Using Clinical Drug Representations for Improving Mortality and Length of Stay Predictions

Usage

https://github.com/tanlab/MIMIC-III-Clinical-Drug-Representations.git
cd MIMIC-III-Clinical-Drug-Representations

Run MIMIC-Extract Pipeline as explained in https://github.com/MLforHealth/MIMIC_Extract.
Copy the output file of MIMIC-Extract Pipeline named all_hourly_data.h5 to mimic-extract folder.
Copy the ADMISSIONS.csv, PRESCRIPTIONS.csv, ICUSTAYS.csv files into mimic-iii folder.
Run 01-MIMIC-III-Drugs-Names-To-Pubchem-ID.ipynb to convert MIMIC-III Drug names into Pubchem ID.
Download drug information via FDA. https://www.fda.gov/drugs/drug-approvals-and-databases/national-drug-code-directory.
Run 02-Create-Cohort.ipynb to select correct drugs for patients and create the final cohort.
Run 03-Embeddings.ipynb to get embeddings of drugs.
Run 04-Timeseries.ipynb to run timeseries baseline model to predict 4 different clinical tasks.
Run 05-ECFP-1024-TimeSeries.ipynb to run ECFP multimodal baseline to predict 4 different clinical tasks.
Run 6-Smiles-Transformer-TimeSeries.ipynb to run Transformers multimodal baseline to predict 4 different clinical tasks.

Download the MIMIC-III dataset via https://mimic.physionet.org/