Application and limitations of large language models in epilepsy care

Abstract

Epilepsy is a common neurological disease that is sometimes not well understood and hard to diagnose or treat. Doctors often do not have the time to read many of patient records that might contain useful information for diagnosis or treatment. Researchers therefore have used computer programs called ‘natural language processing’ (NLP) algorithms to automatically analyze these texts to retrieve important information. Large language models (LLMs) are a form of NLP and have recently become popular worldwide for their ability to do many different tasks, including in health care. LLMs have not yet been used for epilepsy, but like other forms of NLP, doctors can also use them to retrieve patient information from patient records. In addition, LLMs can take over mundane tasks such as creating patient letters or they can even teach people about epilepsy. However, people should be cautious not to rely excessively on LLMs, as they often make mistakes in the texts they produce and they have several other drawbacks. It can therefore be dangerous to patients if such mistakes are not noticed by doctors. The goals this paper is therefore to explain how LLMs work and how they can be used for epilepsy while keeping their drawbacks in mind.