OBJECTIVES:

Numerous studies have confirmed an increased risk of Chronic Kidney Disease (CKD) progression after an incidence of Hospital-Acquired Acute Kidney Injury (HA-AKI). To date, no methods exist evaluating a patient’s progression from new onset AKI to CKD. Therefore, the objective of this study was to develop a datamining tool to identify key determining temporal and clinical phenotypes for kidney disease progression.

METHODS:

We conducted a retrospective study using electronic health record (EHR) data from TriNetX in patients (≥ 18 years of age) with CKD in West Virginia. The study samples were divided into 3 cohorts: AKI within 90 days of hospitalization, random AKI, and no AKI identified. The binary features were extracted temporally (i.e., years 1-3 prior CKD diagnosis) from the demographics, diagnosis, procedures, medications, vitals, and laboratory tables. A Latent class analysis (LCA) random forest-based machine learning approach was used to identify and categorize the phenotypes within each cohort.

RESULTS:

Among 75,033 CKD patients, (28.72%) experienced AKI prior to CKD, (17.79%) had AKI after CKD, and (52.35%) never experienced AKI. Cohorts 1-3 contains 7,442 patients (10%), 6,408 patients (8.5%), and 39,280 patients (52%), respectively. Temporal LCA analysis generated five distinct patient phenotype profiles within each cohort. When comparing cohort 3 (C3) to cohort 1 (C1) and cohort 2 (C2), a high prevalence of hypertensive disorders (C1-92%, C2-94.4%, C3-79.4%), type 2 diabetes (C1-59.4%, C2-62.9%, C3-43.5.%), disorders of lipid disorders (C1-78.4%, C2-83.4%, C3-71.1.4%), long-term drug therapy (C1-86.1%, C2-91.5%, C3-62.1%), and chronic obstructive pulmonary disease (C1-43.3%, C2-48.3%, C3-23.4%) were observed.

CONCLUSIONS:

Our results support phenotypes vary occur among different cohorts and across time. Thus, our proposed AI tool has the potential to identify and forecast distinct kidney disease trajectories to better allocate healthcare resources leading to improved clinical outcomes.