Economic disadvantages

CTN research assumes a certain level of economic privilege to participate in experimental treatments involving neuroscience technologies [Reference Kam, Badhwar and Borghesani21]. For example, research has focused on making fMRI useful for precision psychiatry, both in predicting treatment response [Reference Siegle, Thompson and Collier22] and augmenting the effects of therapy, e.g., with real-time fMRI neurofeedback [Reference Compere, Siegle and Riley23]. Precision psychiatry is an emerging area in neuroscience and medicine that considers individual characteristics to maximize the effectiveness of treatment and interventions. Bringing this work into the community has been met with consistent challenges across insurance agency and clinician stakeholders, such as the availability of relevant hardware (research scans are frequently acquired on 3 T scanners – whether these results will generalize to the 1.5 T scanners more often available clinically is unclear), the ambiguity regarding who would pay for these uses of fMRI (often ∼$500/hour), and the lack of normative data stratified by variables such as race, age, and gender. Other nominally more available technologies are characterized by inherent biases that prevent ethical clinical translation. For example, because electroencephalography (EEG) is difficult to collect in dense hair, the vast majority of EEG data have been collected from white people (only 5/81 examined EEG articles reported having any black people in their samples; [Reference Choy, Baker and Stavropoulos24]); available EEG-based technologies, which might be used for precision medicine or intervention are not even available to communities with denser hair [Reference Choy, Baker and Stavropoulos24–Reference Penner, Wall and Guan26]. EEG hardware costs range considerably, with higher end systems that yield better quality data costing ∼$80,000, again, creating barriers for implementation in disadvantaged communities. Similarly, the tools of psychophysiology, such as electrodermal activity, have long been known to vary with or perform poorly with darker skin (e.g., [Reference Korol, Bergfeld and McLaughlin27]). There are differences, by race, in many baseline psychophysiological measures (e.g., [Reference Adzika Nsatimba, Pathak and Soares28]) whereas norms in psychophysiology are generally for white samples, yielding biases in interpretation. Exclusion criteria common to neuroimaging and psychophysiology studies disproportionately rule out minority and disadvantaged samples, based on increased prevalence in marginalized individuals (e.g., head trauma, blood pressure medicines, drug dependence). Indeed, numerous studies have excluded or diminished the importance of racially minoritized participants because the data were believed to be possibly not usable due to different culture norms or that there is a deficit-based explanation for racialized differences in findings [Reference Webb, Cardenas-Iniguez and Douglas6,Reference Carter, Gibbons and Beach29]. For example, biases have been well-documented in biomedical optics CTN research: paradoxically, skin tone bias in diagnostics and neuroimaging has been ignored on the one hand. Black participants have been disproportionately excluded from skin conductance data in fear conditioning studies on the other [Reference Webb, Etter and Kwasa18].

Diversity versus stimulus set standardization

Another fundamental barrier to effective translation of neuroscience to practice lies within the stimulus sets that are frequently employed in neuroscience paradigms. Exclusion of ethnically minoritized people is acutely obvious in studies that use facial displays of affect or affective scenes as stimuli in cognitive neuroscience paradigms paired with a neuroscience tool, such as fMRI or EEG. With recent exceptions (e.g., [Reference Conley, Dellarco and Rubien-Thomas30,Reference Weierich, Kleshchova, Rieder and Reilly31]), representations of facial affect typically employ stimuli sets comprised of white faces. In some cases, this homogeneity is explicitly acknowledged as a methodological decision (i.e., researchers may argue that this exclusion serves to reduce variability in stimuli). However, homogeneity in stimuli is increasingly recognized as a limitation and a threat to generalizability [Reference Hsu, Shumake and Caffey32]. Neural response to faces can vary based on higher-order perceptions of identity, emotion, similarity, and trustworthiness. These functions are influenced by the social environment, including social experiences of racism, sexism, etc. and individual-level cultural norms. White-standardized stimulus sets may thus be classified as more foreign, novel, or threatening to non-White individuals. These may have a significant effect on data analysis, results, and interpretation [Reference Chen, Zhao, Song, Guan and Wu33]. Because of the ubiquity of racially homogenous stimuli sets, this barrier permeates a range of translational neuroscience methods. A growing consensus among neuroscientists and related fields, however, recognizes the significance of this issue across paradigms such as neuroimaging [Reference Ricard, Parker, Dhamala, Kwasa, Allsop and Holmes19], computerized cognitive training [Reference Peckham34], early childhood assessment [Reference Kamboukos, Ursache and Cheng35], and as discussed below, have called for the adoption of more representative and diverse sets of facial affect stimuli.

