Title: Joint modeling for longitudinal data and binary outcome via h-likelihood
Authors: Toshihiro Misumi - Yokohama City University (Japan) [presenting]
Abstract: Joint modeling techniques of longitudinal covariate and binary outcome have attracted considerable attention in medical research areas. Joint models provide a powerful tool to explore how strongly associated is a longitudinal trajectory of biomarker with an event of interest. The strategy for estimating joint models is to define a joint likelihood based on two sub-models with shared random effects, i.e. linear random effects models for the longitudinal sub-model, and logistic models with random effects for the binary sub-model. A numerical integration, however, is required in the estimation algorithm for the joint likelihood, and a computational problem arises as the assumed sub-models become more complex. In order to overcome the issue, we propose a joint modeling procedure by using a h-likelihood to avoid the numerical integration in the estimation algorithm. The estimating procedure is expected to reduce the computational cost. We conduct some Monte Carlo simulation studies to examine the effectiveness of our proposed modeling procedure, and then apply the method to the analysis of the real data.