How AI Is Accelerating Patient Recruitment in Clinical Trials

Around 80% of clinical trials fail to meet their initial enrollment targets within the planned timeline, and delays can result in lost revenue of as much as $8 million per day for drug-developing companies [1]. That figure has circulated in clinical research discussions for years without meaningfully changing the underlying behavior. Sponsors have tried the obvious fixes: more sites, extended recruitment windows, broader geographic footprints, larger advertising budgets. None has produced consistent improvement.

The structural problem sits deeper than budget. Finding patients who qualify, confirming eligibility against 20 to 40 protocol criteria, and converting that confirmation into enrollment has always been slow, manual, and documentation-heavy. It is also systemically incomplete. Eligible patients exist at most research sites in volumes the coordinator workforce cannot review in time.

What has changed over the past three to four years is that a specific, evidence-based response has begun to take shape, built on AI-assisted eligibility screening, EHR-based patient matching, and predictive analytics applied at the population level. The evidence is strong enough to take seriously. The limitations are real enough to deserve equal attention. This article covers both.

Why Clinical Trial Recruitment Fails at the Scale It Does

The problem is not a shortage of willing patients. Fewer than 4% of U.S. adults participate in clinical trials, and that participation rate has remained essentially flat since 1994 despite growing trial volumes and expanded recruitment investment [5]. Up to 85% of trials fail to recruit or retain a sufficient sample, and nearly $1.9 billion is spent annually on recruitment efforts despite those outcomes [5]. The average dropout rate among enrolled participants sits around 30%, meaning trials that clear the initial enrollment threshold often still lose statistical power before completion [13].

Site-level data reinforces the picture. In a benchmark study of Phase II and III trials, approximately 11% of activated sites ready to begin enrollment never recruited a single patient [14]. The most commonly used strategies at sites, including in-person recruitment and manual EHR review, were described by principal investigators in a cross-sectional survey as the most effort-inefficient methods available [5].

The workflow behind these numbers is fundamentally manual. A coordinator reviews patient records against protocol eligibility criteria, often navigating fragmented EHR systems, then schedules pre-screening visits for candidates who appear to qualify. At sites with high patient volumes and limited coordinator capacity, eligible patients are missed not through negligence but through volume. A patient who would qualify for a Phase II oncology trial, and whose EHR contains all the clinical documentation to confirm eligibility, may never be identified because no one had the bandwidth to look.

How AI Is Being Applied to Patient Recruitment

NLP-Based Eligibility Screening and EHR Integration

The most evidence-supported application of AI to recruitment is automated eligibility screening through natural language processing applied to electronic health records. A 2024 study in JMIR AI by Lee et al. analyzed 3,281 industry-sponsored Phase 2 and 3 interventional trials across six disease areas (including non-small cell lung cancer, prostate cancer, breast cancer, multiple myeloma, ulcerative colitis, and Crohn's disease) [8]. The research team developed an NLP pipeline that converted complex free-text eligibility criteria into machine-readable structured conditions, matched against real-world patient data. The approach generated a standardized eligibility criteria knowledge base and, through a prototype interface integrating real-world patient information, allowed assessment of each criterion's impact on the number of qualifying patients [8].

A parallel implementation at the Medical University of South Carolina tested the TAES (TriAl Eligibility Surveillance) system across five open cardiovascular and cancer trials. Applying NLP and machine learning to detect patients eligible for those trials, the pilot linked information extracted from trial protocols to corresponding clinical information in the EHR and sent automated notifications to providers without interrupting clinical workflow [9]. The TAES prototype achieved moderate accuracy in its initial implementation (recall up to 0.778; precision up to 1.000), demonstrating proof-of-concept that automated eligibility surveillance is feasible in a live healthcare system [9]. The investigators noted that an optimized version could substantially reduce the burden of manual EHR review on research teams.

FHIR R4 standardization, required for certified health information technology under ONC's Health Data, Technology, and Interoperability (HTI-1) final rule [18], provides a standardized API layer that can reduce the technical burden of accessing patient demographics, diagnoses, laboratory values, and medication histories where institutional access authorization, data-sharing agreements, and compatible endpoint implementations are in place. FHIR standardization does not automatically grant sponsors cross-institution query access; local mapping, authorization, governance agreements, and implementation work remain necessary. Data governance requirements, including de-identification, access control documentation, and audit trails, remain sponsor responsibilities depending on contractual roles and data flows.

