Towards a practical use of text mining approaches in electrodiagnostic data

Area of Science:

• Computational Medicine
• Medical Informatics
• Neurology

Background:

• Healthcare generates vast amounts of unstructured textual data.
• Extracting insights from this data is crucial for advancing computational medicine.
• Text mining and statistical methods can unlock valuable clinical information.

Purpose of the Study:

• To demonstrate combining text mining and statistical methods for neurological and neuromuscular health insights.
• To identify patient groups with similar diagnostic attributes.
• To examine demographic and past medical history differences between patient groups.

Main Methods:

• Retrospective study of patients undergoing electrodiagnostic (EDX) evaluation (May 2016 - Feb 2022).
• Data extraction included demographics, test results, and unstructured summary reports.
• Topic modeling (Latent Dirichlet Allocation) and statistical analysis were used to analyze clinical impressions, age, sex, and past medical history.

Main Results:

• Identified 25 diagnostic topics from clinical notes.
• Discovered sex-related differences in 7 topics (3 male-associated, 4 female-associated).
• Found significant age and sex differences in Brachial plexopathy, Myasthenia gravis, and NMJ Disorders.
• Established associations between past medical history keywords and diagnostic topics (e.g., diabetes with polyneuropathy, chemotherapy with sensory polyneuropathy).

Conclusions:

• Text mining and statistical analysis effectively extract insights from clinical text.
• This approach accelerates clinical research and aids in developing decision-making processes.
• Identified specific associations between past medical history and neurological/neuromuscular diagnoses.