Two things are happening at the same time, about forty miles apart, and neither party knows about the other.
At an academic medical center, a Phase III trial is behind schedule. The sponsor has activated 150 sites. A meaningful number of them have not enrolled a single patient and never will. The study team is on a weekly call discussing recruitment tactics: more advertising, a patient-finding vendor, maybe a social media campaign. The trial is burning roughly $55,000 a day just by continuing to exist.
Forty miles away, in a community oncology practice, a physician is sitting with a patient who meets every inclusion criterion in that protocol.
She does not know the trial exists.
If she did know, she would face a second problem, which is that she has no relationship with anyone at that center, no idea who to call, and a well-founded fear that if she sends this patient away, the patient may simply not come back to her.
So the sponsor spends money on advertising to find patients who are already sitting in front of physicians who would refer them, if anyone had ever asked.
This is the clinical trial enrollment crisis. It is discussed everywhere as a recruitment problem. It is much more accurately a search and trust problem between clinicians, and understanding the difference changes what you would build to fix it.
The waste, quantified
Start with the site side, where the numbers are the most concrete.
Analysis from the Tufts Center for the Study of Drug Development, widely cited across the industry, found that:
- About 11 percent of activated sites enroll zero patients.
- Roughly 37 percent under-enroll, meaning 48 percent of approximately 16,000 sites examined failed to hit target.
- Zero-enrollment rates vary by region: about 13 percent in North America, 7 percent in Western Europe, and 20 percent in Latin America.
Each of those sites was not free. Site activation runs roughly $20,000 to $25,000, plus about $1,500 per month to maintain. Multiply across the non-performing share of a 150-site trial and you have several hundred thousand dollars of pure waste before anything else.
But the direct cost is the small part. The real cost is time. Phase III trial conduct runs approximately $55,716 per day by Tufts CSDD's 2024 analysis. Enrollment timelines have been moving the wrong way: IQVIA data indicates median enrollment duration rose from 14.9 months to 16.3 months between 2024 and 2025.
A thirty-day slip attributable to sites that were never going to enroll costs roughly $1.67 million on that arithmetic. And a large share of trials miss their enrollment projections entirely.
Then there is the pre-selection waste. The industry sends the same redundant feasibility questionnaires to large numbers of candidate sites, with the total industry-wide feasibility burden estimated at around $1.6 billion. ASCO has published on the need to streamline exactly this process.
So: enormous, well-documented, expensive waste, in a step that everyone agrees is the cheapest place in the entire development process to fix.
The finding that reframes everything
Here is the study that should change how the industry thinks about this.
Research published in Trials in 2019 surveyed 83 decision-makers, 43 from pharmaceutical sponsors and 40 from CROs, about how they actually choose sites. The findings:
- The top-ranked selection factor, at 42 percent, was "patients ready" post-initiation. Not experience. Not infrastructure. Access to eligible patients.
- 75 percent said they would use an inexperienced site if it had a large eligible patient population.
- 88 percent said they would prefer faster recruitment over a 20 percent cost reduction.
Read those together and the conclusion is unavoidable. Sponsors are not attached to their existing site lists. They have said explicitly, in survey research, that they would happily work with a site they have never used before if that site had the patients.
They are not choosing relationship lists because they prefer them. They are choosing relationship lists because they have no other way to find anybody.
The signal they actually want, which is who has the eligible patients and the bandwidth this quarter, lives in two places: inside clinicians' heads, and inside EHRs that nobody can query across institutions. It exists nowhere that a sponsor can reach.
So they fall back on what is available: historical performance databases, which describe a prior trial with a prior competitive landscape and a prior patient mix, and personal relationship lists, which describe who somebody at the CRO happens to know.
And then they send everyone a self-reported feasibility questionnaire, which asks sites to estimate their own eligible patient counts. Sites have every incentive to be optimistic and no ability to verify, which is why "overoptimistic investigators" is the industry's own standing phrase for the problem.
Nobody is lying and the data is still worthless, because there is no neutral party holding verified capacity information.
Half of trained investigators are never asked again
Now the second half of the waste, which is even less examined and, to my mind, more indefensible.
Research published by CTTI in Contemporary Clinical Trials Communications examined investigators who file an FDA Form 1572. A separate analysis of 172,453 unique investigators between 1999 and 2015 found that 49.5 percent were "one-and-done": they conducted exactly one trial and never appeared again. Roughly 12.6 percent were stop-and-go, and 37.8 percent were stayers.
