Exclusive research is bad science: the epidemiologic argument for diversity, equity, and inclusion in research

Diversity, equity, and inclusion (DEI) is a social movement that focuses on identifying and addressing structural and interpersonal inequities. DEI issues in research are often framed as underrepresentation of marginalized groups among study participants or research team members. While this epistemic exclusion, referring to the exclusion of scholars from marginalized backgrounds, preventing their participation in knowledge generation [1] is important, exclusion in medical research has more far-reaching and overlooked impacts on the integrity of medical research. We conceptualize exclusive research as a lack of application of DEI when selecting the research approach, from study design to dissemination, and research paradigm, including for quantitative studies. Exclusive research can manifest as a study that does not adequately consider inequities in structural, environmental, or biological phenomena that influence study design, recruitment, or the interpretation of findings. Applying DEI throughout an entire research program requires resources, skills, and time—and it is tempting to consider DEI a ‘nice-to-have’ rather than a non-negotiable foundation of medical research. Recent ideological attacks on DEI in research may encourage investigators to avoid the work of engaging with DEI throughout their work. However, exclusive research has led to important errors in the interpretation of epidemiologic and clinical research, resulting in harm to patients and communities. Using examples from the medical literature, we outline two types of exclusion research in medical research based on a misunderstanding of identities; first, imprecisely defining the aspect of biological identity that is associated with a disease or outcome, and second, conflating a social identity with a biological mechanism of disease or outcome during causal inference. The Multicenter Automatic Defibrillator Implantation Trial II (MADIT-II) is a randomized trial comparing implantable cardiac defibrillators (ICDs) to conventional medical therapy in patients with heart failure with reduced ejection fraction (HFrEF) [2]. This landmark trial found that patients who had received an ICD had improved survival compared to those who did not, and American, Australian, European, and Canadian cardiology guidelines have generally recommended ICD insertion for patients with HFrEF on maximal medical therapies since [3]. As ICD insertion was adopted as the standard of care, observational data began to emerge that female patients may have more implantation-related adverse events [4] and fewer appropriate ICD shocks [5] compared to male patients [6]. These findings are inconsistent across observational studies [7], and no randomized trials have enrolled only female patients—and none are currently registered on ClinicalTrials.gov—resulting in a lack of high-quality evidence to guide clinical practice for female patients. When re-examining the results from the original trial considering these emerging data, investigators noted that more than 85% of the original trial population was male [2]. This suggests that the original study was underpowered to detect the signal for harm in female patients and/or overestimated benefit in the general population when benefit existed primarily for male patients [8]. This sex-based inequality in enrollment is not explained by differences between males and females in indications for ICD insertion, and observational data suggest that female patients are undertreated with ICDs compared to males, even after accounting for indication, underlying heart disease, and other comorbidities [9]. Though contemporary trialists and regulatory bodies emphasize diversity in participant recruitment, the proportion of female participants in cardiology trials has not increased over time [10]. Further, even with increased attention to clinical trial diversity, the legacy effect of landmark trials like MADIT-II persists; most international guidelines continue to recommend ICDs for patients with HFrEF without clarifying that clinicians should consider patient sex when balancing the risks and benefits of ICD implantation. Sex may influence outcomes of ICD implantation in patients with HFrEF, but our understanding of this causal relationship is incomplete (Fig. 1) because it is not clear what aspect(s) of being female or male is responsible for this difference [6, 11]. Though sex is often conceptualized, collected, and reported as a binary variable in medical research [12], in reality, biological sex is a multidimensional composite variable (Table 1) [11, 13]. Simplified causal diagrams for the relationship between (A) patient sex, (B) external genitalia observed at birth, (C) chromosomes, (D) hormone exposures, and (E) body size and/or composition with outcomes after implantable cardiac defibrillator insertion. Examples of ways to re-conceptualize measurement and analysis of associations between race, sex, and gender identities with exposure and outcome in clinical studies Gender roles (e.g., caregiving, breadwinner) There is a close and overlapping relationship between race and ethnicity. We have separated these concepts here, but racial groups often share cultural practices and ethnic groups often experience racism or other forms of discrimination (e.g. antisemitism). Though not specified in the MADIT-II trial, participant sex was likely determined by a participant’s legal