Health equity in the era of large language models

Area of Science:

• Health Informatics
• Artificial Intelligence Ethics
• Health Equity

Background:

• Large language models (LLMs) are increasingly used in healthcare.
• Ensuring equitable design and implementation of these AI tools is critical.
• Existing regulations offer guidance on addressing potential disparities.

Purpose of the Study:

• To summarize major US regulations and guidelines impacting equitable healthcare LLM deployment.
• To identify key equity domains and proposed solutions for LLMs in healthcare.
• To highlight the opportunity for LLMs to enhance health equity.

Main Methods:

• Review and synthesis of 8 major US regulations and guidelines.
• Categorization of LLM equity issues into linguistic/cultural bias, accessibility/trust, and oversight/quality control.
• Identification of common solutions across regulations.

Main Results:

• Regulations address diverse data representation and development teams.
• Emphasis on evaluating AI performance against real-world data.
• Mandates for non-discriminatory AI, access for limited English proficiency patients, and human oversight.
• Guidelines promote safe, accessible, and beneficial AI tools respecting privacy.

Conclusions:

• Thoughtfully designed and deployed LLMs can prevent embedding existing healthcare disparities.
• Enhancing health equity requires proactive integration of regulatory principles into AI development.
• Adherence to guidelines ensures LLMs are beneficial and equitable for all patients.