A rule-based electronic phenotyping algorithm for detecting clinically relevant cardiovascular disease cases

Area of Science:

• Health Informatics
• Epidemiology
• Cardiology

Background:

• Electronic Medical Records (EMR) are increasingly adopted, improving efficiency and data availability for research.
• While EMRs enhance data quantity, ensuring data quality for accurate event detection remains a challenge.
• Accurate identification of cardiovascular (CaVD) and cerebrovascular (CeVD) disease cases from EMR data is crucial for epidemiological studies.

Purpose of the Study:

• To assess the sensitivity, specificity, and agreement of a coded-term algorithm for detecting CaVD and CeVD cases.
• To evaluate the performance of EMR data in identifying clinically relevant cardiovascular and cerebrovascular events.
• To establish a reliable method for extracting CaVD and CeVD data from electronic health records.

Main Methods:

• A codes-based algorithm was developed using ICPC-2, ICD-10, and SNOMED-CT terms for CaVD and CeVD.
• The algorithm was applied to a randomly selected patient sample (40-79 years old) with at least one year of HMO seniority.
• Manual chart review served as the gold standard for performance validation, with reviewers assessing both coded and free-text EMR sections.

Main Results:

• The algorithm achieved high sensitivity (0.99) and acceptable specificity (0.86) in detecting combined CaVD and CeVD cases.
• Performance was evaluated against manual chart review, a robust validation method.
• A qualitative analysis of false positives and negatives was conducted to understand algorithm limitations.

Conclusions:

• A straightforward algorithm using coded terms in EMRs can effectively detect clinically relevant CaVD and CeVD events.
• Future improvements may involve integrating Natural Language Processing (NLP) for free-text analysis to enhance detection accuracy.
• This algorithm provides a valuable tool for epidemiological research using EMR data.