willtsai / dlh-sp23-team53

Repository for Team 53 from UIUC CS 598 DLH Spring 2023

Repository for term project: UIUC CS 598 DLH, Spring 2023, Team 53

This repository contains the code for the term project of UIUC CS 598 DLH, Spring 2023, Team 53. The project is based on the paper Context-aware Health Event Prediction via Transition Functions on Dynamic Disease Graphs (Chang et. al., 2022). The goal of this project is to reproduce the original experiments that predict the health events of patients based on their medical history and context information. We also selected the CGL model (Chang et. al., 2021) to reproduce as the baseline comparison.

Project authors:

• Sherry Li ([email protected])
• Wei-Lun Tsai ([email protected])

Original authors (Chet):

• Chang Lu ([email protected])
• Tian Han ([email protected])
• Yue Ning ([email protected])

Original authors (CGL):

• Chang Lu
• Chandan K. Reddy
• Prithwish Chakraborty
• Samantha Kleinberg
• Yue Ning

Original sourcecode repositories:

• github.com/LuChang-CS/Chet
• github.com/LuChang-CS/CGL

Main reproduction experiment results

Instructions

Clone this repository

git clone https://github.com/willtsai/dlh-sp23-team53.git
cd dlh-sp23-team53

Download the data

The data can be downloaded from https://mimic.physionet.org/. You will need to obtain permissions for downloading the MIMIC-III (Johnson et. al., 2016) and MIMIC-IV (Johnson et. al., 2023) datasets. Once you have acquired the necessary credentials, you can download the data into the required directories by running the following commands and script:

export PHYSIONET_USERNAME=yourusername
export PHYSIONET_PASSWORD=yourpassword
bash initialize.sh

Install dependencies

Dependencies for both our main repro model (Chet) and the baseline model (CGL) are captured in requirements.txt. To install the dependencies, run the following command:

pip install -r requirements.txt

Set up and train the model

Run each cell in order from the Jupyter notebooks chet.ipynb and cgl.ipynb to preprocess and load the data, train the models, and evaluate accuracy for diagnosis and heart failure prediction tasks.

References

Alistair Johnson, Tom Pollard, and Roger Mark. 2016. Mimic-iii clinical database.

Alistair Johnson, Lucas Bulgarelli, Tom Pollard, Steven Horng, Leo Anthony Celi, and Roger Mark. 2023. Mimic-iv.

Chang Lu, Tian Han, and Yue Ning. 2022. Context- aware health event prediction via transition functions on dynamic disease graphs. Proceedings of the AAAI Conference on Artificial Intelligence, 36(4):4567–4574.

Chang Lu, Chandan K Reddy, Prithwish Chakraborty, Samantha Kleinberg, and Yue Ning. 2021. Collabo- rative graph learning with auxiliary text for temporal event prediction in healthcare. In Proceedings of the Thirtieth International Joint Conference on Artificial Intelligence, IJCAI-21, pages 3529–3535. International Joint Conferences on Artificial Intelligence Organization. Main Track.