Development and Validation of a Rule-Based Natural Language Processing Algorithm to Identify Falls in Inpatient...

Area of Science:

• Gerontology
• Medical Informatics
• Public Health

Background:

• Underestimation of falls by International Classification of Diseases (ICD) codes in clinical settings necessitates improved detection methods.
• Natural Language Processing (NLP) offers a potential solution by extracting information from clinical notes.
• The application of NLP to inpatient notes for fall identification remains under-investigated.

Purpose of the Study:

• To develop and validate a rule-based NLP algorithm for identifying falls.
• To utilize inpatient admission notes from older patients for fall detection.
• To compare NLP-identified falls with those identified by ICD codes.

Main Methods:

• Retrospective analysis of 12-year electronic inpatient records of patients aged ≥65 years in Hong Kong.
• Development of a rule-based NLP algorithm using a random sample of 1000 patients.
• Performance evaluation using manual review as the gold standard (sensitivity, specificity, precision, F1-score) at record, episode, and patient levels.

Main Results:

• The NLP algorithm demonstrated excellent performance: sensitivity (93.3%), specificity (99.0%), precision (87.5%), and F1-score (0.903) at record/episode levels.
• Patient-level performance was also high: sensitivity (92.9%), specificity (98.3%), precision (89.7%), and F1-score (0.912).
• A combined strategy using ICD codes and NLP yielded the most comprehensive capture of fall-related episodes and fallers.

Conclusions:

• The NLP method is efficient and accurate for detecting falls from inpatient clinical notes.
• Combining NLP with ICD codes is recommended for comprehensive fall identification in future studies and clinical practice.
• This approach can aid in identifying high-risk groups for targeted fall interventions.