Predictive Patient Matching

Beyond screening existing patients at enrolled sites, AI models are being applied to predict which patients in a health system are most likely to qualify for a trial and which among them are most likely to enroll if approached. A 2022 Frontiers in Public Health scoping review by Cascini et al. noted that deep learning approaches showed promise in identifying individuals more likely to participate, which could focus outreach resources toward higher-probability candidates [16].

In oncology, where only 2% to 4% of all cancer patients are ultimately enrolled in a clinical trial, precision gains from better targeting may support faster accrual [13]. A retrospective analysis by Calaprice-Whitty et al. applied an AI-powered prescreening system (Mendel.ai) retroactively to two completed oncology trials and one study that failed to enroll. Compared to standard prescreening practices, AI identified 24% and 50% more patients as correctly potentially eligible across the two enrolling trials, and also reduced elapsed time between eligibility and identification [4]. Study limitations include its retrospective design and single-center setting.

The cross-sectional study by Williamson et al. at Moorfields Eye Hospital (published June 2024) provides a well-controlled example in ophthalmology. A deep learning model trained on optical coherence tomography scans identified patients potentially eligible for geographic atrophy trials, cross-referencing imaging output against EHR eligibility data. Compared to a keyword search of the same EHR records, the AI identified 64% more shortlisted patients (1,139 versus 693), with higher imaging-criterion precision (positive predictive values of 63% for AI versus 40% for keyword search) [3].

Large Language Models for Protocol-Level Eligibility Assessment

The use of large language models to assess clinical trial eligibility against patient records is newer and the prospective evidence base thinner, though growing. The PANCR-AI retrospective comparative pilot study, published February 2026 in JMIR Cancer, evaluated three LLMs (GPT-4.5, Claude 3.7 Sonnet, and Mistral-7B-Instruct v0.3) against a human gold standard (two blinded oncologists) for pancreatic cancer trial screening [15]. The study retrospectively reviewed 85 patients from a single institutional database across 12 candidate trials (341 reviewed patient-trial pairs). Mistral-7B-Instruct v0.3 demonstrated the highest sensitivity (92.2%), followed by GPT-4.5 (83.9%) and Claude 3.7 Sonnet (83.3%) [15].

As a single-center retrospective pilot, the PANCR-AI study does not establish generalizable benchmarks. What it demonstrates is that LLM-assisted screening can evaluate a volume of patient-trial pairs that manual review at routine tumor board frequency cannot cover, provided outputs feed into a human review step rather than substituting for one [15].

Digital Outreach and AI-Assisted Targeting

Identifying eligible patients through existing clinical records addresses one part of the challenge. Reaching patients who have not presented to a research site requires different tools. AI-assisted digital targeting platforms use machine learning models trained on behavioral, demographic, and clinical signals to identify individuals in the general population who may match disease and eligibility profiles.

A 2020 systematic review and meta-analysis by Brøgger-Mikkelsen et al. in JMIR found that online recruitment methods produced enrollment at 4.17 times the rate of offline methods on a per-active-recruitment-day basis, with a cost per enrollee of $72 versus $199 for offline recruitment [1]. The same analysis found no statistically significant advantage across the full recruitment period when measured in months, and 69% of studies showed better conversion rates from offline rather than online approaches [1]. Online recruitment accelerates the early identification phase; converting initial contact to enrollment still tends to favor in-person methods.

What the Evidence Base Actually Shows

The 2024 scoping review published in JAMIA by Lu, Yang, Liang, Hu, Zhong, and Jiang is a major recent synthesis of AI recruitment evidence [2]. Drawing on 5,731 initial articles and including 51 studies published between 2004 and January 2024, the review documented consistent positive signals: AI applied to recruitment improved efficiency, reduced costs, improved accuracy, and increased patient satisfaction across the included literature [2]. Oncology accounted for most of the covered studies, consistent with the concentration of Phase II and Phase III trial activity in that therapeutic area.

The same review was direct about what the evidence does not yet support. Study populations were frequently small. Outcome measures varied substantially across studies, making pooled comparison difficult. Eleven of the 51 included studies raised fairness, discrimination, and selection bias as concrete ethical concerns [2]. The reviewers concluded that further validation using standardized outcome measures and more rigorous prospective study designs is needed before strong generalizability claims can be made [2].

A narrative review by Olawade et al. published online in October 2025 in the International Journal of Medical Informatics, drawing on searches of PubMed, Embase, IEEE Xplore, and Google Scholar from January 2015 to December 2024, reported that across analyzed studies, patient recruitment tools showed enrollment rate improvements of 65% and predictive analytics models showed 85% accuracy in forecasting trial outcomes [6]. Both figures aggregate across a heterogeneous literature base and should be interpreted as indicative rather than as generalizable benchmarks across trial types or AI architectures [6].

