Feb 3, 2025DOI: 10.1161/circulationaha.124.070299Journal ArticleAHA
Written by: Allen SanyiSamuel ByiringiroSanaz DabiriMireille JacobsonAmanda BoydModele O OgunniyiAlanna A MorrisRachel KohnNeal W DickertMeghan B Lane-FallEldrin F LewisScott D HalpernAlexander C Fanaroff
Find these authors on Roon
MO
Dr. Modele Ogunniyi
General Cardiology · Emory University
AF
Dr. Alexander Fanaroff
Interventional Cardiology · UPenn
Introduction
Randomized clinical trials (RCTs) are the cornerstone of evidence-based medicine and are critical for understanding how preventive, diagnostic, and treatment modalities affect patient and policy outcomes1,2. However, patients who enroll in clinical trials are often substantially different from the population of patients with the disease of interest. For example, though Black and Hispanic individuals are disproportionately more likely to have uncontrolled risk factors for cardiovascular disease and more likely to develop cardiovascular disease,3,4 they are substantially underrepresented in cardiovascular RCTs. In one review of 143 clinical trials conducted between 2008 and 2017 to evaluate 35 novel cardiometabolic drugs, just 2.1% of enrollees were Black and 2.1% were Hispanic.5This lack of representativeness is important for both practical and ethical reasons.6,7 From a practical standpoint, the lack of racial and ethnic diversity in RCTs may hinder external validity. Specifically, even though there are few biological differences across races and ethnicities, socioeconomic factors and lived experiences of different groups could make it problematic to extrapolate the risks and benefits of therapies and strategies tested largely in White patients to Black and Hispanic patients.8–10 In addition, representativeness may enhance trust in and engagement with the medical enterprise among historically underrepresented groups. There can also be direct benefits that accrue to clinical trial participants, including access to novel, potentially life-saving therapies, and better clinical outcomes, which will not be equitably distributed if clinical trials do not enroll representative populations.For these reasons, the National Institutes of Health (NIH) and the U.S. Food and Drug Administration (FDA) issued mandates in the 1990s that trial sponsors aim to enroll patient populations in RCTs that reflect the racial and ethnic diversity of real-world populations of patients with the disease under study.11–13 Since then, RCTs have more reliably reported data on the participation of individuals from underrepresented racial and ethnic groups, though meta-analyses examining change in participation in RCTs by race and ethnicity over time have found mixed results.14–19 FDA draft guidance published in 2022 further recommends that trial sponsors set goals to enroll a diverse and representative population and provide specific plans for how to accomplish these goals.20 However, no consensus has been reached on how to quantitatively define an appropriately representative population. This is particularly relevant in the context of increasing “globalization” of RCTs. As stakeholders—including regulators, funders, and journals—mull stronger policies to enforce representativeness, how representativeness is defined and measured will become even more important.
Multiple metrics have been proposed to quantitatively assess representativeness in clinical trials. This review focuses on a selection of these measures of representativeness, chosen by the authorship team, which includes members from each of the five centers in the American Heart Association’s Strategically Focused Research Network on the Science of Diversity in Clinical Trials. The authors selected these measures based on the measures’ frequent use in the literature, their relevance to current clinical trial practices, and their use by investigators working in the field of improving representativeness in cardiovascular RCTs. Each metric has strengths and weaknesses, and each may be appropriate in certain situations, depending on why representativeness is important for the RCT in question. In this review, we will scrutinize various methods to measure representativeness in clinical trials, considering how each method may be relevant for both the practical and ethical reasons for measuring representativeness, how different metrics can be best utilized by key stakeholders, and unintended consequences related to quantitative measurement of representativeness. We conclude with a section on how considering quantitative measures can help researchers choose existing and emerging RCT design features that optimize representativeness.
