4 AI Misconceptions in Clinical Trials You Cannot Afford to Ignore

A pediatric patient received approximately ten times the maximum daily dose of a study medication in August 2024. The cause was not a misread label or a distracted coordinator. The FDA's Warning Letter to the Clinical Integrative Research Center of Atlanta identified the problem as an electronic dispensing algorithm that, in the investigator's own words, "lacked safety guards to prevent errors" ^[1]. The agency cited a failure to conduct the study in accordance with the investigational plan under 21 CFR 312.60.

That incident is not an argument against AI in clinical trials. It is an argument against deploying algorithmic systems without the controls, human review structures, and regulatory awareness that the technology actually requires. And that distinction matters now, because the misconceptions surrounding AI in clinical research are not abstract. They shape protocol decisions, vendor selections, budget allocations, and regulatory submissions.

Four of those misconceptions are worth addressing in depth. Each one is common enough to appear in boardroom presentations and RFP responses. Each one is wrong in ways the published evidence makes clear.

Why These Misconceptions Persist

The clinical trial ecosystem is operating in a genuine moment of AI adoption. By 2023, the FDA had received more than 500 submissions incorporating AI components across various stages of drug development, a figure the agency's Center for Drug Evaluation and Research tracked from 2016 onward ^[2]. The EMA published its Reflection Paper on AI in the medicinal product lifecycle in September 2024 ^[3]. The FDA issued its draft guidance on AI supporting regulatory decision-making in January 2025 ^[4]. In January 2026, the FDA and EMA jointly published their Guiding Principles on Good AI Practice in Drug Development, signaling that both agencies expect transparency, human oversight, and documented validation for AI used in regulatory contexts ^[5].

The pace of policy development is producing exactly the conditions in which misconceptions take hold: the technology is moving fast, the regulatory frameworks are still catching up, and commercial pressure to adopt AI is intense. A 2025 analysis found that AI-related partnerships in clinical trials rose 30% between 2022 and 2024, with sector investments estimated between $2 billion and $4 billion USD ^[6].

Against that backdrop, four specific misconceptions deserve careful examination.

Misconception 1: AI Removes the Need for Human Oversight in Trial Operations

The most persistent and consequential misconception is that sufficiently capable AI removes the requirement for human oversight. This view conflates automation with accountability, and the two are not the same thing.

ICH E6(R3) reached Step 4 final adoption by the International Council for Harmonisation on January 6, 2025. The EMA made Annex 1 and the Principles effective on July 23, 2025. The UK implemented E6(R3) alongside revised national clinical trial regulations on April 28, 2026. The FDA posted the guideline in the Federal Register on September 9, 2025, establishing it as the U.S. reference standard, though the FDA has not yet set a formal domestic compliance date ^{[7, 8a, 8b]}. The guideline's structure is important to understand correctly. Sponsor responsibilities are organized in Annex 1, Section 3, with Section 3.9 covering Sponsor Oversight and Section 3.10 covering Quality Management. Section 4 of Annex 1 introduces new Data Governance obligations for both investigators and sponsors ^[7].

The core principle is not new, but E6(R3) applies it to the modern trial environment explicitly: sponsors must maintain appropriate oversight of all trial activities, including those conducted through automated or outsourced systems. Sponsors' duties in E6(R3) are expanded precisely to address the risks of distributed, technology-driven operations: they must document how oversight is maintained even when tasks are delegated to CROs or technology vendors ^[7].

The EU AI Act (Regulation 2024/1689) classifies certain AI systems used in regulated healthcare contexts as high-risk, with requirements for documented validation, transparency, and human oversight mechanisms. This applies to AI components that function within medical-device-regulated software functions or that directly influence safety-critical decisions, rather than to all clinical research AI categorically. That distinction matters: a sponsor's internal analysis tool and an AI-enabled dispensing algorithm operate under meaningfully different risk classifications ^[10].

The FDA's January 2025 draft guidance on AI in regulatory decision-making (FDA-2024-D-4689) is draft guidance and does not carry enforceable weight until finalized. What it does signal clearly is that the agency will evaluate AI tools used to generate data or content supporting regulatory submissions against a risk-based credibility assessment framework. The burden of demonstrating that credibility sits with the sponsor ^[4].