Across all this literature, several implementation barriers appear consistently: data interoperability limitations, concerns about algorithmic bias, regulatory uncertainty, and limited stakeholder trust [6, 2]. These are not marginal concerns; they represent the primary friction between proof-of-concept and routine deployment.

Regulatory and Documentation Considerations

ICH E6(R3) and Patient-Centric Trial Design

ICH E6(R3), adopted by ICH in January 2025 and effective in the European Union on July 23, 2025, with the FDA publishing a final guidance in September 2025 confirming alignment with the guideline, is the most comprehensive revision to Good Clinical Practice standards since the E6(R2) addendum in 2016; the original ICH E6 guideline dates to 1996 [10, 11]. The guideline's patient-centricity provisions are directly relevant to recruitment planning. E6(R3) calls on sponsors to consider participant burden in trial design, including factors such as visit frequency and practical accessibility, and accommodates decentralized trial elements such as remote assessments and local data collection that can broaden patient access [10].

E6(R3) does not establish specific AI-documentation mandates. Its quality management and data integrity requirements, which have always applied to any system used in trial conduct, would reasonably extend to AI-assisted eligibility tools. Sponsors deploying such tools should document their function, validation, and performance as part of the broader quality management system already required under GCP principles, rather than treating documentation as an AI-specific regulatory obligation [10].

ICH E6(R3) Annex 2, which addresses non-traditional interventional designs including decentralized, platform, pragmatic, and registry-based trials, was formally adopted by ICH at Step 4 on June 3, 2026 [19]. Sponsors building AI-assisted recruitment into decentralized trial frameworks should review Annex 2 and monitor regional implementation timelines, as adoption by individual regulatory authorities follows ICH Step 4 endorsement.

FDA Diversity Action Plans

The FDA released draft guidance on Diversity Action Plans on June 26, 2024, under the mandate of the Food and Drug Omnibus Reform Act of 2022 (FDORA, Section 3602) [7]. The draft guidance proposed that sponsors of certain clinical studies submit plans to improve enrollment from historically underrepresented populations as part of IND applications and premarket submissions [7]. As the FDA noted, clinical studies that include diverse participants are more likely to produce results applicable to the full patient population that will eventually use the product [7].

That draft was removed from active FDA circulation in 2025 following a federal executive order on DEI-related programs [12]. FDORA's underlying statutory requirements remain in place as enacted legislation. Sponsors should track the FDA's current guidance posture through regulatory counsel for up-to-date compliance requirements.

AI-assisted recruitment can support demographic breadth when explicitly designed to do so, by querying eligible patients across a wider range of institutions and geographies. The same tools, trained on narrowly representative EHR datasets, can entrench existing demographic gaps. Algorithmic auditing for subgroup performance is a necessary step in deployment planning, not an optional one.

Data Governance

Sponsors and CROs that qualify as covered entities or business associates and handle protected health information in AI patient-matching systems are subject to HIPAA's Privacy Rule and Security Rule. Not all AI recruitment tools automatically trigger HIPAA obligations; applicability depends on the entity's regulatory status, the nature of the data processed, and any applicable business associate agreements. Systems deployed in clinical trial contexts should fit within the sponsor's quality management framework under ICH E6(R3) [10], with documentation of data access, de-identification methodology, audit trail architecture, and access controls. Sponsors should assess data governance and privacy requirements in each applicable jurisdiction before deploying cross-institution AI recruitment tools.

Limitations and Oversight Requirements

Three categories of limitation deserve operational weight before deployment decisions are made.

Data completeness is the first constraint. AI eligibility screening performs at the quality ceiling set by the underlying EHR data. Incomplete records, inconsistently applied diagnosis codes, or clinically relevant findings documented only in free-text notes without structured capture will produce false negatives. Patients with complex comorbidities, recent medication changes, or diagnostic history distributed across multiple institutions may be missed entirely unless the system integrates across all relevant data sources.

