A phenotyping algorithm for classification of single ventricle physiology using electronic health records

Area of Science:

• Cardiology
• Medical Informatics
• Computational Biology

Background:

• Congenital heart disease (CHD) patients with single ventricle physiology (SVP) exhibit diverse characteristics, complicating accurate cohort classification.
• Electronic health record (EHR) data offers a rich resource for patient phenotyping, but requires robust algorithms for precise identification.

Purpose of the Study:

• To develop and validate a high-performance phenotyping algorithm for identifying SVP patients within EHR data.
• To compare the developed algorithm's accuracy against a previously published SVP identification method.

Main Methods:

• Utilized International Classification of Diseases (ICD-9/10) codes, enhanced with clinical notes, imaging reports, and expert domain knowledge.
• Developed and tested the algorithm on distinct cohorts of CHD patients (1020 for development, 2500 for testing, 22,500 for validation).
• Evaluated algorithm performance using accuracy, precision, sensitivity, and F1 score.

Main Results:

• The developed algorithm achieved 99.24% accuracy, 89.39% F1 score in the testing cohort, significantly outperforming a published method (95.20% accuracy, 58.04% F1 score).
• In the validation cohort, the algorithm demonstrated 93.82% precision compared to 43.00% for the published method.
• The algorithm effectively identified false positives and missed cases by integrating clinical notes and ICD codes.

Conclusions:

• The automated phenotyping algorithm, combined with physician input, surpasses existing methods for SVP classification in EHR data.
• This algorithm has the potential to enhance research and clinical management of SVP patients by enabling automated cohort development for prognostication and monitoring.