Effect of Smoothing in a Generalized Linear Mixed Model Context on Estimation of Covariance Structures for Clustered or Longitudinal Data

• Activity Number: 550
• Type: Contributed
• Date/Time: Wednesday, August 7, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Muhammad Mullah*+ and Andrea Benedetti
• Companies: McGill University and McGill University
• Keywords: Smoothing ; Generalized Linear Mixed Models ; Clustered Data ; Fractional Polynomials ; Regression Splines ; Variance Components

Abstract

Generalized linear mixed models (GLMM), primarily used for analysing correlated data, can also be used for smoothing. In the present study, an attempt has been made to see the effect of smoothing in the parameter estimates when dealing with correlated (clustered or longitudinal) data using the GLMM to take care of both smoothing and correlation simultaneously. We compare smoothing in a GLMM via restricting the knots in a regression spline, to simpler methods for estimating the nonlinear dose-response association such as fractional polynomials and parametric nonlinear function (e.g., quadratic) in data with correlated responses. Through a number of simulations, the effect of smoothing on variance parameters (that describe the correlation between clustered observations) has been evaluated by calculating bias, variance and mean squared error of the estimates. These methods then have been applied to a practical data set with the goal of modelling the association between CD4+ cell count and time since seroconversion for 369 HIV infected men enrolled in the Multicenter AIDS Cohort Study.