Clinical trials are critical for determining the safety and efficacy of drugs, devices, and treatment approaches, but as the branch of medical science most closely integrated with patient care, they also play an important role in representing medical science to patients and the public. Accordingly, the goals of increasing diversity and representativeness in clinical trial participation include not only generating biomedical knowledge but also promoting fair access to novel treatment modalities, earning and building trust, and promoting engagement (Table 1).7
From the standpoint of advancing biomedical knowledge, a report from the National Academies of Science, Engineering, and Medicine emphasized that increasing representativeness in RCTs will improve the generalizability of findings, produce novel biologic insights, and lead to therapeutics targeted to different racial and ethnic groups.21 It’s critical, however, to approach the expectation of treatment effect differences with nuance, acknowledging that such differences are infrequent. When observed, they are often attributable to ancestral or genetic variants, or to socioeconomic disparities and structural racism, rather than inherent differences between groups historically under-or-over-represented in clinical research. Consequently, the argument for significantly enlarging RCTs to detect these potential heterogeneities in treatment effects is not merely a question of resource allocation, but an investment in ensuring the precision and relevance of medical research. Moreover, RCTs would need to be substantially larger to detect such heterogeneity in treatment effects, which would strain already limited resources for conducting RCTs seeking answers to patient-centered questions. Furthermore, over-emphasis on heterogeneous effects of treatments by race is in tension with important efforts to avoid inappropriate and potentially harmful use of race as a biological variable.22 By contrast, heterogenous treatment effects by other social determinants of health may be more plausible. Individuals of different sexes and genders have important biological differences that may affect treatment response. In RCTs of care delivery interventions or care strategies, the social context in which the intervention is delivered—including rurality, socioeconomic status, health literacy, and other social determinants of health—is also critical to determining its success or failure.
The ability of increased representativeness in RCTs to promote goals related to fair access to the benefits of clinical research participation and to earning and building trust in the medical research enterprise is more straightforward. RCTs should be representative of the populations with the disease the trial seeks to treat to align with the principle of justice, as outlined in the Belmont Report. Black and Hispanic patients are less likely to be treated with novel, potentially life-saving therapies,23–27 which may relate to their lack of inclusion in RCTs showing the benefit of these therapies. Moreover, some evidence suggests that patients enrolled in RCTs or admitted to hospitals that participate in RCTs have better outcomes, suggesting that participation in RCTs may improve clinical outcomes via closer attention from clinical care teams.28–32 Representativeness is thus important to ensure that both direct and more downstream, population-level benefits are equitably distributed.
RCTs are also important for building trust in the medical enterprise, which is particularly important considering historical research abuses and present-day restrictions on equitable access to clinical services. Black patients shown hypothetical RCT results for an effective medication are 20 percentage points more likely to believe the drug would be effective for them when the trial population enrolled a representative proportion of Black patients.33
The multiple purposes of representativeness in RCTs, especially its role in promoting equitable access to the benefits of research participation and trust in the research enterprise, highlight the fact that representativeness is not an end, but a step in the process of advancing harder-to-measure goals such as diversity, equity, inclusion, and belonging (DEIB) (Figure 1). As such, representativeness is a useful measure specifically because it is quantifiable. Quantitative measurement of representativeness offers a clear, objective lens through which DEIB can be viewed and analyzed. By converting these intangible concepts into measurable data, it is possible to measure progress as well as identify gaps or areas where DEIB is lacking. Objective measurements can then inform decisions, guiding researchers and organizations on where to focus their efforts to enhance inclusion.
In addition to quantitative measures of representativeness, qualitative methods—including patient interviews and focus groups—offer valuable insights into participant experiences and perceptions. Though these insights are essential for a comprehensive understanding of DEIB in clinical trials, qualitative methods are beyond the scope of this review.
There are multiple ways to quantitatively measure representativeness in clinical trials, which all have strengths and weaknesses. We describe these methods below and summarize them in Table 2.
