This code is for paper "Interpretable Disease Prediction via Path Reasoning over Medical Knowledge Graphs and Admission History".
The key files are:
-
train.py: Used for training and validating the models. All parameters use default values, but can also be set at runtime. -
metrics.py: Contains all evaluation metrics used in the paper, which referenced https://github.com/luchang-cs/chet -
The
modeling/folder contains the model implementations. -
The
util/folder contains utility functions like data processing and transformations. -
The
data/folder is meant for storing the data, but we cannot provide it due to restrictions. Please download it yourself following the instructions provided.You can download the MIMIC III from : https://physionet.org/content/mimiciii/1.4/
MIMIC IV from: https://physionet.org/content/mimiciv/2.2/
To obtain the data, you need to register as a credentialed user on PhysioNet, complete the required CITI training on research ethics, and sign the PhysioNet Data Use Agreement. Please refer to the database overview pages for complete instructions.
To run:
python train.py
The code shows an example using the MIMIC III dataset, but can be adapted for MIMIC IV by changing the data paths.
- NOTE: Please download the MIMIC dataset yourself before running. Contributions and improvements are welcome!
In our work, part of the code is referenced from the following open-source code:
-
QA-GNN: Question Answering using Language Models and Knowledge Graphs. https://github.com/michiyasunaga/qagnn
-
HiTANet: Hierarchical Time-Aware Attention Networks for Risk Prediction on Electronic Health Records. https://github.com/HiTANet2020/HiTANet
-
Chet: Context-aware Health Event Prediction via Transition Functions on Dynamic Disease Graphs. https://github.com/luchang-cs/chet
Many thanks to the authors and developers!