Leveraging undecided cases in chart-reviewed phenotypes to enhance EHR-based association studies.

Area of Science:

• Biomedical Informatics
• Clinical Epidemiology
• Health Data Science

Background:

• Electronic health record (EHR) phenotyping algorithms offer efficient patient classification but suffer from misclassification errors.
• Manual chart review by clinicians is the gold standard but is labor-intensive and limited to small patient subsets.
• Indeterminate phenotypes in manual reviews can introduce a third category, complicating analysis.

Purpose of the Study:

• To develop a method for integrating algorithm-derived binary phenotypes and clinician-reviewed trinary phenotypes from EHR data.
• To enhance the accuracy and efficiency of association studies using combined phenotyping data.
• To address challenges in rare disease research with a cost-effective sampling strategy.

Main Methods:

• Proposed an augmented estimation method combining algorithm-derived phenotypes (entire cohort) with chart-reviewed phenotypes (subset).
• Implemented a cost-effective, outcome-dependent sampling strategy for rare disease scenarios.
• Evaluated the TriCA (trinary chart-reviewed phenotype integrated cost-effective augmented estimation) method in simulations and real-world EHR data.

Main Results:

• The augmented method improved mean square error by up to 28.3% compared to random sampling in simulations.
• Efficiency gains of up to 33.3% (ADRD data) and 50.8% (SBCE data) were observed compared to using only chart-reviewed phenotypes.
• Demonstrated improved statistical efficiency and unbiased estimates in both simulation and real-world applications.

Conclusions:

• The proposed TriCA method effectively integrates algorithm and manual chart review data for EHR-based association studies.
• This approach enhances statistical efficiency and provides unbiased estimates, outperforming existing methods.
• Applicable to a broad range of studies, improving risk factor identification in EHR data.