The participation-to-prevalence ratio (PPR) is calculated by dividing the proportion of trial participants who are members of a particular group by the proportion of patients with the disease or exposure of interest who are members of that same group. For instance, if 12% of U.S. patients with heart failure are Black, and a hypothetical trial enrolling U.S. patients with heart failure had a population that was 6% Black, then the PPR for Black participants in that trial would be 6%/12%, or 0.5. The ideal, completely representative trial would have a PPR of 1; values < 1 indicate under-representation of a given group and values > 1 indicate over-representation. By convention, a PPR between 0.8 and 1.2 is interpreted as proportionate enrollment.34
PPR’s major advantage is that it collapses the concept of representativeness into a single easily understandable number that can be directly compared across RCTs. However, it also has several methodologic challenges and disadvantages.
First, it is entirely dependent on accurate and relevant population prevalence data by race, ethnicity, and other characteristics of interest. Due to factors ranging from limited healthcare access to societal stigmas around certain conditions, many diseases go underreported, especially in historically marginalized populations, leading to underestimated or skewed prevalence figures.35 To the extent that trials can only be expected to enroll patients who know they have a condition, how PPR defines representativeness on published estimates of disease prevalence allows trialists to focus on the set of issues they can control. However, for a trial with multiple inclusion and exclusion criteria, it may be impossible to arrive at reasonable estimates of the racial or ethnic distribution of patients with the trial’s exact inclusion and exclusion criteria.
Second, the geographic area where the population prevalence should be determined is often undefined. Most would agree that a U.S.-based clinical trial should use U.S. population prevalence data, whereas a trial enrolling in Europe or the Middle East would need to use local population prevalence data. But in a multicenter study, should one define the relevant geographic area as the combined catchment areas of the enrolling sites, as the countries (or states) of the enrolling sites, or some other regional definition? While it is not sensible to expect a trial conducted in New York to enroll many Alaskan Americans, it is also problematic to simply allow a “pass” for a trial conducted in predominantly White portions of the U.S. Midwest that failed to enroll Black persons, especially since trial investigators and sponsors are often able to choose sites.
Representativeness of an RCT may also be assessed by simply reporting the proportion of participants that belong to a particular group as a fraction of the total trial population. These proportions can then be compared with prior trials, effectively capturing the evolution of representativeness over time. The major advantage of this approach is that there is no need to identify population prevalence, with all the challenges this process involves. However, this approach’s reliance on historical data introduces potential weaknesses, such as inconsistencies or biases in records of completed RCTs, adding a level of uncertainty to the analysis.36 Such an approach also fails to define the proper historical controls to be used for comparison. Comparing RCTs across time is made challenging by differences in inclusion and exclusion criteria, the sites at which the RCTs are conducted, and the intervention being studied itself. In addition, though comparison with historical controls explicitly focuses on making progress toward representativeness, it risks giving RCTs too much credit for marginal improvements on poor past performance and does not offer an obvious definition of representative enrollment. Raw proportions of patients enrolled in each RCT from a particular group of interest might also be compared to the proportion of patients from that group in the country or countries in which the RCT is conducted in a crude approximation of PPR. However, this type of comparison is less rigorous than PPR, failing to account for differences in the prevalence of disease in subgroups.
Representativeness may also be assessed by reporting the enrollment fraction, or the proportion of individuals approached for participation who enroll in the RCT by racial, ethnic, gender-based, or another subgroup. The resulting proportions can then be compared between subgroups. This approach has several advantages. If both the numerator and denominator for these proportions are reported, then it affords a nuanced approach to understanding the enrollment process. The denominator, representing the number of potential participants from each group approached for participation, is an indicator of how well the RCT’s design (including sites selected and methods for identifying potential participants) considered equitable enrollment. The numerator, representing how many participants enrolled from each group, indicates how attractive the RCT was to participants from different racial and ethnic groups.
