Disparities in AI-Based Prior Authorization for Head and Neck Reconstruction: A Large Language Model Analysis

Area of Science:

• Artificial Intelligence in Healthcare
• Medical Informatics
• Health Equity Research

Background:

• Large language models (LLMs) are increasingly used in healthcare for administrative tasks.
• The potential for bias in AI-driven insurance authorization decisions is not well understood.
• This study investigates bias in LLM-based prior authorization for otolaryngology, focusing on oral cavity squamous cell carcinoma (SCC).

Purpose of the Study:

• To assess bias in Large Language Model (LLM)-driven prior authorization for head and neck cancer reconstruction.
• To simulate insurance coverage decisions using GPT-4o for oral cavity squamous cell carcinoma (SCC) treatment.
• To identify demographic disparities in LLM recommendations for surgical reconstruction.

Main Methods:

• Utilized OpenAI's GPT-4o to simulate 19,900 insurance authorization decisions for T2N2 oral cavity SCC.
• Compared two reconstruction methods: radial forearm free flap (RFFF) and split-thickness skin graft (STSG).
• Systematically varied patient profiles by age, sex, race/ethnicity, income, socioeconomic status (SES), and substance use.

Main Results:

• Significant disparities observed: RFFF was favored for younger, Asian, or white patients from high-income/SES backgrounds (p < 0.0001).
• Older, Black, and Hispanic patients, and those from lower-income areas or with substance use history, received fewer RFFF approvals (p < 0.0001).
• Including tumor-specific data in prompts reduced sociodemographic bias, favoring RFFF across groups.

Conclusions:

• LLM outputs demonstrated significant bias based on patient demographics when clinical information was limited.
• Detailed and pertinent clinical data are crucial for LLM inputs to mitigate bias in decision-making.
• Increased regulatory oversight, bias recognition, and safeguards are essential for equitable AI in insurance authorization.