SemanticFind: Locating What You Want in a Patient Record, Not Just What You Ask For

Area of Science:

• Medical Informatics
• Natural Language Processing
• Information Retrieval

Background:

• Traditional patient record search relies on literal text or synonym matching.
• Existing methods often fail to retrieve clinically relevant information that is not explicitly queried.

Purpose of the Study:

• To introduce SemanticFind, a novel model for patient record search.
• To improve the retrieval of clinically relevant patient data beyond explicit user queries.

Main Methods:

• Utilized the Unified Medical Language System (UMLS) semantic network.
• Employed distributional semantics and Natural Language Processing (NLP) techniques.
• Matched query terms across multiple dimensions within patient records.

Main Results:

• SemanticFind identified twice as many relevant matches compared to traditional literal search.
• Achieved high precision and recall in retrieving clinically related concepts.
• Organized returned matches along several dimensions for better usability.

Conclusions:

• SemanticFind significantly improves patient data retrieval accuracy.
• The model has potential applications in clinical practice and retrospective chart reviews.
• SemanticFind can aid in the automated extraction of quality metrics from patient records.