This approach also has several weaknesses. First, this approach requires the capture and recording of demographic data from individuals who are approached but do not enroll in RCTs. Adopting this approach requires active collaboration with Institutional Review Boards to convey the critical need for gathering demographic information from all persons approached for research. This data is crucial for understanding the dynamics of trial enrollment and given its minimal risk to subjects, aligns with federal criteria for waivers from informed consent. The more pressing limitation, however, lies in the inherent selection bias in whom researchers approach for enrollment, which is a major factor limiting enrollment in certain clinical trials. Addressing this limitation demands a substantial shift in the methodology of RCTs, focusing on inclusive and equitable approaches to participant recruitment.37
Raw participation numbers in particular groups of interest allow for the examination of meaningful subgroup sizes, which is necessary for powering heterogeneity of treatment effect analyses to understanding of how different populations within trials may respond variably to treatments.38 However, it is limited by lack of contextual information about the demographic and clinical characteristics of the target population. Moreover, powering RCTs to detect heterogeneity of treatment effect across multiple subgroups would require massive sample sizes, increasing the costs of conducting RCTs (which may affect the ability of sponsors to conduct other important RCTs), delaying market availability of novel drugs and devices or implementation of novel care delivery interventions, and needlessly exposing more patients to potentially inferior treatment options after enrolling enough patients to adequately power the trial for its primary outcome. These considerations must be weighed against the possibility of important heterogeneity of treatment effects when deciding how much importance to place on the raw numbers of participants enrolled from subgroups of interest.
Beyond the specific strengths and weaknesses of existing methods, there are overarching limitations of the quantitative measurement of representativeness. All the above methods fail to consider intricate variations within communities, which can lead to a superficial understanding of representativeness. Elements such as income levels, educational achievement, and occupation play vital roles in healthcare and RCTs access, and racial, ethnic, gender, and socioeconomic characteristics intersect in important ways, with implications for equity, inclusion, and belonging that cannot be collapsed into any single number.39 These issues may be particularly acute when attempting to contextualize findings, as there are inherent difficulties in determining population prevalence of diseases by intersectional characteristics.40 To some extent, this barrier can be overcome by calculating measures of representativeness, but this risks unstable estimates due to small sample sizes in each group, will produce a complex and difficult to parse matrix of representativeness values, and still will not account for all types of intersectionality.
The rationale behind measuring representativeness in RCTs is crucial for guiding the choice of metric (Figure 2). If the intent of representativeness is to measure heterogeneity of treatment effects by racial or ethnic group, then raw participation numbers provide a straightforward way to ensure adequate numbers of each subgroup are present for analysis. If the intent is to build confidence in the research enterprise among historically underrepresented groups or to ensure fair access to the benefits of RCT participation, then PPR or comparison of participation proportions with historical data are more fitting, because these metrics demonstrate how well the RCT’s participant makeup reflects the broader population with the condition of interest. However, for communities that are not very large or whose members are unlikely to have a given disease, ratio-based measures like PPR or participation proportions may not advance this goal, since the numbers of patients enrolled from these communities will be low. If building confidence in the research enterprise among members of historically underrepresented groups with relatively small numbers is an important goal for a given study, then researchers may be best served by conducting clinical trials specifically among members of the relevant groups, seeking to inflate enrollment of this group above population prevalence.41 The task of measuring representativeness in clinical trials is complex, and no single method provides a comprehensive solution. A multifaceted approach may offer a refined perspective, with the use of PPR, raw participation proportions, enrollment fraction, and raw participation numbers all potentially appropriate in certain situations.
In most cases, each of these measures provides important context and can compensate for the weaknesses of the other three measures. By reporting all four measures, an RCT will provide information on how well DEIB is incorporated into the RCT design phase (through the denominator from enrollment fraction), whether the enrollment processes themselves appealed to different racial and ethnic groups (through the numerator of enrollment fraction), how well it achieved these goals relative to historical or contemporary comparators (through participation proportions), whether it will be useful in assessing heterogeneity of treatment effects in various subgroups (through raw enrollment numbers), and whether it was able to achieve a trial population representative of patients with the disease in the population (through PPR). These different measures can be informative regarding where in the sequence of processes that lead to enrollment (or non-enrollment) efforts to achieve DEIB are going well or poorly.