gender in their medical record, which is typically assigned based on the appearance of their external genitalia at birth. However, it is biologically implausible that observed sex differences in outcomes after ICD implantation would be related to the morphology of a person’s external genitalia (Fig. 1B). This relationship could be interrogated by examining outcomes of ICDs in intersex patients with HFrEF—who are notably absent from most trial and observational data on ICD outcomes [21], despite making up an estimated 0.5%–2.0% of the global population [22]. There could be a genetic explanation for the difference in outcomes observed between male and female patients (a surrogate for XY and XX chromosomes, respectively; Fig. 1C) [11]. For example, people with Turner’s syndrome (45XO) are known to have a greater incidence of left-sided structural heart disease, and this risk decreases in people with Turner’s syndrome who have partial absence of a second X chromosome rather than X chromosome monosomy [23]. Similarly, people with Klinefelter’s syndrome (47XXY) may have a greater prevalence of diastolic dysfunction and chronotropic incompetence [24]. A relationship between hormone status and outcomes of ICD implantation seems physiologically plausible, given the well-documented relationship between exposure to sex hormones and the pathophysiology of many types of heart disease, including channelopathies, heart failure, and dysrhythmias [6, 11]. In this causal diagram, female sex would act as a surrogate marker for a greater estrogen-to-testosterone ratio (Fig. 1D)—though this relationship would differ throughout the lifespan for participants who were premenopausal, on hormone replacement therapy, or postmenopausal. Rather than a binary variable, hormone status is a continuous, dynamic variable that ranges from hypogonadal to supraphysiologic due to exogenous hormones or tumours. Understandably, the clinical relationship between female and male sex hormone profiles and heart disease is not straightforward [25]. Female or male sex may be a surrogate variable for body size or composition (Fig. 1E) [6]. Body size influences heart size and thickness, cardiac output, and the size of vascular structures, all of which could confound the relationship between patient sex and outcomes for ICDs [26]. In this case, male patients with smaller body sizes may have more adverse events related to implantation or female patients could have a greater cardiac function after correction for body size, reducing the benefits of ICDs seen at lower ejection fractions. This simplified example demonstrates how researchers have dichotomized a group of continuous biological variables—anatomy, genetics, hormones, and body size, among many others [11], including gender—into two clusters of traits represented by the binary variables ‘female’ and ‘male’. While this approach may reveal insights when studying population-based samples, without a more nuanced understanding of how sex operates as a surrogate variable, true associations will be missed. Further, this example demonstrates how exclusive research precludes good science. First, the pattern of worse outcomes and less benefit for female patients who received ICDs was missed due to the under-representation of female patients in the landmark clinical trial. Second, due to the ongoing lack of precision in defining sex, the true association between ‘female’ sex and worse outcomes has not been identified. This leaves clinicians to make decisions about who may benefit from ICDs without data about which aspect of the composite variable named ‘female’ is associated with potential harm. When reported as a binary variable, researchers must remember that sex is operating as a surrogate variable for some other type of difference [11] that is more often found in females or males. When researchers forget that sex is operating as a surrogate variable, errors in causal inference lead to missing true associations. Investigators should carefully consider the proposed mechanism of how sex may influence their outcome of interest, collect relevant patient data based on this mechanism, and adjust for or stratify by these variables in addition to sex in their analysis (Table 1). While small sample sizes may result in underpowered statistical analysis in individual studies, reporting results separately for sex minority participants (e.g. intersex participants) can avoid erasure of this patient group, facilitate meta-analysis, justify dedicated clinical studies in underrepresented subgroups, and/or direct post-marketing surveillance for these patients. ‘Race’ is associated with the incidence, prevalence, and prognosis of a range of diseases. For example, the increased incidence and severity of uterine fibroids among Black women compared to white women has been reported in the literature [27] and taught in medical schools and textbooks for decades (Supplementary Box S1). However, race has no biological, genetic, or physiologic meaning—races are social identities based on observed physical attributes and are used to enforce social norms, hierarchies, and laws [28]. Racial categories are notoriously difficult to define and have shifted over time, further reinforcing their lack of biological meaning. Because there is no physiologic or genetic meaning to being Black, Black race must be a surrogate marker for some other factor in the relationship between Black women and the development of uterine fibroids (Fig. 2A), with implications for causal inference. In