Reduce Patient Screening Failures with CTMS Pre-Screening

Why Patient Screening Failures Cost More Than They Appear on a Budget Sheet

The financial case against high screen failure rates is well documented, though the full operational cost rarely appears in a single budget line. The direct cost of a screen failure, including staff time for chart review, informed consent administration, scheduling, and lab work that yields a disqualification, runs roughly $1,200 per candidate across the industry [5]. In a trial that screens 200 patients to enroll 80, a 40% failure rate means approximately $96,000 in direct screening costs before a single enrolled patient completes a visit.

The indirect costs compound quickly. A 2024 peer-reviewed analysis by Smith, DiMasi, and Getz at Tufts CSDD, published in Therapeutic Innovation and Regulatory Science, found that a single day of trial delay costs sponsors approximately $500,000 in lost prescription drug or biologic sales, and that the direct daily cost to conduct a Phase II or III clinical trial averages approximately $40,000 per day [6]. The authors drew on 645 drugs launched since 2000 and 409 clinical trial budgets, replacing earlier figures that had been based on anecdotal estimates. A companion Tufts CSDD white paper reported that Phase III trials specifically carried a higher direct cost of approximately $55,716 per day, with Phase II closer to $23,737 [7]. Both cost dimensions clarify what a week of enrollment slippage, driven by preventable screen failures, actually costs.

The systemic nature of the problem is visible in published literature across therapeutic areas. A 2026 systematic review of systemic sclerosis randomized controlled trials examined 127 studies involving more than 10,000 screened patients and found a median screen failure rate of 36%, with an interquartile range spanning 20% to 58% [2]. In early-phase oncology, a 2025 multicenter analysis published in ESMO Open covering 202 failures across three French cancer centers found rates between 21.4% and 26.4%, with radiological ineligibility accounting for 29.2% of failures and biological criteria failure for 23.8% [3]. In a 2024 Phase II anti-amyloid monoclonal antibody Alzheimer's trial, the screen failure rate reached 74.1% across two sites [4].

What these figures share is the fingerprint of late-stage detection. By the time a radiological or biological ineligibility is confirmed, the patient has already consented, traveled to a site, undergone baseline assessments, and consumed coordinator hours. Intervening earlier, before consent where clinically and ethically appropriate, is the operational imperative that this article addresses.

Protocol amendments, which are themselves partly downstream of high failure rates, add another cost dimension. Tufts CSDD data shows that the percentage of Phase III protocols with at least one substantial amendment rose from 66% in 2013-2015 to 82% in 2018-2021, with the mean number of amendments per protocol increasing from 2.3 to 3.5 over the same period [8]. Many amendments follow directly from mid-enrollment realization that eligibility criteria were drawn too tightly or operationalized inconsistently across sites.

What "Screen Failure" Actually Means: Four Distinct Stages

The term is used loosely across the industry, which leads to miscommunication about where in the pipeline failures occur and what can realistically be prevented. Four stages carry distinct operational and regulatory implications:

CONSORT 2025, published simultaneously in The Lancet, BMJ, JAMA, and other journals in April 2025, requires that published trial reports account for patient numbers and flow at least through the formal screening and randomization stages, including the number failing screening and the reasons [9]. It does not specifically extend to the pre-screening or enrollment-failure stages as defined in this table, though the principle of transparent patient flow accounting applies more broadly. ICH E6(R3), finalized by FDA in September 2025, reinforces data integrity expectations across the full pre-enrollment period [10]. CTMS-based interventions are most effective at reducing formal screen failures and preventing pre-screening rejections from reaching consent; they have less impact on protocol-driven disqualifications.

The Root Causes of Manual Screening Failure

Understanding why manual screening fails at scale requires separating protocol-level causes from operational ones. CTMS-connected pre-screening addresses the operational side most directly, though it can also surface data that informs protocol amendment decisions.