Beyond depictions of white faces, a potentially more insidious issue with stimuli used in neuroscience paradigms involves linguistic stimuli, such as sentences, phrases, or single words used in many cognitive neuroscience paradigms, or cognitive training paradigms such as interpretation bias modification. Many such stimuli sets are normed within convenience samples of college students in North America, the majority of whom are white [36]. Few studies have explicitly examined how intersecting factors of race, ethnicity, class, gender, or sexual orientation may influence ratings for such stimuli, yet the ubiquity of college student samples and the limited diversity of such samples raise the significant possibility that translation of paradigms using such verbal stimuli is inherently limited in their potential for uptake in more diverse settings. One study examined racial bias in stimulus sets used in multiple research studies, largely finding that corpi of affective stimulus sets tend to have photos rated, and thus selected, as “negative” which include more ethnically minoritized people, but photos that are rated and used as exemplars of “positive” content include more apparently white people. Explicitly including racially diverse content across valences may require going outside standard normed picture corpi. Racial bias in using standardized stimulus sets is problematic, given well-known demand characteristics associated with responding to nonrepresentative researchers and stimuli and, indeed, studies have shown that researchers who are not diverse tend to propagate perspectives that do not account for diversity [Reference Roberts, Bareket-Shavit, Dollins, Goldie and Mortenson37] and may introduce demand characteristics based on their own positionality (e.g., [Reference Sue38]).

Limitations of predictive and normative modeling

A major obstacle to progress in precision psychiatry is that systematic algorithmic biases in current machine learning approaches are overly tuned to majority populations and often fail to generalize to minority populations. For example, predictive models formed on neuroimaging data collected from white participants break down in their performances when used to predict phenotypes for Black participants [Reference Li, Bzdok and Chen39]. Another statistical approach increasingly applied in precision psychiatry, called normative modeling, maps variation in quantitative brain metrics associated with psychiatric conditions. Normative models employ a strategy similar to growth charts in pediatric medicine, where an individual child’s physical measurements are compared to a reference population, and significant deviations may indicate a medical concern. Similarly, in the sibling discipline of neuropsychology, individual cognitive performance is compared to normative reference standards that include faulty separate sets of norms for different racial groups [Reference Conley, Dellarco and Rubien-Thomas30]. These have led to calls for removing race as a factor in normative modeling [40], and for taking extra care in determining when a neuropsychological construct can be expected to generalize (i.e., is influenced by race-related psychosocial factors), and when it should not [Reference Malik and Norman41]. In CTN, normative models may be used to identify the extent of neural variation of an individual diagnosed with a clinical condition relative to a normative (nondiagnosed) population [Reference Shan, Uddin and Xiao42]. However, similar to growth charts and neuropsychological reference norms, such approaches have inherent flaws. First, the ecological validity of the reference group must be considered. When reference groups fail to include individuals from diverse sociodemographic perspectives in the normative population to which individuals are compared, healthy individual differences can be conflated with pathology and further disadvantage marginalized groups. This concern has been discussed by some of the groups who are leading the field of normative modeling [Reference Rutherford, Fraza and Dinga43], but not yet functionally addressed. Second, it remains unclear what percentage or standard deviation difference from a normative model is cause for concern and whether this varies by sociodemographic measurements. Third, the clinical importance of observed differences remains unclear and is subject to clinician bias. If an individual’s pattern of neural variation differs from a normative model, who determines it is cause for concern, and what clinical workup follows is not defined objectively through well-established clinical guidelines. Without objective strategies for clinical decision-making, the known microaggressions and discrimination in clinical bias can impact implementation.

