Deep Research

Eligibility Criteria and Screening Burden: A Study-Build Evidence Review (2026)

30 min readEClinCloud Editorial Team
Eligibility and screening evidence — EClinCloud research cover

The difficult part of an eligibility build is deciding what each criterion requires someone to know, when they must know it, and what evidence supports the decision. A completed checklist can conceal an expired laboratory result, a misunderstood time window or an undocumented clinical judgment. A useful build makes those dependencies visible before anyone tries to randomize a participant.

Evidence reviewed: September 18, 2026. This report addresses protocol, clinical-operations and data-management teams. Its build examples and workload scenarios are hypothetical and require adaptation to the approved protocol. The analysis combines published eligibility research, regulatory sources and a new check of FDA inspection-observation frequencies.

TL;DR

  • Challenge unnecessary criteria before configuring them. FDA's December 2025 final guidance addresses representative enrollment and unnecessary exclusions. The first decision is whether a criterion serves the trial's scientific or safety purpose; the second is how to implement the retained criterion reliably. [1]
  • Use inspection data with its actual denominator. Across FY2020–FY2025, the reviewed BIMO sheets contain 479 frequency rows totaling 1,715 citation occurrences. The 522 occurrences under 312.60 include 481 protocol-compliance and 41 informed-consent citations. These data support targeted review questions, not a measured screening-error rate.
  • Separate the evidence, evaluation and authorization. Capture the source result and its context, apply the approved criterion, and record the authorized eligibility decision. A missing result should remain unresolved; a software permission should reflect the documented decision and the applicable protocol version.

Which criteria deserve a place in the protocol?

Study build begins with a scientific question. For every proposed inclusion or exclusion, ask which part of the trial's purpose or participant protection it serves and whether the restriction is necessary for that purpose. FDA's December 2025 final guidance encourages approaches that improve representation while maintaining safety and effectiveness standards. Its scope includes demographic characteristics and matters such as comorbidities, organ dysfunction and disability. The appropriate response is a documented, product-specific review of the criteria before the build freezes them. [1]

ICH E8(R1) places attention on factors critical to study quality and on designing quality into a study. That provides a useful reason to bring investigators, operational staff, data managers and relevant participant perspectives into protocol development. Each group sees a different failure mechanism. A statistician may identify a criterion needed to interpret the outcome; an investigator may question whether its wording can be applied consistently; a coordinator may identify a scheduling dependency; a participant may identify a visit burden that the protocol team overlooked. [2]

Keep the decision record short and concrete. For a retained criterion, state its purpose, the information needed, the permitted source, the timing requirement and the responsible decision maker. For a removed or broadened criterion, preserve the scientific and safety rationale and the approval route. This is a proposed design aid. It helps prevent operational convenience from quietly becoming an unexamined restriction on the study population.

Published research shows why a criterion-by-criterion review can be more useful than copying a predecessor protocol. Liu and colleagues used real-world data from 61,094 patients with advanced non-small-cell lung cancer to emulate oncology trials and evaluate eligibility criteria. Their 2021 study illustrates a data-driven approach to examining inclusion and outcome implications. Its findings depend on the disease, treatments, available variables and observational methods. A sponsor still needs a trial-specific scientific and safety justification before changing a criterion. [3]

This distinction affects the build discussion. Evidence that a restriction deserves reconsideration should reach the protocol decision makers before programming. Once the approved protocol defines the population, the system must implement that version faithfully. Quietly relaxing a threshold in an edit check creates a different problem from proposing a justified protocol amendment. Preserve both the protocol decision and its implementation history so the trial team can distinguish a design improvement from an execution error.

A practical meeting output is a list of unresolved criterion questions with named owners. “Adequate function,” “recent treatment” and “clinically significant” may each require a different kind of clarification. Some need a defined measure and interval; others intentionally require clinical judgment. The team should resolve which is which before turning every sentence into a mandatory yes/no field. Otherwise, the form may look complete while different sites apply different rules.

What does the FDA citation census support?