The Warning Letter in Atlanta illustrates the enforcement reality. The electronic dispensing algorithm was task-specific, not a frontier AI model. Its failure was traced to the absence of safeguards that human review would have caught. Investigators who deploy AI in clinical study operations without maintaining adequate oversight structures and error-prevention controls face the same regulatory exposure under 21 CFR 312.60 that investigators have always faced for protocol deviations, regardless of whether a human or a system made the error ^[1].

Human-in-the-loop design is not a philosophical position. Both the EU AI Act and the FDA/EMA Guiding Principles on Good AI Practice treat meaningful human oversight as a baseline expectation for AI systems operating in high-risk or safety-relevant clinical contexts ^{[5, 10]}. Frameworks analyzing AI oversight in clinical research similarly describe active human involvement as essential where errors carry significant consequence for patient safety or trial integrity ^[11]. For adverse event processing, for instance, AI can handle non-serious cases at volume while routing those involving coding discrepancies, seriousness determinations, or causality questions to human review. That delineation requires deliberate workflow design. It does not emerge automatically from a capable model.

E6(R3) also strengthens quality management expectations in ways that govern how AI-assisted workflows must be documented. A 2024 Tufts CSDD survey of 206 sponsors, CROs, and biotechnology organizations found that companies had implemented RBQM in an average of 57% of their clinical trials, with lower adoption among smaller organizations and barriers concentrated in organizational knowledge, change management, and unclear value propositions ^[9]. E6(R3)'s introduction of quality-by-design principles and documented quality tolerance limits raises the bar for the nearly half of organizations not yet at full RBQM adoption. Inspectors under E6(R3) will examine whether sponsors have implemented a quality culture with evidence, not just a technology stack with documentation.

Misconception 2: AI Automatically Produces More Diverse and Unbiased Trial Populations

AI-driven patient matching and recruitment tools have genuine potential for expanding trial access to underrepresented populations. The evidence for that potential is real. So is the evidence that AI, when trained on historical clinical data, can reproduce and in some cases amplify the demographic gaps it is supposed to correct.

The starting point matters here. A 2025 peer-reviewed analysis published in Communications Medicine and available in PMC evaluated enrollment data from 341 Phase III clinical trials supporting FDA drug approvals from 2017 to 2023, drawing on the FDA Drug Trial Snapshots Program. Only 6% of those pivotal trials achieved enrollment aligned with the racial and ethnic distribution of the U.S. population. Enrollment of Black and Hispanic participants declined over the study period, even as the field's attention to equity increased ^[12].

Those numbers reflect decades of structural barriers in clinical research: geographic site concentration, eligibility criteria that inadvertently filter out patients with comorbidities common in minority populations, language barriers in consent documentation, and logistical constraints that disproportionately affect lower-income participants. AI models trained on electronic health records carry that history into their outputs.

A scoping review published in the Journal of the American Medical Informatics Association in 2024 covering AI applications in clinical trial recruitment and retention found that 11 of the reviewed studies specifically identified fairness, discrimination, and selection bias as problems with AI recruitment tools. Machine learning models can inadvertently perpetuate biases present in training data and produce outcomes that disadvantage specific patient groups. The review also noted that EHR data is subject to accuracy and completeness problems that compound these effects ^[13].

A scoping review published in npj Digital Medicine covering 142 AI clinical trial risk assessment studies from 2013 to 2024 similarly identified selection bias and data quality issues as persistent challenges even among models achieving high predictive performance ^[14]. An AUROC of 0.96 on a validation dataset tells you about model discrimination on that particular cohort; it does not tell you whether that cohort reflects the patient population the trial is designed to serve.

FDORA amended FD&C Act sections 505(z) and 520(g) to require Diversity Action Plans for specified drug, biologic, and device studies. FDA's June 2024 draft guidance (not for implementation) describes the proposed form, content, timing, and waiver process for satisfying that requirement ^[15]. The obligation is not satisfied by deploying an AI recruitment tool. Responsible deployment additionally involves examining whether the tool's training data reflects those populations and monitoring enrollment demographics against the plan's targets throughout the trial, practices aligned with the spirit of both FDORA and the FDA's proposed guidance, even where the draft document is not yet enforceable.