Eligibility Criteria That Accumulate Without a Feasibility Check

Protocol complexity has grown steadily. A Tufts CSDD analysis comparing Phase III protocols from 2002 and 2012 documented a rise from 31 to 50 eligibility criteria on average, alongside an increase from 106 to 167 total study procedures per trial [11]. That trajectory continued through the following decade. A 2022 JAMA Oncology paper from a multi-stakeholder group that included FDA authors found that overly restrictive inclusion and exclusion criteria have created substantial barriers to patient access, hindered trial recruitment, and limited the generalizability of results [12].

From an operational standpoint, eligibility criteria that multiply without a concurrent assessment of their prevalence in the target patient population produce predictably high failure rates. A patient might have the right diagnosis and the right biomarker but be on a concurrent medication that triggers an exclusion, or have a lab value that falls just outside an arbitrarily set threshold. When criteria are reviewed sequentially during manual screening rather than evaluated simultaneously against a patient's full record, the disqualifying criterion surfaces late.

Chart Review as a One-Person, One-Patient Task

In standard practice, a clinical research coordinator manually reviews each patient's EHR, checking demographics, diagnosis codes, lab values, medication history, and comorbidities against a printed or PDF eligibility list. A prospective real-time effort-tracking study of cancer clinical trial eligibility screening by Penberthy et al., published in the Journal of Oncology Practice in 2012, found that evaluation time spanned a wide range, from under 10 minutes for straightforward cases to more than 120 minutes for complex ones, with the most common category being 30 to 60 minutes per patient [23]. The authors estimated annual institutional screening costs exceeded $90,000 and that the cost per enrolled patient ranged from $129 to $337 depending on study type. When a site is running multiple concurrent trials, coordinators deprioritize systematic pre-screening in favor of responding to patients who self-refer or are referred by treating clinicians.

A time-and-motion study evaluating an automated screening system at a pediatric emergency department documented that coordinators in the manual-screening cohort spent nearly 47.6% of their working hours on electronic screening-related tasks; after automation was introduced, that proportion dropped to 32.5% [13]. The saved time was redirected to patient communication and administrative tasks.

No Structured Pre-Screening Workflow Before Consent

Pre-screening remains inconsistently applied across sites. When it occurs, it is often informal: a phone call, a brief record check, a coordinator's judgment call. The absence of a structured pre-screening step means that candidates who would fail quickly detectable criteria, such as age, prior therapy, geographic distance, or concurrent trial enrollment, proceed through the full consent and onboarding process before being disqualified.

This is a workflow design failure as much as a resource problem. Properly sequenced, a pre-screening step that filters against the highest-prevalence failure criteria can narrow the candidate pool substantially before formal screening resources are committed.

Missing Enrollment Analytics

Most sites running on spreadsheets or basic CTMS implementations cannot answer fundamental enrollment diagnostic questions in real time: which criteria are causing the most failures, which patient subgroups are systematically ineligible, whether failure rates are trending differently across sites, or whether a specific lab test is repeatedly returning values just outside the inclusion window. Without that analytics layer, corrective actions are reactive and delayed.

The Step-by-Step CTMS Framework for Reducing Avoidable Screen Failures

The following five steps describe how CTMS-driven workflows, integrated with EHR-connected pre-screening tools, address each failure mode described above. An important clarification before beginning: a CTMS is a trial management and project tracking platform. It structures workflows, tracks subject status, generates audit trails, and surfaces analytics. EHR-based eligibility matching, particularly for unstructured clinical data, requires a dedicated pre-screening system, an AI eligibility engine, or FHIR-compatible integration layer connected to the CTMS. The steps below reference both the CTMS layer and the pre-screening integration as distinct but complementary components.

Step 1: Map Eligibility Criteria to Pre-Screening Logic Before Site Activation

The earliest point of intervention is protocol design and site activation, not enrollment. Before a site begins screening, each inclusion and exclusion criterion should be assigned a feasibility weight: how prevalent is this characteristic in the site's patient population, how easily can it be verified from structured EHR data, and how frequently does it appear as a failure cause in analogous trials?

A CTMS that integrates with site-level patient databases or EHR data exports can run a pre-activation feasibility query: given the site's historical diagnosis codes, lab value distributions, and concurrent trial participation records, how many patients in the existing pool plausibly meet the top-level inclusion criteria? This is not eligibility confirmation; it is a signal about whether the site's patient population is appropriately matched to the protocol's requirements before the first coordinator hour is spent.