We inspected the shared data catalog and the FDA inspection-observation descriptor and manifest before analyzing the pinned Bioresearch Monitoring subset. The source snapshot is August 23, 2026. Its compressed data checksum matches the manifest. We selected fiscal years 2020 through 2025, retained all BIMO program-area rows and summed the published frequency field. The resulting 479 rows represent 1,715 citation occurrences. The public calculation supplement includes the selected rows, exact grouping rules and annual totals. [4]

The unit matters. These are frequency aggregates from BIMO sheets covering multiple oversight areas, including clinical investigators, sponsors, institutional review boards and laboratory-related work. They are not 1,715 clinical-investigator inspections, unique sites, trials or participants. One inspection can contribute several observations. FDA's inspection-observation page describes observations recorded during inspections, while the Form 483 FAQ explains that a Form 483 is not a final agency determination of a violation. [5] [6]

Grouping by regulation reference gives 522 occurrences under 21 CFR 312.60 and 271 under 312.62(b), totaling 793, or 46.2% of all selected BIMO frequencies. However, 312.60 contains two citation identifiers in these files. Citation 7560 concerns investigator-statement and protocol compliance, with 481 occurrences. Citation 7562 concerns informed consent, with 41. Calling all 522 “protocol noncompliance,” or calling the combined 793 “screening failures,” would erase distinctions present in the source. [4]

Separate the two citation identifiers under 312.60

FY2020–FY2025 BIMO frequencies. Denominator: 1,715 occurrences across all selected BIMO program rows, not investigator inspections.

Unit · Citation occurrences

Scroll sideways for the full figure.

Separate the two citation identifiers under 312.60312.60: protocol comp…312.60: protocol compliance (7560), Frequency: 481 Citation occurrences481312.60: informed cons…312.60: informed consent (7562), Frequency: 41 Citation occurrences41312.62(b): case histo…312.62(b): case histories, Frequency: 271 Citation occurrences271
View chart data
CategoryFrequency
312.60: protocol compliance (7560)481 Citation occurrences
312.60: informed consent (7562)41 Citation occurrences
312.62(b): case histories271 Citation occurrences
Source: FDA inspection-observation workbooks; EClinCloud calculation, snapshot August 23, 2026.

The annual denominator also changes. Total selected BIMO frequencies are 221 in FY2020, 249 in FY2021, 298 in FY2022, 301 in FY2023, 312 in FY2024 and 334 in FY2025. The corresponding 312.60 totals are 58, 90, 77, 104, 104 and 89. Case-history frequencies are 30, 48, 36, 49, 56 and 52. These series describe published observation frequencies under an inspection program, with changing activity and case selection. They leave the prevalence of eligibility errors across all trials unresolved.

A separate FDA source supplies the specific screening connection. The FY2024 clinical-investigator observation-trends presentation lists eligibility themes including unmet inclusion criteria, met exclusion criteria and randomization before eligibility was established. It also includes many other protocol themes. This supports examining the eligibility workflow directly. The presentation's list of themes provides no basis for assigning all protocol observations to eligibility or estimating the reduction achievable through a particular EDC feature. [7]

The decision for a build team is therefore to review the timing and evidence of eligibility determinations, then measure its own process with defined study data. The census helps justify the question. It supplies neither a universal ranking of software controls nor a promised return on investment. That boundary keeps the article useful to teams whose studies differ substantially from the inspected activities in the public records.

How should a criterion become a build specification?

For US IND investigations, 21 CFR 312.60 establishes investigator responsibilities for conducting the investigation according to the investigational plan and protecting participants. Section 312.62(b) addresses adequate and accurate case histories for the individuals within its stated scope. Those obligations make it useful to connect an eligibility decision to the evidence on which it rests. The build specification below is a proposed operational method for creating that connection. [8] [9]

Give each criterion a stable identifier within a controlled protocol version. Preserve the original wording, then describe the operational interpretation in a separate field. That separation allows a reviewer to identify an unresolved ambiguity instead of mistaking a programmer's interpretation for approved protocol language. Link any clarification to the process by which the sponsor and investigator teams approved it, and determine whether the clarification requires a formal amendment or another controlled document change.