A 2026 conceptual framework published in Trials describes how AI and machine learning can support equity, diversity, and inclusion across the full trial lifecycle, with one necessary precondition: training datasets must be curated with that goal in mind ^[16]. Predictive modeling to assess the inclusivity of eligibility criteria, reinforcement learning to adaptively redesign protocols for underrepresented cohorts, and natural language processing to identify patients from EHR narrative notes are all legitimate applications. Each requires deliberate attention to what the training data contains and whose experience it captures.

The misconception is not that AI cannot improve diversity in trials. It is that diverse and unbiased outputs emerge automatically from a capable model. They do not. Approaches such as FHIR-based eligibility screening with human eligibility confirmation can help close that gap when paired with deliberate bias monitoring.

Misconception 3: AI Performance in Controlled Evaluations Transfers Directly to Trial Operations

Published benchmarks for AI tools in clinical trials are often striking. A 2025 narrative review published in the International Journal of Medical Informatics covering research from 2015 to 2024 reported that AI patient recruitment tools improved enrollment rates by 65% in analyzed studies and that AI integration accelerated trial timelines by 30 to 50% while reducing costs by up to 40% in some implementations ^[17]. These figures come from selected deployments under favorable conditions and should be read as illustrative of what well-integrated AI can achieve, not as a performance guarantee that transfers to any new implementation.

The gap between controlled evaluation performance and real-world trial operations is well documented in the clinical AI literature. A comprehensive review covering research from 2015 to 2024 found that implementation barriers consistently include algorithmic bias from homogeneous training datasets, workflow misalignment at the site level, interoperability gaps across EHR systems, and inconsistent reporting standards between sites, all of which diminish AI effectiveness when it moves from controlled evaluations to diverse clinical environments ^[17].

Algorithmic drift is a specific and underappreciated manifestation of this problem. The FDA has explicitly named it as a lifecycle monitoring concern. In a published Q&A on AI in clinical trial design, FDA's Dr. ElZarrad noted that model performance can degrade over time, particularly in learning systems where new data inputs cause outputs to shift: what the agency describes as a data drift problem ^[18]. The FDA's January 2025 draft guidance on AI in regulatory decision-making addresses lifecycle monitoring as a core framework expectation, not an optional enhancement ^[4].

Research on temporal performance degradation in AI models, including healthcare operations datasets, confirms that model quality can decline over deployment timelines as patient demographics, documentation practices, coding systems, and operational environments change ^[19]. The FDA's 2025 draft guidance on AI/ML-enabled device software functions (FDA-2024-D-4488) introduces specific expectations for predetermined change control plans and performance monitoring frameworks, including re-validation when model performance drifts below established thresholds ^[20].

For sponsors relying on AI-generated outputs for protocol decisions, site selection, or patient matching, this means that vendor benchmarks from controlled evaluations are a starting point for due diligence, not a substitute for site-specific validation. Performance validation against the actual patient population, site infrastructure, and data quality of the planned trial is required before go-live. Ongoing monitoring after deployment is required throughout the trial. These are not operational aspirations. For AI uses that affect patient safety, trial integrity, or regulatory decision-making, they reflect the documented risk-based expectations of both agencies in their current guidance frameworks.

For AI-assisted document generation specifically, the implication is the same: the accuracy of an AI tool across a general document benchmark population does not predict its accuracy in a novel indication with atypical eligibility criteria or a complex safety monitoring structure. Expert review against the specific clinical and regulatory context of the trial is a quality management responsibility, not a formality.

Misconception 4: AI Reduces Implementation Burden from Day One

The ROI framing that surrounds AI adoption in clinical trials consistently leads with efficiency gains and cost reductions. Those gains are real in established, well-integrated implementations. The operational reality of getting there is consistently underweighted in adoption pitches.

A 2025 comprehensive narrative review in the International Journal of Medical Informatics identified data quality issues affecting 50% of clinical trial datasets, regulatory uncertainty, and algorithmic bias as the primary barriers to AI implementation in trial settings ^[17]. Data quality is not simply a technical limitation. It is an organizational challenge that precedes any AI deployment: tools that depend on structured EHR inputs will produce unreliable outputs wherever EHR documentation is inconsistent, incomplete, or coded differently across sites. Remediation of those inputs is rarely scoped into AI vendor contracts.

