Structure Discovery for Joint Models Through Variable Selection Approaches

• Activity Number: 628
• Type: Invited
• Date/Time: Thursday, August 7, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Zangdong He and Wanzhu Tu and Zhangsheng Yu*+
• Companies: Indiana University Fairbanks School of Public Health and Indiana University School of Medicine and Indiana University School of Medicine

Abstract

To assess the linear and potentially nonlinear effects of covariates in the joint model setting, we introduce a linear mixed effect model for the longitudinal component and general frailty model for the survival component with nonparametric additive covariate functions. To perform data-driven method for selecting linear and non-linear component, we decompose the B-spline estimate for nonparametric function into linear and non-linear portions orthogonally and they used penalized likelihood method (Adaptive LASS or SCAD) to select linear and non-linear coefficients respectively. The tuning parameters for penalty terms are selected using a BIC approach. This approach allows the functional form (no effect, linear, or non-linear effect) of each covariates to be determined by the observed data. Simulation studies show excellent performance in determining the covariate functional form with minimal bias. We apply the proposed method to analyze a pain study dataset.