Identify the decision type. A criterion may depend on a fixed value, a result relative to a laboratory reference range, a time interval, a previous treatment, a documentary review, a clinical judgment or a combination of these. One compound sentence can require several inputs. Conversely, several criteria may use the same underlying measurement. Mapping those dependencies helps avoid collecting duplicate information while preserving the separate evaluations that the protocol requires.

For each input, specify the source and context. A laboratory value needs the analyte, specimen or relevant method context, collection date, result date where relevant, unit and applicable reference limit. A treatment history may need the treatment identity and dates sufficient to evaluate the specified interval. A clinical judgment needs the authorized assessor and enough supporting information to understand the conclusion. These are examples of an evidence model, not a requirement to upload every source document into the sponsor's database.

Turn each criterion into an explicit evidence decision

Proposed study-build aid; the approved protocol and applicable requirements control the actual workflow.

Scroll sideways for the full figure.

Criterion typeInputs to defineDecision evidence
Fixed thresholdValue, unit and permitted sourceApproved rule and evaluated result
Relative laboratory limitResult, reference limit, laboratory and effective dateDerived ratio and applicable threshold
Time intervalStart event, end event and counting conventionDates, calculated interval and rule version
Clinical judgmentRelevant information and authorized assessorAttributable assessment with appropriate support
Compound conditionLogical operators, exceptions and subgroup scopeEvaluation of the complete approved condition
Source: EClinCloud implementation analysis informed by ICH E6(R3) and 21 CFR 312.60/312.62.

Define the evaluation separately from authorization. The system may calculate a ratio or flag a missing result, while a qualified investigator or other properly authorized person makes the clinical determination required by the protocol. ICH E6(R3) addresses investigator responsibilities, proportionate oversight and reliable trial information. Delegation and computerized support should be reflected in the trial's actual responsibilities, rather than inferred from whoever can click a button. [10]

Finally, specify the downstream consequence. A criterion can generate a query, keep an eligibility assessment pending, prevent a configured randomization action or require an authorized review. The consequence should fit the decision and the system architecture. If a clinically important determination lives in a site record, the interface may carry its verified status and provenance instead of duplicating sensitive narrative. That design still needs a tested way to establish which decision was in force when the participant proceeded.

Distinguish preliminary identification of potential participants from procedures performed solely for research eligibility. FDA's screening guidance explains the importance of consent before research-only screening tests and of making their purpose clear to prospective participants. Existing clinical information can raise different questions from a new procedure performed specifically for the trial. The applicable law, ethics review, privacy basis and approved protocol determine the permitted workflow. A form labeled “prescreening” resolves none of those questions by itself. [11]

The operational design should identify when research procedures begin, which information may be accessed beforehand and what authorization supports that access. Record the approved consent document version, language, process and timing where required. FDA's informed-consent guidance discusses the respective roles of IRBs, investigators and sponsors and treats consent as more than possession of a signed form. A system workflow should preserve the evidence required by the approved process and provide a route for questions before participation. [12]

Avoid turning an incomplete consent record into an ordinary eligibility failure. The two states have different meanings and may require different follow-up. A potential participant who declines participation, a person who has consented but is awaiting a test, and a person who fails a protocol criterion should remain distinguishable in the operational record. This helps the site manage the next action and helps analysts choose a meaningful denominator later.

A staged screening plan can reduce unnecessary procedures when the protocol and ethical requirements permit that sequence. Start with the checks whose results are needed to decide whether later activities should occur, while respecting safety, scientific and scheduling dependencies. Some assessments must occur together; others rely on information that takes several days to obtain. The proposed order should be reviewed with sites before it becomes a rigid workflow that makes an otherwise feasible visit impossible.