The integration timeline compounds this. A 2024 Tufts CSDD empirical study estimated that the direct cost to conduct a single day of a Phase II or III clinical trial is approximately $40,000, with lost prescription sales adding roughly $500,000 more per delayed day [21]. Against that backdrop, even a two-week acceleration from well-implemented AI in regulatory document generation represents a material financial return. But the same calculation makes the cost of a poorly integrated AI implementation visible: delays from unreliable outputs, protocol amendments driven by AI-assisted errors, and compliance remediation from inadequate oversight documentation all counteract the efficiency case.

AI tools used in regulated trial workflows must also meet documentation standards that are not trivial to establish. Under ICH E6(R3) Annex 1, Section 4, sponsors must document data governance processes proportionate to data criticality across the full data lifecycle. Where electronic records are used to satisfy FDA regulatory requirements under applicable predicate rules, 21 CFR Part 11 governs their creation, modification, maintenance, and transmission; the FDA's Part 11 guidance applies a narrow scope and enforcement-discretion framework, with audit trail and validation requirements tied to whether those records and signatures substitute for paper records and handwritten signatures under the relevant regulation ^{[7, 22]}.

The EU AI Act (Regulation 2024/1689) classifies certain AI systems used in regulated healthcare and drug development contexts as high-risk. For those systems, the Act creates obligations around risk management, technical documentation, human oversight, accuracy, and post-market monitoring ^[10]. This applies to the applicable subset of AI tools in clinical research, not to all software involved in trial operations. Sponsors with AI deployments that fall within that classification need to account for compliance overhead that is not reflected in a tool's purchase price or a vendor's ROI model.

AI implementation that fully accounts for data quality requirements, integration timelines, validation overhead, and ongoing monitoring costs will frequently show a genuine ROI, particularly in document generation and site selection, where quality improvements compound across submission timelines. AI implementation that treats those factors as manageable after go-live will often underperform its business case and, in some cases, produce compliance exposure that costs more to remediate than the efficiency gains were worth.

Regulatory and Documentation Considerations

The regulatory context for AI in clinical trials is changing faster than most SOPs can track, but several things are now settled. ICH E6(R3) reached Step 4 final adoption by ICH on January 6, 2025. In the EU, Annex 1 and the Principles became effective on July 23, 2025. In the UK, E6(R3) took effect alongside revised national clinical trial regulations on April 28, 2026. The FDA posted it as a final guidance document in the Federal Register on September 9, 2025, making it the U.S. reference standard, though a formal U.S. compliance date has not been set ^{[7, 8a, 8b]}. FDA guidance documents represent the agency's current thinking and are not legally binding regulations; however, inspectors reference them in practice.

The FDA's January 2025 draft guidance on AI supporting regulatory decision-making (FDA-2024-D-4689) is draft guidance, not currently enforceable. Sponsors should monitor FDA's guidance docket for finalization and treat the credibility assessment framework it describes as the likely direction of submission requirements for AI-derived data ^[4]. The joint FDA-EMA Guiding Principles on Good AI Practice, published in January 2026, provide additional directional clarity on what both agencies will expect sponsors to demonstrate regarding transparency, human oversight, and validation ^[5].

For AI-generated regulatory documents, the obligations under existing frameworks are already concrete. ICH E6(R3) Annex 1, Section 4 requires documented data governance processes across the full data lifecycle, proportionate to data criticality. Where sponsors use electronic records or electronic signatures to satisfy applicable FDA records requirements, 21 CFR Part 11 may apply. Sponsors should document the applicable predicate-rule context, justify risk-based controls, and address access controls, audit trails, validation evidence, and signature controls where the predicate rule and record type require them ^{[7, 22]}.

How Kitsa Fits Into This Problem

These four misconceptions share a common thread: they treat AI as something that happens to a trial, rather than something that must be integrated deliberately within it. Kitsa's approach to AI-native clinical research infrastructure is designed with that distinction in mind. KScribe, Kitsa's regulatory document generation platform, is designed to support expert medical writer review as part of the workflow, not to replace it. KScreener, Kitsa's FHIR-connected patient pre-screening tool, is built to surface eligible candidates for human eligibility confirmation rather than to automate enrollment decisions. The design intent reflects what GCP, ICH E6(R3), and the FDA's evolving AI governance framework actually require.

