Using Natural Language Processing to Improve Fall Documentation in VA Nursing Home Residents

Area of Science:

• Gerontology
• Health Informatics
• Data Science

Background:

• Falls are a significant risk for older adults, necessitating accurate documentation for effective prevention and quality improvement.
• The Minimum Data Set (MDS) is utilized in nursing homes for fall documentation but is criticized for underreporting, misclassification, and lack of fall dates.
• Electronic Health Records (EHRs) contain detailed patient information, including potential fall events.

Purpose of the Study:

• To develop and validate a natural language processing (NLP) algorithm for extracting fall dates from EHRs.
• To compare the NLP algorithm's fall data with the MDS to evaluate misclassification and underreporting of falls.
• To assess the accuracy and utility of NLP in improving fall detection and documentation in long-term care settings.

Main Methods:

• A retrospective cohort study was conducted involving veterans in long-term care.
• A rule-based NLP algorithm was developed to identify fall dates within EHRs.
• The NLP algorithm was validated against manual chart review and compared with MDS v3.0 assessments.

Main Results:

• The NLP algorithm achieved 96% accuracy in identifying fall dates compared to manual chart review.
• The MDS correctly identified 86.1% of residents with falls documented in EHRs, but 6.3% of falls were missed by the MDS.
• A median delay of 17 days was observed between a fall event and its documentation in the MDS.

Conclusions:

• The MDS underreports falls and lacks specific fall dates, limiting its effectiveness for prevention and quality improvement.
• Supplementing MDS fall reporting with NLP can enhance fall detection sensitivity and provide crucial temporal information.
• NLP offers a promising approach to improve the accuracy and completeness of fall data in long-term care settings.