Consider a hypothetical scheduling example. A preliminary document review takes 10 minutes. A later assessment bundle takes 40 minutes, and 60% of candidates are expected to proceed to it. Expected staff time for those two activities is 10 plus 0.60 times 40, or 34 minutes per candidate entering the review. If the proportion proceeding is 80%, the estimate becomes 42 minutes. These are planning assumptions, not measured performance or a recommendation to withhold required procedures. The calculation supplement records the formula and scenarios. [4]

Expected staff time depends on who reaches the next stage

Hypothetical: 10-minute preliminary review plus a 40-minute later bundle. Travel, waiting, repeats and clinical review are excluded.

Unit · Minutes per candidate

Scroll sideways for the full figure.

Expected staff time depends on who reaches the next stage010.52131.54240% proceed, Expected staff time: 26 Minutes per candidate40% proceed60% proceed, Expected staff time: 34 Minutes per candidate60% proceed80% proceed, Expected staff time: 42 Minutes per candidate80% proceed
View chart data
CategoryExpected staff time
40% proceed26 Minutes per candidate
60% proceed34 Minutes per candidate
80% proceed42 Minutes per candidate
Source: EClinCloud illustrative arithmetic; no observed effectiveness claim.

The model omits participant travel, waiting, repeat tests, medical review and administrative follow-up. Add those components explicitly when they affect the decision. A ten-minute task completed during an existing visit and a ten-minute task requiring another trip impose very different participant burdens. Separating staff time from elapsed time and participant effort gives the protocol team a better basis for revising the workflow than counting the number of eligibility lines alone.

How should laboratory thresholds and time windows be represented?

A threshold rule should preserve the value on which it operates. In a hypothetical protocol, suppose a laboratory criterion is expressed as a result no greater than 1.5 times the applicable upper limit of normal. A result of 72 with an upper limit of 50 yields 1.44; the same result with an upper limit of 45 yields 1.60. The difference comes from the reference limit, not a change in the measured result. The example illustrates data dependencies and carries no clinical recommendation about an actual analyte or threshold. [4]

Capture the original unit and reference range with the result, identify the laboratory and preserve relevant effective dates. If the system standardizes units, retain the original value and the controlled conversion rule as well as the derived value. A conversion should be specific to the measurement and validated for the intended use. Treating every numerical laboratory result as directly comparable because it appears in the same column can hide clinically meaningful differences in context.

Define how the protocol handles repeated measurements. The first result, the most recent result and a permitted confirmatory result answer different questions. The specification should identify which result may support eligibility and under what conditions another result is allowed. An operator selecting whichever value passes can create an unplanned decision rule even when every individual value was transcribed correctly. Preserve the sequence and the reason for the selected result.

Time-window rules need an equally explicit reference event. “Within 14 days” may depend on consent, randomization, dosing or another protocol-defined event. Specify whether the calculation uses collection time or report time, calendar days or elapsed hours, and how boundary dates are treated. Where the protocol leaves an important ambiguity, obtain a controlled clarification before programming. The site should be able to see both the input dates and the interpretation used by the system.

Test a hypothetical boundary case in which a sample is collected on September 1 and randomization is planned for September 15. A calendar-date difference is 14 days, while inclusive counting of named calendar dates gives 15 dates. The system must apply the approved convention. Similarly, a late report can arrive after a permitted collection window even though the sample itself was timely. Distinguish an expired assessment from an unavailable report, because they may lead to different follow-up under the protocol.

FDA's electronic-systems guidance addresses trustworthy electronic records and the controls relevant to their use in clinical investigations. For this build, a practical application is to keep the raw input, relevant metadata, derived assessment and changes traceable. Record the effective laboratory range and rule version so that a later range update leaves earlier decisions interpretable. A recalculated display should make clear whether it represents the historical decision or a current reassessment. [13]

Which rules should block randomization?