Key Takeaways

• ICH E6(R3) (Step 4 adopted January 6, 2025; FDA posted September 9, 2025) places sponsor oversight obligations in Annex 1, Sections 3.9, 3.10, and 4. Those obligations apply to AI-assisted workflows and cannot be delegated to an automated system.
• The August 2024 FDA Warning Letter to the Clinical Integrative Research Center of Atlanta demonstrates that algorithmic systems operating without adequate human safeguards create direct 21 CFR 312.60 exposure for investigators.
• AI recruitment tools trained on historically unrepresentative EHR data reproduce and can amplify demographic gaps; a 2025 Communications Medicine analysis found that only 6% of 341 pivotal trials achieved enrollment reflecting the U.S. racial and ethnic population.
• The gap between controlled evaluation performance and real-world trial operations is documented across the clinical AI literature; vendor benchmarks require site-specific validation before go-live, not after.
• Algorithmic drift causes AI model performance to degrade after deployment; the FDA has named this explicitly as a lifecycle monitoring obligation under its AI guidance frameworks.
• Data quality issues affect an estimated 50% of clinical trial datasets and represent an organizational challenge that precedes any AI deployment; integration timelines of several months to over a year are common in real-world implementations.
• Sponsors using AI in regulated workflows should establish validation documentation, audit trails, and human review protocols before deployment; remediation after an inspection finding costs more than implementation overhead.

