Identifying Patients with Depression Using Free-text Clinical Documents.

Area of Science:

• Clinical Informatics
• Natural Language Processing
• Machine Learning

Background:

• Depression affects 1 in 10 adults, incurring substantial personal, societal, and economic costs.
• Accurate identification of depression is crucial for effective patient management and resource allocation.

Purpose of the Study:

• To evaluate the effectiveness of the MTERMS Natural Language Processing (NLP) system and machine learning algorithms in identifying depression from clinical discharge summaries.
• To assess the performance of these methods in detecting cases not captured by traditional coded diagnosis lists.

Main Methods:

• Application of the MTERMS NLP system and various machine learning classification algorithms to a dataset of discharge summaries.
• Manual review and confidence-based classification (high, intermediate, low) of depression cases by domain experts.
• Performance evaluation using metrics such as F-measure, precision, and recall.

Main Results:

• For high-confidence depression cases, MTERMS achieved an F-measure of 89.6%, slightly outperforming machine learning classifiers.
• MTERMS demonstrated the highest F-measure (70.6%) for intermediate-confidence cases.
• The NLP approach successfully identified approximately 20% of depression cases missed by coded diagnosis lists.

Conclusions:

• NLP and machine learning methods, particularly MTERMS, are effective tools for identifying depression in clinical notes.
• These computational approaches can uncover cases of depression that might otherwise go undiagnosed through standard coding practices.
• The findings suggest potential for improved depression screening and diagnosis using automated text analysis of electronic health records.