Interpretable RL Framework for ICU Hypotension Management

This repo contains notebooks, code and files associated with paper An Interpretable RL Framework for Pre-deployment Case Review, Improvement, and Validation Applied to Hypotension Management in the ICU

Requirements

• python 3.7 or above
• pytorch version 1.10 or above
• numpy 1.21 or above
• sklearn 1.0.1
• pandas 1.3.4

Code Access

• notebook: the folder contains jupyter notebooks required to produce our research results. To reproduce the results in our manuscript, you can execute the ordered notebooks. Note, before running 03.identify_and_evaluate_cluster.ipynb, you need to run the code launch_P_calc_four_action.py to determine which points among the train and test data are decision points.
• src: the folder contains helper functions.
• results: the folder contains intermediate, visualizations and evaluation results from running the notebooks.
• mimic_pipeline_tools: the folder contains scripts that extract query data from raw data and scripts that do data cleaning.

Data Access

The data we used for this project was derived from MIMIC-III - Medical Information Mart for Intensive Care using mimic_pipeline_tools. Regarding questions around running the scripts, please contact Jiayu Yao at jiy328@g.harvard.edu or Dr.Finale Doshi who runs Data to Actionable Knowledge (DtAK) Lab at finale@seas.harvard.edu.