NLP for healthcare

Predicting 30-day ICU readmissions using structured and unstructured data in MIMIC III

Structured data

• Logic for the structured network modeling is contained within files in the Structured directory
• The ETL logic for structured network can be found in structured_etl_part1.scala and structured_etl_part2.py
• Modeling logic is contained within struc_net.py
• Modeling scripts are run through the wrapper py_train_struc.py which also allows for random search for hyperparameter tuning for the network.

Unstructured data

Data Processing for unstructured text

• All data processing scripts for unstructured data are contained in the dataproc directory.
• Process NOTEEVENTS to get word vectors using data_processing_script.py.
  • Write Discharge summaries using get_discharge_summaries.py
  • Build vocab from discharge summaries using build_vocab.py.
  • Train word embeddings on all words using word_embeddings.py.
  • Write trained word embeddings with our vocab using gensim_to_embeddings method in extract_wvs.py.

Modeling scripts

• Modeling scripts can be found in the models directory.
• To run the modeling scripts, arguments ca