Phase III trials now collect an average of 5.9 million data points per protocol, growing 11% annually since 2020, and that number is precisely why the FDA’s January 2025 draft guidance on AI in regulatory decision-making shifts real cost onto sponsors. The guidance does not treat all AI tools equally. It introduces a risk-based credibility framework, meaning sponsors must calibrate how much validation work they do based on how consequential the AI output is. The higher the stakes of a given AI-generated result for a regulatory decision, the more documentation, testing, and justification the sponsor owns.
That burden is not theoretical. Sponsors using AI to clean data, flag protocol deviations, or model safety signals now face a formal question: can you demonstrate, with audit-ready evidence, that the tool produced a reliable output for this specific use? The guidance does not require that a tool be perfect. It requires that sponsors understand where it can fail, document that understanding, and show the FDA the validation logic behind any AI-generated data that feeds a submission. In practice, that means validation plans written before deployment, not after the data are already in a dossier.
The friction point for most trial teams is that AI tools in clinical operations have typically been adopted for efficiency, with formal validation treated as a software IT function. FDA’s existing data integrity standards already require sponsors to maintain quality assurance systems and written SOPs covering clinical study conduct, but the new AI guidance extends that accountability into the algorithm layer. A sponsor cannot hand the FDA an output from an AI-assisted EDC cleaning process and simply point to the vendor’s documentation. The sponsor must demonstrate fit-for-purpose validation specific to their protocol’s data and the decision that AI output informed.
Operationally, this creates a sequencing problem that will surface first at the largest sites running high-volume Phase III programs. Teams that adopted AI tooling in 2023 or 2024 without a prospective validation strategy now face retrospective documentation work before those tools can support a submission. The concrete pressure point to watch is whether FDA, in finalizing the guidance, provides worked examples or a defined threshold for what constitutes sufficient credibility evidence, because without that, sponsors are writing their own standards in a vacuum.
Source link: https://www.eclinicalsol.com/news/fdas-ai-risk-based-guidance-shifts-validation-burden-to-sponsors/
Moe Alsumidaie, MBA, MSF, is founder and Chief Editor of Vanguard Publications, which publishes Clinical Trial Vanguard, Pharma Vanguard and BullScope, and Head of Research at CliniBiz. He has two decades in clinical trial operations and data science, with earlier roles at Genentech, Abbott Vascular and Stanford University Medical Center, and is a guest lecturer in clinical trial sciences at Rutgers University.
