In January 2025, draft FDA guidance on clinical trial Diversity Action Plans disappeared from the agency’s website, shortly after an executive order targeting diversity, equity and inclusion programmes. It came back on 11 February, reportedly following a court order. The underlying legal requirement, written into the Food and Drug Omnibus Reform Act of 2022, never went away: sponsors of certain late-stage trials must submit plans setting enrollment goals by age, sex, race and ethnicity.
The episode generated a lot of heat about whether diversity requirements would survive. It generated very little discussion about whether they were ever aimed at the right target.
Here is the argument of this post. Most efforts to diversify trial enrollment focus on participants: outreach campaigns, translated consent forms, community advisory boards, transport vouchers. Those things help. But the bottleneck sits further upstream, in who designs trials, who runs trial sites and who sits across the table when a patient is asked to take part.
The numbers on both sides of the table
The FDA’s own Drug Trials Snapshots for the 53 novel drugs approved in 2020 reported that, on average, 75% of trial participants were white, 11% Hispanic, 8% Black and 6% Asian. For a country where Black Americans make up around 13% of the population, and for diseases such as hypertension and diabetes that hit some minority groups harder, that is a structural mismatch, not a rounding error.
Now look at the other side. According to the Association of American Medical Colleges, only about 5.7% of active US physicians identify as Black or African American. Among investigators funded by the National Institutes of Health, the share who are Black is under 4%, a figure cited in recent research on the issue. Principal investigators at industry trial sites are drawn largely from the same academic and specialist pool.
The participant gap and the workforce gap are usually treated as two separate problems, handled by different offices with different budgets. They are more likely the same problem.
Trust is not a messaging issue
A study published in the Journal of Health Economics in March 2025, by Marcella Alsan and colleagues, tested this directly. Researchers showed more than 320 Black American participants photographs of real NIH researchers, randomly varying the investigator’s race, and asked how willing they would be to take part in that person’s study. Willingness rose by 12.6% when the investigator was Black. The strongest explanatory factor was perceived trustworthiness.
It is a survey experiment, not a trial of actual enrollment, and it should be read that way. But it lines up with what many trial coordinators say privately: people join studies because they trust a person, not because they read a well-designed leaflet.
That has an uncomfortable implication for sponsors. A great deal of money is spent on recruitment vendors, digital advertising and patient-facing materials, all intended to generate trust at scale. If trust is mostly relational, much of that spending is working against the grain.
Where the trial sites are is who gets enrolled
The second workforce issue is geographic. Trials are placed where experienced investigators and research infrastructure already exist, which in most countries means large academic centres in major cities. Patients who get care from community hospitals, safety-net clinics or rural practices are rarely offered a trial at all.
Sponsors know this. Site selection still favours proven sites because they enroll quickly, produce clean data and pass audits. A new site in an underserved community might take longer to start and require more monitoring. Under the pressure of development timelines, where each month of delay has a real cost, the proven site wins almost every time.
So the diversity problem is partly an incentive problem. Nobody in the development chain is paid to build research capacity in places that do not already have it.
The site-selection arithmetic
It helps to look at a site decision the way a clinical operations team does. The figures below are illustrative, not drawn from any particular sponsor, but the shape of the calculation is familiar to anyone who has run one.
A proven academic site might activate in a few months and enroll steadily from its existing patient base. A new community site might take twice as long to open, need extra training and monitoring visits, and enroll slowly at first while its staff learn the protocol. On a single study, the proven site wins on every metric an operations team is judged by: time to first patient, cost per patient, data queries per case.
Now change the time horizon. The community site, once running, keeps its staff and its relationships. On the second study it opens faster. By the third, it may enroll patients the academic centre never sees at all, because they do not travel to the city. The investment is front-loaded and the returns arrive on later studies, often for a different programme, sometimes for a different sponsor.
That is the whole problem in miniature. The cost sits in one study’s budget. The benefit is spread across a portfolio, and across an industry. Nobody’s scorecard captures it, so nobody pays for it.
What a workforce-first approach looks like
If the constraint is people and places rather than messaging, the interventions change:
- Invest in new investigators, not only new participants. Fund training and mentoring for clinicians in community settings to become sub-investigators and then principal investigators, and commit to using their sites across multiple studies.
- Pay for the slower start. Budget explicitly for longer activation and heavier support at new sites, instead of treating it as a cost overrun.
