Is ChatGPT Listening in On Your Next Informed Consent?

Jeri Burr, MS, RN, PED-BC, CCRC, FACRP, Program Director, University of Utah

You may be happily using ChatGPT to produce meeting minutes or literature reviews, and your children may be secretly using it to complete essays and term papers, but is the world of medical research ready to have such artificial intelligence (AI) tools producing informed consent forms (ICFs) for clinical trial participants?

Ready or not, various applications of AI are openly and not-so-openly filtering into the mainstream operations of clinical research sponsors and study sites, as in so many other professional settings. When it comes to the importance of the informed consent process and the forms used to explain the purpose, risk/benefit calculation, and scope of a clinical trial to participating patients, speakers slated to dive into the topic at the ACRP 2024 conference urge caution from trial team members who may be hoping for ChatGPT to ease ICF-related burdens on their list of chores.

For one thing, Jeri Burr, MS, RN, PED-BC, CCRC, FACRP, Program Director for the HEAL Pain Effectiveness Research Network and the Data Coordinating Resource Center at the University of Utah, notes that trial team members should be aware that the companies offering different ChatGPT products may, behind the scenes, be using summaries of a site’s pre-existing ICFs and the contents of new ICF drafts under development to “train” their AI models to handle such contents.

“It’s important to go into the ChatGPT settings and turn OFF the history so that the company will not use your online conversations about your ICFs to train its model,” Burr says. “If you are using ‘pretend’ consent documents to develop your next form, this is not an issue, but if we are looking at real consent documents, just know that the contents of your conversations are being used to train the model. Similarly, if you put up a document and ask for a summary, the document is now part of the training data.”

Burr, along with University of Utah colleagues Ann Johnson, PhD, MPH, Director of the Institutional Review Board and Human Research Protection Program, and John VanBuren, PhD, Associate Professor of Biostatistics, aim to answer the question of whether the use of ChatGPT can improve the quality of consent forms and create consent forms that are more understandable to everyday people while meeting regulatory requirements. They will also present practical guidance on effectively utilizing ChatGPT to review already-created consent forms—not asking ChatGPT to write the document, but to help with other tasks that are hard for research staff—and on using ChatGPT to pull concise summaries out of long documents and to decrease the complexity (reading grade level) of documents.

With an eye toward institutional compliance concerns tied to ChatGPT, Johnson notes that, while researchers can employ it to explore ethical questions, assess public opinion, and conduct virtual experiments, “this raises concerns about breach of confidentiality and ensuring appropriate informed consent from users whose data are used to train or fine-tune the model” following such uses. “Institutions must establish stringent protocols to safeguard against potential misuse and ensure compliance with data privacy regulations when employing ChatGPT in research settings,” she adds.

Revolutionizing Informed Consent Through Responsible Use of Artificial Intelligence

Join Jeri, Ann, and John at ACRP 2024 [May 3 – 6; Anaheim, CA], where they will discuss how ChatGPT has the potential to revolutionize the create of informed consent forms and related documents for clinical trials. View complete schedule.

Meanwhile, VanBuren, a Faculty Biostatistician and principal investigator (PI) for many trials, aims to examine the role of ChatGPT in research beyond informed consent and its impact on institutional compliance.

“PI oversight of how ChatGPT is being used at a site is vital for ensuring the validity of changes being made to ICFs,” he says. “The last thing we want is for staff to just ‘Accept All’ when the program suggests revisions and call the consent form good to go. But beyond those kinds of concerns, we should keep up to speed on institutional expectations about whether and how ChatGPT is being used in other areas tied to our trials—such as for grant writing, reporting results to sponsors and regulatory authorities, and authoring journal articles.”

Edited by Gary Cramer