However, RCTs are complex and challenging undertakings, and asking trialists to report each of these measures will increase the complexity and challenges. A more parsimonious approach could focus on the combination of subgroup enrollment fractions and a complete reporting of the demographics of all potential participants approached to offer enrollment. These two measures directly assess the efforts trials make to offer enrollment to diverse populations and the success they have in these individuals, while avoiding challenges related to determining population prevalence. For trials where the population prevalence of disease is well-defined, including registry-based clinical trials and other pragmatic designs with broad inclusion criteria, PPR may be a useful single-number measure of representativeness.
Quantitative measures of representativeness can and should be used by stakeholders to increase DEIB in the field of RCTs. Though reporting representativeness itself does not increase representativeness, quantitative measures are useful for two specific reasons. First, quantitative measures of representativeness can identify gaps and areas for improvement, guiding stakeholders in implementing strategies to enhance DEIB in clinical trials. Second, quantitative measures of representativeness allow stakeholders to design structures to enforce representation. Rather than simply mandating that researchers consider representativeness and incorporate features in RCTs’ design phase to increase representativeness, stakeholders can set standards and demand results. Currently, RCT operations—whether the trial is funded by industry, government, or non-profits—are evaluated by funders primarily based on cost and speed, two easily quantifiable metrics. Quantifiable measures are necessary for representativeness to become part of the primary criteria by which RCT operations are judged. Adopted criteria, however, should account not only for the specific goals of representativeness but also the tradeoffs with other design priorities.
Regulatory agencies, medical journals, clinicians, trialists, trial sponsors, and patient groups have called for improvements in the representativeness of clinical trials,42–46 but policy prescriptions for improving representativeness have largely focused on the study design phase. Quantitative metrics can be used in conjunction with design phase requirements to “keep score” and/or enforce goals related to representativeness, by rewarding groups running RCTs for enrolling representative populations, penalizing them for not achieving representative enrollment, and/or forcing mid-trial changes in recruitment strategies if a representative population is not being enrolled. The use of quantitative metrics enables clear communication between regulatory agencies, funders, journals, and RCT sponsors and investigators with respect to what will be considered adequate representativeness. Regulators, funders, and journals can then develop policies to enforce goals related to representativeness, and sponsors and investigators will have fair warning to make the necessary changes to recruitment strategies to achieve the goal of their RCT.
To date, guidance from the NIH and FDA has focused largely on the processes by which sponsors might attempt to increase representativeness rather than outcomes. For example, in an April 2022 guidance, the FDA recommended that trial sponsors develop a “Race and Diversity Plan” in the pre-trial planning phase.20 The FDA recommended that such a plan should include the scope of the medical product development program, goals for enrollment of underrepresented racial and ethnic participants, and a specific plan of action to enroll and retain diverse participants and review the status of meeting enrollment goals. However, this guidance stops short of setting explicit quantitative targets for diversity or representativeness. Similarly, in June 2023, the Centers for Medicare and Medicaid Services (CMS) published a draft guidance document which proposes that post-coverage studies conducted under the Coverage with Evidence Development program reflect the demographic and clinical diversity of Medicare beneficiaries who are the intended population to receive the intervention.47 The guidance also proposes that sponsors plan to analyze demographic and clinically relevant subgroups. However, like FDA guidance, this draft document does not reference quantitative standards for representativeness. This approach contrasts with the FDA’s more prescriptive stance on pulse oximeter trials, which mandates that at least 15% of participants must have “darkly pigmented skin,” a directive that acknowledges the impact of skin pigmentation on device performance and sets a concrete benchmark.48 Though we are not aware of key government agencies assessing RCT representativeness using quantitative measures in other circumstances, the pulse oximeter precedent demonstrates agencies’ capacity to enforce minimum standards for representativeness and monitor representativeness throughout trial conduct, and illustrates how quantitative measures of representativeness can facilitate clear communication of expectations between government agencies and study sponsors. The NIH, for example, could refuse to release funds for future study conduct if the chosen metric for measuring representativeness lags behind a certain target and investigators do not make substantive changes to recruitment approaches. If a certain demographic group is underrepresented in a drug trial or critical post-marketing study, the FDA could refuse to grant approval or CMS could refuse coverage if trial sponsors do not reach quantitative representativeness targets. Similarly, though most medical journals now insist that major RCTs report baseline demographic characteristics of participants by race, ethnicity, gender, and age, these raw participation numbers and proportions may be inadequate to help readers assess the representativeness or external validity of a study. Furthermore, context-free participation numbers cannot be used to judge the representativeness of a trial in any objective way. Journals could require the use of at least one of these metrics, such as PPR or other measures of representativeness, and could preferentially consider papers that use multiple, providing readers with the necessary context to assess representativeness. Similarly, just as journals are required to preregister RCTs as a condition for publication,49 journal editors could require RCTs to meet standard PPR targets by race, ethnicity, and gender as a condition for publication.
