same sex. Estimating sex differences in effects will generally require the inclusion of an interaction term in regression models, necessitating the combined interpretation of main effects and interactions.

This study also provides a snapshot of the gender composition of PIs and authors in NIH-funded research. The proportion of men to women PIs in our sample (64% vs. 33%) is analogous to data reported by the NIH for the same time period¹⁹. While we observed gender parity among first authors of the resulting publications, disparities became evident when analyzing author dyads. The most common pairing in publications featured men as both first and last authors. Typically, first authors contribute the most to technical and experimental work, whereas last authors secure funding, provide oversight, and drive experimental design²⁰.

Notably, we found that research teams led by women were more likely to analyze data by sex. These findings are consistent with previous studies, which have found a strong positive correlation with women-led publications and sex inclusion and analysis^21,22, and in our own previous work exploring sex and gender analysis in publications regarding COVID-19²³. This suggests that gender dynamics within research teams may influence the prioritization of sex-based analyses, further underscoring the importance of diversity in scientific leadership.

This study also examines patterns of sex inclusion, reporting, and analyses across NIH institutes and Centers. Variation in these practices by funding Institute or Center may reflect discipline-specific norms, training, and education. For example, nearly one-quarter of NIAID-funded studies did not report the sex of their subjects. Among studies that included both sexes, only 58% provided a description of the sample size by sex. This may be due to the high reliance on animal models in these studies. In contrast, studies funded by NIA consistently reported the sex of their subjects and provided a description of the sample size by sex. They also demonstrated a much lower reliance on animal models (12% compared to 62% for NIAID). Interestingly, Carmody and colleagues recently examined sex-based reporting and analyses in a subset of NIA-funded publications that exclusively utilized mouse models and found that 9% of studies failed to report sample size by sex²⁴. This suggests that the use of animal models versus human subjects may contribute to variability in sex-based reporting by the funding Institute or Center. We also observed variations in sex-based analyses by Institute or Center. These differences may reflect the type of research funded and common research practices. In our previous work examining sex inclusion and sex-based analyses across biological disciplines, we found that the fields of reproductive biology and endocrinology had higher proportions of studies that conducted sex-based analyses compared to disciplines such as immunology or neuroscience². Research in reproductive biology or endocrinology may focus on sex differences or hormonal influences as core components of study design. Given that NICHD is a primary funder of reproductive research, its higher proportion of sex-based analyses is consistent with these disciplinary norms. Collectively, these data offer insight into where additional resources and guidance may be needed to improve adherence to SABV principles.

This study focuses exclusively on the NIH SABV policy, yet similar initiatives have been implemented by other major funding agencies worldwide. For example, both the Canadian Institutes of Health Research (CIHR)²⁵ and UK Research and Innovation (UKRI)²⁶ have introduced policies requiring consideration of sex and gender in research design and reporting. Kim and colleagues found that these policies, including the NIH SABV policy, have contributed to an increase in grants and publications addressing sex and gender considerations, although implementation varies across regions and disciplines²⁷. Their findings underscore the potential for funding policies to drive meaningful change in research practices and highlight the need for continued efforts to harmonize these approaches across the global research landscape

It is important to note that while the SABV policy aims to enhance the rigor and reproducibility of biomedical research, funding agencies such as the NIH are not responsible for micromanaging individual scientists. Instead, the responsibility for accountability and transparency in research practices falls upon the broader scientific community. Publishers, journal editors, and peer reviewers play a critical role in upholding rigorous research standards. The peer-review process remains a cornerstone of quality assurance in research, but its effectiveness relies on reviewers’ familiarity with best practices in sex-reporting and sex-based analyses. Encouragingly, some journals have implemented policies to reinforce these standards. Most notably, the Sex and Gender Equity in Research (SAGER) guidelines provide a comprehensive framework for reporting sex and gender information in study design, data analysis, results, and interpretation²⁸. Other guidelines, such as ARRIVE for animal research and the ICMJE recommendations, were similarly developed to improve transparency and reproducibility, yet their adoption remains inconsistent. For example, some journals incorporate selected elements of the SAGER guidelines without adopting them in full, while others have introduced more structured requirements, such as dedicated sections in the methods to describe approaches to sex-based analyses^29,30. Future research is needed to determine how the adoption of SAGER or other sex-based reporting guidelines impacts research practices over time.

