High-throughput multimodal automated phenotyping (MAP) with application to PheWAS.

Area of Science:

• Biomedical Informatics
• Translational Research
• Computational Biology

Background:

• Accurate and efficient patient phenotyping is crucial for translational studies but remains a significant bottleneck.
• Electronic health records (EHR) linked with biorepositories offer a powerful platform for research.
• Current phenotyping methods often rely on International Classification of Diseases (ICD) codes, which may lack granularity.

Purpose of the Study:

• To develop an automated, high-throughput phenotyping method.
• To integrate International Classification of Diseases (ICD) codes and narrative data extracted using natural language processing (NLP).
• To improve the accuracy and efficiency of patient phenotyping for large-scale studies.

Main Methods:

• Developed a mapping method using the Unified Medical Language System to identify relevant ICD and NLP concepts.
• Employed latent mixture models to jointly analyze aggregated ICD and NLP counts alongside healthcare utilization.
• Created the Multimodal Automated Phenotyping (MAP) algorithm to predict phenotype probability and classify patients.

Main Results:

• The MAP algorithm demonstrated superior or comparable performance (AUC, F-score) to ICD codes alone across 16 phenotypes.
• Automated feature extraction achieved accuracy comparable to manual curation (AUCMAP 0.943 vs. AUCmanual 0.941).
• Phenome-wide association studies (PheWAS) using MAP showed higher power in detecting validated associations compared to ICD-based methods.

Conclusions:

• The MAP approach significantly enhances phenotype definition accuracy and scalability.
• This automated method facilitates large-scale phenotyping essential for studies like PheWAS.
• MAP represents a significant advancement in leveraging EHR and NLP for translational research.