Clinicians and patients have an important role in using quantitative measures to apply research findings in real-world scenarios and agitate RCT sponsors and investigators to enroll representative study populations.
However, as with any measurement of a surrogate outcome that becomes a target for efforts to improve quality, there are potential pitfalls to using quantitative measures of representativeness (Table 3).
First, stakeholders must set clear expectations of representativeness, making plain what the relevant metrics and their thresholds and targets are. Ideally, these targets should be empirically derived, with evidence generated on how different representativeness thresholds achieve the various goals of increasing representativeness in clinical trials. Expectations for how targets are achieved must also be clarified: For multi-center trials, is it acceptable to fix problems with representativeness by choosing sites with higher representation of historically under-represented populations, or does each site need to demonstrate suitable enrollment fractions among the groups it is able to approach? Multi-national clinical trials also pose a challenge with respect to developing quantitative targets for trial representativeness, as stakeholders from different countries might have different priorities for representativeness, and trialists may not be able to comply with the standards set by each one.
Another concern is the potential for “gaming” recruitment strategies. When specific quantitative targets are set, researchers might prioritize meeting these targets over the genuine inclusivity and integrity of the study. For example, to achieve an adequate PPR, a study might over-recruit from certain demographics, even if not scientifically warranted, merely to satisfy the ratio. Furthermore, perverse incentives can also arise in this context. If funding, publication, or regulatory approval is tied to achieving specific representative metrics, researchers may be motivated to design studies that are more likely to meet these metrics rather than those that are more scientifically relevant. This could result in the prioritization of easier-to-recruit populations, possibly sidelining hard-to-reach, but important demographic groups.50 Furthermore, creating real penalties for the under-representation of historically marginalized groups might incentivize researchers to employ ethically problematic tactics to recruit members of these groups, potentially exposing them unjustly to the risks of research participation. Any attempt to create incentives for enrolling representative populations into RCTs must pre-emptively consider the potential for off-target effects that could harm study participants, and work to limit their salience. Broadening clinical trial enrollment necessarily means directly confronting differences in socioeconomic status and health literacy that have classically been thought to raise challenges related to informed consent and exploitation.
Another potential pitfall is the risk of oversimplification. Trials may be inclined to report an overall number that masks important nuances. For example, two studies may report the same PPR for race, but one may have a more diverse age range, while another might have a better geographical representation.51 In the case of representativeness in clinical trials, the multiple quantitative measures must be integrated into a qualitative whole to get a full picture of DEIB. In some cases, quantitative metrics, even in combination, may not be wholly adequate. While quantitative measures of representativeness are useful tools, they are inherently limited as surrogates for true DEIB in clinical trials. No single metric can capture all aspects of DEIB, underscoring the need for a multifaceted approach that combines several metrics and qualitative insights.
