Phenotyping in distributed data networks: selecting the right codes for the right patients

Area of Science:

• Health Informatics
• Observational Data Analysis
• Pharmacovigilance

Background:

• Observational data is crucial for drug surveillance, treatment pathway investigation, and patient outcome prediction.
• Developing accurate phenotype algorithms for large, heterogeneous health data networks presents significant challenges.
• Existing methods struggle with patient representation differences and data heterogeneity.

Purpose of the Study:

• To present a novel process for creating comprehensive concept sets for phenotype algorithms.
• To introduce the PHenotype Observed Entity Baseline Endorsements (PHOEBE) recommender system.
• To enhance the identification and early capture of patient cohorts in observational studies.

Main Methods:

• Utilized code utilization data from 22 electronic health record and claims datasets across 6 countries.
• Mapped datasets to the Observational Health Data Sciences and Informatics (OHDSI) Common Data Model.
• Developed PHOEBE to recommend semantically and lexically similar codes, integrated with Cohort Diagnostics.

Main Results:

• PHOEBE successfully recommends relevant codes, facilitating the creation of comprehensive concept sets.
• The system is now integrated into major OHDSI network studies.
• Using PHOEBE for cohort creation identified more patients and captured them earlier in disease progression.

Conclusions:

• PHOEBE significantly improves the efficiency and comprehensiveness of phenotype algorithm development.
• The system enhances patient cohort identification in large-scale observational health data studies.
• PHOEBE contributes to more effective drug surveillance and patient outcome research using real-world data.