Identifying and Characterising a Chronic Kidney Disease Electronic-Phenotype Using Electronic Health Record-Derived Data: A Narrative Review of Strategies and Applications.

Abstract

Chronic kidney disease (CKD) represents a significant and growing healthcare burden. As CKD is defined and staged using laboratory values, it can be readily identified and characterised via data points derived from the electronic health record (EHR). This narrative literature review describes various strategies that have been employed to develop such a CKD 'e-phenotype,' evaluating accuracy, fidelity, and practicality. Methods discussed include the use of International Classification of Diseases (ICD) codes, estimated glomerular filtration rate (eGFR) and proteinuria criteria, free-text analysis and natural language processing (NLP), and machine learning techniques. Considerable variability in algorithm performance and complexity exists, with the use of eGFR and proteinuria criteria likely constituting the most practical and reliable basis for a CKD e-phenotype. In addition, promising current and future applications of the CKD e-phenotype have been outlined, such as characterising the burden of CKD complications and comorbid disease, and use as a tool to encourage optimisation of CKD management with quality, guideline-directed care. Future directions and challenges may involve integration of risk stratification and clinical decision support systems, alongside applications across public health resourcing and clinical trial recruitment.