Start from the approved decision and the consequences of proceeding with unresolved information. A deterministic check can help prevent a known invalid state, such as a required assessment being absent. A clinical judgment may need an authorized determination and supporting record. A purely administrative query may be compatible with a completed eligibility decision, depending on its content and the protocol. The categories should be agreed by the trial team and tested in the intended workflow. [8] [10]

Use explicit statuses: not assessed, awaiting information, assessed eligible, assessed ineligible, and other protocol-defined states where necessary. A blank field should retain its missing-information meaning. Likewise, “not applicable” needs a defined condition. Defaulting every unanswered exclusion to “No” can create a complete-looking form from incomplete evidence. A useful interface shows which unresolved dependencies prevent the authorized decision and who is responsible for resolving them.

The randomization interface should carry enough information to establish the participant, protocol version, relevant assessment status and authorization. Define which system owns each item and what happens if an update is delayed. A displayed green status in EDC and an old status in RTSM can disagree even when each system is functioning according to its own rules. The integration specification needs an explicit rule for freshness, acknowledgments and failure handling.

Test the timing of changes. If a laboratory result is corrected after eligibility is approved but before randomization, identify how the assessment is reopened or flagged. If randomization has already occurred, preserve the historical record and route the new information for the appropriate clinical and protocol-deviation assessment. Changing a status should not silently erase the event that already occurred. The operational response depends on the circumstances and responsible clinical judgment.

A medical-monitor discussion can clarify an ambiguity or inform a safety decision. It should not be represented as an unrestricted power to waive the approved eligibility criteria. Define the permitted review process, documentation and escalation route. Where a protocol change is needed, use the applicable amendment and approval process. Software should support that governance and preserve its evidence instead of treating an override button as the authority for an exception.

Plan for downtime before the first participant. A controlled contingency process needs to establish which information is required, who can authorize the action, how records are retained and how reconciliation occurs when systems recover. Some circumstances may require postponing the action. The correct result follows from the protocol, participant protection and validated process, not from a general promise that enrollment can always continue offline. Document and test the selected contingency alongside the normal pathway.

What should be retained for a screen failure?

Define the purpose and scope of the screening record before deciding how much information to collect. A minimal operational record may identify the screening episode, consent status where applicable, assessments undertaken, disposition and coded reason for ending the episode. The source location and responsible assessor may be needed to reconstruct a particular determination. Collection should fit the approved study, applicable privacy requirements and the purpose of the record. Avoid collecting a full participant database merely because a screening number exists. [10] [12]

The text of 21 CFR 312.62(b) specifically describes case histories for individuals administered the investigational drug or employed as controls. It should not be paraphrased as a universal rule that every person considered for a study has the identical case-history obligation. Other requirements, the protocol, ethics conditions and the actual activities performed can affect screening documentation and retention. Establish the applicable policy for the study rather than making a blanket statement from that subsection alone. [9]

Differentiate the reason a person stopped from the criterion that was assessed. Withdrawal of interest, inability to attend, unavailable information, a failed inclusion criterion and a met exclusion criterion can have different operational meanings. If several criteria fail, define whether the record holds the first identified reason, the primary reason or all assessed reasons. That choice changes the resulting statistics. Preserve an “assessment incomplete” state when later criteria were never evaluated.

Rescreening needs a controlled identity relationship. The same person may have several screening episodes under the protocol's permitted process. Retain the connection without reusing an episode identifier in a way that overwrites earlier results. Count people and episodes separately when reporting them. A person-level recruitment estimate and an episode-level laboratory workload estimate can both be useful, but they answer different questions.

ClinicalTrials.gov registration definitions distinguish structured eligibility fields, such as age and sex, from narrative eligibility criteria. Registry design fields can help describe a study, but the number of arms or countries is not a validated estimate of screening burden. A multicountry study may use harmonized assessments; a single-country study may have difficult, highly specific eligibility procedures. For build planning, read the actual criteria, schedule and implementation requirements. [14]

Results reporting also needs its own definition check. Under 42 CFR 11.48, participant flow includes information about progress through a trial by arm, including numbers starting and completing. This should not be converted into a blanket assertion that every trial must publish every prescreened person's reason for screen failure. Map the study's operational counts to the applicable results fields and reporting requirements, keeping the underlying definitions available for reconciliation. [15]

How can teams measure burden without confusing the denominators?