- Staff coordinators from the community. Research coordinators and nurses often do the actual consent conversation. Hiring them locally, and paying them properly, may matter more than who the principal investigator is.
- Measure the team, not just the cohort. Diversity Action Plans set enrollment goals. Sponsors could also report the make-up and location of their investigator network.
None of this will produce results in one development cycle. That is exactly why it gets neglected. Enrollment targets are measured study by study. Workforce capacity pays off over a decade.
The strongest objection
A fair critic would say this argument leans too heavily on concordance. Patients do not need an investigator who looks like them; they need one who is competent, honest and available. Plenty of patients trust clinicians of other backgrounds, and building a trial strategy around demographic matching risks tokenism, or worse, the suggestion that minority patients can only be reached by minority clinicians.
That objection has real force. The Alsan study measured stated willingness in a survey, and trust can be earned across any line of difference. A white research nurse who has worked in the same community clinic for fifteen years may carry more trust than a newly hired coordinator of any background.
But notice that the objection, followed through, still lands on the workforce. What builds trust in that example is not the nurse’s demographics but her presence: long tenure, local relationships, a known face. That is exactly what trials placed only at distant academic centres lack. Whether the lever is concordance or continuity, it is a staffing lever, not a marketing one.
Can technology route around the problem?
The obvious rejoinder is decentralisation. If patients cannot reach the trial site, bring the trial to the patient: remote consent, home visits by mobile nurses, wearable devices, local laboratories and telemedicine follow-up. These approaches expanded during the pandemic, and they genuinely lower some barriers, especially travel time and time off work.
But decentralisation shifts the workforce problem rather than removing it. Someone still has to explain the study, answer the questions a frightened patient asks at the kitchen table, and notice when a participant is struggling with side effects. A video call with a distant coordinator is a thinner version of that relationship, and for people whose distrust of research is rooted in history, thinner is not better. Remote tools also assume broadband, a smartphone and a quiet space for a call, which are least available to exactly the populations trials miss.
The more promising model is hybrid: local clinicians and coordinators who hold the relationship, supported by remote monitoring and central data systems that reduce the paperwork burden on small sites. Technology works best here as a way to make community sites viable, not as a substitute for them. Used that way, it could cut the activation cost that makes sponsors reluctant to try new sites in the first place.
There is a regulatory dimension too. The more a trial relies on local, non-academic staff, the more sponsors worry about data quality and protocol deviations. Clear, proportionate expectations from regulators about what oversight is required at small sites would lower the perceived risk, and with it one of the quiet reasons proven sites always win.
The side effects nobody budgets for
Workforce-first strategies carry their own risks. The small number of investigators from under-represented groups are already asked to sit on every committee, advise every sponsor and mentor every trainee. Academic medicine has a name for this burden, sometimes called the minority tax, and a diversity strategy that leans on the same few people will exhaust them.
There is also the question of what happens to a community site between studies. If a sponsor builds capacity for one trial and then moves on, the coordinator is let go, the equipment sits idle and the community learns that research arrives, takes and leaves. That history is part of why trust is low in the first place. Capacity that is not sustained can do more harm than no capacity at all.
Who could move this
Sponsors and contract research organisations could commit to multi-study site networks in under-served areas, with a guaranteed flow of studies that makes it worth training staff. Academic medical centres could treat community affiliates as research partners with their own investigators, not as referral pipelines. Public funders such as the NIH could fund research infrastructure, coordinators and training, in community settings rather than only project grants. And regulators, whatever the fate of specific guidance documents, can keep asking the scientific question: does the trial population resemble the people who will use the drug?
The things worth watching over the next year or two are whether Diversity Action Plans, as they come into force, reveal anything about where trials are sited, and whether any large sponsor starts publicly reporting its investigator network the way it reports enrollment.
The trade-off worth naming
There is a real tension here. Faster trials get effective treatments to patients sooner, and every sponsor can make that argument in good faith. Building new sites and training new investigators slows things down, at least at first.
But a trial that enrolls quickly from a narrow population produces evidence that may not hold for the patients who will eventually take the drug. That is not a political concern. It is a scientific one, and regulators have been saying so for years.
Whatever happens to the guidance documents in Washington, the fix does not live there. It lives in who gets trained, hired and funded to run research, and where.
Has your organisation tried building trial capacity in communities it has not worked with before? What got in the way? Tell us in the comments.
