When a Phase II trial misses its enrollment target, the post-mortem almost always points to sites. Low patient volume, slow chart turnaround, CRAs stretched thin across too many protocols. It's a reasonable diagnosis — and it's almost always incomplete. The deeper problem was locked in weeks before the first site activated, at the moment a feasibility coordinator opened a spreadsheet and started estimating how many eligible patients a site might have based on prior experience and gut feel.
Manual chart review is not a site execution problem. It's an infrastructure problem — and it sits directly upstream of every enrollment timeline decision a CRO makes.
What a Feasibility Assessment Actually Involves
The standard feasibility questionnaire asks sites to self-report: How many patients matching this general profile do you see per year? Do you have experience with this indication? How long did your last comparable trial take to reach target enrollment? These are useful questions. They are not the same as asking how many patients in your EHR right now meet these 14 specific inclusion and exclusion criteria.
The gap between those two questions is where delays are manufactured.
A site coordinator answering a feasibility questionnaire for a Phase II NASH trial — non-alcoholic steatohepatitis, a notoriously narrow population — might estimate 30–50 eligible patients based on their hepatology volume. That estimate rarely accounts for the compounding effect of exclusion criteria: prior liver biopsy within 12 months, documented cirrhosis (SNOMED CT 19943007), HbA1c ≥ 10%, or current use of thiazolidinediones (RxNorm code class). When those criteria are applied to the actual EHR population, the eligible cohort frequently shrinks by 60–75%.
The estimate isn't dishonest — it's structural. Site coordinators aren't running FHIR R4 queries against their Epic patient population. They're drawing on experience, which doesn't account for a protocol they've never run before.
Where the 40% Figure Comes From
Industry enrollment literature consistently places 40–50% of trial delays as originating in the feasibility and site activation phase — not in post-activation screen failures. The mechanism is straightforward: if a feasibility assessment overestimates eligible patient volume by a factor of two, the CRO negotiates an enrollment timeline with the sponsor that is physically impossible to meet at the selected sites. By the time this becomes visible — typically 8–12 weeks into enrollment when randomization rates plateau — the damage to the trial timeline is already done.
We're not saying that site execution problems don't exist. They do. Screen-fail rates are real; protocol deviations add complexity; site staff turnover disrupts enrollment momentum. But these problems compound an already incorrect feasibility baseline. They don't cause the problem from scratch.
The distinction matters operationally because the remedies are different. A site execution problem calls for more CRA support, protocol amendment, or site replacement. An incorrect feasibility baseline calls for a different approach to patient identification before the first site is selected.
The Anatomy of Manual Chart Review
Consider what actually happens when a CRO feasibility team tries to validate a site's patient population estimate. A CRA or feasibility coordinator manually pulls records from the site's EHR — if they have access — or requests a manual pull from site staff. They scan encounter records, progress notes, and lab values looking for patients who appear to match the protocol criteria. For a medium-complexity oncology protocol with 12 inclusion criteria and 18 exclusion criteria, a thorough single-patient review takes 15–25 minutes.
Across a 10-site feasibility assessment, at 40 patients reviewed per site, that is 400 chart reviews — roughly 100–170 hours of clinical review time, spread across a process that typically runs in parallel with protocol finalization and site contracting. The review is performed independently at each site, by different people, interpreting the same free-text protocol criteria in slightly different ways. Two reviewers at two different sites will not code the same ambiguous criterion identically.
A mid-size Phase II cardiometabolic program at a Durham-area CRO running feasibility across nine sites in the Southeast estimated in early 2025 that their feasibility team was spending approximately 3.5 CRA-days per site on chart review and documentation — nearly 32 days total across the site network, before a single patient had been consented.
What Structured EHR Matching Changes
Structured EHR matching against protocol criteria doesn't eliminate chart review — it redirects it. The identification phase, where a CRA works through a list of potentially eligible patients without knowing who actually qualifies, can be replaced by a structured query against coded EHR data: ICD-10-CM diagnosis codes, LOINC lab values, RxNorm medication histories, and SNOMED CT condition codes that already exist in the patient record.
The output isn't a consent — it's a ranked cohort list. Patients whose records contain structured data matching the protocol criteria surface at the top; patients whose records have gaps or contradictory findings score lower. A CRA doing chart review on the top 15 candidates from a structured match run is doing confirmatory review, not discovery review. That is a fundamentally different task, both in time and in error profile.
FHIR R4 standardizes how this data is accessed across EHR systems. A Patient resource paired with Condition, Observation, and MedicationStatement resources contains the structured data needed to evaluate most Phase II inclusion and exclusion criteria — without transferring identified patient records outside the health system. See our overview of how the matching engine works for a walkthrough of this data flow.
The Limitation EHR Matching Doesn't Solve
Structured EHR matching works when the criteria can be expressed in coded clinical terminology and when the relevant data is documented in structured EHR fields. That describes the majority of common criteria in cardiometabolic, oncology staging, and chronic condition trials. It does not describe every trial.
Criteria that depend on symptom duration documented only in free-text progress notes, or eligibility determined by a specialist's clinical judgment that isn't captured in any code, cannot be matched from structured data alone. For these cases, natural language processing (NLP) on clinical notes — a meaningfully more complex technical undertaking with its own accuracy trade-offs — becomes necessary. The structured matching approach is the right tool for a large fraction of feasibility assessments; it is not a universal solution for every protocol complexity.
Protocol design itself also affects matchability. Criteria written with vague temporal qualifiers ("recent prior therapy") or dose thresholds that require calculation across multiple medication records ("cumulative anthracycline exposure ≥ 450 mg/m²") require additional logic that goes beyond direct field matching. We discuss how structured criteria design at the protocol level affects downstream matching accuracy in a separate piece.
Reclaiming the Feasibility Timeline
The practical implication for a CRO feasibility team is not that manual chart review should be abandoned — it's that the effort should concentrate where it adds value. Confirmatory review of a pre-filtered, high-probability cohort is the right use of a CRA's clinical expertise. Discovery review of an unfiltered EHR population is a use of that same expertise that a structured query can handle more quickly and more consistently.
Sites that can provide FHIR R4 API access through their EHR — Epic via SMART on FHIR, Oracle Health (formerly Cerner) via FHIR R4, and increasingly Athenahealth and eClinicalWorks — can participate in a structured feasibility match run that produces a cohort size estimate with more precision than a questionnaire response, before any manual chart review is performed.
That changes what a feasibility assessment is. Instead of asking a site to guess its eligible patient population and then spending CRA time validating or disproving that guess, a structured match run produces the denominator directly. Site selection decisions built on that denominator are more defensible to sponsors, more accurate in timeline projections, and less likely to produce the 8-week enrollment plateau that triggers crisis management calls.
Forty percent of trial delays start at feasibility not because CRO teams aren't trying hard enough, but because the information they've been working with is inherently imprecise. The EHR data to answer the question more precisely already exists at the sites. Accessing it structurally — rather than case-by-case through manual review — is the point where the feasibility process changes. For more on how Cohortbridge structures this access, visit the EHR integrations overview.