Extracting Deep Phenotypes for Chronic Kidney Disease Using Electronic Health Records

Area of Science:

• Nephrology
• Biostatistics
• Health Informatics

Background:

• Chronic kidney disease (CKD) is a prevalent global health issue with variable progression rates.
• Phenotypic subtyping of CKD is essential for improved risk stratification and tailored patient treatments.
• Electronic health records (EHR) offer a rich data source for identifying CKD subtypes.

Purpose of the Study:

• To leverage EHR data for identifying meaningful phenotypic subtypes of CKD.
• To extract disease severity profiles for CKD while controlling for confounding factors.
• To analyze the clinical relevance of discovered CKD subtypes.

Main Methods:

• Utilized a probabilistic model to identify precise phenotypes from CKD patient EHR data.
• Decomposed estimated glomerular filtration rate (eGFR) trajectories into disease subtype, covariate, and individual long-term/short-term effects.
• Analyzed clinical relevance of identified disease subtypes.

Main Results:

• Discovered distinct CKD disease subtypes using the probabilistic subtyping model.
• Presented clinical health markers associated with identified CKD subtypes.
• Demonstrated the utility of large EHR datasets for retrospective deep phenotype identification.

Conclusions:

• The identified CKD subtypes and associated markers can inform risk prediction.
• The probabilistic model provides a framework for deep phenotyping in CKD.
• Further research can expand the model for enhanced CKD management.