From Hours to Seconds: End-User Insights on GenAI-Supported Data Review in Budget-Optimised Clinical Trials

• Understand how generative AI can be safely integrated as a decision-support tool for manual data review in regulated clinical trial environments.
• Learn practical strategies for designing compliant GenAI workflows, including prompt engineering, risk mitigation, and maintaining human oversight to protect data integrity and study confidentiality.

ACDM members and non-members who would like to join an open discussion concerning a Hot Topic in Data Management.

Clinical trials have played a significant role in advancing medical science and improving the healthcare outcomes of today. A key contributor to this success are the stringent regulatory frameworks that governs trial operations to ensure patient safety, data integrity, and scientific validity. However, the emergence of high value but relatively new tools such as generative artificial intelligence (GenAI) challenges the status quo of this highly regulated sector. Thus, AI adoption in clinical trials have been particularly delicate, as the technology continues to evolve within a sector that demands long-term validation, robust information governance, and well-defined risk management processes.

This abstract describes the pragmatic implementation of GenAI as a decision-support tool for manual data review within a budget-optimised clinical research setting. As the sole Clinical Data Manager supporting multiple concurrent Phase IV studies in a high-performing research team, GenAI-assisted workflows were implemented to enhance efficiency and data quality. This was not to showcase technological novelty, but to support reviews where natural language processing (NLP) was essential. In these areas, conventional automated data review tools, such as edit checks and cross-form logic were unsuitable.

Standardised iterative prompts were applied to GenAI tools to optimise their NLP features for manual review of de-identified clinical data only, including medication listings, medical history terms, verbatim adverse event terms, and unavoidable free-text fields. For data management review of pre-randomisation eligibility checks, this approach reduced overall review time from approximately one hour to twenty minutes per subject. GenAI response times ranged from milliseconds to seconds, even for large volumes of data. This ensured protocol compliance and effective sponsor oversight, without delaying patient approval for randomisation or access to study treatment.

During routine monthly data cleaning, GenAI facilitated the review of 5,750 CRFs (115 CRFs per subject) across 50 subjects within one month, compared with the previous timeline of three months. It enabled faster and more accurate identification of safety trends, data entry errors, and reporting anomalies in areas where EDC auto-coders were unavailable, with average response times in milliseconds. Implementing the data review workflow in Appendix 1 below enabled timely soft locking of subject records in batches and allowed sufficient time for site query resolution within existing database lock timelines. This approach also improved overall data quality from approximately 60% to 90% prior to final lock.

To ensure compliance with GCP, GDPR, and institutional AI governance policies, the subject and site identifiers, though already anonymised, were deleted from the datasets used for GenAI-supported reviews. The data manager retained responsibility for reviewing complex checks requiring identifiable study context. However, the contribution of GenAI in reducing review timelines was substantial and cannot be overlooked. Human oversight was maintained throughout, with all GenAI outputs reviewed and validated prior to final decision-making.

Key risks of AI in regulated industries include inaccurate outputs, algorithm bias, fabricated sources, reduced human oversight, and data privacy issues. In this workflow, the first three limitations were mitigated by well-designed AI prompts, while continued human review and use of de-identified data addressed oversight and privacy concerns. Hence, these measures provide a safe and compliant way to use GenAI in CDM, despite limited visibility into vendor data-handling practices compared with traditional trial EDC systems. Until clearer regulatory guidance emerges, this workflow represents one of the most compliant approaches currently applied in practice.

In conclusion, the vital step in harnessing the potential of GenAI is equipping trial teams with effective prompt engineering skills to ensure accurate and reliable outputs. Furthermore, acknowledging AI as a “decision-support tool” rather than a substitute for human judgement. With this foundation in place, focus can shift to addressing broader challenges such as vendor data handling, designing AI-friendly clinical trials, and maintaining regulatory compliance.

MEET THE PRESENTER

Comfort Ajani is certified clinical data manager and EDC implementation professional, with over five years of experience delivering regulated clinical trials in CROs, NHS, and academic research environments across the United States of America and United Kingdom. She specialises in developing clinical data management processes, with a strong focus on optimising data quality, workflow efficiency, and regulatory compliance.

Comfort is particularly interested in how emerging technologies, including artificial intelligence, can enhance efficiency and decision-making in clinical research. She is passionate about knowledge-sharing and regularly contributes to professional development initiatives within the clinical data management community.