fact, despite efforts to identify a genetic or biological link between the Black race or African ancestry with fibroid development [29], there is no genetic or molecular explanation for the increased prevalence of fibroids in Black people [30]. Despite this, the association of uterine fibroids and the Black race continues to be taught and reported as an unmodifiable, biological, risk factor for Black women. (A) Simplified causal diagram for the association of Black race and uterine fibroid development and (B) proposed simplified causal diagram for the association of racism and uterine fibroid development. In cross-sectional studies, uterine fibroids are more common among women with greater levels of urinary metabolites for phthalates with a dose–response relationship between urinary phthalate metabolites and fibroid volume [31]. Further, certain phthalates speed the growth of abnormal uterine tissue In vitro [32]. Phthalates are a group of endocrine-disrupting chemicals that are found in a range of medical devices and beauty products—including hair straightening products (chemical relaxers) [33] and other products more often used by women of colour [18]. This differential environmental exposure to chemicals based on race and gender has been termed the ‘intersectional exposome’ by population health scientists Zota and VanNoy [31]. Despite topical application, hair straightening products pose a risk of systemic absorption due to their abrasive nature and tendency to disrupt the skin barrier; this concern is supported by observational data that suggests that the risk of uterine fibroids was greater in women who had experienced chemical burns from hair straightener product use [34]. Though there are many potential environmental sources of phthalates, the evidence for hair straightening products as an important contributor to the observed increase in uterine fibroid prevalence for Black women is strong and building [35]. Exposure to phthalates via hair straightening products is associated with uterine fibroids through a biologically plausible mechanism and is associated with the Black race via a social mechanism, namely, hair discrimination (Fig. 2B). Discrimination based on natural hair textures is a common reason for Black students to be suspended from school [36] and lack of conformity to Eurocentric standards for hair is an important barrier to education, workplace participation, and acceptance for Black people [37]. It is so common that it has been addressed in landmark legislation in the USA [38] with other countries such as the UK possibly following suit [39]. In this framing, the Black race is a confounder rather than the cause of uterine fibroids (Table 1). This has important clinical implications, as social mechanisms of disease require sociocultural and policy solutions rather than medical treatments (Fig. 2B). For example, policies that more strictly regulate harmful ingredients in personal care products or additional legal protections for Black people who experience hair discrimination would address factors on the causal pathway of uterine fibroids. Further, stating that the Black race is a biological risk factor for uterine fibroids along with ‘age… nulliparity, time since last birth, and premenopausal status’ [27] frames this association as unmodifiable and, more concerningly, reinforces the false notion that being Black is a biological rather than social phenomenon [28]. We do not advocate that researchers be ‘colorblind’ in their data collection, analysis, and reporting, since race is a real social experience with important biological and clinical outcomes [40]. Critical insights like the relationship between uterine fibroids and phthalate exposure rely on examining these effects. Instead, we advocate for the application of social epidemiology [41] and DEI in research, where conceptualize race accurately by explicitly stating that race cannot be the direct, causal exposure with disease. Researchers should emphasize that further studies to understand the biological causality and structural influences of disease are needed (Table 1; Box 1). Researchers who wish to make the case that race is a surrogate for a shared ancestry or genetic history [29] should clearly state and defend this assumption [42] in their writing. Similarly, researchers should consider the relationship between female patients and increased adverse events and less benefit from ICDs as hypothesis generating: What characteristic is more common among the female population that may be responsible for this finding, and how can this be investigated? Box 1. Steps for investigators and research teams to address exclusive research. Teams should consider their resources and skills when selecting which steps will be feasible for their projects. 1. Review extant literature in your field for epistemic exclusion and injustice: Which social and demographic identities have been excluded from studies in this field? How have associations between identities and outcomes been explored? Are there biologically plausible mechanisms to better explain associations between identity and outcomes? 2. Collect data on potential explanatory variables that may characterize associations between identities and outcomes: Are there more precise explanatory variables already available in your datasets for exploratory analysis? What instruments or tools could be added to your study protocol to capture other explanatory variables? 