Use of Natural Language Processing Algorithms to Identify Common Data Elements in Operative Notes for Knee...

Area of Science:

• Medical Informatics
• Clinical Data Extraction
• Artificial Intelligence in Healthcare

Background:

• Manual review of clinical notes is time-consuming and expensive.
• Electronic health records contain valuable unstructured data.
• Automated data extraction can improve efficiency and data quality.

Purpose of the Study:

• Develop and evaluate rule-based natural language processing (NLP) algorithms.
• Automatically extract surgery-specific data from knee arthroplasty operative notes.
• Reduce the need for manual chart review in orthopedic research.

Main Methods:

• Utilized a cohort of 20,000 knee arthroplasty operative notes (2000-2017).
• Developed NLP algorithms using expert rules and keywords.
• Extracted data on surgery category, laterality, constraint type, patellar resurfacing, and implant model.
• Evaluated algorithms against institutional registry data as the gold standard.

Main Results:

• High accuracy achieved for surgery category (98.3%), laterality (99.5%), constraint (99.2%), and patellar resurfacing (99.4%).
• Implant model extraction yielded an F1-score of 99.9%.
• Demonstrated the effectiveness of NLP in capturing specific surgical details.

Conclusions:

• Rule-based NLP algorithms are a viable and promising alternative to manual chart review.
• Automated extraction of information from knee arthroplasty notes enhances data capture efficiency.
• Further validation in diverse settings is recommended for widespread implementation.