Choose the question before selecting a rate. A site may need to understand how many people considered for the trial reach consent. A sponsor may need the proportion of completed screening episodes that lead to randomization. A data manager may need the frequency of missing eligibility evidence at the planned decision time. A participant-engagement team may need the number of extra visits or elapsed days. Each measure requires a different numerator, denominator and observation window. [4]

For a hypothetical cohort, suppose 100 people begin a defined screening episode. At the reporting cutoff, 30 have been randomized, 40 have been found ineligible, 10 have withdrawn and 20 remain unresolved. Dividing 40 by 100 gives the proportion already found ineligible among all episodes started. Dividing 40 by the 70 with an eligibility determination gives the ineligible share among determined episodes, approximately 57.1%. Both can be calculated correctly while describing different states. Label the cutoff and unresolved group rather than presenting either number as a universal screen-failure rate.

The same screening cohort supports different rates

Hypothetical cutoff: 100 episodes started; 30 randomized, 40 ineligible, 10 withdrawn and 20 unresolved.

Scroll sideways for the full figure.

MeasureNumeratorDenominatorResult
Already found ineligible among all episodes started4010040.0%
Ineligible among episodes with an eligibility determination407057.1%
Unresolved among all episodes started2010020.0%
Source: EClinCloud illustrative arithmetic; definitions must be prespecified for the study.

Now suppose several people were rescreened. The number of screening episodes can exceed the number of people. A person might be ineligible in one episode and eligible in a later permitted episode. Report the rule for assigning their final status in a person-level analysis. For an episode-level workload analysis, retain both episodes because both consumed resources. The relationship between those views should be explicit enough for operations and statistics teams to reconcile their reports.

Site comparisons need context. A site receiving more preliminary referrals may show more unsuccessful episodes while contributing useful recruitment work. Another may complete extensive prescreening outside the captured workflow and appear to have a high success rate. Differences in disease severity, access to testing, reporting delay or recruitment sources can affect the comparison. Use the metric to identify a question for investigation before attributing performance to staff or software.

FDA's risk-based monitoring questions and answers supports directing oversight toward important risks and data. For eligibility, a study-specific monitoring plan might examine unresolved evidence near randomization, late authorizations, repeated corrections or discrepancies between source and derived status. These are proposed indicators. The appropriate thresholds, review frequency and escalation route should reflect the protocol and the consequences of an incorrect determination. [16]

Keep burden and quality measures together. A faster workflow may still be incomplete; a lower screen-failure proportion may reflect a change in referrals rather than better execution. Track the evidence needed to interpret the apparent improvement, including protocol amendments, site activation dates and any change in screening definitions. This turns the screening log into a usable operational dataset while preserving the uncertainty around causal claims.

What should user acceptance testing demonstrate?

Build testing should reproduce the decisions that matter, including ambiguous and interrupted paths. FDA's electronic-systems guidance provides a basis for proportionate validation and controlled use of systems. The following test set is a proposed starting point for eligibility configuration, to be expanded or narrowed according to the study. A passed generic platform test is useful background; the study-specific rules and integrations still need applicable evidence. [13]

Start with boundary values. Test a result exactly at the threshold, immediately on either side, missing, corrected and reported with an unexpected unit. Include a changed reference range and a result from an unapproved or incorrectly identified laboratory where that distinction matters. The expected outcome should be documented before execution. A tester should be able to explain why the system's response follows the approved rule rather than accepting whatever the current configuration produces.

Test time boundaries independently. Include the earliest and latest permitted assessment, an expired assessment, a delayed report, a corrected collection date and a planned randomization date that changes. Where time zones or daylight-saving transitions affect the chosen rule, include them explicitly. A study using calendar dates may need different cases from one using elapsed hours. Preserve that choice in the specification and the test evidence.

Test the decision boundaries and interrupted paths

Proposed examples to adapt to the study-specific validation plan.

