Clinical Documents Clustering Based on Medication/Symptom Names Using Multi-View Nonnegative Matrix Factorization.

Area of Science:

• Health Informatics
• Natural Language Processing
• Machine Learning

Background:

• Clinical documents contain valuable unstructured data on medications and symptoms.
• Extracting and utilizing this information can significantly enhance healthcare.
• Current methods may not fully leverage the semantic content of clinical notes.

Purpose of the Study:

• To develop an integrated system for extracting medication and symptom names from clinical text.
• To apply Nonnegative Matrix Factorization (NMF) and multi-view NMF for clinical document clustering.
• To compare the effectiveness of feature extraction methods for document clustering.

Main Methods:

• Developed a system for automated extraction of medication and symptom entities from clinical notes.
• Employed Nonnegative Matrix Factorization (NMF) and a multi-view variant for document clustering.
• Clustering was performed using sample-feature matrices derived from extracted entities and general words.

Main Results:

• Multi-view NMF demonstrated superior performance in clustering clinical documents compared to standard NMF.
• Clustering clinical documents using extracted medication and symptom names significantly outperformed using only general words.
• The integrated system effectively captures and utilizes key clinical information for improved document organization.

Conclusions:

• Multi-view NMF is a highly effective technique for clustering clinical documents.
• Leveraging extracted medication and symptom names enhances the accuracy and meaningfulness of clinical document clustering.
• This approach offers a promising pathway for better organization and analysis of clinical data.