Could an artificial intelligence approach to prior authorization be more human?

Area of Science:

• Health Informatics
• Artificial Intelligence in Healthcare
• Clinical Decision Support

Background:

• Prior authorization (PA) is a critical but burdensome process in healthcare.
• Current automated PA methods, like HL7 DaVinci, face limitations.
• PA contributes to physician burnout and can delay patient care.

Purpose of the Study:

• To propose an alternative, human-centric approach to PA using artificial intelligence (AI).
• To leverage existing electronic health data and AI for more efficient and just authorization decisions.
• To refine AI models with expert input and few-shot learning to mitigate bias.

Main Methods:

• Utilizing modern data access and exchange for electronic health records.
• Applying AI methods trained on expert panel judgments, including patient representatives.
• Employing few-shot learning techniques to minimize bias in AI models.
• Simulating human appropriateness assessments using AI.

Main Results:

• AI can efficiently simulate human appropriateness assessments from existing data.
• This approach has the potential to reduce administrative burdens and bottlenecks in PA.
• The proposed method aims to preserve the benefits of PA in limiting inappropriate care.

Conclusions:

• AI offers a promising alternative to current rule-based PA systems.
• A human-centric AI approach can create a more just and efficient healthcare system.
• Integrating AI with expert knowledge and patient perspectives can optimize PA processes.