Extracting Clinical Features From Dictated Ambulatory Consult Notes Using a Commercially Available Natural Language...

Area of Science:

• Clinical Informatics
• Natural Language Processing
• Health Data Management

Background:

• Electronic health records (EHRs) offer valuable data but often contain unstructured text.
• Manual data abstraction from EHRs is time-consuming and labor-intensive.
• Natural Language Processing (NLP) can automate the extraction of structured data from unstructured clinical text.

Purpose of the Study:

• To evaluate the accuracy of a commercial NLP tool.
• To assess NLP's ability to extract clinical features from free-text consult notes.
• To determine NLP performance across varying feature complexity.

Main Methods:

• Pilot, retrospective, cross-sectional study design.
• Utilized dictated consult notes from a tuberculosis clinic.
• Compared NLP extraction against manual chart abstraction as the gold standard.
• Analyzed 15 clinical features categorized by complexity (simple, moderate, complex).

Main Results:

• Overall accuracy for simple features: 96%.
• Accuracy for moderate features: 93%.
• Accuracy for complex features: 91%.

Conclusions:

• NLP is effective for extracting clinical features from dictated notes in a tuberculosis clinic setting.
• Findings support NLP adoption for structured data extraction from EHRs.
• Further validation studies are recommended to establish NLP's broader utility.