3. Review whether diagnostic tests used in your study have correction factors for sex or race which could lead to misclassification or measurement bias: Run and report the results of sensitivity analysis with and without correction factors to understand how these may influence results or associations. Use diagnostic tests that do not use correction factors when possible. Report the potential for measurement bias in your study as a limitation when diagnostic tests that include correction factors are used. 4. Report results by granular identities to avoid erasure of underrepresented groups. This will also increase the opportunity to combine participants across multiple studies in future meta-analysis. 5. Edit academic writing to ensure that relationships, mechanisms, and associations are correctly described: Does your presentation of results accurately describe potential associations? Are social variables like race presented as biological facts? By reporting only female and male participants in studies, medical science has erased the existence of trans, gender diverse, and intersex people. This erasure in medical research has been used by transphobic and homophobic groups to argue that science has defined only two genders: female and male (Supplementary Box S1). Not only has this ‘scientific’ argument been used to restrict participation of trans, gender diverse, and intersex people in sports [43], education [44], and public spaces [45], but access to gender-affirming and transition-related care often requires people to conform to Eurocentric binary gender norms and undergo medical procedures that lead to sterilization [46]. Further, this erasure pathologizes non-binary sex and gender identities which can lead to harmful practices like the medically unnecessary ‘corrective’ procedures performed on the genitals of intersex children [46]. Similarly, a lack of understanding of social identities like race, racism, and critical race theory in medical research undermines epidemiology and mirrors previous ‘scientific’ racist movements that aimed to use biology to support racial differences between bodies such as phrenology and eugenics. In this way, addressing exclusive research in medicine is a moral for While and regulatory bodies have been structural to participant and study team additional are to address exclusive research. and should ensure that researchers are accurately reporting the association of disease with demographic rather than conflating social identities with biological of health science researchers and social could be by with DEI should and on to ensure that associations between identity and disease are being could consider to the mechanisms of and including trials that only underrepresented to epistemic of DEI in research should be taught in and as skills, on with other critical skills like study design, and A research program at study and through to but investigators can on research by of research resources or 1). researchers should use to describe their findings and the of their attention to whether associations include surrogate variables that further study rather than the causal factors In addition to the literature which typically the study researchers should carefully their field of study for exclusive research 1). In where certain demographic groups have been underrepresented among study participants (e.g. the MADIT-II research teams could to participants with underrepresented identities to potential differences in disease epidemiology or outcomes between groups. Researchers should how associations between demographic social and of disease causality or have been in the extant literature and consider whether the proposed causal mechanisms of these reported associations are biologically This aspect of the literature may study for example, in a field where there is an association between sex and researchers could consider which biological variable (e.g. hormone body this association in their proposed variables may already be available datasets (e.g. status based on body but investigators may to additional to their or to the case report to capture the true variable of (Table 1). Not all will be relevant or for all investigators should consider the given their study design, and of their teams should also consider a DEI to their study design, data and analysis, and reporting, to ensure that associations are and (e.g. hormone status sex The use of causal diagrams (e.g. direct may be in clarifying the of to a lack of DEI it can be to identify with appropriate Investigators should consider from relevant or in the social There are many to addressing exclusive research. identities is Second, despite decades of many marginalized underrepresented as research participants and to address this have had is a to the of clinical and further or in are to be to and Further, DEI is a dynamic field that requires dedicated that many health and medical research teams Despite these medical research must address epistemic not do so is not only harmful to patients but also science. is a and health with a on medical is an medicine and with an in is the for the a Canadian that research and to health for gender people. is a in at the to advocate for people with and gender in the and in medical research. is an and the of health at the for is a general medicine and with a on in the medical to the literature and writing of this will act as the for this work. data is available at There is no associated with this There is no data associated with this was not used at of this work.

Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.