Phenotype discovery in patients exhibiting extended length of stay (LOS) in ICU settings

COMP90089 Machine Learning Applications in Health Final Project.
Group #13, Semester 2, 2022.

Group members

• Bushra Aldosari
• Joshua Hoyle
• Yvonne Tao
• Johnson Zhou

Quick access supplementary information to the report

Extracted datasets have been excluded from the repository as per PhysioNet's data use agreement.

They can be re-constructed by running the jupyter notebooks with credentialled access to MIMIC.

Notebooks:

• Feature extraction SQL
• Data exploration
• Data processing
• Dimensionality Reduction PCA
• Clustering K-Means
• Clustering DBSCAN
• Feature importance: largest cluster
• Feature importance: second largest cluster

Organisation

• template: template code to use for queries
• mimic: MIMIC exploration scripts
• data: data sources used in the project
• sql: sql scripts used to query MIMIC
• data_exploration: exploratory data analysis and data processing
• cluster: scripts associated with dimensionality reduction and clustering
• feature_importance_analysis: scripts associated with feature importance analysis
• img: extracted images and plots from notebooks
• src: python helper functions

Resources

MIMIC docs

• Github MIT-LCP/mimic-iv-website (more comprehensive)
• MIMIC-IV docs website

Conda environments

conda_env_gbq.yml: To access and query MIMIC on Google BigQuery locally.
conda_env_tensor_m1.yml: To enable tensorflow with GPU on M1 macs.

conda env create -f <file>.yml

Visual Studio Code

Local access to MIMIC and BigQuery via the following extension:
https://marketplace.visualstudio.com/items?itemName=GoogleCloudTools.cloudcode

Licensing

Usage of data within this repository must comply with PhysioNet's data usage policy.