JMIR Publications

ABSTRACT

Background and

Objective:

Large language models (LLMs) can provide benefits for most medical applications. However, while most studies investigate how LLMs are used, they neglect the importance of selecting suitable LLM architectures.

Methods:

We conduct a scoping review, to identify which LLMs are used in healthcare. Our search included manuscripts from PubMed, arXiv, and medRxiv. We used open and selective coding to assess the 114 identified manuscripts regarding 11 dimensions related to usage and technical facets.

Results:

We identify four foci that emerged previously in manuscripts with a focus on investigating LLM performance. We find that GPT-based models are used for general purposes like exam preparation or communicating symptoms and treatments, while BERT-based models are used for knowledge discovery and model improvements.

Conclusions:

Healthcare professionals should consider the intended purpose to select the best-suited LLM architecture. Our study suggests that GPT-based model can aid general-purpose tasks in healthcare like report generation or treatment recommendations. However, other LLM families like BERT seem to allow easier extensions of their models for domain-specific tasks.