The challenges with developing and implementing quantitative measures of representativeness in clinical trials illustrate how the world of clinical research is inextricably connected to the healthcare system in which it is integrated. Groups that are historically under-represented in clinical trials are also historically under-diagnosed and under-treated for multiple diseases.25 Under-diagnosis and under-treatment are health system issues that drive problems with representative enrollment and are very difficult for research teams to fix alone. Setting quantitative thresholds for representativeness in clinical trials creates a mandate for researchers to engage in efforts to build a more equitable health system with fewer disparities and offers an opportunity to align resources toward that end.
A number of existing and emerging clinical trial design features will be critical components in trialists’ toolboxes in a world where researchers are expected to strive for and achieve quantitative thresholds for representativeness.
Pragmatic RCTs, registry-based RCTs, and community-based participatory research (CBPR) have grown in popularity in the cardiovascular research community for efficiencies they offer with respect to data collection and participant burden, which can help reduce costs. Properly deployed, these strategies may also help researchers achieve representative populations.
Pragmatic RCTs are designed to evaluate the effectiveness of interventions among broadly selected populations in real-world settings, in contrast to traditional RCTs, which evaluate efficacy, and therefore often have strict inclusion criteria, take place in controlled environments that may not reflect everyday clinical practice,52 and place a high burden on participants by requiring frequent visits to the clinical trial site. Pragmatic RCTs lower participant burden by not requiring frequent—or any—site visits. This may be especially important for individuals from groups historically under-represented in clinical trials, who note that family and professional responsibilities are a key reason they are unable to participate in traditional RCTs.53–55 Broad inclusion criteria and real-world settings may allow researchers to conduct RCTs outside of the “ivory tower” of academic medical centers, including at sites with fewer resources to conduct traditional clinical trials, which necessarily care for patients that are not well-represented in traditional clinical trials. However, choice of sites alone is unlikely to create entirely representative trials: In the realm of valvular heart disease, for example, sites that conduct clinical trials care for more Black patients and patients with low socioeconomic status, even as clinical trials in valvular heart disease systematically under-include Black patients.56,57 Another potential advantage of pragmatic RCTs is that they often recruit by direct contact with patients via email or text message, avoiding clinicians’ unconscious biases in whom they choose to recruit.54,58–61 However, no study of direct-to-patient recruitment strategies has demonstrated that this strategy leads to more representative enrollment, and gains in representativeness from avoiding unconscious bias may be countered by patients from groups historically under-represented in clinical trials being less likely to respond to unsolicited outreach from researchers they do not know or trust.53,62–64
Registry-based RCTs, a subset of pragmatic RCTs, use disease- or procedure-based registries to identify eligible participants and collect baseline and follow-up data, lowering participant burden.65 Registry-based RCTs also have the advantage of collecting baseline data on all participants eligible for the trial, allowing researchers to set clear goals for representativeness before the trial starts, facilitating calculation of PPR, and improving researchers’ and other stakeholders’ understanding of the representativeness of the cohort enrolled. However, because groups historically under-represented in clinical trials are also under-diagnosed and under-treated, they may also be under-represented in registries.25 Furthermore, participation in clinical registries requires financial and operational resources that lower resourced hospitals may not have.66,67 Nevertheless, the potential for registry-based RCTs to facilitate pragmatic designs with lower participant burden may help increase representativeness, though this should be tested empirically.
