Causality patterns and machine learning for the extraction of problem-action relations in discharge summaries

Area of Science:

• Natural Language Processing
• Clinical Informatics
• Medical Record Analysis

Background:

• Clinical narrative texts contain vital patient medical history, including problem progression and treatments.
• A chronological view of patient history aids clinical audits and improves care quality.

Purpose of the Study:

• To develop a method for extracting Problem-Action relations from clinical narrative text.
• To present a chronological view of patient problems and physician actions.

Main Methods:

• Defined clinical semantic units as problem and/or action relations.
• Extracted clinical events using conditional random fields and external knowledge.
• Classified semantic units into Problem-Action relations using support vector machines and event causality patterns.

Main Results:

• Achieved 78.8% performance in F1-measure on Korean discharge summaries.
• Demonstrated effective classification of clinical Problem-Action relations.

Conclusions:

• The proposed method successfully extracts and classifies clinical Problem-Action relations.
• This facilitates the creation of chronological patient medical histories for better clinical insights.