Patient recruitment in clinical trials: challenges and solutions

Recruitment is the dominant operational bottleneck in clinical research, and it's the subject where the gap between circulating statistics and traceable evidence is widest. Every figure below carries a measurement date, and two of the most repeated ones are flagged rather than reproduced. What emerges from the data is a different problem: not patient reluctance, but trial availability, protocol design, and a referral loop that stays open.

Do most trials really fail to recruit?

The claim that 80% to 85% of trials fail to enroll enough participants shows up in peer-reviewed papers, trade press, and vendor content alike. Following the citation chain never reaches a primary measurement, and the claim quietly merges two separate statements: missing the enrollment target, and missing the original timeline.

The best-documented site-level dataset, nearly 16,000 investigative sites across 151 phase II and III trials, reported the opposite headline: 89% of trials met enrollment, while timelines slipped and about half of sites under-enrolled. Only 11% of ready-to-recruit sites enrolled nobody. That measurement dates from 2008 to 2010 and has no comparable public update.

The $600,000 to $8 million per day figure fares worse. It traces back to a mid-2000s press release rather than a published methodology, and it describes lost sales for a hypothetical blockbuster, not the cost of a recruitment shortfall.

What can be said with confidence is that protocol complexity has risen sharply. Between 2010 and 2020, endpoints in a typical protocol nearly doubled, procedures rose 40.6%, sites per trial rose 60.1%, enrollment duration rose 36.9%, and dropout rates rose 105.1%.

Where are patients actually lost?

The most useful breakdown of the pathway comes from a meta-analysis of the oncology decision route, and it shifts responsibility away from the patient entirely.

Structural and clinical factors account for 77.1% of nonparticipation, so more than three patients in four never reach the point of discussing a trial with their physician. When a trial is actually offered, 55% of patients enroll, with no significant difference by race or ethnicity, which places observed participation disparities upstream of the patient's decision.

Public opinion data point the same way. Among 12,887 respondents surveyed worldwide in 2025, 87% were somewhat or very willing to take part. What has eroded is awareness, not willingness: self-reported understanding of what a clinical study is fell from 91% in 2023 to 78% in 2025.

Why does the referral chain break down?

A survey of 2,000 physicians and nurses found that 90% of physicians felt comfortable discussing trials, yet referred fewer than 0.2% of their patients. Worse for the feedback loop: nearly 30% of physicians and 45% of nurses reported never hearing back from site staff after making a referral, and a referral that disappears is a referral that won't be repeated.

A channel mismatch compounds the problem. Patients say they would rather hear about trials from their own doctor. Among those who actually participated, though, 22% first learned about the study online against 16% from their physician, and 48% of that online group found it through social media. Tracking referrals and enrollment status in one place lets sites report back to referring clinicians instead of leaving them in silence.

[[cta]]

Is the protocol itself the constraint?

Eligibility criteria are the lever with the clearest quantified effect, and the quantification comes from real-world data. Applied retrospectively to 10,500 patients with advanced non-small cell lung cancer, broadened ASCO and Friends of Cancer Research criteria produced a striking result: three commonly used criteria excluded 47.7% of the cohort, and broadening them cut exclusion to 1.5%. The eligible pool nearly doubled, from 5,495 to 10,346 patients.

The added patients were also more representative: women rose from 40% to 48% and patients aged 75 and over from 16% to 29%, with stage IV disease climbing from 59% to 72%. Within the same analysis, restrictive criteria in 231 phase I trials were associated with enrollment periods of 26 months against 17, and with more terminations.

The consensus position inverts the default: patients are eligible unless a scientific or safety rationale excludes them. Broader populations do increase heterogeneity, but any protocol review that fails to attach an estimated population cost to each exclusion criterion leaves the largest measurable lever unpriced.

Which solutions actually have evidence behind them?

Fewer than the discourse suggests. The reference Cochrane review covered 68 trials and more than 74,000 participants; of 72 comparisons tested, only three are supported by high-certainty evidence. They are running an open trial rather than a blinded placebo-controlled one (+10 percentage points), telephoning nonresponders to a mailed invitation (+6 points), and custom user-tested information leaflets, whose +1 point means little or no effect.

Decentralized elements show promising but uncontrolled signals. A pooled analysis of 1,069 participants across six sponsor studies reported 0.46 participants per site per month in oncology against a 0.40 benchmark. Completeness was strong on mandatory baseline fields (97.9% to 99.0%) and much weaker on optional assessments (47.5% to 49.2%).

The authors declined to claim causation, and adoption remains marginal at roughly 1% of new trials since 2022. Geography nonetheless keeps the rationale real: 40% to 50% of US patients with advanced cancers live at least an hour from the nearest recruiting site. That distance is where decentralized and hybrid designs earn their place, provided enrollment endpoints are prespecified rather than assumed.

Artificial intelligence performs well at classification and remains largely unmeasured on enrollment. Pooled sensitivity reaches 0.91 for patient screening and 0.92 for classifying eligibility criteria, but it falls to 0.70 for identifying patients or cohorts. More telling: 77% of the studies reviewed were retrospective, and fewer than half were applied in an actual trial. Models read eligibility text accurately; whether deploying them enrolls more patients is a different endpoint.

What do national performance figures reveal?

France offers the most granular public picture. In 2025, the median time from submission to first patient enrolled was 185 days, 116 of which were authorization. Even a flawless regulatory process can't compress the timeline below roughly 70 days, and the contracting block is larger: 79 days for the coordinating site contract, then 104 days from signature to first patient.

Site-level performance is more sobering. Among trials authorized in 2024, 83% of sites enrolled at least one patient, 40% reached half their target, and only 12% reached it in full. The mean rate was 0.39 patients per site per month.

Spain reaches first patient in about 162 days and reports roughly 95% of recruitment objectives met. That performance is attributed to a reflexive referral culture, dedicated trial offices, and a near-daily public registry. In South Korea, first patient comes in 120 to 150 days, helped by near-systematic molecular profiling in routine care that enables prescreening before activation.

One regulatory point deserves correcting: US Diversity Action Plans are a statutory requirement whose trigger has not fired. Mandatory submission applies only to studies enrolling more than 180 days after a final guidance, and none has been published. Describing the plans as currently mandatory, or as repealed, is inaccurate either way.

[[cta]]