Repeated Measures Design with Generalized Linear Mixed Models for Randomized Controlled Trials is the first book focused on the application of generalized linear mixed models and its related models in the statistical design and analysis of repeated measures from randomized controlled trials. The author introduces a new repeated measures design called S:T design combined with mixed models as a practical and useful framework of parallel group RCT design because of easy handling of missing data and sample size reduction. The book emphasizes practical, rather than theoretical, aspects of statistical analyses and the interpretation of results. It includes chapters in which the author describes some old-fashioned analysis designs that have been in the literature and compares the results with those obtained from the corresponding mixed models.
The book will be of interest to biostatisticians, researchers, and graduate students in the medical and health sciences who are involved in clinical trials.
Author Website:Data sets and programs used in the book are available at http://www.medstat.jp/downloadrepeatedcrc.html
Table of Contents:
Table of Contents
Introduction
Repeated measures design
Generalized linear mixed models
Model for the treatment effect at each scheduled visit
Model for the average treatment effect
Model for the treatment by linear time interaction
Superiority and non-inferiority
Naive analysis of animal experiment data
Introduction
Analysis plan I
Analysis plan II
each time point
Analysis plan III - analysis of covariance at the last time point
Discussion
Analysis of variance models
Introduction
Analysis of variance model
Change from baseline
Split-plot design
Selecting a good _t covariance structure using SAS
Heterogeneous covariance
ANCOVA-type models
From ANOVA models to mixed-effects repeated measures models
Introduction
Shift to mixed-effects repeated measures models
ANCOVA-type mixed-effects models
Unbiased estimator for treatment effects
Illustration of the mixed-effects models
Introduction
The Growth data
Linear regression model
Random intercept model
Random intercept plus slope model
Analysis using
The Rat data
Random intercept
Random intercept plus slope
Random intercept plus slope model with slopes varying over time
Likelihood-based ignorable analysis for missing data
Introduction
Handling of missing data
Likelihood-based ignorable analysis
Sensitivity analysis
The Growth
The Rat data
MMRM vs. LOCF
Mixed-effects normal linear regression models
Example: The Beat the Blues data with 1:4 design
Checking missing data mechanism via a graphical procedure
Data format for analysis using SAS
Models for the treatment effect at each scheduled visit
Model I: Random intercept model
Model II: Random intercept plus slope model
Model III: Random intercept plus slope model with slopes varying over time Analysis using SAS
Models for the average treatment effect
Model IV: Random intercept model
Model V: Random intercept plus slope model
Analysis using SAS
Heteroscedastic models
Models for the treatment by linear time interaction
Model VI: Random intercept model
Model VII: Random intercept plus slope model Analysis using SAS
Checking the goodness-of-_t of linearity
ANCOVA-type models adjusting for baseline measurement
Model VIII: Random intercept model for the treatment effect at each visit
Model IX: Random intercept model for the average treatment effect
Analysis using SAS
Sample size
Sample size for the average treatment effect
Sample size assuming no missing data
Sample size allowing for missing data
Sample size for the treatment by linear time interaction
Discussion
Mixed-effects logistic regression models
The Respiratory data with 1:4 design
Odds ratio
Logistic regression models
Models for the treatment effect at each scheduled visit
Model I: Random intercept model
Model II: Random intercept plus slope model
Analysis using SAS
Models for the average treatment effect
Model IV: Random intercept model
Model V: Random intercept plus slope model
Analysis using SAS
Models for the treatment by linear time interaction
Model VI: Random intercept model
Model VII: Random intercept plus slope model
Analysis using SAS
Checking the goodness-of-_t of linearity
ANCOVA-type models adjusting for baseline measurement
Model VIII: Random intercept model for the treatment effect at each visit
Model IX: Random intercept model for the average treatment effect
Analysis using SAS
The daily symptom data with 7:7 design
Models for the average treatment effect
Analysis using SAS
Sample size
Sample size for the average treatment effect
Sample size for the treatment by linear time interaction
Mixed-effects Poisson regression models
The Epilepsy data with 1:4 design
Rate Ratio
Poisson regression models
Models for the treatment effect at each scheduled visit
Model I: Random intercept model
Model II: Random intercept plus slope model
Analysis using SAS
Models for the average treatment effect
Model IV: Random intercept model
Model V: Random intercept plus slope model
Analysis using SAS
Models for the treatment by linear time interaction
Model VI: Random intercept model
Model VII: Random intercept plus slope model
Analysis using SAS
Checking the goodness-of-_t of linearity
ANCOVA-type models adjusting for baseline measurement
Model VIII: Random intercept model for the treatment effect at each visit
Model IX: Random intercept model for the average treatment effect
Analysis using SAS
Sample size
Sample size for the average treatment effect
Sample size for Model
Sample size for Model V
Sample size for the treatment by linear time interaction
Bayesian approach to generalized linear mixed models
Introduction
Non-informative prior and credible interval
Markov Chain Monte Carlo methods
WinBUGS and OpenBUGS
Getting started
Bayesian model for the mixed-effects normal linear regression Model V
Bayesian model for the mixed-effects logistic regression Model IV
Bayesian model for the mixed-effects Poisson regression Model V
Latent pro_le models - classification of individual response pro_les
Latent pro_le models
Latent pro_le plus proportional odds model
Number of latent pro_les
Application to the Gritiron Data
Latent pro_le models
R, S-Plus and OpenBUGS programs
Latent pro_le plus proportional odds models
Comparison with the mixed-effects normal regression models
Application to the Beat the Blues Data
Applications to other trial designs
Trials for comparing multiple treatments
Three-arm non-inferiority trials including a placebo
Background
Hida-Tango procedure
Generalized linear mixed-effects models
Cluster randomized trials
Three-level models for the average treatment effect
Three-level models for the treatment by linear time interaction
Solutions to Exercises
About the Author :
Toshiro Tango is the Director of Center for Medical Statistics, Tokyo. His research interests include various aspects of biostatistics including design and analysis of clinical trials and spatial epidemiology. He has served as associate editor for several journals including Biometrics and Statistics in Medicine, and is the author of Statistical Methods for Disease Clustering.
Review :
"The main focus of this book is to introduce the generalized linear mixed-effects models with S:T repeated measures design, which provide a flexible and powerful tool to deal with longitudinal data with heterogeneity or variability among subject-specific responses and missing data. This book illustrates theoretical methodologies with a focus on the practicality with a wealth of real-life examples making it easy to understand the topics. It is well organized and contains SAS codes and outputs as useful references. In summary, this is an excellent book with a very good selection of examples. It is clearly written and is enjoyable to read."
~Misoo C. Ellison, Merck & Co., Inc., Kenilworth, NJ