The criteria most likely to produce early failures, those that are binary and verifiable from structured data, such as age range, diagnosis code, documented performance status, or major organ dysfunction lab thresholds, should anchor the pre-screening questionnaire. Criteria that require clinical judgment, such as radiological assessment or investigator evaluation of functional status, cannot be automated but can be staged to occur after high-frequency exclusions have already been ruled out.

Step 2: Implement Structured EHR-Based Pre-Screening Before Consent

Published evidence on automated EHR-based pre-screening is consistent: connecting eligibility criteria to patient records before formal screening substantially reduces coordinator workload and narrows the candidate pool to plausibly viable patients. The key word is "before": the operational goal is to filter candidates against the most prevalent disqualifying criteria prior to informed consent, not after.

A 2014 study published in JAMIA, covering 202,795 patient records across 13 trials in a pediatric emergency department, found that NLP and machine learning-based automated eligibility screening reduced coordinator workload by 92% compared to manual review, though the authors measured workload in terms of candidate records reviewed, not actual screen failure rates [14]. A separate pediatric oncology study covering 215 patients across 55 active trials reduced the average number of patients a coordinator needed to review from 163 per trial to 24, representing an 85% workload reduction [15]. The 2019 prospective evaluation of ACTES at Cincinnati Children's Hospital found that automated screening reduced total patient screening time by 34%, with the saved time redirected to patient communication and administrative work [13].

These efficiency gains translate into fewer formal screen failures when the pre-screening step correctly identifies and routes out candidates who would fail common criteria before they reach the consent stage. They do not guarantee a specific reduction percentage in formal screen failure rates, because that depends heavily on protocol design, indication, and the specific criteria causing failures at a given site.

AI-assisted tools extend these gains to unstructured clinical data. A Cleveland Clinic study published in March 2026 reported that an AI system deployed within a unified EHR environment achieved approximately 96% accuracy in assessing both structured data and unstructured clinical notes for a cardiac amyloidosis trial's 32 eligibility criteria [16]. A 2024 preprint from Stanford Medicine demonstrated LLM-based zero-shot patient-to-trial matching across a combined structured and unstructured record set, at a reported cost of approximately $34.75 per patient screened [17]. These results are promising but represent single-site or early-stage deployments; they should be treated as directional evidence rather than generalizable performance benchmarks.

FHIR-based interoperability is the technical layer enabling this across heterogeneous EHR systems. A 2024 proof-of-concept published in a PMC-indexed study evaluated FHIR Genomics combined with HL7 Clinical Quality Language (CQL) for molecularly-guided trial matching and found that even complex molecular eligibility criteria can be represented in computable form when the semantics of each criterion are formalized in base FHIR specifications [24]. Separately, a 2026 unified pre-screening framework analysis in npj Precision Oncology noted that AI approaches using FHIR protocols can support standardization across EHR systems, but also flagged that data privacy concerns and the potential for algorithmic bias must be carefully managed to ensure ethical patient selection [18].

For sites and sponsors implementing this step, the minimum viable configuration includes: a structured pre-screening questionnaire mapped to the five to seven highest-frequency exclusion criteria, integration with the site's EHR to pull structured data fields (diagnosis codes, lab values, medication lists, demographics), and a CTMS dashboard that logs pre-screening outcomes with timestamps and criterion-level failure reasons.

Step 3: Build a Sequential Eligibility Gate Into the Formal Screening Workflow

Even after pre-screening, formal screening visits must surface disqualifying criteria as early in the visit sequence as possible. Manual screening workflows often run eligibility checks in the order the protocol lists them, which may not correspond to the order most likely to detect failures quickly.

A CTMS-driven screening workflow restructures this sequence. The system flags which criteria are verifiable at baseline assessment (vital signs, weight, documented diagnoses) versus which require additional workup (specialist assessment, imaging, central lab results). Criteria in the first category are checked first. If a patient fails a baseline criterion, the CTMS records the failure immediately and notifies the coordinator without requiring the patient to undergo procedures relevant only to later-stage criteria.