Understand structural barriers to mental health

The structural competency framework associated with critical consciousness fosters an understanding of systems-level factors that can perpetuate social oppression and mental health disparities [Reference Metzl and Hansen9]. Systems-level determinants of mental health may include sociodemographic factors, housing insecurity, transportation availability, and neighborhood factors, such as poverty and unemployment [Reference Alegria, Alvarez, Cheng and Falgas-Bague53]. Structural competency is an extension of cultural competency and implies that the scope of understanding cultural influences on health outcomes must be expanded to the systems level to understand policy-related patterns that shape differential access to mental health resources and interventions. Social barriers to receiving neuroscience-informed interventions must be addressed to succeed in making clinically significant improvements on a population scale. For individual research studies, understanding such structural barriers may translate to shifting the timing and access of research study visits to accommodate those who may need more support to attend visits (e.g., extending study visits beyond typical business hours, providing transportation or meal vouchers for longer visits). At an institutional level, researchers can work with their departmental units and local IRBs to shape recruitment and retention practices and improve equitable access to research for marginalized groups.

Integrate CBPR approaches and build trust

CBPR can serve as a guiding framework for conducting inclusive CTN. “Nothing about us without us” is a simple and strong guiding principle. CBPR approaches enhance the broad reach of science and reduce health disparities for marginalized groups in the process but does not apply only to health equity research. It encourages collaboration with individuals and groups outside of academia, and attention to participants as decision-makers in the research process. These can be family members, caregivers, teachers and other educators, paraprofessionals, spiritual leaders, and policymakers. Not having to remove street clothes during fMRI scans, being able to access and understand their own data, or hearing about incidental findings that allow them to take clinical action, may open opportunities for modifying study procedures are some examples of how input can shape study procedures. Involving additional stakeholders helps to develop a shared language and build trust in the research process, in the value of neuroscience in informing psychological and physical health, and in developing a common understanding of the causes of mental health disorders. Scientists who lack prior experience examining health disparities should be cognizant of quick solutions that may fail them in the long term. Sustainable, proactive solutions will come from an acquired understanding of health equity work that has already been conducted with communities of interest. To avoid erasure of historical efforts, community partners should be acknowledged on a timeline and in a manner that is beneficial to them. This may not necessarily lead to a metric such as authorship in a peer-reviewed manuscript that is only unidirectionally meaningful if results are not further shared in outreach or given back to those – individuals and communities – who offered their primary data.

Develop mindful theoretical models

One major challenge in neuroscience is how researchers use socially defined groups in analyses and then hypothesize a biological – rather than social – basis for any observed group differences. One example of this is in research comparing people by sexual orientation. While significant research has shown very limited evidence for a biological basis for sexual orientation [Reference Ganna, Verweij and Nivard54], comparing neural structure and function between heterosexual and nonheterosexual people has been an ongoing topic of interest in neuroscience (for systematic review and critique, see [Reference Nicholson, Siegel and Wolf55]). However, recent research has shown that it is the impact of differential social experiences (i.e., victimization) and not orientation itself that influences differences in neural functioning [Reference Eckstrand, Silk and Nance56]. Other examples of biological embedding have emerged in the threat processing literature [Reference Carter, Gibbons and Beach29], which underscores the bi-directional relationship between life experiences and biological processes.