Scroll sideways for the full figure.

Test familyExampleExpected evidence
Value boundaryAt threshold, above, below, missing or correctedApproved rule and traceable outcome
Time boundaryCollection date, report delay and changed randomization dateCorrect reference event and counting convention
AuthorizationRequired reviewer pending or role withdrawnPermitted actions and attributable decision
IntegrationDelayed, duplicate or rejected status messageDefined recovery with correct participant and episode
AmendmentSites on different approved versionsVersion-specific behavior and historical traceability
Source: EClinCloud implementation analysis informed by FDA electronic-systems guidance.

Test authority and state transitions. A coordinator may enter information while the required investigator authorization remains pending. A user may lose a delegated role. A previously approved assessment may reopen after a correction. A participant may be rescreened under a new episode. The test should demonstrate the permitted actions for each role and state, including the information retained in the audit trail and the status received by downstream systems.

Test integration failure as a first-class case. Delay a message, repeat it, reject it or interrupt the connection, using a controlled test environment. Verify that the participant remains associated with the correct episode and protocol version, that a duplicate message causes the defined response, and that an operator can identify a failed handoff. The recovery process should avoid issuing a second randomization merely because an acknowledgment was delayed.

Finally, test the reviewer experience. Can a qualified person locate the source context, understand the calculation and determine who authorized progression? Can a later correction be distinguished from the original information available at the decision time? Can the team export or retain the necessary record when the system is retired? These questions connect the build to its intended evidence use rather than limiting acceptance to whether forms render and buttons respond.

How should extracted text and clinical judgment be reviewed?

Automated extraction can help turn a long protocol or laboratory report into a proposed data structure. The useful output is a draft mapping with its source location, uncertainty and unresolved questions. The responsible team then reviews the mapping against the controlled document. FDA's electronic-systems guidance and ICH E6(R3) both make the reliability and intended use of trial information relevant to the assessment of the supporting process. The following review approach is a proposed implementation aid. [13] [10]

Start with logical structure. A criterion containing “and” differs from one containing “or,” and an exception can reverse the meaning of an otherwise simple exclusion. A statement such as “previous treatment is excluded unless the specified interval has elapsed” needs the treatment history, the applicable interval and the reference event. A text-extraction tool that captures the treatment name but drops the exception produces an apparently structured rule with the wrong meaning. Preserve the sentence context for human review.

Next, check scope. A threshold may apply only to a defined subgroup, a particular screening stage or a specific measurement method. The page containing a table may refer to a footnote elsewhere in the protocol. A list item may inherit a condition from its parent heading. The reviewer should trace the proposed rule back through those dependencies and record any unresolved interpretation. Testing only whether the extracted number matches the printed number misses these errors.

For laboratory reports, review identity and metadata alongside values. A correctly recognized result attached to the wrong screening episode can be more consequential than an obvious transcription error. Check the participant or source identifier permitted by the workflow, collection date, laboratory, analyte, unit and reference limit. Where identifiers are redacted before transfer, the reconciliation process still needs a controlled way to preserve the correct association without exposing unnecessary personal information.

A review queue should distinguish extraction uncertainty from an eligibility concern. A low-confidence character in a report means the captured value needs verification. A verified value outside a protocol threshold means the eligibility assessment needs the protocol-defined response. Combining both into a generic red flag makes it harder for staff to know whether to inspect the document, repeat a permitted assessment or seek the responsible clinical review. Record the resolution of each type of issue separately.

Clinical judgment requires its own supporting record. Some criteria deliberately ask an investigator to assess clinical significance or suitability in context. The system can present relevant information and require an attributable assessment, while the authorized clinician remains responsible for that judgment. A mandatory free-text justification for every routine assessment may add burden without improving the evidence. Agree on the documentation needed for the particular criterion and the circumstances requiring additional explanation.