This is not a novel concept operationally, but it is implemented more reliably in a rule-based CTMS workflow than through coordinator judgment alone, particularly at sites running three or four concurrent trials with overlapping screening populations.

Step 4: Use Real-Time Enrollment Analytics to Identify and Address Failure Clusters

A functioning CTMS generates granular screen failure data as a byproduct of normal operation. The analytical value depends on how that data is structured and surfaced. The minimum data fields that should be captured for every screen failure: the criterion or criteria that caused the failure, the point in the screening sequence at which the failure occurred (pre-screening, formal screening, post-consent), the time elapsed between first contact and failure determination, and the patient subgroup characteristics relevant to the failure reason.

When this data is available across sites and accumulated over weeks rather than months, a sponsor or CRO can identify patterns that are invisible in aggregate enrollment numbers. If 40% of screen failures at one site are driven by a single laboratory criterion, that is a signal to investigate whether the site's patient population is mismatched to the protocol, whether the lab threshold is unnecessarily restrictive, or whether the lab measurement itself is being conducted under conditions that systematically produce out-of-range values.

To illustrate: consider a sponsor whose CTMS dashboard shows that 23 of 41 screen failures across four sites over eight weeks triggered on a single exclusion criterion, a creatinine threshold set at 1.5 mg/dL. The data shows the failure is concentrated at two of the four sites, both recruiting primarily from an older urban population with a higher baseline prevalence of chronic kidney disease. The sponsor can bring that criterion-level failure distribution to the next protocol review meeting with a concrete question: does the 1.5 mg/dL threshold reflect a pharmacokinetic safety boundary, or was it carried over from a prior protocol without re-evaluation? If the latter, a targeted amendment or site-level protocol clarification may be warranted. Without the criterion-level analytics layer, this pattern would surface only through anecdotal coordinator reports, if at all.

Formal reduction in formal screen failure rate must be measured site-by-site, not assumed from aggregate workload improvements. Recommended tracking metrics include: pre-screen-to-consent conversion rate (the proportion of pre-screened candidates who proceed to formal screening), formal screen failure rate by site and by criterion, false-negative rate on pre-screening (candidates who pass pre-screening but fail formal screening), and criterion-level failure distribution updated at least monthly.

Tufts CSDD research on protocol complexity found that the most complex protocols had volunteer screen-to-completion rates approximately 50% lower than simpler protocols, and required timelines nearly 73% longer from protocol completion to last patient visit [11]. Real-time failure analytics are the mechanism that makes those patterns actionable before the trial is already months behind schedule.

Step 5: Connect Screen Failure Data to Protocol Amendment Decisions

The most impactful step, and the one most rarely formalized, is creating a structured feedback loop between screen failure analytics and the protocol development and amendment process.

When CTMS data shows that a specific criterion, for example a washout requirement for a commonly prescribed medication, is causing 18% of all screen failures across sites, that data constitutes a concrete basis for evaluating whether a protocol amendment is warranted. Without the CTMS analytics layer, sponsors often learn this from informal coordinator feedback or end-of-study retrospective analyses, by which point the window for a cost-effective amendment has passed.

Connecting CTMS failure data to sponsor review meetings, with a standing agenda item for criterion-level failure analysis, is a workflow design decision that requires no additional technology. It requires only that the CTMS data be structured to produce criterion-level reports rather than aggregate enrollment summaries.

Regulatory and Documentation Requirements for Screening Data

Regulatory expectations for screening documentation have sharpened in recent years. FDA finalized ICH E6(R3) Good Clinical Practice guidance in September 2025, incorporating risk-proportionate monitoring frameworks and explicit recognition of technology-enabled trial designs and data sources [10]. The finalized guidance, described by FDA as a significant evolution in global GCP principles, reinforces that sponsors bear responsibility for the quality and completeness of data across the full trial lifecycle, including the pre-enrollment screening period.

