EHR_2_vec

Experiences to develop EHR2vec : make a patient cohort pre selection based on clinical notes

---

The data on which we based our experiments comes from a healthcare database, and is therefore not shareable.

Two methods can be tested and implemented for patient pre-selection:

1. a document-oriented method: documents are translated into vector representation by 3 different models:

• the classic TFIDF model, which the user can implement with the tfidf_expe1.ipynb and tfidf_expe2.ipynb notebooks
• the doc2vec model, for use with the doc2vec_expe2.ipynb and doc2vec_expe2.ipynb notebooks
• the docBERT model, for use with notebooks docbert_expe1.ipynb and docbert_expe1.ipynb

2. A query expansion method :

• Firstly, by creating a list of synonyms using the CODER model via the notebook: CODER_synonym.ipynb and, for example, the query proposed in the query_4types.txt file.
• then by searching for documents containing these terms after creating the list of synonyms in the notebook word_search_doc.ipynb.

All codes are available after installation of requirements.txt