Scalable and interpretable alternative to chart review for phenotype evaluation using standardized structured data...

Area of Science:

• Health Informatics
• Clinical Research Methodology
• Data Science in Healthcare

Background:

• Manual chart review is the established standard for phenotype evaluation but is not scalable for large electronic health record (EHR) and claims data research.
• Observational research requires efficient and interpretable methods for phenotype evaluation at scale.

Purpose of the Study:

• To evaluate the efficacy of structured data for efficient and interpretable phenotype evaluation as an alternative to manual chart review.
• To assess the performance of a novel tool, Knowledge-Enhanced Electronic Profile Review (KEEPER), in phenotype evaluation.

Main Methods:

• Developed KEEPER to extract and standardize patient-specific structured data elements based on clinical reasoning principles.
• Compared KEEPER's performance (interrater agreement, accuracy, time) against manual chart review for four distinct clinical conditions using a crossover design.

Main Results:

• KEEPER achieved case ascertainment twice as fast as manual chart review.
• Concordant patient classification between KEEPER and manual review was 88.1%, with higher clinician agreement (91.2%) using KEEPER versus manual review (76.3%).
• Discrepancies were primarily due to missing data and interpretation differences.

Conclusions:

• Structured data, when curated and organized by clinical reasoning, enables efficient and interpretable phenotype evaluation.
• KEEPER demonstrates non-inferior performance to manual chart review, achieving significant time savings and improved clinician agreement.