Why most clinical trials miss enrollment deadlines — and what fixes it

The most expensive part of many clinical trials is not the drug, the imaging, or the statisticians. It is finding the people. Industry analyses have repeated the same figure for two decades — roughly 80 percent of trials fail to meet their initial enrollment timelines — and it has held up well enough to become shorthand. Behind the shorthand: coordinators working phone lists, flyers in waiting rooms, and sites that enroll one patient. Or none.

The numbers behind the shorthand

Recruitment is the most common operational reason trials run late, and the failure is broad rather than concentrated. ClinicalTrials.gov lists hundreds of thousands of registered studies; when trials terminate early, failure to accrue participants is among the most frequently recorded reasons. In multicenter trials, a large share of sites under-enroll against their targets, and some never randomize anyone at all — a site stood up, trained, monitored, and paid, for zero participants.

Sponsors know this, so they plan for it: open more sites than the math requires, extend timelines, budget for rescue campaigns. Which is to say, the industry's standard answer to funnel waste has been a bigger funnel. The waste is institutionalized, not solved.

The clock compounds it. Every month of delay is a month of fixed costs — sites, staff, monitoring — and, for a sponsor, a month of patent life spent before the product earns anything. For patients, the arithmetic is simpler: a month longer until the question has an answer.

Why the funnel leaks

The funnel leaks because the standard model is advertising to strangers. Walk it stage by stage.

Awareness comes first, and it fails first. For most patients, learning that a relevant trial exists depends on their clinician remembering it during a fifteen-minute visit — and most clinicians are not investigators on any trial. Only a small minority of patients ever participate in research; surveys consistently find that far more would consider it if asked. The asking is the missing step.

Eligibility fails next. Criteria written to produce clean answers — narrow age bands, few comorbidities, no recent medication changes — exclude a large share of the real-world patients who actually have the disease. The people who pass the screen then meet the burden: travel to a distant academic center, repeated visits, time away from work. Each stage sheds people, and the consent conversation arrives only at the very end of an exhausting path.

The funnel's last insult is the screen failure: a patient who said yes, traveled in, and was then found ineligible on a lab value that already sat in their records weeks earlier. Screen failures are demoralizing for patients, expensive for sponsors, and almost entirely an artifact of checking eligibility last instead of first.

What failure costs

An under-enrolled trial is not a cheaper trial; it is an unanswered question purchased at nearly full price. A study that closes below its statistical power can neither confirm nor rule out the effect it was built to measure. The participants it did enroll accepted risk, needles, and lost afternoons for a study that cannot conclude — which research ethicists rightly treat as a moral problem, not merely a financial one.

The slower costs compound. Therapies reach patients years later than they should. Some questions are simply never answered, and the burden falls hardest on rare and underfunded conditions, where the eligible population was small to begin with and a failed trial can set a field back a decade.

Recruitment as matching, not advertising

The structural fix is to invert the direction of search: instead of broadcasting a trial and hoping eligible patients hear about it, query consented records for the people who already match.

The pieces exist. Eligibility criteria can be written as computable logic over diagnosis codes, medications, and labs — the same discipline behind computable phenotypes and cohort definitions. Standing pools of consented, well-characterized patients already exist in patient registries. Put together, feasibility becomes a count instead of a guess: how many people actually meet these criteria is a question answerable in days, against real data, before a protocol is finalized or a single site is budgeted. Protocols can be tuned while they are still soft — if one criterion eliminates half the eligible population for marginal scientific gain, the count says so while the criterion can still be changed.

What it asks of the data — and of patients

Matching only works, and is only defensible, if the consent underneath it is real. Being findable for research is a permission a patient grants, scopes, and can take back — not a default they must discover and escape; we unpack what genuine revocability requires in what revocable consent really means. The burden side has its own fix: decentralized elements — eConsent, home visits, televisits — shrink the geography problem, a shift the FDA has supported in guidance and one we cover in decentralized clinical trials. Registries feeding the World Health Organization's international trial registry network have made trials far easier to find; the harder half — making patients findable without making them exposed — is the infrastructure question.

It is the question Mithras Health is built around. Patients across 28 chronic conditions hold their own records and grant each study a scope-limited, revocable consent; researchers running feasibility see aggregates only, never raw records, and never a cohort below the k = 5 floor; participation fees post when an IRB-approved research-use event occurs, so patients are paid when their data does the work. Recruitment becomes a property of the infrastructure: the people exist, the permission exists, and the match is a query.

Trials do not miss their deadlines because the patients do not exist. They miss them because the patients are invisible to the trial and the trial is invisible to the patients. That was never really an advertising problem. It is an information problem — and information problems have infrastructure answers.