Identifying Patient-Reported Outcome Measure Documentation in Veterans Health Administration Chiropractic Clinic...

Area of Science:

• Health Informatics
• Natural Language Processing
• Clinical Quality Measurement

Background:

• Patient-reported outcome measures (PROMs) are crucial for high-quality chiropractic care.
• The Veterans Health Administration (VHA) faces challenges in monitoring PROM use due to unstructured clinical notes.
• Natural Language Processing (NLP) offers a solution for extracting PROM data from text.

Purpose of the Study:

• To evaluate NLP techniques for identifying PROMs in VHA chiropractic notes.
• To assess the feasibility of using NLP for quality metric monitoring in chiropractic care.

Main Methods:

• Utilized VHA chiropractic notes (2017-2020) from a large patient cohort.
• Developed and evaluated a rule-based NLP model using medspaCy and spaCy.
• Compared rule-based models with machine learning approaches (bag-of-words, CNNs, ensembles) for note categorization.

Main Results:

• The rule-based NLP model demonstrated high performance in text matching and note categorization.
• Rule-based model achieved 88.2% F-measure for text matching and 94.7% for note categorization.
• Documented PROM use in VHA chiropractic notes was low, at 17.0% prevalence.

Conclusions:

• A rule-based NLP approach is optimal for identifying PROMs in unstructured clinical notes.
• NLP can overcome data extraction challenges, enabling consistent monitoring of PROM use.
• Low PROM documentation indicates a need for quality improvement initiatives in veteran chiropractic care.