Large language models in hepatology: A systematic review

Area of Science:

• Hepatology
• Artificial Intelligence
• Medical Informatics

Background:

• Generative artificial intelligence (AI), particularly large language models (LLMs), is rapidly advancing and presents new opportunities in healthcare.
• LLMs have emerging implications for the field of hepatology, impacting clinical practice and research.

Purpose of the Study:

• To systematically review the current research on the application of LLMs in hepatology.
• To evaluate the capabilities, limitations, and future directions of LLMs in real-world clinical hepatology settings.

Main Methods:

• A systematic literature search was conducted across MEDLINE, EMBASE, and OVID up to January 2025.
• Studies were included if they investigated LLM clinical utility and performance in hepatology with a comparison to a defined ground truth.
• Risk of bias was assessed using the ROBINS-I tool.

Main Results:

• Twenty-one studies were included, showing LLMs can process textual and visual data for liver diseases like hepatocellular carcinoma and cirrhosis.
• LLMs demonstrated utility in radiological image interpretation, clinical decision support, and generating patient education materials.
• Variable accuracy, "hallucinations," data quality dependence, and ethical concerns were identified limitations.

Conclusions:

• Generative AI shows feasibility in hepatology applications, but significant challenges in accuracy, reliability, and safety persist.
• Further research is needed on training methods, data quality, ethical considerations, and real-world validation against benchmarks for safe integration.