Login
Donate
image

Discussion Guide and Transcript

Season Three - Episode Ten

image

“AI in Clinical Research and Ethical Oversight" Guests: Jonathan Helfgott

  • In this episode of PRIM&R's podcast, "Research Ethics Reimagined," we explore the use of artificial intelligence in clinical research and ethical oversight with Jonathan Helfgott, senior lecturer and program coordinator for the Master of Science in Regulatory Science program at Johns Hopkins University and former Associate Director for Risk Science in the Office of Scientific Investigations at the US Food and Drug Administration's Center for Drug Evaluation and Research. Helfgott discusses why transparency is the central risk when AI is used in research, where accountability sits when AI supports decisions, and how institutional review boards (IRBs) can approach new tools without losing sight of fundamentals. He also considers how AI may speed product development, what meaningful governance could look like, and how new approaches might broaden who can take part in research. Listen on Spotify | Listen on Apple| Listen on Amazon
This podcast is part of PRIM&R's AI Ethics and Governance Initiative, supported by a grant from Advarra. Advarra's support reflects a shared commitment to advancing ethical, responsible research and helping PRIM&R expand timely educational resources and foster informed discussion on the ethical and responsible use of artificial intelligence in research.

Discussion Questions

Transparency and Accountability When AI Is Used in Research
  • Helfgott names transparency as "the biggest risk," noting that there is no standardized way to disclose whether, how, and by whom AI is being used, and that IRBs and applicants are often unsure what needs to be disclosed. How can IRBs and investigators develop shared expectations about disclosing AI use, and what details would make that disclosure meaningful rather than a formality?
  • He stresses that there is "no exemption to the rule" just because AI is involved, and that sponsors, clinical investigators, and IRBs are responsible, individually and collectively, for ensuring that tools are fit for purpose. How should institutions think about accountability when AI is used by applicants, reviewers, and vendors at different points in the same study?
Practical Approaches for IRBs and Institutions
  • Helfgott cautions IRBs not to "let the tail wag the dog" when vendors offer "a new bell and a whistle" every other day, urging them to match each tool to a specific activity rather than treating AI as equivalent across every phase of review. He also reminds listeners that tools can hallucinate and that an expert must catch the errors, and that compliance is "never a one person show." How can an IRB office decide which AI tools fit its needs, and who should be involved in those decisions?
  • He suggests that PRIM&R and the broader community consider developing a best-practices checklist for good clinical practice (GCP) compliance when using AI tools, while also saying that compliance is "just as much of an art as it is a science" and "not a litmus test." How can a checklist support consistent review without encouraging box-checking in place of critical thinking?
Speed, Innovation, and Inclusion
  • Helfgott expects AI to shorten development timelines one step at a time, beginning with administrative tasks, rather than as "a quantum leap," and says safety and efficacy review can never be shortchanged. How should research institutions and oversight bodies respond to growing pressure to move faster while protecting participants and the quality of evidence?
  • He describes using digital twins to strengthen a trial analysis, while warning that synthetic control arms can raise concerns if comparison patients are selected in ways that influence outcomes. He also describes a hybrid model that brings patients from primary care practices to research sites, reaching people "hiding in plain sight." What ethical questions should IRBs ask about such approaches, and how might they weigh the potential for greater inclusion against the risks of new methods?
Key Terms
  • Good Clinical Practice (GCP): An international ethical and scientific quality standard for designing, conducting, recording, and reporting clinical trials involving human participants. Helfgott repeatedly frames AI use in terms of compliance with GCP.
  • ALCOA: A shorthand for the attributes of reliable research data: attributable, legible, contemporaneous, original, and accurate. Helfgott notes that these expectations apply equally to paper records, electronic records, and AI tools.
  • Digital Twins and Synthetic Control Arms: Computer-generated or data-matched comparison patients, often built from information about people who have already received a treatment, that can supplement or reduce the need for a traditional placebo group. Helfgott cautions that how comparison patients are chosen can influence a study's outcome.
  • Predetermined Change Control Plan (PCCP): An FDA mechanism that allows the maker of an AI-enabled medical device to gain agreement in advance on certain planned updates, such as retraining its model, so that each change does not require a new submission.
Resources

Transcript

Please note, a transcript generator was used to help create written show transcript. Listen to the full conversation on PRIM&R’s Research Ethics Reimagined podcast.