Algorithmic bias is a documented risk. The JAMIA scoping review found that 11 of 51 included studies raised fairness, discrimination, and selection bias as ethical concerns, noting that ML models may inadvertently perpetuate biases present in training data [2]. A 2024 review in PLOS Digital Health by Cross, Choma, and Onofrey found that biases in medical AI arise throughout the development lifecycle, including in data features, model development and evaluation, deployment, and publication. Insufficient sample sizes for certain patient groups produce suboptimal performance that can be clinically consequential [17]. An AI recruitment tool that underperforms for a demographic subgroup does not merely fail to help that group; it narrows their trial access if coordinators reduce manual review in reliance on the system.

Regulatory validation is expected, not assumed. Depending on intended use and regulatory classification, an AI-based recruitment tool may be subject to applicable FDA requirements for software used in clinical development; sponsors should assess the status of each tool with qualified regulatory counsel, as no single FDA policy covers every category of recruitment-support software. As a risk-control practice consistent with ICH E6(R3)'s quality management principles, sponsors should document model performance metrics, version histories, training data characteristics, and how AI outputs are reviewed by clinical staff before enrollment decisions are made.

How Kitsa Fits Into This Problem

Effective AI-assisted recruitment depends on activating the right sites before the screening algorithms run. KScout, Kitsa's site selection product, is designed to match sponsor protocols with research sites whose patient populations align with specific eligibility criteria, providing site-level feasibility intelligence at the selection stage. KScreener, Kitsa's FHIR-connected patient pre-screening product, is designed to automate eligibility screening against EHR data at enrolled sites, reducing the manual review burden on coordinators so their time can be directed toward candidate engagement. Both products address distinct bottlenecks in the recruitment workflow: which sites to activate, and which patients within those sites to prioritize.

Key Takeaways

• Around 80% of clinical trials miss initial enrollment targets, with delays costing as much as $8 million per day. Up to 85% fail to recruit or retain a sufficient sample despite nearly $1.9 billion in annual recruitment spending in the United States [1, 5].
• A retrospective analysis by Calaprice-Whitty et al. found 24% to 50% improvements in correctly identified eligible patients using AI prescreening versus standard methods across two oncology trials; the Moorfields cross-sectional study found 64% more patients shortlisted by AI versus EHR keyword search [4, 3].
• The 2024 JAMIA scoping review of 51 studies (Lu, Yang, Liang et al.) documented consistent AI benefits across efficiency, cost, and accuracy metrics, while explicitly identifying algorithmic bias, data quality, and generalizability as unresolved challenges requiring further validation [2].
• A narrative review by Olawade et al. (published online October 2025) reported aggregate enrollment rate improvements of 65% and 85% predictive accuracy across analyzed studies, but these figures are heterogeneous estimates from a broad literature base rather than benchmarks from controlled trials [6].
• ICH E6(R3), effective in the EU from July 2025 and FDA-adopted September 2025, calls on sponsors to consider participant burden in trial design and accommodates decentralized trial elements; the guideline does not establish AI-specific documentation mandates, but existing GCP quality management principles apply to AI-assisted screening tools [10, 11].
• Algorithmic bias in AI recruitment tools is a documented risk: Cross et al. (2024) found that insufficient sample sizes for certain groups produce suboptimal model performance that can compound existing healthcare disparities [17].
• Sponsors deploying AI recruitment tools should document model performance, training data characteristics, version history, and human oversight procedures as part of their GCP quality management system; applicable FDA requirements depend on intended use and regulatory classification and should be assessed with qualified regulatory counsel.

FAQ

Why do so many clinical trials fail to hit their enrollment targets?+

The JAMIA scoping review by Lu et al. describes a convergence of factors: limited physician referrals, geographic and financial barriers, restrictive eligibility criteria that narrow the qualifying pool, and manual screening workflows that cannot keep pace with the volume of potential candidates at busy sites [2]. Among enrolled patients, a roughly 30% dropout rate further erodes the data needed to achieve statistical endpoints [13].

What does AI actually do to speed up patient recruitment?+

The main applications are NLP-based eligibility screening that automates the comparison of patient records against protocol criteria; machine learning models that predict patient-trial match probability from EHR data; LLM-based protocol assessment applied at tumor board or case review; and AI-assisted digital targeting for outreach to trial-unaware populations. Each addresses a different stage of the funnel, from population-level identification through site-level candidate prioritization.

Is AI-assisted eligibility screening compatible with HIPAA and GCP regulations?+

Sponsors and CROs that qualify as covered entities or business associates and handle protected health information in AI patient-matching systems are subject to HIPAA's Privacy Rule and Security Rule; not every recruitment tool automatically triggers HIPAA obligations. Under ICH E6(R3), sponsors are called on to apply their quality management system to processes that affect trial data or patient selection [10]. Depending on the tool's intended use and regulatory classification, AI-assisted recruitment software may be subject to applicable FDA requirements for software used in clinical research; sponsors should assess tool-specific status with qualified regulatory counsel. Human review of AI-generated candidate lists before enrollment decisions is a recommended risk-control practice, consistent with ICH E6(R3)'s risk-based quality principles.