The industry's standard explanation for this is regulatory burden. Trials are onerous, paperwork is crushing, and physicians decide it is not worth it. That explanation is partly true. In the CTTI survey of 201 principal investigators, workload was cited by 63.8 percent, time by 63.4 percent, and reporting requirements by 56.5 percent.
But the same survey contains a finding that should have redirected two decades of industry effort, and did not.
Among one-and-done investigators, 44.4 percent said they wanted to run another trial and no opportunity ever came.
Not "it was too hard." Nobody asked.
The entry paths explain why. Investigators reported entering research through a direct sponsor request (36.6 percent) or through a colleague (29.0 percent). Which means that for a large share of investigators, their entire connection to clinical research ran through one specific human relationship.
When that person changed jobs, retired, or moved to a different therapeutic area, the investigator vanished from every list simultaneously. They did not opt out. They were dropped, silently, by a system with no memory.
Meanwhile the pool is shrinking. The number of active US principal investigators declined from about 28,292 in 2014 to 26,115 in 2020, while trial volume increased.
An intellectually honest note: this figure is contested. Analysts including Glass have argued the true one-and-done rate is likely below 20 percent, because the 1572 database is incomplete and misses investigators who appear under different filings. That criticism is fair and worth stating. But it does not touch the core finding, which comes from directly surveying investigators about their own experience: a large share of people who ran one trial wanted more work and were never contacted again.
Every one of those investigators represents sunk training capital. GCP certification, site qualification, protocol training, regulatory familiarity. All of it paid for, all of it discarded, while sponsors simultaneously complain about a shortage of qualified sites and pay to onboard novices.
CTTI's fourth recommendation, published years ago, was to build mechanisms for "discovering opportunities for conducting additional trials."
As far as I can determine, no product has ever been built to do it.
The patient's own doctor is the missing node
The third failure is the largest, and it is the one where the numbers are most stark.
Roughly 90 percent of patients say they want their own physician to tell them about clinical trials. It is the single most trusted channel for this information, by a wide margin.
Fewer than 1 percent of trial participants are actually referred by their physician.
Survey data from CISCRP found the share of participants who learned about their trial from their own primary care physician or specialist moved from 14 percent in 2013 to 16 percent in 2019. Only about 10 percent of physicians regularly discuss trials with patients at all.
The consequences show up directly in enrollment. Research published in the Journal of Clinical Oncology in 2024, examining 1,200 Commission on Cancer programs, found:
- 7.1 percent of adult cancer patients enroll in treatment trials overall.
- 4.1 percent at community programs versus 21.6 percent at NCI-designated comprehensive centers.
A five-fold difference by site of care. Most American cancer patients are treated in community settings.
And there is a funnel finding worth sitting with: only 62 percent of eligible patients are ever even considered for a trial. The largest single drop-off in the entire enrollment funnel happens before anyone talks to a patient.
What community physicians actually say
When you ask community oncologists directly why they do not refer, they give three consistent answers, documented in JCO Oncology Practice interviews and in a 2023 Cancer Control study:
- They do not know which trials exist. Cited by 54 percent in one analysis. A clinician described the situation precisely: "Everything in our system is put on the front-line clinician to be aware of the trial, and there are relatively few resources to help identify which trials are available."
- They do not know where to refer. Cited by 48 percent. Even knowing a trial exists is insufficient without a person to call.
- They do not have time. Cited by 33 percent. Screening protocols against a patient in a fifteen-minute visit is not realistic.
Those three are the stated barriers. But the fourth one, the one that appears in qualitative interviews rather than checkbox surveys, is the one that actually decides behavior.
They are afraid the patient will not come back.
Clinicians describe worrying that patients "may feel abandoned if care is transitioned to a clinical trial physician." Underneath the phrasing about patient feelings is a straightforward professional and economic reality: if you send your patient to an academic center, that center may keep them. You lose the relationship, the continuity, and the revenue.
Referring your patient into a trial at another institution is an act that costs you and benefits someone else. Under those conditions, the surprising thing is not that fewer than 1 percent of patients are physician-referred. The surprising thing is that anyone does it at all.
