Natural language processing to ascertain two key variables from operative reports in ophthalmology

Area of Science:

• Ophthalmology
• Medical Informatics
• Surgical Safety Research

Background:

• Antibiotic prophylaxis is essential in ophthalmic surgery.
• Accurate data extraction is crucial for comparative effectiveness research.
• Identifying specific variables like intracameral antibiotic injection and posterior capsular rupture (PCR) is vital for studying cataract surgery outcomes.

Purpose of the Study:

• To apply natural language processing (NLP) to operative notes for extracting key variables for cataract surgery research.
• To ascertain the exposure variable (intracameral antibiotic injection) and a potential confounder (posterior capsular rupture).
• To describe the NLP protocol and lessons learned for broader research application.

Main Methods:

• Utilized SAS Text Miner and SAS text-processing modules on 743,838 operative notes.
• Developed lexicons from a training set of 10,000 notes to identify misspellings, abbreviations, and link terms into concepts.
• Iteratively confirmed NLP tool accuracy using random samples of 2,000 notes.

Main Results:

• Successfully identified approximately 60,000 intracameral antibiotic injections and 3,500 cases of PCR.
• Achieved high accuracy: positive and negative predictive values exceeded 99% for antibiotic injections.
• Demonstrated strong performance for PCR detection, with predictive values exceeding 94%.

Conclusions:

• Natural language processing (NLP) is a valid and feasible method for extracting critical variables for surgical safety research.
• The developed NLP tools proved effective for the study sample.
• Further validation is recommended for external or future datasets.