Can AI improve diversity in clinical trial enrollment, or does it widen existing gaps?+

Both outcomes are possible, and which occurs depends on how the system is built and monitored. AI tools trained on EHR data from populations with incomplete or inconsistent records tend to underperform for those groups, potentially compounding existing access disparities. When applied across broader, more diverse data networks and subjected to ongoing demographic performance monitoring, AI-based tools can surface eligible patients missed by site-centric manual workflows. The difference lies in design intent and operational monitoring, not the technology itself [2, 17].

What does the evidence base actually say about AI's effectiveness in recruitment?+

A major 2024 scoping review is Lu, Yang, Liang et al. in JAMIA (PMID 39259922, DOI 10.1093/jamia/ocae243), covering 51 studies through January 2024. The review documented consistent improvements in efficiency, cost savings, and accuracy, while explicitly flagging small study populations, variable outcome measures, and algorithmic bias concerns as methodological limitations [2]. Study-level results, such as the 24-50% improvement in Calaprice-Whitty et al. and the 64% increase in shortlisted patients in Williamson et al., came from retrospective single-center analyses and should be interpreted accordingly [4, 3].

What does ICH E6(R3) mean practically for sites using AI recruitment tools?+

ICH E6(R3) calls on sponsors to consider participant burden in trial design and accommodates decentralized and technology-enabled trial elements where appropriate [10]. For sites using AI-assisted recruitment, the practical implications include documenting the tool's role in eligibility determination within the quality management system, maintaining validation records, and confirming that clinical judgment is applied to AI-generated candidate lists before enrollment decisions are finalized.