FAQ

References

1. [1]U.S. Food and Drug Administration. "Warning Letter: Julio R. Flamini, M.D. / Clinical Integrative Research Center of Atlanta, 691123." FDA, August 20, 2024. https://www.fda.gov/inspections-compliance-enforcement-and-criminal-investigations/warning-letters/julio-r-flamini-mdclinical-integrative-research-center-atlanta-691123-08202024
2. [2]U.S. Food and Drug Administration, Center for Drug Evaluation and Research. "Artificial Intelligence in Drug Development." FDA CDER, 2024. https://www.fda.gov/about-fda/center-drug-evaluation-and-research-cder/artificial-intelligence-drug-development
3. [3]European Medicines Agency. "Reflection Paper on the Use of Artificial Intelligence (AI) in the Medicinal Product Lifecycle." EMA/CHMP/CVMP/83833/2023. Adopted September 9, 2024; published September 30, 2024. https://www.ema.europa.eu/en/use-artificial-intelligence-ai-medicinal-product-lifecycle-scientific-guideline
4. [4]Food and Drug Administration. "Considerations for the Use of Artificial Intelligence to Support Regulatory Decision-Making for Drug and Biological Products." FDA Draft Guidance (FDA-2024-D-4689), January 2025. https://www.fda.gov/regulatory-information/search-fda-guidance-documents/considerations-use-artificial-intelligence-support-regulatory-decision-making-drug-and-biological
5. [5]Food and Drug Administration / European Medicines Agency. "Guiding Principles on Good AI Practice in Drug Development." FDA/EMA, January 2026. https://www.fda.gov/about-fda/artificial-intelligence-drug-development/guiding-principles-good-ai-practice-drug-development
6. [6]Clinical Leader. "Global AI In Clinical Trials: Market Trends and Current Partnerships." Clinical Leader, February 2025. https://www.clinicalleader.com/doc/global-ai-in-clinical-trials-market-trends-current-partnerships-0001
7. [7]International Council for Harmonisation. "ICH E6(R3): Good Clinical Practice Guideline." Step 4 final adoption January 6, 2025; FDA publication September 2025. https://database.ich.org/sites/default/files/ICH_E6%28R3%29_Step4_FinalGuideline_2025_0106.pdf
8. [8a]Association of Clinical Research Professionals. "FDA Publishes ICH E6(R3): What it Means for U.S. Clinical Trials." ACRP, September 2025. https://acrpnet.org/2025/09/16/fda-publishes-ich-e6r3-what-it-means-for-u-s-clinical-trials
9. [8b]UK Medicines and Healthcare products Regulatory Agency. "Clinical Trials for Medicines: Guidance on Compliance with ICH E6 Good Clinical Practice in the United Kingdom." GOV.UK, published March 27, 2026; updated April 28, 2026. https://www.gov.uk/guidance/clinical-trials-for-medicines-guidance-on-compliance-with-ich-e6-good-clinical-practice-gcp-in-the-united-kingdom
10. [9]Dirks, A., Florez, M., Torche, F. et al. "Comprehensive Assessment of Risk-Based Quality Management Adoption in Clinical Trials." Therapeutic Innovation and Regulatory Science, 58(5), 996, September 2024. https://pmc.ncbi.nlm.nih.gov/articles/PMC11043178/
11. [10]European Union. "Regulation (EU) 2024/1689 of the European Parliament and of the Council: the Artificial Intelligence Act." Official Journal of the European Union, 2024. https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai
12. [11]Kandikatla, L. and Radeljic, B. "AI and Human Oversight: A Risk-Based Framework for Alignment." arXiv preprint, 2025. https://arxiv.org/pdf/2510.09090
13. [12]Zaaijer, S. and Groen, S.C. "Longitudinal Clinical Trial Enrollment Trends across 341 US FDA-Approved Drugs and Their Guiding Role in Precision Medicine Strategies." Communications Medicine (Nature Portfolio), 5, Article 514, December 5, 2025. https://www.nature.com/articles/s43856-025-01270-2
14. [13]Khozin, S. et al. "Artificial Intelligence for Optimizing Recruitment and Retention in Clinical Trials: A Scoping Review." Journal of the American Medical Informatics Association, 31(11), 2749-2761, November 2024. https://academic.oup.com/jamia/article/31/11/2749/7755392
15. [14]Tahir, H. et al. "A Scoping Review of Artificial Intelligence Applications in Clinical Trial Risk Assessment." npj Digital Medicine, 2025. https://www.nature.com/articles/s41746-025-01886-7
16. [15]Food and Drug Administration. "Diversity Action Plans to Improve Enrollment of Participants from Underrepresented Populations in Clinical Studies." FDA Draft Guidance (not for implementation), June 2024. https://www.fda.gov/regulatory-information/search-fda-guidance-documents/diversity-action-plans-improve-enrollment-participants-underrepresented-populations-clinical-studies
17. [16]Akinduro, O. et al. "AI/ML-Based Strategies for Enhancing Equity, Diversity, and Inclusion in Randomized Clinical Trials." Trials, Springer Nature, 2026. https://link.springer.com/article/10.1186/s13063-026-09537-2
18. [17]Olawade, D.B. et al. "Artificial Intelligence in Clinical Trials: A Comprehensive Review of Opportunities, Challenges, and Future Directions." International Journal of Medical Informatics, 206, 106141, February 2026 (epub October 2025). https://www.sciencedirect.com/science/article/pii/S1386505625003582
19. [18]Food and Drug Administration. "Q&A: Role of Artificial Intelligence in Clinical Trial Design and Research, Dr. ElZarrad." FDA Website. https://www.fda.gov/drugs/news-events-human-drugs/role-artificial-intelligence-clinical-trial-design-and-research-dr-elzarrad
20. [19]Vela, D., Sharp, A., Zhang, R., Nguyen, T., Hoang, A. and Pianykh, O.S. "Temporal Quality Degradation in AI Models." Scientific Reports, 12, 11654, July 8, 2022. https://www.nature.com/articles/s41598-022-15245-z
21. [20]Food and Drug Administration. "Artificial Intelligence-Enabled Device Software Functions: Lifecycle Management and Marketing Submission Recommendations." FDA Draft Guidance (FDA-2024-D-4488), January 7, 2025. https://www.fda.gov/regulatory-information/search-fda-guidance-documents/artificial-intelligence-enabled-device-software-functions-lifecycle-management-and-marketing
22. [21]Smith, Z., DiMasi, J. and Getz, K. "New Estimates on the Cost of a Delay Day in Drug Development." Therapeutic Innovation and Regulatory Science, 2024. https://link.springer.com/article/10.1007/s43441-024-00667-w
23. [22]Food and Drug Administration. "Part 11, Electronic Records; Electronic Signatures: Scope and Application." 21 CFR Part 11, FDA Guidance. https://www.fda.gov/regulatory-information/search-fda-guidance-documents/part-11-electronic-records-electronic-signatures-scope-and-application

Related Articles

• ROI of AI Regulatory Document Generation in Clinical TrialsRead article
• How AI Is Transforming Clinical TrialsRead article
• 7 Questions to Ask Before Choosing an AI Patient Screening SolutionRead article
• Why Patient Screening Failures Become Trial CostRead article
• Hidden Costs of Manual Patient ScreeningRead article