Our sample included 574 articles published between 2017 and 2024. This is a small fraction of all the human and vertebrate subject biomedical research published in this timeframe. The number of articles that could be coded was limited due to the manual coding and review process conducted by a small team of reviewers. In addition, manual data coding has the potential to introduce bias. However, we believe that having researchers manually code each article, rather than relying on AI or other automated methods, may result in more accurate coding that reflects the interpretations of individuals with expertise in research and experimental design. This view is supported by a recent study comparing manual and AI-based data extraction in systematic reviews³¹, and we anticipate that additional research comparing these methods will continue to emerge.

Although many NIH grant programs exist, this study focused on R01s only. While this is only one of many activity codes the NIH uses to differentiate the wide variety of research-related funding programs they support, the R01 is NIH’s most commonly used grant program for independent research projects. By focusing on R01 projects, we aimed to analyze the methods of research that were held to scientifically rigorous standards.

The choice of FY2017 and FY2018 was strategic for two reasons. First, these years represent a timeframe in which investigators would likely be aware of the NIH’s SABV policy, which was implemented on January 25, 2016. By focusing on these fiscal years, we aimed to capture projects funded in a period where adherence to the SABV policy would be expected. Second, selecting awards from these years allowed sufficient time for the resulting work to be published, providing an opportunity to evaluate project outcomes, including manuscripts and articles that describe the research stemming from these awards. This approach ensured a representative sample of investigator-driven projects with both adherence to SABV policy and publicly available scientific outputs.

To identify the gender of each study’s first and last authors, we utilized Gender API, a Name-to-Gender Assignment tool. When determining the gender of a name, Gender API will return an accuracy or probability value that measures how sure the model is that the name is either male or female, calculated by the number of names in the database. This means there may be cases in which Gender API misreports the gender of a name if the author’s name is less common or of a different national origin (e.g., the name Andrea is more commonly associated with men in Italy but may be associated with either men or women in the U.S.). Additionally, Gender API is trained on a gender binary and Western-oriented framework, resulting in more frequent inaccuracies in non-Western, unisex names.

Lastly, this study focuses on the SABV policy and does not address gender-based considerations in research practices. This is a limitation, and an important distinction to make, particularly in human subjects research, where gender can influence health and health behaviors in dynamic ways across the lifespan. Yet, only a handful of biomedical journals require authors to report both the sex and gender of human subjects³². Until these practices improve, it will remain difficult to assess gender reporting and gender-based analyses at scale.

Conclusions

Thus, it becomes even more important for other segments of the biomedical research enterprise, not just funders, to address SABV in the context of rigorous and reproducible science. Academic journals need to be involved in this process as the final gatekeepers prior to publication. Educating peer reviewers and journal editors, and ensuring the wide dissemination of policy updates regarding SABV to reviewers, could improve overall adherence to the NIH SABV policy and increase the rigor and generalizability of biomedical research funded by the NIH.

These findings have important implications for research policy and funding agencies. While NIH policies mandate the inclusion of SABV, enforcement of sex-based analyses remains inconsistent. Our results suggest that additional efforts are needed to encourage researchers to analyze and report data by sex, particularly in preclinical studies. Funding agencies and journals should consider implementing stricter guidelines and accountability measures to ensure that sex-based analyses are conducted when both sexes are included.

Furthermore, our findings highlight the potential impact of gender diversity in authorship on research practices. Increasing the representation of women in leadership roles within research teams may contribute to more comprehensive analyses of sex differences. Future studies should explore the underlying mechanisms driving these associations and identify strategies to promote equity in authorship and research design. Continued, collaborative efforts from funding agencies, journals, and researchers are needed to ensure that biomedical research fully accounts for SABV, ultimately leading to more effective healthcare outcomes for all.

Author contributions

Conceptualization: N.C.W. Project administration: N.C.W. Methodology: N.C.W., J.H.W., and M.P. Data curation: N.C.W., J.H.W., and M.P. Formal analysis: N.C.W. Writing—original draft: L.J.W., J.H.W., M.P., N.C.W. Writing—review & editing: L.J.W., N.C.W., J.H.W., M.P. Supervision: N.C.W.

Peer review

Peer review information

Communications Medicine thanks Laure Marignol, Tina Findley, Shirin Heidari and the other, anonymous, reviewer(s) for their contribution to the peer review of this work.

Data availability

R01 data examined in this manuscript are publicly available from the NIH Reporter (https://reporter.nih.gov/). In accordance with NIH policy, most grant‑related information submitted by applicants or recipients becomes public information once an award is made, except for proprietary or private information that cannot be released^33,34. The datasets generated and analyzed during the current study are available in Northwestern University’s Prism repository and can be accessed at https://prism.northwestern.edu/records/rmzgr-h2v84. Full names have been removed; however, the corresponding author will provide the identified data upon reasonable request.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s43856-026-01547-0.