This is why patient-facing advertising campaigns keep underperforming. They are aimed at the wrong node. The decision does not happen when a patient sees an advertisement. It happens in an exam room, with a trusted clinician, who currently has no information, no counterparty, and a strong disincentive.
Why this matters more right now than it did five years ago
Two forces have converged.
The regulatory one. The Food and Drug Omnibus Reform Act made Diversity Action Plans a statutory requirement, obliging sponsors to plan for representative enrollment. The regulatory path has been turbulent: draft guidance issued in June 2024 was removed from the FDA site in January 2025 following an executive order, then restored by court order in February 2025. The statutory obligation persists throughout.
The operational implication is unavoidable. Representative enrollment requires reaching patients who are overwhelmingly treated in community settings, by community clinicians, whom sponsors have never worked with and cannot currently find. The enrollment gap that mattered ethically for decades now matters legally.
The economic one. Enrollment durations are lengthening. Costs per day are rising. And the historical fallback of activating more sites has diminishing returns when one in nine of them enrolls nobody.
The reframe, and what follows from it
State the reframe plainly:
The zero-enrolling site is not a recruitment failure. It is a search failure. The right investigator existed and was never asked.
Everything changes once you accept that.
If it is a recruitment problem, you buy advertising, hire patient-finding vendors, and build prediction models on historical enrollment data. That is what the industry does now, and the results speak for themselves.
If it is a search problem, you need three things that do not currently exist:
One: verified, member-controlled capacity signals. A clinician stating what patient phenotypes they actually see and whether they have research bandwidth this quarter, held by a neutral party rather than by a counterparty in a negotiation. Sponsors have told us they would use an unknown site with the right patients. They simply cannot find one.
Two: a discoverable willingness signal for investigators. The 44.4 percent who wanted another trial need a way to be findable that is not "advertise availability on LinkedIn," which no physician will do. Willingness is a live, private state, and no sponsor database stores intent, only history.
Three: a peer covenant that makes community referral rational. The clinician's fear of losing the patient is legitimate and cannot be argued away. It can only be answered by a commitment from a specific named counterpart: your patient returns to you for ongoing care, you are acknowledged, and the relationship runs both ways. That is not a technology feature. It is an agreement between two clinicians who can identify each other.
Note that all three are clinician-to-clinician trust problems. None is a patient-outreach problem. Which is why fifteen years of patient-facing recruitment innovation has moved physician referral from 14 percent to 16 percent.
What you can do
If you are a community clinician
Find your three nearest trial sites and make one call each. Ask a specific question: "What phenotypes are you actively looking for right now, and who do I call?" Most community clinicians have never done this. It takes about an hour total and it converts a diffuse awareness problem into three named relationships.
Ask about the return-of-care commitment explicitly. "If I refer, does the patient come back to me for their ongoing non-protocol care?" This is a reasonable, professional question, and the answer tells you a great deal about whether that site is a good partner. Get the answer from a person, not a brochure.
Keep a list of your own hardest-to-place phenotypes. The patients you see and cannot find options for. That list is exactly what a sponsor is spending millions trying to locate.
If you have ever been an investigator
Tell someone you are available. If you ran one trial and would run another, the evidence says you are one of a very large group whose only barrier is that nobody knows. Contact the CRO or sponsor medical monitor from your prior study, and contact the research office at your nearest academic center. Both are unusually cheap actions relative to how they are currently distributed.
Keep your own record. Trials run, indication, enrollment achieved, dates. Sponsors' databases lose you when their staff turn over. Yours will not.
Look at who introduced you. If your research career came through one person, as it did for roughly two-thirds of investigators, that is a single point of failure. Build a second connection deliberately.
If you select sites
Ask the question you actually care about. Sponsors rank "patients ready" first at 42 percent, then select on historical performance. That gap is the whole problem. Design feasibility around verified current phenotype access rather than prior trial counts.
Re-contact your one-and-done investigators. You have a list of people who completed a trial and were never approached again, and roughly 44 percent of them wanted more work. This is the cheapest site-identification exercise available to any sponsor and it requires no new technology at all.
Fund the return-of-care commitment in writing. If community referral is blocked by loss aversion, and the qualitative research says it is, then a formal, written commitment to return patients for ongoing care is the intervention. It costs nothing and addresses the actual barrier rather than the stated one.
