Automating case definitions using literature-based reasoning

Area of Science:

• Medical Informatics
• Computational Epidemiology
• Natural Language Processing

Background:

• Establishing Case Definitions (CDefs) is crucial for epidemiological and clinical activities.
• Current CDef application involves manual steps, leading to inefficiencies in surveillance and research.

Purpose of the Study:

• To describe the need for automating the representation of medical condition CDefs.
• To propose an approach for automated CDef generation and application.

Main Methods:

• Translated an existing anaphylaxis CDef using NLM MetaMap for synonym identification.
• Generated a new CDef from PubMed abstracts using text mining and semantic network analysis.
• Classified 6034 H1N1 reports for anaphylaxis using generated and translated CDefs with similarity approaches.

Main Results:

• Overall classification performance was similar across CDef generation methods.
• The generated CDef with vector space model and cosine similarity achieved the highest accuracy (0.825 ± 0.003).
• The semi-automated approach with vector space model and cosine similarity yielded the highest recall (0.809 ± 0.042), though precision was low across all methods.

Conclusions:

• Automating the representation of CDefs is complex but offers significant efficiency gains.
• Automated CDefs can enhance safety and clinical surveillance effectiveness.