A0222
Title: Spatiotemporal multilevel joint modeling of generalized longitudinal and survival outcomes in end-stage kidney disease
Authors: Esra Kurum - University of California, Riverside (United States) [presenting]
Abstract: Individuals with end-stage kidney disease (ESKD) on dialysis experience high mortality and excessive burden of hospitalizations over time relative to comparable Medicare patient cohorts without kidney failure. A key interest in this population is to understand the time-dynamic effects of multilevel risk factors that contribute to the correlated outcomes of longitudinal hospitalization and mortality. Multilevel data from the United States Renal Data System (USRDS) is utilized, where repeated measurements/hospitalizations over time are nested in patients and patients are nested within (health service) regions across the U.S. A novel spatiotemporal multilevel joint model (STM-JM) is developed that accounts for the aforementioned hierarchical structure of the data while considering the spatiotemporal variations in both outcomes across regions. The proposed STM-JM includes time-varying effects of multilevel (patient- and region-level) risk factors on the correlated outcomes and incorporates spatial correlations across the spatial regions via a multivariate conditional autoregressive correlation structure. Efficient estimation and inference are performed via a Bayesian framework. An application of the proposed method to the USRDS data highlights significant time-varying effects of risk factors on hospitalization and mortality and identifies specific time periods on dialysis and spatial locations across the U.S. with elevated hospitalization and mortality risks.