From a documentation standpoint, CONSORT 2025, the updated reporting standard published simultaneously in The Lancet, BMJ, JAMA, and other journals, requires that published trial reports account for patient numbers and flow before randomization, including the number failing screening and the reasons for failure [9]. These reporting requirements apply to trial publications; they inform what sponsors should track operationally, since regulatory reviewers and ethics committees may query this information during review, but CONSORT 2025 does not itself impose operational mandates on sponsors.

For electronic screening records, FDA's 21 CFR Part 11 framework establishes requirements for systems used to capture clinical trial data electronically, including audit trails that record the creation, modification, and deletion of records with timestamps and user identifiers, access controls, and validation documentation [19]. A CTMS designed and validated in accordance with 21 CFR Part 11 can support compliance with these requirements, but "designed and validated" is the operative phrase: the regulation imposes obligations on sponsors and sites that extend to how a system is deployed and maintained, not only whether the software has audit-trail functionality. No CTMS automatically satisfies Part 11 requirements by virtue of its features alone.

The pre-screening phase presents a distinct documentation challenge. Candidates who are pre-screened and found implausibly ineligible before formal consent may not appear in the formal screening log. ICH E6(R3)'s emphasis on quality management systems and data integrity across the trial lifecycle is consistent with the practice of retaining pre-screening records in a retrievable format, even where candidates did not proceed to formal screening [10]. This is a reasonable interpretation of the guidance's intent rather than a specific stated requirement; sponsors should review their own quality management procedures and consult legal and regulatory counsel to confirm the appropriate scope of pre-screening documentation at their institutions.

What AI-Assisted Pre-Screening Changes in Practice

AI-assisted pre-screening differs from rule-based EHR queries in its capacity to handle unstructured clinical data. Most of a patient's clinically relevant information lives in free-text physician notes, radiology reports, pathology summaries, and discharge documentation. Standard CTMS integrations can pull structured fields automatically; interpreting free text requires NLP or LLM-based processing.

The Cleveland Clinic implementation described above achieved 96% accuracy for a single cardiac amyloidosis trial using a system that evaluated both structured fields and unstructured notes in combination [16]. The Stanford preprint reported LLM-based matching at approximately $34.75 per patient screened across a combined dataset [17]. Both results represent specific deployment contexts; neither generalizes automatically to other indications, eligibility criterion structures, or EHR environments.

A 2026 analysis in npj Precision Oncology examining unified pre-screening frameworks for oncology explicitly flagged that AI approaches using FHIR protocols address interoperability challenges but that data privacy concerns and the potential for algorithmic bias must be carefully managed [18]. A study of screening failures in Alzheimer's disease and related dementias trials found that 68.7% of candidates screen-failed for eligibility-related reasons were minority older adults [25]. A broader cancer trial enrollment analysis found that Black patients and individuals in socioeconomically disadvantaged areas were systematically less likely to reach enrollment even when they had expressed interest [20]. Both findings point to the same underlying problem: eligibility criteria calibrated to historically studied populations may exclude others not because they are clinically ineligible but because the criteria were not designed with their prevalence patterns in mind. AI pre-screening tools inherit the biases embedded in eligibility criteria and in the EHR data they query. Human review of AI-generated candidate recommendations remains necessary.

The ACRP has described AI-powered EHR alerts as tools that expand coordinator capacity for patient communication and trust-building rather than substituting for clinical judgment in enrollment decisions [21]. That framing reflects the practical integration path: AI pre-screening narrows the candidate pool and surfaces criterion-level information; coordinators make the enrollment calls.

Implementation Risks Worth Tracking

Any organization implementing CTMS-connected pre-screening should account for the following:

False negatives: A pre-screening system that incorrectly flags an ineligible candidate as eligible can generate false confidence and delay formal failure detection. Recall rates, not just precision, should be tracked in any validated pre-screening implementation.

EHR mapping quality: Eligibility criteria mapped to the wrong structured data fields, or to fields that are inconsistently populated across sites, will produce unreliable pre-screening outputs. Data mapping validation is a prerequisite, not an afterthought.

Site burden: Implementing a new pre-screening workflow requires coordinator training and ongoing support. Sites with limited staff capacity may struggle with the change management required, particularly if the CTMS adds rather than replaces steps in an already burdened workflow.

