Preface

Clinical trials have long been one of the most important tools in the arsenal of clinicians and scientists who help develop pharmaceuticals, biologics, and medical devices. It is reported that nearly 10,000 clinical studies are conducted every year around the world. One can find many excellent books that address fundamental statistical and general scientific principles underlying the design and analysis of clinical trials, such as those by Pocock (1983), Fleiss (1986), Meinert (1986), Friedman, Furberg and DeMets (1996), Piantadosi (1997) and Senn (1997). Numerous references can be found in these fine books.

The aim of this book is unique in that we focus in great detail on a set of selected and practical problems facing statisticians and biomedical scientists conducting clinical research. We discuss solutions to these problems based on modern statistical methods and review computer-intensive techniques that help clinical researchers efficiently and rapidly implement these methods in the powerful SAS environment.

It is a challenge to select the few topics that are most important and relevant to the design and analysis of clinical trials. Our choice of topics for this book was guided by the International Conference on Harmonization (ICH) guideline for the pharmaceutical industry entitled “Structure and Content of Clinical Study Reports” (this document is commonly referred to as ICH E3). The document states that

Important features of the analysis, including the particular methods used, adjustments made for demographic or baseline measurements or concomitant therapy, handling of dropouts and missing data, adjustments for multiple comparisons, special analyses of multicenter studies, and adjustments for interim analyses, should be discussed [in the study report].

Following the ICH recommendations, we decided to focus in this book on the analysis of stratified data, incomplete data, multiple inferences, and issues arising in safety and efficacy monitoring. We also address other statistical problems that are very important in a clinical trial setting, such as reference intervals for safety and diagnostic measurements.

One special feature of the book is the inclusion of numerous SAS macros to help readers implement the new methodology in the SAS environment. The availability of the programs and the detailed discussion of the output from the macros help make the application of new procedures a reality. The authors are planning to make the SAS macros compatible with new SAS products such as SAS Enterprise Guide. Enterprise Guide tasks that implement the statistical methods discussed in the book will be published on the Business Intelligence SAS Users Group Web site at http://www.biopharmnet.com/bisug/.

The book is aimed at clinical statisticians and other scientists who are involved in the design and analysis of clinical trials conducted by the pharmaceutical industry, academic institutions, or governmental institutions such as the National Institutes of Health (NIH). Graduate students specializing in biostatistics will also find the material in this book useful because of its applied nature.

Because the book is written for practitioners, it concentrates primarily on solutions rather than theory. Although most of the chapters include some tutorial material, this book was not intended to provide a comprehensive coverage of the selected topics. Nevertheless, each chapter gives a high-level description of the theoretical aspects of the statistical problem at hand and includes references to publications that contain more advanced material. In addition, each chapter gives a detailed overview of the underlying statistical principles.

There are some exceptions to the presentation of minimum theory in the book. For example, Chapter 5 discusses the analysis of incomplete data and covers comparatively complex statistical concepts such as multiple imputation. Although the theoretical part is written at a higher statistical level, examples and applications are prepared in such a way that they can be easily understood.

Examples from real trials are used throughout the book to illustrate the concepts being discussed and to help the reader understand their relevance in a clinical trial setting. Most of the data come from real clinical trials. In several cases, because of confidentiality concerns, we relied on simulated data that are representative of real clinical trial data. Although simulated data might lack authenticity, using them does afford us the opportunity to see how close to the truth we can get using the proposed methodology. In this regard, we echo Hastie and Tibshirani’s (1990, page 239) statement that “an advantage of using simulated examples is that you know the truth.”

Chapter 1: Analysis of Stratified Data. This chapter discusses the analysis of clinical outcomes in the presence of influential covariates. It reviews stratified analyses of continuous, categorical, and time-to-event endpoints. The chapter also introduces statistical methods for studying treatment-by-stratum interactions.

Chapter 2: Multiple Comparisons and Multiple Endpoints. This chapter reviews statistical strategies for handling multiplicity issues arising in clinical trials. It covers basic single-step multiple tests as well as more advanced closed, fixed-sequence and resampling-based multiple testing procedures. The chapter also describes strategies for handling multiple objectives and dose-finding studies.

Chapter 3: Analysis of Safety and Diagnostic Data. This chapter describes methods for constructing reference intervals for safety and diagnostic measures with various clinical trials applications, such as computation of reference limits based on sample quantiles and tolerance limits. The second part of the chapter reviews statistical methods for the analysis of shift tables produced after reference intervals have been applied to safety and diagnostic data.

Chapter 4: Interim Data Monitoring. The first part of this chapter reviews the popular approaches for designing and monitoring group sequential trials, such as repeated significance tests and the error spending function approach. The second part introduces stochastic curtailment methods, including frequentist, mixed Bayesian-frequentist, and fully Bayesian methods.

Chapter 5: Analysis of Incomplete Data. This chapter discusses basic and more advanced statistical methods for the analysis of incomplete longitudinal data. These methods include complete case analysis, last observation carried forward, likelihood-based methods, and multiple imputation.

We would like to thank the following individuals (listed alphabetically) for their valuable comments:

Caroline Beunckens (Limburgs Universitair Centrum)

Prof. Raymond Carroll (Texas A&M University)

Dr. Brenda Gaydos (Lilly)

Dr. Jeffrey Helterbrand (Genentech)

Dr. Joseph Heyse (Merck)

Prof. Jason Hsu (Ohio State University)

Ivy Jansen (Limburgs Universitair Centrum)

Prof. Michael Kenward (London School of Hygiene and Tropical Medicine)

Prof. Richard Kryscio (University of Kentucky)

Dr. Gordon Lan (Aventis)

Dr. Ilya Lipkovich (Lilly)

Dr. Nuwan Nanayakkara (Amylin)

Dr. Gerhardt Pohl (Lilly)

Dr. Andreas Sashegyi (Lilly)

Prof. Stephen Senn (Glasgow University)

Dr. Herbert Thijs (Limburgs Universitair Centrum)

Prof. Geert Verbeke (Katholieke Universiteit Leuven)

Prof. Peter Westfall (Texas Tech University)

Dr. Ilker Yalcin (Lilly)

We are grateful to Grace Ying Li (Lilly) who has carefully tested most of the SAS programs in this book and helped us optimize some of the complicated SAS macros.

Alex Dmitrienko and Walt Offen would like to sincerely thank Lilly Global Statistical Science senior management, in particular Drs. Carlos Alatorre, Todd Sanger and Mike Wilson, for their enthusiastic and continual support of this endeavor.

Geert Molenberghs would like to thank his colleagues for their support.

Christy Chuang-Stein wishes to thank Drs. Mohan Beltangady and Debra Gmerek for their support and encouragement.

We also thank Donna Faircloth, our editor at SAS Books by Users Press, for her support and assistance in preparing this book.