Measure referring clinicians, not just sites. Nobody currently tracks which community physicians referred patients into trials. That is the highest-value graph in enrollment and it is entirely unbuilt.
Frequently asked questions
What percentage of clinical trial sites enroll no patients? Tufts CSDD analysis found approximately 11 percent of activated sites enroll zero patients, with about 37 percent under-enrolling, meaning roughly 48 percent of some 16,000 sites examined failed to meet enrollment targets. Zero-enrollment rates were about 13 percent in North America and 20 percent in Latin America.
How much does a trial delay cost? Phase III trial conduct runs approximately $55,716 per day according to Tufts CSDD 2024 analysis, so a thirty-day delay costs roughly $1.67 million in direct conduct costs alone, before lost or delayed revenue. Site activation adds roughly $20,000 to $25,000 per site plus about $1,500 monthly.
What is a one-and-done investigator? An investigator who conducts exactly one clinical trial and never conducts another. An analysis of 172,453 unique Form 1572 investigators from 1999 to 2015 found 49.5 percent fell into this category, though some analysts argue the true rate is below 20 percent because the database is incomplete. Critically, CTTI survey research found 44.4 percent of one-and-done investigators wanted to run another trial and no opportunity was ever offered.
Why don't community physicians refer patients to clinical trials? Four reasons, three stated and one decisive. Physicians cite not knowing which trials exist (54 percent), not knowing where to refer (48 percent), and lack of time (33 percent). Qualitative research adds the barrier that most determines behavior: fear that the patient will be absorbed by the trial institution and not return, which makes referral professionally and economically costly to the referring clinician.
How many cancer patients enroll in clinical trials? About 7.1 percent of adult cancer patients enroll in treatment trials overall, but the site of care matters enormously: 4.1 percent at community programs versus 21.6 percent at NCI-designated comprehensive cancer centers, based on Journal of Clinical Oncology analysis of 1,200 Commission on Cancer programs.
Do patients want to hear about trials from their own doctor? Overwhelmingly yes. Roughly 90 percent say they want their own physician to tell them about trials, and it is the most trusted channel available. Yet fewer than 1 percent of participants are physician-referred, and the share learning about their trial from their own physician moved only from 14 percent in 2013 to 16 percent in 2019.
What are Diversity Action Plans? Requirements established under the Food and Drug Omnibus Reform Act obliging sponsors to plan for representative enrollment in certain clinical trials. Draft guidance issued in June 2024 was removed from the FDA website in January 2025 and restored by court order in February 2025, while the underlying statutory obligation remained in force. Meeting representative enrollment goals generally requires reaching patients treated in community settings, which is precisely where the referral infrastructure is weakest.
The bottom line
The clinical trial enrollment crisis is treated as a problem of persuading patients. Nearly all the money and innovation goes there: advertising, patient-finding platforms, decentralized designs, engagement apps.
The evidence points somewhere else entirely.
Sponsors say they would use unfamiliar sites with the right patients, and cannot find them. Half of trained investigators disappear after one trial, and almost half of those wanted more work and were never asked. Ninety percent of patients want their own doctor to tell them about trials, and fewer than one percent hear it from them. The largest drop in the enrollment funnel happens before a patient is ever approached.
Every one of those is a failure to connect two clinicians who would both benefit from being connected.
Somewhere right now a community physician is sitting with an eligible patient, forty miles from a trial that is burning $55,000 a day for want of that exact person. Neither of them knows the other exists.
That is not a recruitment problem. Nobody needs to be persuaded of anything. They need to be able to find each other.
Part of a series on the missing professional infrastructure of healthcare. Previously: AI Answered. Who Is Accountable?
Evidence note: site performance and cost figures come from Tufts Center for the Study of Drug Development analyses as reported in Applied Clinical Trials and compiled industry sources; site selection preferences from Trials (2019), surveying 83 sponsor and CRO decision-makers; investigator retention from CTTI research in Contemporary Clinical Trials Communications and analyses of FDA Form 1572 filings, with the contested nature of the one-and-done rate noted in the text; enrollment rates from Journal of Clinical Oncology (2024); physician referral barriers from JCO Oncology Practice (2020), Cancer Control (2023), and CISCRP survey data; enrollment duration trends from IQVIA. Some frequently cited Tufts figures date from earlier analyses and are noted where relevant.
