Reproduce “Prompting Large Language Models for Zero-Shot Clinical Prediction with Structured Longitudinal Electronic Health Record Data” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yinghao Zhu, Zixiang Wang, Junyi Gao, Yuning Tong, Jingkun An, Weibin Liao, Ewen M. Harrison, Liantao Ma and 1 more
Researchers developed a prompting framework that enables Large Language Models to perform zero-shot clinical predictions on structured longitudinal Electronic Health Record data, achieving an AUROC of 93.55 for in-hospital mortality prediction on the TJH dataset with GPT-4, significantly outperforming traditional models trained with limited data.
Key insights
- A feature-wise input format for longitudinal data, coupled with explicit clinical context (units and reference ranges), significantly enhances LLM comprehension, compliance, and predictive accuracy for EHR data.
- Integrating 1-2 simulated in-context learning examples into the prompt substantially boosts LLM performance, while too many examples can lead to performance degradation.
- LLMs, particularly GPT-4 with optimal prompting, demonstrate superior zero-shot performance for clinical prediction tasks on structured EHR data compared to traditional models trained with limited (few-shot) examples.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.