Computer-assisted expert case definition in electronic health records

Area of Science:

• Medical Informatics
• Clinical Data Analysis
• Machine Learning in Healthcare

Background:

• Electronic health records (EHRs) contain vast amounts of data.
• Developing precise case definitions for specific conditions within EHRs is challenging.
• Automating the extraction of clinical insights from EHRs is a key area of research.

Purpose of the Study:

• To describe a computer-assisted technique for experts to infer machine-implementable rules for case definitions in EHRs.
• To apply this technique to define acute liver dysfunction (ALD) in patients with inflammatory bowel disease (IBD).

Main Methods:

• An iterative process involving expert review and machine-learning rule refinement.
• Sampling case candidates, classifying them with machine rules and expert judgment.
• Updating rules to incorporate expert-identified distinctions and exceptions.

Main Results:

• The technique was applied to 29,336 adults with IBD using structured data and NLP-derived terms.
• Iterative refinement improved rule predictive value and sensitivity over three rounds.
• The final ALD case definition achieved 92% sensitivity and 79% positive predictive value in an independent test.

Conclusions:

• Machine-supported expert review can create transparent, data-driven, and improvable case definitions.
• The derived rules successfully defined ALD in IBD patients, outperforming previous methods.
• Expert updates to the rules provide valuable insights into clinical distinctions.