Every enrollment forecast I have seen in clinical trial planning starts from the same place: the protocol-defined inclusion and exclusion criteria, the site investigator's estimate of how many eligible patients they see per month, and a target enrollment number that backs into a timeline. The math is usually right. The assumptions underneath it almost always underestimate screen-fail rates -- and that underestimation is where enrollment forecasts go wrong.
Screen failures are not a rounding error. In Phase II oncology trials, screen-fail rates above 40% are common. In autoimmune disease trials with restrictive biomarker criteria, rates of 60 to 70% are not unusual. When the model assumed 20%, the enrollment timeline built on that assumption becomes unreliable from day one. The question is not whether screen failures will happen -- they always do. The question is whether your operational infrastructure can detect the pattern early enough to respond before the schedule slips.
Why Screen-Fail Rate Estimates Are Systematically Too Optimistic
Site investigators giving enrollment estimates are usually well-intentioned and often experienced. But their estimates are based on their clinical patient population, not on the population that will actually pass screening for a specific protocol. There are three consistent sources of optimism bias in these estimates.
First, investigators count patients who meet the primary diagnosis criteria, not patients who meet the full inclusion/exclusion criteria. For a study with 15 to 20 eligibility criteria, the cumulative exclusion rate across all criteria compounds quickly. A patient might represent 1-in-5 on the primary diagnosis, then 1-in-3 on the required lab value range, then 1-in-4 on the washout period from prior therapy. The actual eligible fraction can be 3 to 5 times smaller than the investigator's intuitive estimate.
Second, estimates are rarely adjusted for the specific visit structure of the screening period. Studies with complex screening procedures -- multiple visits, invasive procedures, central laboratory requirements -- accumulate a category of failures that are not really eligibility failures: patients who qualify medically but withdraw during screening due to the burden of the screening process itself. These procedural withdrawals are structurally different from eligibility failures and require different remediation strategies, but they show up in the same screen-fail rate number.
Third, site estimates are almost never informed by historical screen-fail data from similar studies at that site. Most CROs do not systematically extract and share site-level screen-fail benchmarks from prior trials to inform enrollment projections for new studies. The information exists in completed study databases. It is rarely used.
The Compounding Effect on Timeline Projections
To understand why screen-fail rate errors are so damaging to enrollment timelines, consider a simplified example. A Phase II study needs 120 patients across 10 sites, with a 12-month enrollment window. The model assumes a screen-fail rate of 25%, so it projects that sites will need to screen 160 patients to randomize 120.
The actual screen-fail rate turns out to be 50%. To randomize 120 patients at a 50% fail rate, sites need to screen 240. At the same monthly screening rate, that is a 50% increase in time required. The 12-month enrollment window becomes 18 months. The study database lock moves out by 6 months. The IND filing slips by a quarter.
Now add the real-world complexity: screen-fail rates are not uniformly distributed across sites. Typically, two or three sites will perform significantly below average on screening conversion -- either because their patient population is less well-matched to the protocol criteria than expected, or because their screening workflow has gaps that cause eligible patients to fall out early. If those sites account for 40% of planned enrollment, the aggregate effect on timeline is larger than the average screen-fail rate suggests.
What Early Warning Looks Like Operationally
The operational window for managing screen-fail rates is the first 60 to 90 days after sites open. This is when patterns are most visible and most correctable. Sites that are failing at a high rate early are likely to fail at a high rate throughout the study. Sites that are converting well early tend to maintain that performance.
The signals to watch in this window are not just the aggregate screen-fail rate. They are the reason-for-failure distribution across sites. Most eCRF systems capture the primary reason a patient failed screening against the protocol's eligibility criteria list. That distribution is diagnostic:
- If failures cluster on a single inclusion criterion, the eligibility criterion may be set more restrictively than the realistic patient population supports. This is a protocol amendment discussion, not a site remediation problem.
- If failures are distributed across many criteria at one site but concentrated on one criterion at other sites, the first site may be misapplying a criterion or may have a genuinely different patient population than modeled.
- If failures are disproportionately labeled as "patient withdrawal" during screening rather than "eligibility failure," the screening visit burden may be the actual problem -- something addressable through site support, not protocol change.
This kind of reason-level analysis requires that the failure data be aggregated across sites in near-real-time and that someone is actually looking at the distribution. In our experience, most CROs produce site-level enrollment reports weekly or bi-weekly, but those reports show randomization counts and screen numbers -- not the reason-code distribution that reveals why the numbers are what they are.
The Operational Levers Before Timelines Slip
When early screening data reveals a problem, the response options depend on its root cause. These are not decisions to delay until the 3-month enrollment review. They are decisions that need to be made in the first 8 to 10 weeks.
If the issue is a single overly restrictive criterion: initiate a protocol amendment early. Protocol amendments take time to execute -- typically 6 to 10 weeks from submission to site implementation. Starting this process at week 8, when the pattern is first visible, is very different from starting it at week 20, after the timeline has already slipped significantly.
If the issue is site-specific screening conversion: assign a clinical research associate to work intensively with underperforming sites on patient identification and pre-screening workflows. Sites often fail not because their patient population is wrong, but because their referral patterns within the clinical practice are not well-aligned with the protocol. A clinical operations specialist who understands both the protocol and the site's workflow can often identify the disconnect quickly.
If the issue is patient burden during the screening period: consider whether any screening procedures can be consolidated, resequenced, or moved to post-randomization assessments. This requires sponsor agreement and potentially a protocol amendment, but it can substantially reduce procedural withdrawals without affecting the scientific validity of the study.
If the issue is insufficient eligible patient volume at current sites: add sites. This sounds obvious, but it requires lead time. Site activation for a new site added mid-study typically takes 8 to 12 weeks, including IRB approval and regulatory document review. Sites added in response to an enrollment problem that was identified at week 10 can begin contributing by week 22. Sites added in response to a problem identified at week 20 cannot contribute meaningfully until the study is already significantly behind schedule.
Building Screen-Fail Analysis into Your Enrollment Dashboard
The practical takeaway is that screen-fail monitoring should not be an ad-hoc analysis triggered by a sponsor call. It should be a default feature of your enrollment operations view, built into the weekly data review from day one of site activation.
At minimum, your real-time enrollment dashboard should show, per site: screens initiated, screens completed, randomizations, screen-fail rate (rolling 30-day and cumulative), and top three reasons for failure by frequency. Threshold alerts should trigger when a site's 30-day screen-fail rate exceeds the study average by more than 20 percentage points, or when the cumulative screen-fail rate exceeds the model assumption by more than 10 points.
These are not sophisticated analytics. They are basic operational visibility. The difference between teams that catch enrollment problems early and teams that are still troubleshooting them at month 6 is usually not analytical capability. It is whether someone has set up the dashboard to surface the right numbers in the right time frame -- and whether the escalation path when those numbers go red is defined before the numbers go red, not after.
Screen-fail rates are predictable with better data and correctable with faster action. Both of those things are operational decisions, not scientific ones.
