Leveraging Large Language Models to Identify In-Hospital Cardiac Arrest.

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Clinical Data Analysis

Background:

• Manual chart abstraction is the current standard for identifying in-hospital cardiac arrest (IHCA) but is labor-intensive.
• Diagnosis codes offer an accessible, automated alternative but suffer from low accuracy (sensitivity and positive predictive value).

Purpose of the Study:

• To develop and evaluate a novel large language model (LLM) for automated identification of IHCA.
• To determine the accuracy and efficiency of LLMs in extracting IHCA events and their locations from clinical notes.

Main Methods:

• A novel approach utilizing large language models (LLMs) was developed to process clinical notes.
• The LLM was trained and applied to identify instances of in-hospital cardiac arrest (IHCA) and their specific locations within the patient's record.

Main Results:

• The large language model approach demonstrated potential for rapid and accurate IHCA identification.
• This method highlights the capability of LLMs to automate the extraction of critical clinical events from unstructured text.

Conclusions:

• Large language models offer a promising solution for overcoming the limitations of manual abstraction and diagnosis codes in IHCA identification.
• Automated LLM-based methods can significantly improve the efficiency and accuracy of identifying in-hospital cardiac arrest events.