Construction of the Design Matrix for Generalized Linear Mixed-Effects Models in the Context of Clinical Trials of Treatment Sequences

Abstract

The problem of constructing a design matrix of full rank for generalized linear mixed-effects models (GLMMs) has not been addressed in statistical literature in the context of clinical trials of treatment sequences. Solving this problem is important because the most popular estimation methods for GLMMs assume a design matrix of full rank, and GLMMs are useful tools in statistical practice. We propose new developments in GLMMs that address this problem. We present a new model for the design and analysis of clinical trials of treatment sequences, which utilizes some special sequences called skip sequences. We present a theorem showing that estimators computed through quasi-likelihood, maximum likelihood or generalized least squares, or through robust approaches, exist only if appropriate skip sequences are used. We prove theorems that establish methods for implementing skip sequences in practice. In particular, one of these theorems computes the necessary skip sequences explicitly. Our new approach allows building design matrices of full rank and facilitates the implementation of regression models in the experimental design and data analysis of clinical trials of treatment sequences. We also explain why the standard approach to constructing dummy variables is inappropriate in studies of treatment sequences. The methods are illustrated with a data analysis of the STAR*D study of sequences of treatments for depression.