Understanding Clinicians' Informational Needs for AI-Driven Clinical Decision Support Systems: Qualitative Interview...

Area of Science:

• Health Informatics
• Artificial Intelligence in Medicine
• Clinical Decision Support Systems

Background:

• Artificial intelligence (AI) is transforming healthcare with AI-driven clinical decision support systems (AI-CDSS).
• Clinician adoption of AI-CDSS is hindered by concerns regarding algorithm misuse, misinterpretation, and transparency.
• Understanding clinician informational needs is crucial for effective AI-CDSS integration.

Purpose of the Study:

• To explore clinicians' informational needs and preferences for using AI-CDSS.
• To gather AI experts' perspectives on essential information for safe AI-CDSS use.
• To identify optimal reporting standards for AI-CDSS in clinical practice.

Main Methods:

• Qualitative descriptive study using semistructured interviews with 16 participants (8 clinicians, 8 AI experts).
• Exploration of experiences with AI, informational needs, and feedback on reporting standards (Model Cards, Model Facts, TRIPOD-AI).
• Analysis of interview transcripts using codebook thematic analysis.

Main Results:

• Clinicians require clear data on AI training sets (origin, size, criteria) for applicability assessment.
• Performance metrics beyond AUC are needed, focusing on clinical relevance.
• Specific warnings and limitations are essential to prevent AI-CDSS misuse.
• Information delivery should be layered, customizable, jargon-free, and integrated into clinical workflows.

Conclusions:

• Reporting standards for AI-CDSS must prioritize clinician comprehension and usability.
• Enhanced transparency in training data and performance metrics can improve AI-CDSS assessment.
• A clinician-centered, layered information approach integrated into workflows is vital.
• Co-creation with clinicians during AI-CDSS development is key for practical usability.