IRB and privacy constraints: Querying EHR data for pre-screening purposes involves patient privacy considerations that vary by jurisdiction and institution. Sponsors should confirm that pre-screening data use is covered under the trial's IRB approval and any applicable data use agreements before implementation.

How Kitsa Fits Into the Screening Workflow

The following descriptions reflect Kitsa's stated product capabilities and design intent as described in its public product documentation (kitsa.ai). They are included here as orientation for readers, not as independently validated performance claims.

Kitsa's KScreener is designed to address the EHR-to-eligibility gap. The platform is designed to use FHIR-based connectivity to pull structured patient data from site EHR systems and evaluate candidates against protocol-specific inclusion and exclusion criteria before formal screening begins, with the intent of providing sites a pre-screened candidate list rather than a full chart review request.

On the protocol side, KScribe is designed to generate ICFs and protocol documents with eligibility criteria in structured formats, which Kitsa describes as supporting downstream use in screening workflows and reducing the transcription errors that occur when coordinators manually extract criteria from PDF documents into screening checklists.

Key Takeaways

• Published screen failure rates range from approximately 20% in some Phase II and III oncology settings to above 70% in biomarker-gated indication areas, with a median of 36% across some disease categories [2][3][4].
• A 2024 peer-reviewed Tufts CSDD analysis found that each day of trial delay costs sponsors approximately $500,000 in lost prescription drug or biologic sales and approximately $40,000 per day in direct Phase II and III trial conduct costs, with Phase III specifically higher at approximately $55,716 per day per a companion Tufts white paper [6][7].
• Automated EHR-based pre-screening has reduced coordinator workload by 85% to 92% in published studies, and patient screening time by 34% in a prospective evaluation; a 2012 Journal of Oncology Practice study with real-time effort tracking found complex evaluations commonly ran 30 to 60 minutes per patient, with costs exceeding $90,000 annually per institution; these are efficiency benchmarks, not guaranteed screen-failure-rate reductions, and results vary by protocol and site [13][14][15][23].
• CTMS and EHR pre-screening tools are distinct and complementary: CTMS structures workflows and generates analytics; eligibility matching requires dedicated pre-screening systems or AI tools with EHR integration.
• The five-step framework covers criteria-to-pre-screening mapping, structured EHR pre-screening before consent, sequential eligibility gating, real-time analytics, and a failure-to-amendment feedback loop.
• ICH E6(R3), finalized by FDA in September 2025, and CONSORT 2025 reinforce expectations for documented, auditable screening data; 21 CFR Part 11 sets electronic records requirements that a validated CTMS can be designed to support, but validation is an obligation on sponsors and sites, not an automatic feature of any software [9][10][19].
• AI pre-screening achieves strong results in documented single-site deployments but carries risks of algorithmic bias, false negatives, and EHR mapping errors that require human oversight and ongoing monitoring [16][17][18][25].

