What is it about?
Can ChatGPT really help review Japanese clinical research documents for ethics? Our pilot study suggests it just might. We examined whether generative AI models—specifically GPT-4 and GPT-4o—could accurately and consistently identify key elements in Japanese-language clinical research documents, such as research protocols and informed consent forms. These elements are crucial for ethical review and include the study’s objectives and background, research design, and participant-related risks and benefits. We compared two versions of ChatGPT and tested whether using custom-designed prompts would improve performance over standard prompts. The results showed that GPT-4o, especially when guided with well-crafted prompts, produced more accurate and consistent outputs. This pilot study lays the foundation for developing explainable AI tools to support (but not replace) institutional review board (IRB) decision-making in Japan and potentially other non-English-speaking contexts.
Featured Image
Photo by Jonathan Kemper on Unsplash
Why is it important?
This study represents an important first step toward developing explainable AI tools that aim to support, not replace, the ethical review of clinical research documents. We evaluated how accurately and reliably GPT models can extract key components—such as the study’s purpose, design, and potential risks and benefits to participants—that are considered essential in ethical review, using original Japanese documents and custom-designed prompts written in Japanese. This approach preserves the original wording and nuance, making it especially practical in non-English research contexts. Our findings demonstrate the potential for generative AI to assist in pre-review of Japanese-language research documents, contributing to greater transparency, consistency, and efficiency in IRB decision-making, and laying the groundwork for broader application of AI in global research ethics review.
Perspectives
Why I started this research In Japan, IRB reviewers often spend hours reading long documents just to find basic information like the study’s goal or risks. I wondered if AI could help by quickly pointing out these key parts—without needing translation. What surprised me With carefully written Japanese prompts, GPT-4o could extract essential details—like purpose, design, and risk–benefit balance—with impressive consistency. But it still struggles with technical terms, so human review remains essential. What comes next I hope AI can handle routine checks so people can focus on tough ethical decisions. I’m now working on improving accuracy with new methods and want to collaborate with others in the field.
Yasuko Fukataki
Read the Original
This page is a summary of: Developing artificial intelligence tools for institutional review board pre-review: A pilot study on ChatGPT’s accuracy and reproducibility, PLOS Digital Health, June 2025, PLOS,
DOI: 10.1371/journal.pdig.0000695.
You can read the full text:
Contributors
The following have contributed to this page
