Large Language Model-Based Critical Care Big Data Deployment and Extraction: Descriptive Analysis

Area of Science:

• Biomedical Informatics
• Clinical Data Management
• Artificial Intelligence in Healthcare

Background:

• Publicly accessible critical care databases hold vast clinical data but require advanced programming skills for utilization.
• Complexity of large and unstructured datasets poses challenges for clinicians lacking data analysis expertise.

Purpose of the Study:

• To simplify the deployment and extraction of critical care databases using large language models.
• To empower clinicians with easier access to critical care data without extensive technical expertise.

Main Methods:

• Developed an automated database deployment platform using Docker with Metabase and Superset analytics interfaces.
• Created the intensive care unit-generative pretrained transformer (ICU-GPT), a large language model fine-tuned on ICU data, integrating LangChain and Microsoft AutoGen.

Main Results:

• The platform enables user-friendly, automated deployment of databases across various environments.
• ICU-GPT generates SQL queries and extracts insights from ICU data, overcoming token limits and supporting multischema data.
• A front-end interface allows for code-free SQL generation and data visualization.

Conclusions:

• The automated platform and ICU-GPT enhance efficiency and flexibility in visualizing, extracting, and arranging critical care data.
• This approach can reduce time and effort associated with complex bioinformatics methods, advancing clinical research.