METHODS: A feasibility study was conducted to assess AI/LLM (large language models) in attribute identification for PPR, exploring whether AI with human involvement can increase efficiency in PPR.

RESULTS: The study demonstrates that LLM with human validation supports attribute identification, refinement, and early survey designs. AI leverages extensive databases of existing studies and social media content on patient/caregiver/stakeholder preferences to provide a broad pool of evidence, enabling human researchers to identify relevant attributes. It also provides more nuanced attributes that may be relevant to sub-group populations, or those less documented in published studies that might be overlooked in traditional literature reviews. Additionally, AI produces multiple versions of a single attribute for pretesting with patients and co-creating PPR. Lastly, using automated survey production, piloting, and AI-based validation, AI creates effective preference surveys for validation by researchers and patients.

CONCLUSIONS: Our feasibility assessment demonstrates that AI can effectively generate numerous attributes that are likely important to patients in their decision-making processes, including nuances often missed with traditional methods. These AI-identified attributes can then be validated with patients to build trust, allowing for more patient conversations—thanks to the time saved in the attribute identification phase. Strategically leveraging AI while maintaining transparency in this integration process can optimize efficiency and resource use (quoted as primary reason for underutilization in HTAs), thereby promoting greater utilization of PPR.