Development of research ethics guidelines for healthcare generative artificial intelligence: deriving expert...

Area of Science:

• Healthcare Ethics
• Artificial Intelligence
• Generative AI

Background:

• Generative AI is rapidly expanding in healthcare research and clinical settings.
• Existing AI ethics standards lack specificity for generative AI's unique features like hallucination and autonomy.
• Previous studies focused on universal principles, facing challenges in expert consensus due to broad AI concepts.

Purpose of the Study:

• To systematically identify ethical issues specific to generative AI in healthcare research.
• To develop practical ethical guidelines and a checklist for the entire AI research and development lifecycle.
• To provide sector-specific guidance for healthcare AI under the Framework Act on AI.

Main Methods:

• A three-stage modified Delphi method was employed, adhering to the Conducting and REporting DElphi Studies (CREDES) guidelines.
• Thirty-five experts from medicine, law/ethics/policy, and AI technology participated across in-person and online survey rounds.
• Consensus was measured using Likert scales with criteria for high and strengthened consensus (IQR ≤ 1.5/1.0, CV < 0.5).

Main Results:

• High agreement was reached on guideline adoption (97.7% in Round 2).
• Key ethical considerations included documentation standards, safety measures, and evaluation methods.
• Explainable AI, diverse training data, and human-in-the-loop approaches were identified as top-priority strategies.

Conclusions:

• A differentiated ethical framework and practical checklist tailored to generative AI's technical features were successfully developed.
• These resources offer sector-specific guidance for healthcare AI applications.
• The framework and checklist can be utilized as criteria for institutional review board (IRB) assessments.