High-throughput phenotyping with electronic medical record data using a common semi-supervised approach (PheCAP).

Area of Science:

• Biomedical Informatics
• Computational Biology
• Genetics

Background:

• Phenotypes are crucial for clinical and genetic studies of disease risk.
• Biobanks linked to electronic medical record (EMR) data necessitate efficient phenotyping methods.
• Current EMR phenotyping faces challenges with data accuracy and manual input.

Purpose of the Study:

• To develop a high-throughput, semi-supervised phenotyping pipeline called PheCAP.
• To address challenges in EMR data accuracy and manual labor in phenotyping.
• To create robust approaches for phenotyping millions of patients.

Main Methods:

• PheCAP integrates structured EMR data and natural language processing (NLP) from clinical notes.
• The pipeline employs standardized, automated procedures to minimize manual input.
• Machine learning is utilized for training the phenotyping algorithm.

Main Results:

• PheCAP can be executed in 1-2 days, with chart review being the main time constraint (approx. 2 weeks).
• The pipeline generates a phenotype algorithm, patient phenotype probabilities, and classifications.
• PheCAP offers a semi-supervised, high-throughput solution for phenotyping.

Conclusions:

• PheCAP provides an efficient and accurate method for large-scale phenotyping using EMR data.
• The pipeline reduces manual effort and improves the robustness of phenotype identification.
• This approach facilitates genetic and clinical studies of disease risk and outcomes.