A0746
Title: Joint modeling with integrated fractional Brownian motion
Authors: Seongho Song - University of Cincinnati (United States) [presenting]
Anushka Palipana - University of Cincinnati (United States)
Rhonda Szczesniak - Cincinnati Children Hospital Medical Center (United States)
Nishant Gupta - University of Cincinnati (United States)
Abstract: Biomarker data are often used to understand a disease's progression over time and characterize the relationship between biomarker data and the event outcome simultaneously using joint models. Motivated by being unable to effectively capture a biological process' variations using conventional random effects longitudinal sub-model, we propose a five-component longitudinal sub-model for a joint model. The most novel development is the scaled integrated fractional Brownian motion (IFBM) which has shown to reasonably depict biological processes measured with error. Other model components are the random intercept, fixed effects, and measurement error. Cox proportional hazards model serves as the event sub-model, which includes a time-dependent true longitudinal trajectory, and a set of baseline covariates. We use Markov chain Monte Carlo (MCMC) methods for Bayesian posterior computation and inference. We perform a simulation study and a comparative study of our joint model with IOU process from literature, and a joint model without a stochastic process. Our novel approach is then applied to the National Heart, Lung, and Blood Institute (NHLBI) lymphangioleiomyomatosis (LAM) registry data set and the Cystic Fibrosis (CF) data from 2 selected CF centers recorded in the US CF foundation patient registry (CFF-PR). We use forced expiratory volume in one second (FEV1) in liters and FEV1 pct-predicted as longitudinal biomarkers in LAM, and CF applications respectively.
