A Multivariate Single Index Model for Longitudinal Data with Application in Clinical Investigation

• Activity Number: 368
• Type: Contributed
• Date/Time: Tuesday, August 6, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Jingwei Wu*+ and Wanzhu Tu
• Companies: Indiana University, School of Medicine, Department of Biostatistics and Indiana University School of Medicine
• Keywords: index measure ; multivariate outcomes ; semiparametric modeling ; single index model ; p-spline ; mixed effect model

Abstract

Use of index measures for quantifying risks associated with study subjects is common in medical research. For example, body mass index (BMI), as a measure of adiposity, has been shown to correlate well with the risk of multiple disease developments. Yet, construction of these indices is often based on heuristic arguments. In this research, we propose a new approach for deriving index measures by extending the traditional single index model to a multivariate setting, so that the resulting indices would work for multiple outcomes. Herein, the model is developed for longitudinal data. We use penalized cubic spline to characterize the index components, while leaving the other subject characteristics as additive components. P-splines are estimated by penalizing nonlinear least squares and the estimation is implemented using R software. To illustrate, we use the approach to construct a new pediatric body mass index using height and weight, that correlates with both systolic and diastolic blood pressure. We assess the performance of the method through a simulation study, and showed that the new index outperformed the standard BMI.