Using natural language processing to link patients' narratives to visual capabilities and sentiments.

Area of Science:

• Ophthalmology and Vision Science
• Computational Linguistics
• Data Science

Background:

• Analyzing unstructured clinical narratives is challenging.
• Understanding patient experiences with low vision is crucial for effective care.
• Existing methods may not fully capture the nuances of patient-reported visual function.

Purpose of the Study:

• To demonstrate the feasibility of a framework combining NLP and machine learning for analyzing patient medical records.
• To investigate associations between daily living activities narratives and vision quality in low vision patients.
• To explore links between daily living narratives and patient sentiments towards assistive items.

Main Methods:

• Utilized a dataset of 616 low vision patient records.
• Extracted keywords related to activities of daily living from patient complaints.
• Applied NLP to convert narrative data into numerical formats for machine learning analysis.
• Classified narratives based on acuity, contrast sensitivity, and patient sentiments toward assistive items.

Main Results:

• The framework successfully predicted categories of interest (acuity, contrast sensitivity, sentiments) from patient narratives.
• Demonstrated strong associations between narratives and visual factors for specific activities (e.g., driving).
• Showcased the model's ability to predict patient sentiments and vision quality.

Conclusions:

• The proposed framework effectively extracts semantic patterns from medical narratives.
• This approach can predict patient sentiments and assess vision quality from text data.
• The findings suggest potential for large-scale optometric research using NLP and machine learning on big data.