Preface to the first edition

This book is primarily concerned with clinical trials planned and conducted within the pharmaceutical industry. Much of the methodology presented is in fact applicable on a broader basis and can be used in observational studies and in clinical trials outside of the pharmaceutical sector; nonetheless, the primary context is clinical trials and pharmaceuticals. The development is aimed at non-statisticians and will be suitable for physicians, investigators, clinical research scientists, medical writers, regulatory personnel, statistical programmers, senior data managers and those working in quality assurance. Statisticians moving from other areas of application outside of pharmaceuticals may also find the book useful in that it places the methods that they are familiar with, in context in their new environment. There is substantial coverage of regulatory aspects of drug registration that impact on statistical issues. Those of us working within the pharmaceutical industry recognise the importance of being familiar with the rules and regulations that govern our activities, and statistics is a key aspect of this.

The aim of the book is not to turn non-statisticians into statisticians. I do not want you to go away from this book and 'do' statistics. It is the job of the statistician to provide statistical input to the development plan, to individual protocols, to write the statistical analysis plan, to analyse the data and to work with medical writing in producing the clinical report, and also to support the company in its interactions with regulators on statistical issues.

The aims of the book are really threefold. Firstly, to aid communication between statisticians and non-statisticians; secondly, to help in the critical review of reports and publications; and finally, to enable the more effective use of statistical arguments within the regulatory process. We will take each of these points in turn.

In many situations, the interaction between a statistician and a non-statistician is not a particularly successful one. The statistician uses terms such as power, odds ratio, p-value, full analysis set, hazard ratio, non-inferiority, type II error, geometric mean, last observation carried forward and so on, of which the non-statistician has a vague understanding, but maybe not a good enough understanding to be able to get an awful lot out of such interactions. Of course, it is always the job of a statistician to educate and every opportunity should be taken for imparting knowledge about statistics, but in a specific context, there may not be time for that. Hopefully this book will explain, in ways that are understandable, just what these terms mean and provide some insight into their interpretation and the context in which they are used. There is also a lot of confusion between what on the surface appear to be the same or similar things: significance level and p-value, equivalence and non-inferiority, odds ratio and relative risk, relative risk and hazard ratio (by the way this is a minefield!) and meta-analysis and pooling to name just a few. This book will clarify these important distinctions.

It is unfortunately the case that many publications, including some in leading journals, contain mistakes with regard to statistics. Things have improved over the years with the standardisation of the ways in which publications are put together and reviewed. For example, the CONSORT statement (see Section 16.5 [this is Section 21.5 in the 2nd edition]) has led to a distinct improvement in the quality of reporting. Nonetheless mistakes do slip through, in terms of poor design, incorrect analysis, incomplete reporting and inappropriate interpretation - hopefully not all at once! It is important therefore when reading an article that the non-statistical reader is able to make a judgement regarding the quality of the statistics and to notice any obvious flaws that may undermine the conclusions that have been drawn. Ideally, the non-statistician should involve their statistical colleagues in evaluating their concerns, but keeping a keen eye on statistical arguments within the publication may help to alert the non-statistician to a potential problem. The same applies to presentations at conferences, posters, advertising materials and so on.

Finally, the basis of many concerns raised by regulators, when they are reviewing a proposed development plan or assessing an application for regulatory approval, is statistical. It is important that non-statisticians are able to work with their statistical colleagues in correcting mistakes, changing aspects of the design, responding to questions about the data to hopefully overcome those concerns.

In writing this book, I have made the assumption that the reader is familiar with the general aspects of the drug development process. I have assumed knowledge of the phase I to phase IV framework, of placebos, control groups, and double-dummy together with other fundamental elements of the nuts and bolts of clinical trials. I have assumed however no knowledge of statistics! This may or may not be the correct assumption in individual cases, but it is the common denominator that we must start from, and also it is actually not a bad thing to refresh on the basics. The book starts with some basic issues in trial design in Chapter 1, and I guess most people picking up this book will be familiar with many of the topics covered there. But don't be tempted to skip this chapter; there are still certain issues, raised in this first chapter, that will be new and important for understanding arguments put forward in subsequent chapters. Chapter 2 looks at sampling and inferential statistics. In this chapter, we look at the interplay between the population and the sample, basic thoughts on measuring average and variability and then explore the process of sampling leading to the concept of the standard error as a way of capturing precision/reliability of the sampling process. The construction and interpretation of confidence intervals are covered in Chapter 3 together with testing hypotheses and the (dreaded!) p-value. Common statistical tests for various data types are developed in Chapter 4 which also covers different ways of measuring treatment effect for binary data, such as the odds ratio and relative risk.

Many clinical trials that we conduct are multi-centre and Chapter 5 looks at how we extend our simple statistical comparisons to this more complex structure. These ideas lead naturally to the topics in Chapter 6 which include the concepts of adjusted analyses, and more generally, analysis of covariance which allows adjustment for many baseline factors, not just centre. Chapters 2-6 follow a logical development sequence in which the basic building blocks are initially put in place and then used to deal with more and more complex data structures. Chapter 7 moves a little away from this development path and covers the important topic of 'intention-to-treat' and aspects of conforming with that principle through the definition of different analysis sets and dealing with missing data. In Chapter 8, we cover the very important design topics of power and the sample size calculation which then leads naturally to a discussion about the distinction between statistical significance and clinical importance in Chapter 9.

The regulatory authorities, in my experience, tend to dig their heels in on certain issues and one such issue is multiplicity. This topic, which has many facets, is discussed in detail in Chapter 10. Non-parametric and related methods are covered in Chapter 11. In Chapter 12, we develop the concepts behind the establishment of equivalence and non-inferiority. This is an area where many mistakes are made in applications, and in many cases, these slip through into published articles. It is a source of great concern to many statisticians that there is widespread misunderstanding of how to deal with equivalence and non-inferiority. I hope that this chapter helps to develop a better understanding of the methods and the issues. If you have survived so far, then Chapter 13 covers the analysis of survival data. When an endpoint is time to some event, for example, death, the data are inevitably subject to what we call censoring and it is this aspect of so-called survival data that has led to the development of a completely separate set of statistical methods. Chapter 14 builds on the earlier discussion on multiplicity to cover one particular manifestation of that, the interim analysis. This chapter also looks at the management of these interim looks at the data through data monitoring committees. Meta-analysis and its role in clinical development is covered in Chapter 15, and the book finishes with a general Chapter 16 on the role of statistics and statisticians in terms of the various aspects of design and analysis and statistical thinking more generally.

It should be clear from the last few paragraphs that the book is organised in a logical way; it is a book for learning rather than a reference book for dipping into. The development in later chapters will build on the development in earlier chapters. I strongly recommend, therefore, that you start on page 1 and work through. I have tried to keep the discussion away from formal mathematics. There are formulas in the book but I have only included these where I think this will enhance understanding; there are no formulas for formulas...