It is critical for CTN to explore research strategies that understand how biology interacts with the differential social experiences that come with marginalization to produce mental health disparities [Reference Eckstrand, Singh and Diversity57]. To do this effectively, researchers need to be mindful of methods for (1) collecting and reporting sociodemographic information on people; (2) optimizing community engagement to inform research questions and analyses; (3) interpreting results that neither overinterpret findings towards social categorizations as a function of biology nor ignoring important group differences that could aid understanding of disparities; and (4) translating results equitably to reduce suffering in the communities disproportionately impacted [Reference Call, Eckstrand and Kasparek58].

The most effective research study designs may therefore not translate to group comparisons between marginalized and nonmarginalized groups, as marginalized communities may be valuable to study in their own right without reference to comparative populations. Health equity and developmental psychopathology researchers have long advocated for study designs that focus on within-person differences over time, recognizing that relying solely on cross-sectional, between-group comparisons may obscure important variations in lived experiences and clinical presentations [Reference Carter, Gibbons and Beach29,Reference Cicchetti and Rogosch59–Reference Woody, Bell, Cruz, Wears, Anderson and Price63]. The use of between-group designs should involve careful consideration of how metrics from marginalized populations are compared with those from majority groups to avoid reinforcing inequities by perpetuating the notion of the majority experience as the standard.

Use multimodal assessments of lived experiences

Laboratory-based paradigms designed to capture biopsychosocial influences on health inequities will benefit from the use of personalized and ecologically valid stimuli that capture the unique challenges of minority stress (e.g., laboratory analogs) [Reference Harrell, Hall and Taliaferro64,Reference Jones, Harrell, Morris-Prather, Thomas and Omowale65]. Studies may also include measures of life experiences that theoretically modulate the biological processes or interventions of interest, such as measures of early adverse life experiences, traumatic experiences, or experiences with everyday racism and discrimination. However, there are limitations to the validity, acceptability, and clinical utility of screening measures for adverse childhood experiences, as well as barriers to implementing trauma-informed care in clinical research settings [Reference Danese, Amussen and MacLeod66,Reference Huo, Couzner, WIndsor, Laver, Dissanayaka and Cations67]. Paradigm development should involve incorporating feedback from community experts and insights gleaned from qualitative research.

Utilize diverse stimulus sets

Despite the historical reliance on stimuli that reflect white faces or stimuli normed among predominantly white samples, recent commentaries and reviews have encouraged researchers to seek out and utilize stimulus sets that reflect greater racial diversity [Reference Ricard, Parker, Dhamala, Kwasa, Allsop and Holmes19,Reference Peckham34,Reference Kamboukos, Ursache and Cheng35]. An example of such a database is the Racially Diverse Affective Expression Face Stimuli Set (RADIATE [Reference Conley, Dellarco and Rubien-Thomas30]). Unfortunately, there are few comparable examples of verbal stimuli developed and normed by racially diverse participant pools; developing these stimuli will be a positive accomplishment for future research.

Adopt inclusive psychophysiological methods

Psychophysiological methods such as EEG and electrodermal activity are linked to the unintended exclusion of Black participants from translational neuroscience research [Reference Harrell, Hall and Taliaferro64,Reference Jones, Harrell, Morris-Prather, Thomas and Omowale65]. Although there is increasing emphasis on the introduction of novel hardware to address these disparities, there are immediate steps that researchers can take to improve measurement with existing systems [Reference Choy, Baker and Stavropoulos24,Reference Etienne, Laroia and Weigle25,Reference Bradford, DeFalco and Perkins68]. For example, informed by CBPR, scholars now recommend more inclusive and equitable recruitment materials that directly acknowledge concerns regarding the potential impact of EEG on hair with dense coils, particularly for people with hairstyles requiring substantial time and financial upkeep. Additionally, scheduling research appointments flexibly to accommodate changes in hairstyles can increase inclusion. Moreover, adopting the 15-20 minute braiding technique to increase electrode contact with the scalp can be used with existing systems [Reference Etienne, Laroia and Weigle25]. For instance, as part of the EEG Hair Project, researchers at the Biomechanics, Rehabilitation, and Interdisciplinary Neuroscience (BRaIN) Lab at the University of Central Florida enlisted the expertise of a local Black hairstylist to train research staff in this braiding method, aiming to promote greater equity and inclusion for Black participants [Reference Kopal, Uddin and Bzdok69].

