Multistate Modeling Patient Transitions from AKI to CKD or Death Using Electronic Health Records (EHR)

Background: Patients with AKI are at high risk for CKD and death. The epidemiology of these transitions is poorly understood. This study aims to characterize patient health trajectories from the initial AKI episode to CKD or all-cause mortality using EHR data from NewYork-Presbyterian/Columbia University. Methods: This retrospective study included 20,699 patients. Clinical states were identified by clustering patient vectors derived from temporal medical codes and serum creatinine (SCr) time series using natural language processing models. Transition probabilities between clinical states and to outcomes (CKD or death) were estimated using multi-state modeling. Results: In the AKI cohort, 17% developed CKD and 19% died. We identified 15 clinical states with varying disease burdens. Each state had unique transition probabilities over 5 years (Figure). Risk factors for each outcome were identified within different AKI subpopulations based on clinical state trajectories (Table). Conclusion: This study enhances our understanding of patient trajectories from an initial AKI episode to CKD diagnosis or death by tracking the progression of medical conditions, interventions, treatments, and SCr levels over time. Funding: Other NIH Support - The project was supported by grants from the NIH (grants R01LM012895, R01HG013031, and UL1TR001873). Funding was also provided by the ASN KidneyCure-Harold Amos Medical Faculty Development Program (ASN-AMFDP) Grant (JGN)., Private Foundation SupportFigure: Transition Probabilities Between Clinical States Over a 5-Year PeriodTable: Risk Factors for CKD and All-Cause Mortality for Each Initial State Transition