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

Usage

1. Clone the code to local.

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

2. Run MIMIC-Extract Pipeline as explained in https://github.com/MLforHealth/MIMIC_Extract.

3. Copy the output file of MIMIC-Extract Pipeline named all_hourly_data.h5 to mimic-extract folder.

4. Copy the ADMISSIONS.csv, PRESCRIPTIONS.csv, ICUSTAYS.csv files into mimic-iii folder.

5. Run 01-MIMIC-III-Drugs-Names-To-Pubchem-ID.ipynb to convert MIMIC-III Drug names into Pubchem ID.

6. Download drug information via FDA. https://www.fda.gov/drugs/drug-approvals-and-databases/national-drug-code-directory.

7. Run 02-Create-Cohort.ipynb to select correct drugs for patients and create the final cohort.

8. Run 03-Embeddings.ipynb to get embeddings of drugs.

9. Run 04-Timeseries.ipynb to run timeseries baseline model to predict 4 different clinical tasks.

10. Run 05-ECFP-1024-TimeSeries.ipynb to run ECFP multimodal baseline to predict 4 different clinical tasks.

11. Run 6-Smiles-Transformer-TimeSeries.ipynb to run Transformers multimodal baseline to predict 4 different clinical tasks.

References

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

MIMIC-Extract implementation: https://github.com/MLforHealth/MIMIC_Extract