Performance and improvement strategies for adapting generative large language models for electronic health record applications: A systematic review

Xinsong Du, Zhengyang Zhou, Yifei Wang, Ya-Wen Chuang, Yiming Li, Richard Yang, Pengyu Hong, David W. Bates, Li Zhou

Abstract: Highlights•Only a small portion of studies quantitatively evaluated improvement strategies.•The most assessed performance improvement strategy was fine-tuning.•Few-shot, fine-tuning, and multimodal data integration may degrade the performance.•Fine-tuning a small model can outperform a large model.•Performance biases were only assessed in four studies.

External IDs:dblp:journals/ijmi/DuZWCLYHBZ26