Community-based participatory research (CBPR) strategies engage community stakeholders in the research process, which can improve recruitment and retention of underrepresented populations.68 CBPR emphasizes collaboration with community members, so that researchers can identify questions of interest to the community and design trial protocols—including inclusion and exclusion criteria—that will be acceptable to members of the community. On the broadest scale, CBPR introduces people from diverse backgrounds to the biomedical research process while centering patient and community members’ experiences and values, increasing transparency and trust. On a smaller scale, by partnering with members of the community, researchers can develop study champions among members of communities historically under-represented in clinical trials; these champions enhance trust in the research process among members of their community, increasing the likelihood that they will participate. The NIH has demonstrated a commitment to CBPR though its Community Partnerships to Advance Science (ComPASS) program, which seeks to cultivate community research partnerships and build research capacity in the community and among relevant partners.69
In the future, these RCT design features may be supplemented by artificial intelligence (AI).70,71 AI can assist with patient recruitment by analyzing large databases, including electronic health records and other real-world data sources, to identify potential participants who meet the trial inclusion criteria. Rather than only identifying patients seen in clinical setting by trial investigators, this approach will allow researchers to look at broader populations of patients, including some overlooked by more traditional recruitment methods. However, AI models have the potential to reinforce bias when built with incomplete data or using proxy measures of the target of interest.72 Furthermore, as with other forms of direct-to-patient recruiting, using AI to identify potential participants does not ensure participation, and research is needed to identify the best ways to engage patients historically under-represented in RCTs after direct outreach. AI platforms may also be able to identify sites that see large volumes of patients with the disease of interest but that do not typically participate in RCTs. However, these sites may require additional support—financial and otherwise—to be effective. The Advanced Research Projects Agency for Health’s Advancing Clinical Trial Readiness program and the NIH’s Institutional Development Award (IDeA) Networks for Clinical and Translational Research are examples of programs that will invest in building capacity to conduct RCTs in underserved communities while simultaneously diversifying the clinical trial workforce.73 The NIH’s Trial Innovation Network (TIN) is dedicated to developing scalable and decentralized solutions to help less-experienced trial sites and investigators.74 In the ideal future, AI will automate some features of RCT recruitment, freeing time for diverse, representative, and well-trained teams of humans to focus on engaging with potential participants and making them comfortable with the research process.
Yet none of these trial design features will be a panacea for questions of representation in clinical trials. The only way to ensure DEIB in RCTs is to actively and consciously engage populations historically under-represented in RCTs. Though a complete accounting of how to increase representativeness in RCTs is beyond the scope of this review, and has been covered elsewhere,46 the many programs funded by federal agencies, non-profits, and industry are critical for diversifying the clinical trial workforce and community,73 a necessary first step for including diverse voices in trial planning and protocol development. Though more inclusion in trial leadership and operations will not itself lead to representation, considerable evidence demonstrates that more diverse trial leadership teams enroll more representative populations.75,76 Beyond diverse RCT leadership and operations, improving DEIB will require investigators to engage with patients and potential trial participants from communities historically under-represented in research in the trial planning phase, so that investigators choose to answer clinical questions of interest to these potential participants, as well as understand and mitigate the barriers foreseen by potential participants. Clear quantitative targets for representativeness do not do any of these things on their own but will help stakeholders incentivize the use of trial design elements that enhance DEIB.
Conclusion
Quantitative measures of representativeness in RCTs provide essential, objective data on an outcome related to an RCT’s DEIB. Measures such as PPR, participation proportion, and subgroup-specific enrollment fraction are all useful for defining representativeness in specific contexts, but all have disadvantages and limitations. Though a comprehensive approach to reporting representativeness would include all of these measures, minimum standards to avoid overburdening trialists might include subgroup-specific enrollment fractions and a report of the demographics of potential participants approached by trialists. Ultimately, individual stakeholders—including regulatory agencies, funders, and journals—may set their own reporting standards, based on the overarching goals they hope to achieve by increasing representativeness. Regardless of the method chosen, quantitative assessment of RCT representativeness opens the door for these powerful stakeholders to create strong, transparent, and enforceable regulations that will require or incentivize trial sponsors and investigators to prioritize key principles of DEIB; however, such regulations may create perverse incentives, and thoughtful deployment must take this possibility into account. Despite these limitations, quantitative methodologies to assess RCT representativeness add empirical rigor and a structured framework, providing tangible metrics against which inclusivity can be measured.
MO
Dr. Modele Ogunniyi
Cardiology · General Non-Invasive CardiologyEmory Universitymodele