Frequently Asked Questions

References

1. [1]AutoCruitment. "How to Reduce Screen Fail Rates." AutoCruitment Blog, 2026. Link
2. [2]Foucher Y, et al. "Screening failure in systemic sclerosis randomized trials: reporting, rates, causes, and trends over time." Rheumatology Advances in Practice, 2026. Link
3. [3]Korakis I, Ouali K, Hanvic B, Verlingue L, Allignet B, Lusque A, Clementei S, Magne E, Massard C, Delord J-P, Cassier P, Baldini C, Segier B, Gomez-Roca C. "Addressing screening failures in early-phase clinical trials in oncology: impact on patient outcomes and strategies for improvement." ESMO Open. 2025 Aug 13;10(8):105331. PMC12362514. Link
4. [4]Irvine Clinical Site Network. "Non-inferior screen fail rate for persons under the age of 65 in an anti-amyloid monoclonal antibody clinical trial on Alzheimer's Disease." PMC, 2024. Link
5. [5]Antidote. "25+ useful clinical trial recruitment statistics for better results." Antidote Blog, 2026. Link
6. [6]Smith ZP, DiMasi JA, Getz KA. "New Estimates on the Cost of a Delay Day in Drug Development." Therapeutic Innovation and Regulatory Science, 2024;58(5):855-862. PMID: 38773058. Link
7. [7]Smith ZP, DiMasi JA, Getz KA. "Quantifying the Value of a Day of Delay in Drug Development." Tufts CSDD White Paper, August 2024. Link
8. [8]Clinical Trial Vanguard / Tufts CSDD SCOPE 2024 Presentation. "Tufts CSDD: New Insights on The Clinical Trial Industry." Clinical Trial Vanguard, March 2024. Link
9. [9]Schulz KF, Altman DG, Moher D, CONSORT Group. "CONSORT 2025 Statement: updated guideline for reporting randomised trials." The Lancet, April 2025. Link
10. [10]U.S. Food and Drug Administration. "E6(R3) Good Clinical Practice." FDA Final Guidance for Industry, September 2025. Link
11. [11]Getz KA. "The Cost of Clinical Trial Delays." CTTI Annual Meeting Presentation, Tufts CSDD, 2012. Link
12. [12]Gerber DE, Singh H, Larkins E, Ferris A, Forde PM, Selig W, Basu Roy U. "A New Approach to Simplifying and Harmonizing Cancer Clinical Trials: Standardizing Eligibility Criteria." JAMA Oncology, 2022;8(9):1333-1339. Link
13. [13]Ni Y, et al. "A Real-Time Automated Patient Screening System for Clinical Trials Eligibility in an Emergency Department: Design and Evaluation." Journal of Medical Internet Research, 2019. PMC6685132. Link
14. [14]Ni Y, et al. "Automated clinical trial eligibility prescreening: increasing the efficiency of patient identification for clinical trials in the emergency department." Journal of the American Medical Informatics Association, 2015. PMC4433376. Link
15. [15]Doan S, et al. "Increasing the efficiency of trial-patient matching: automated clinical trial eligibility pre-screening for pediatric oncology patients." Journal of the American Medical Informatics Association, 2013. PMC4407835. Link
16. [16]Cleveland Clinic Consult QD. "AI Can Reliably Unlock EHR Data to Determine Clinical Trial Eligibility, Study Finds." Cleveland Clinic, March 2026. Link
17. [17]Wang S, et al. "Zero-Shot Clinical Trial Patient Matching with LLMs." arXiv preprint, Stanford Medicine and Stanford Healthcare, 2024. Link
18. [18]Fouquier de Lerisson C, et al. "A unified framework for pre-screening and screening tools in oncology clinical trials." npj Precision Oncology, 2026. Link
19. [19]U.S. Food and Drug Administration. "21 CFR Part 11: Electronic Records; Electronic Signatures." FDA Code of Federal Regulations. Link
20. [20]Caston NE, et al. "Ineligible, unaware, or uninterested? Associations between underrepresented patient populations and retention in the pathway to cancer clinical trial enrollment." JCO Oncology Practice, 2023;18(11):e1853-65. Link
21. [21]Gray S, Boyd P, Sternke EA. "Pre-Screening Reimagined: AI, Script Engineering, and the Future of Clinical Trials." ACRP, December 2025. Link
22. [22]Meijer RP, et al. "Screen Failure Rates in Contemporary Randomized Clinical Phase II/III Therapeutic Trials in Genitourinary Malignancies." Urologic Oncology, 2017. Link
23. [23]Penberthy LT, Dahman BA, Petkov VI, DeShazo JP. "Effort Required in Eligibility Screening for Clinical Trials." Journal of Oncology Practice, 2012;8(6):365-370. PMC3500483. Link
24. [24]Kahn MG, et al. "Molecularly-Guided Cancer Clinical Trial Matching using FHIR and HL7 Clinical Quality Language: A Proof of Concept." PMC12099354, 2024. Link
25. [25]Kapoor A, et al. "Understanding Reasons for Screen Failures in Alzheimer's disease and related dementias (ADRD) Trials: Challenges and Implications for Recruitment." PMC12725729, 2025. Link