Finally, assess changes to the extraction or decision-support process. A new model, prompt, document template or laboratory format may alter the output even when the protocol is unchanged. Define which changes trigger review and regression testing, retain the approved version for the study, and monitor the types of corrections reviewers make. Those correction patterns can identify a recurring mapping problem. They should lead to a controlled improvement and an assessment of affected records rather than an unexplained change to rules already in use.

How should amendments and implementation ownership be handled?

An eligibility amendment changes a population definition and may change data collection, interpretation and system behavior. Start with the approved wording and effective dates, then identify which sites and participant states are affected. A criterion may apply differently to new screening episodes, ongoing episodes or already randomized participants. The protocol and applicable approvals determine the treatment; a global software switch should implement that decision precisely. [8] [10]

Create an impact map across the criterion specification, forms, edit checks, laboratory ranges, site instructions, integrations, monitoring indicators and reporting definitions. Preserve the relationship between old and new criterion identifiers where useful, while keeping changed meanings visible. A renamed field and a changed clinical meaning are different changes. Existing data may require reinterpretation, additional collection or no action, depending on the amendment and the participant's state.

Plan site activation for the amendment. Identify training, approvals and configuration conditions required before a site uses the new version. Test a mixed-version period if sites can legitimately transition at different times. The operational dashboard should make version status visible without assuming that document distribution proves implementation. Ask sites to demonstrate the changed workflow with the relevant example cases.

EClinCloud's published EDC material describes study-build support, forms and edit checks, source capture and data-manager review of AI-generated configuration. Its RTSM material describes randomization and supply workflows and study-specific validation. Those are relevant capabilities to evaluate for this evidence model. The study team should confirm the actual configuration, integrations, permissions and validation scope for its deployment, rather than treating a product description as proof that a particular protocol is already implemented correctly. [17] [18]

A useful implementation handoff contains the approved criterion specification, open-question log, source and data-flow map, role matrix, test evidence, contingency process and amendment plan. Assign each item to a named function and identify the person authorized to accept it. The vendor can implement and support agreed functions; the sponsor and investigator teams retain their applicable scientific, clinical and oversight responsibilities.

Before first screening, walk one hypothetical participant through the complete process. Include consent, a pending result, a corrected value, an eligibility decision and the handoff to randomization. Follow the record forward and backward. The intended outcome is a workflow in which people can see what is known, what remains unresolved and which authorized decision permits the next step. That is the practical standard against which the study-specific build should be reviewed.

Sources

1. FDA — Enhancing Participation in Clinical Trials: Eligibility Criteria, Enrollment Practices, and Trial Designs, final guidance, December 2025.

2. EMA — ICH E8(R1), General Considerations for Clinical Studies.

3. Liu et al. — Evaluating eligibility criteria of oncology trials using real-world data and AI, Nature 592, 629–633, 2021.

4. EClinCloud — BIMO frequency recalculation and clearly labeled hypothetical screening models. FDA snapshot August 23, 2026; reviewed September 18, 2026.

5. FDA — Inspection Observations, annual frequency workbooks.

6. FDA — Form 483 frequently asked questions.

7. FDA — FY2024 Clinical Investigator FDA 483 Observation Trends.

8. eCFR — 21 CFR 312.60, General responsibilities of investigators.

9. eCFR — 21 CFR 312.62, Investigator recordkeeping and record retention.

10. FDA — E6(R3) Good Clinical Practice, final guidance, September 2025.

11. FDA — Screening Tests Prior to Study Enrollment.

12. FDA — Informed Consent, guidance for IRBs, clinical investigators and sponsors, August 2023.

13. FDA — Electronic Systems, Electronic Records, and Electronic Signatures in Clinical Investigations: Questions and Answers, October 2024.

14. ClinicalTrials.gov — Protocol registration data-element definitions.

15. eCFR — 42 CFR 11.48, Clinical trial results information and participant flow.

16. FDA — A Risk-Based Approach to Monitoring of Clinical Investigations: Questions and Answers, April 2023.

17. EClinCloud — EDC product and implementation scope. Company description checked September 18, 2026.

18. EClinCloud — RTSM product and implementation scope. Company description checked September 18, 2026.