Disparities in seizure outcomes revealed by large language models

Area of Science:

• Medical Informatics
• Neurology
• Health Services Research

Background:

• Large-language models (LLMs) in healthcare may perpetuate or introduce biases.
• Social determinants of health impact epilepsy care access, but their effect on seizure outcomes among those with specialty care access is unclear.

Approach:

• Evaluated an epilepsy-specific LLM for intrinsic bias across demographic groups (race, ethnicity, sex, income, insurance).
• Utilized LLM-extracted seizure outcomes to assess disparities in seizure freedom across demographic groups.
• Employed univariable and multivariable analyses on data from 84,675 clinic visits (25,612 patients, 2005-2022).

Key Points:

• The epilepsy-specific LLM demonstrated no intrinsic bias in performance across demographic groups.
• Multivariable analysis revealed worse seizure outcomes for female patients, those with public insurance, and individuals from lower-income zip codes.
• Black patients showed worse outcomes in univariable but not multivariable analysis compared to White patients.

Conclusions:

• LLM-extracted seizure data revealed significant disparities in epilepsy care outcomes.
• Findings underscore the urgent need to address and reduce health disparities in epilepsy management.
• Ensuring equitable care for all epilepsy patients remains a critical challenge.