Fine-tuned AI could aid trial design and transparency compliance, study says

Customized, “application specific” artificial intelligence (AI) models could help researchers create more effective trial protocols and comply with transparency requirements, according to a new study.

The study, conducted by a team at the University of Copenhagen in Denmark, looked at how application-specific language models (ASLM) — AI models trained on regulations, scientific literature, and best-practice protocols relevant to trials — could guide protocol design.

“This specialized training allows the ASLM to spot common pitfalls — like poorly chosen endpoints or underrepresented populations — very early in the design process,” co-author Sebastian Porsdam Mann told Clinical Insider.

“Ultimately, an ASLM can guide researchers to craft stronger, more inclusive, and more efficient trial protocols before a single patient is ever enrolled,” he continued.

In the study, Posdam Mann and his co-authors suggest application specific models could help sponsors optimize various aspects of protocol design, citing inclusion and exclusion criteria, endpoints, and safety monitoring plans as examples.

They also argue ASLMs could help clinical trial sponsors in resource-limited settings to “design trials with higher chances of meeting regulatory standards.”

Transparency

As well as improving trial design, ASLMs could help drug firms comply with the complex rules that cover trial data publication, according to Porsdam Mann, who said, “One of the biggest challenges for trial sponsors is juggling multiple transparency obligations.”

At present in the EU, sponsors need to comply with EMA policies 0043 and 70, which cover public access to documents and trial data publication, respectively, as well as separate transparency rules set out in the EU Clinical Trial Regulation (CTR).

In time, the proposed European Health Data Space (EHDS) — a framework of rules, common standards, and governance designed to give EU citizens control of their health data — will likely introduce additional transparency obligations.

And this is where ASLMs could help.

Porsdam Mann said, “ASLMs could automate much of the heavy lifting: generating standardized summaries of trial results, cross-checking compliance with different regulatory requirements, and flagging missing information or inconsistencies.

“By training these models with detailed prompts on EMA guidelines, sponsors could get an AI that ‘speaks’ the regulatory language fluently. As a result, trials can meet evolving data-sharing and publication expectations more smoothly,” he said.

DepositPhotos/plarmzag@gmail.com