References

1. [1] Brøgger-Mikkelsen M, Ali Z, Zibert JR, Andersen AD, Thomsen SF. "Online Patient Recruitment in Clinical Trials: Systematic Review and Meta-Analysis." Journal of Medical Internet Research. 2020;22(11):e22179. doi:10.2196/22179. PMID: 33146627. https://pubmed.ncbi.nlm.nih.gov/33146627/
2. [2] Lu X, Yang C, Liang L, Hu G, Zhong Z, Jiang Z. "Artificial intelligence for optimizing recruitment and retention in clinical trials: a scoping review." Journal of the American Medical Informatics Association. 2024;31(11):2749-2759. doi:10.1093/jamia/ocae243. PMID: 39259922. https://pubmed.ncbi.nlm.nih.gov/39259922/
3. [3] Williamson DJ, Struyven RR, Antaki F, et al. "Artificial Intelligence to Facilitate Clinical Trial Recruitment in Age-Related Macular Degeneration." Ophthalmology Science. 2024;4(6):100566. doi:10.1016/j.xops.2024.100566. https://doi.org/10.1016/j.xops.2024.100566
4. [4] Calaprice-Whitty D, Galil K, Salloum W, Zariv A, Jimenez B. "Improving Clinical Trial Participant Prescreening With Artificial Intelligence (AI): A Comparison of the Results of AI-Assisted vs Standard Methods in 3 Oncology Trials." Therapeutic Innovation & Regulatory Science. 2020;54(1):69-74. doi:10.1007/s43441-019-00030-4. https://doi.org/10.1007/s43441-019-00030-4
5. [5] Idnay B, Butler A, Fang Y, et al. "Principal Investigators' Perceptions on Factors Associated with Successful Recruitment in Clinical Trials." AMIA Joint Summits on Translational Science Proceedings. 2023;2023:281-290. PMID: 37350899. PMC10283115. https://pmc.ncbi.nlm.nih.gov/articles/PMC10283115/
6. [6] Olawade DB, Fidelis SC, Marinze S, Egbon E, Osunmakinde A, Osborne A. "Artificial intelligence in clinical trials: A comprehensive review of opportunities, challenges, and future directions." International Journal of Medical Informatics. 2026;206:106141 (published online October 8, 2025). doi:10.1016/j.ijmedinf.2025.106141. PMID: 41075423. https://pubmed.ncbi.nlm.nih.gov/41075423/
7. [7] U.S. Food and Drug Administration. "Diversity Action Plans to Improve Enrollment of Participants from Underrepresented Populations in Clinical Studies: Guidance for Industry." Draft Guidance. Docket No. FDA-2021-D-0789. June 26, 2024. https://www.regulations.gov/document/FDA-2021-D-0789-0048
8. [8] Lee K, Liu Z, Mai Y et al. "Optimizing Clinical Trial Eligibility Design Using Natural Language Processing Models and Real-World Data: Algorithm Development and Validation." JMIR AI. 2024;3:e50800. doi:10.2196/50800. https://ai.jmir.org/2024/1/e50800
9. [9] Meystre SM, Heider PM, Cates A, et al. "Piloting an automated clinical trial eligibility surveillance and provider alert system based on artificial intelligence and standard data models." BMC Medical Research Methodology. 2023;23:88. doi:10.1186/s12874-023-01916-6. PMID: 37041475. https://pubmed.ncbi.nlm.nih.gov/37041475/
10. [10] ICH E6(R3) Guideline on Good Clinical Practice. ICH Step 4 endorsement: January 6, 2025. Adopted by EMA/CHMP: December 12, 2024; effective: July 23, 2025. EMA/CHMP/ICH/135/1995. https://www.ema.europa.eu/en/documents/scientific-guideline/ich-e6-r3-guideline-good-clinical-practice-gcp-step-5_en.pdf
11. [11] U.S. Food and Drug Administration. "E6(R3) Good Clinical Practice: International Council for Harmonisation Guidance for Industry." Federal Register. 2025;90(173). September 9, 2025. https://www.federalregister.gov/documents/2025/09/09/2025-17311/e6r3-good-clinical-practice-international-council-for-harmonisation-guidance-for-industry
12. [12] AJMC. "FDA Quietly Removes Draft Guidance on Diversity in Clinical Trials Following Executive Order on DEI." 2025. https://www.ajmc.com/view/fda-quietly-removes-draft-guidance-on-diversity-in-clinical-trials-following-executive-order-on-dei
13. [13] Applied Clinical Trials Online. "Streamlining Clinical Trials: Mitigation and Management." December 2025. https://www.appliedclinicaltrialsonline.com/view/streamlining-clinical-trials-mitigation-and-management
14. [14] Applied Clinical Trials Online. "Enrollment Performance: Weighing the 'Facts.'" (Tufts CSDD Phase II/III benchmark data, 11% site non-enrollment.) https://www.appliedclinicaltrialsonline.com/view/enrollment-performance-weighing-facts
15. [15] Claessens A, Simon A, Manchart A, et al. "AI-Driven Patient Screening for Clinical Trials in Pancreatic Cancer: The PANCR-AI Pilot Retrospective Comparative Study." JMIR Cancer. 2026;12:e80268. doi:10.2196/80268. Published February 23, 2026. PMID: 41730173. https://pubmed.ncbi.nlm.nih.gov/41730173/
16. [16] Cascini F, Beccia F, Causio FA, Melnyk A, Zaino A, Ricciardi W. "Scoping review of the current landscape of AI-based applications in clinical trials." Frontiers in Public Health. 2022;10:949377. doi:10.3389/fpubh.2022.949377. PMID: 36033816. https://pubmed.ncbi.nlm.nih.gov/36033816/
17. [17] Cross JL, Choma MA, Onofrey JA. "Bias in medical AI: Implications for clinical decision-making." PLOS Digital Health. 2024;3(11):e0000651. doi:10.1371/journal.pdig.0000651. PMID: 39509461. https://pubmed.ncbi.nlm.nih.gov/39509461/
18. [18] Office of the National Coordinator for Health Information Technology (ONC). "Health Data, Technology, and Interoperability: Certification Program Updates, Algorithm Transparency, and Information Sharing" (HTI-1). Final Rule. Federal Register. 2024;89(7):1192. January 9, 2024. https://www.federalregister.gov/documents/2024/01/09/2023-28857/health-data-technology-and-interoperability-certification-program-updates-algorithm-transparency-and
19. [19] International Council for Harmonisation. "ICH E6(R3) Guideline on Good Clinical Practice, Annex 2." Step 4, adopted June 3, 2026. https://database.ich.org/sites/default/files/ICH_E6%28R3%29_Annex%202_Guideline_Step%204_2026_0603_0.pdf

Suggested internal links: What Is AI Regulatory Writing in Clinical Research | ICH E6(R3): What GCP Changes Mean for Your Trial | KScreener: FHIR-Connected Patient Pre-Screening | KScout: Site Selection Intelligence