Mount Sinai: GPT-4 predicts ER admissions with minimal training

Mount Sinai researchers discovered that generative AI like GPT-4 can predict hospital admissions for emergency room patients with minimal training and limited data.

As part of a study, researchers from the Icahn School of Medicine at New York City-based Mount Sinai analyzed records from more than 864,000 emergency room visits across seven Mount Sinai Health System hospitals. Using structured data, such as vital signs, and unstructured data, such as nurse triage notes, researchers developed models to predict hospital admissions. Out of the total of emergency room visits analyzed, 18.5% resulted in hospital admissions, according to a May 21 news release from Mount Sinai.

The researchers evaluated GPT-4's performance against traditional machine-learning models such as Bio-Clinical-BERT for text analysis and XGBoost for structured data. They explored various scenarios, assessing GPT-4's ability to predict admissions both independently and when combined with traditional methods.

The researchers found that unlike traditional machine-learning models that require millions of records for training, large language models like GPT-4 could effectively learn from just a few examples. 

They also noted that large language models could incorporate predictions from traditional models, essentially enhancing their performance.

The full study was published on May 21 in the online issue of the Journal of the American Medical Informatics Association.

Copyright © 2024 Becker's Healthcare. All Rights Reserved. Privacy Policy. Cookie Policy. Linking and Reprinting Policy.


Featured Whitepapers

Featured Webinars