Identify and mitigate bias in predictive models

There is an urgent need to develop a more complete understanding of the role that sociodemographic factors characterizing aspects of population diversity (e.g., participant gender, race, ethnicity, and income) play in cognitive and mental health phenotype prediction [Reference Kopal, Uddin and Bzdok69]. Identification and explicit modeling of the major sources of population stratification is a necessary milestone towards developing justice, equity, diversity, and inclusion (JEDI)-informed machine learning frameworks. This can be done by constructing predictive modeling solutions that jointly acknowledge a broad portfolio of interindividual differences and is a necessary precondition for the responsible use of population neuroscience data to avoid further disadvantaging underrepresented groups and communities. For example, systematic bias in who gets classified as carrying a particular medical or mental health diagnosis could reinforce disparities in access to health care and services. Underdeveloped practices in predictive modeling and inappropriate handling of demographic stratification can lead to further exclusion and stigmatization of groups that have historically been marginalized in healthcare institutions. Since the reproducibility, generalizability, and clinical utility of published machine learning models have rarely been put to the test in broadly sampled, more diverse cohorts, deepening our understanding of the full extent and consequences of generalization failure in neuroscience is a critical paradigm shift necessary for developing a truly JEDI-informed precision psychiatry that serves the diverse global population. It is also necessary to frame results and questions carefully, especially when considering participant race and ethnicity, to counteract harmful biases [Reference Cardenas-Iniguez and Gonzalez70] and improve health-related predictions at the single-subject level. Documenting the extent of neural variation along with diverse sociodemographic factors may be an important step for making stronger inferences about variation that falls within healthy and pathological ranges, so long as large and diverse datasets can be examined.

Recruit and retain a diverse research workforce

The likelihood that translational insights will be applied beyond well-resourced majority populations will be improved when marginalized communities are represented not just as community partners, but as mental health scientists in their own right Governmental agencies, funding bodies, and other key players in the decision-making process of who enters and advances down the pipeline from trainee to principal investigator have started to attend more to JEDI issues than before. Yet the leakiness of this pipeline, especially for trainees from marginalized communities, is well documented [Reference Berenbaum, Washburn and Sbarra71,Reference Gee, DeYoung and McLaughlin72]. Inclusivity of scientific ecosystems at all levels of training will be improved when these factors come into sustainable focus [Reference De Los Reyes and Uddin73].

Provide training in neuroscience literacy

Supporting the development of neuroscience literacy among clinicians is another important step toward the adoption of neuroscience in mental health clinics in marginalized communities. Clinicians may be hesitant to adopt neuroscience-informed tools in the clinic because they are unfamiliar with neuroscience terminology and because they may not have received training in the tools neuroscience has to offer [Reference Goss and Panell74]. In addition, a training emphasis on CBPR-informed neuroscience studies may also increase the uptake of these tools by emphasizing education modules most likely to positively impact care for marginalized communities. It is important to consider not only doctoral-level clinicians but especially the much greater number of mental health workers at the master’s level who conduct trauma work in marginalized communities. Development of neuroscience literacy can be supported by offering free Continuing Education credits across mental health professions to educate clinicians about how neuroscience can be applied in the clinic, as well as integrating a neuroscience curriculum into training programs. Enhancing neuroscience literacy may also be beneficial for patients and research participants. For example, information on how brain and disorder mechanisms change with treatment, sometimes referred to as neuroeducation, may directly impact clinical care by supporting treatment [Reference Kryza-Lacombe, Richards, Hansen and Goldin75], increase patient openness to participating in neuroscience studies.