Digital Tools for Precision Nephrology

PROJECT SUMMARY/ABSTRACT This K08 application proposes to establish robust phenotyping pipelines for detection of early Collagen Type IV-Associated Nephropathy (COL4A-AN) using electronic health records (EHR). The study aims to overcome diagnostic challenges associated with the diverse manifestations of COL4A-AN, enabling timely, personalized

interventions crucial for delaying chronic kidney disease (CKD) progression. Utilizing EHR data presents an opportunity to support CKD subtype identification but faces hurdles such as data heterogeneity and semantic gaps across health systems. The proposed solution involves leveraging well-established, open-source, Natural

Language Processing (NLP) systems to convert unstructured clinical narratives into structured data, aiming to extract nuanced phenotypic descriptions for early COL4A-AN patient identification. Moreover, these experiments utilize the Unified Medical Language System (UMLS) and Observational Medical Outcomes

Partnership-Common Data Model (OMOP-CDM v5.4) to standardize language and guarantee data interoperability, enabling broader research application. The project endeavors to create precise, transferable EHR prediction models for diverse CKD subtypes, laying the groundwork for automated decision support tools, with a broad impact on COL4A-AN patients, those with

genetic CKD subtypes, and the precision nephrology field. Through two experiments, it aims to develop highly efficient early disease prediction models for COL4A-AN by mining unstructured text and leveraging nuanced phenotype data from clinicians' narratives, distinct from ICD code-based models. The primary goal of this study

is to establish scalable phenotyping pipelines to identify patients with genetic CKD subtypes, using NLP and standardized data models to enhance prediction accuracy and automate decision-support tools. Standardizing the early COL4A-AN prediction models will facilitate research transparency and foster collaboration across

multiple institutions. This project facilitates the creation of EHR-embedded decision support tools. Moreover, it lays the groundwork for a future R01 clinical trial application that aims to investigate the diagnostic yield among patients highly likely to have COL4A-AN, who will subsequently be referred for genetic testing. These efforts

set the stage for future studies examining how early diagnosis influences the progression of CKD. The comprehensive training plan is specifically designed to provide Dr. Jordan Nestor with NLP expertise for extracting nuanced phenotypes from unstructured EHR narratives, creating EHR-based prediction models, and

aiding her transition into an independent NIH-funded investigator. Dr. Nestor's ultimate objective is to advance patient care and enhance CKD outcomes through the direct implementation of precision nephrology at the point of care.