Repository files navigation

This is the code for our paperRAM-EHR: Retrieval Augmentation Meets Clinical Predictions on Electronic Health Records, published in ACL 2024.

[IMPORTANT, PLEASE READ!]In order to facilitate the reproducibility, we provide a small synthetic subset in data for MIMIC-III dataset inmimic folder (hyperedges-mimic-text-{train/valid/test}-example.jsonl). We include them in this repo only to show the format of the two datasts we used. Thus, their experimental results should NOT reflect the performance we report in the paper.

We provide some of the publicly available sources inmesh.txt as well as the code for querying drugbank and pubmed database inquery_drugbank.py andquery_pubmed.py, respectively. The processed information for each medical code can be found in themimic andcradle folder. The drugbank information can be accessed atthis link.

Please refer toprompt_gpt_def.py for details.

For graph neural network component, please follow our previous work onhypergraph neural networks for EHR prediction for details. We have also provided the code of using language models for prediction inmodel_src folder.

If you find this paper useful for your research, please cite the following in your publication. Thanks!

@inproceedings{xu2024ram,  title={RAM-EHR: Retrieval Augmentation Meets Clinical Predictions on Electronic Health Records},  author={Xu, Ran and Shi, Wenqi and Yu, Yue and Zhuang, Yuchen and Jin, Bowen and Wang, May D and Ho, Joyce C and Yang, Carl},  booktitle={Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics},  year={2024}}