Statistical Monitoring of Clinical Trials

The approach taken in this book is, to studies monitored over time, what the Central Limit Theorem is to studies with only one analysis. Just as the Central Limit Theorem shows that test statistics involving very different types of clinical trial outcomes are asymptotically normal, this book shows that the joint distribution of the test statistics at different analysis times is asymptotically multivariate normal with the correlation structure of Brownian motion ("the B-value") – irrespective of the test statistic. Thus, this book offers statisticians an accessible, incremental approach to understanding Brownian motion as related to clinical trials. The main body of the chapters is accessible to anyone with knowledge of a standard mathematical statistics text. More mathematically advanced readers will find rigorous developments in appendices at the end of chapters. Reading the book will develop insight into not only monitoring, but power, survival analysis, safety, and other statistical issues germane to clinical trials.

The approach taken in this book is, to studies monitored over time, what the Central Limit Theorem is to studies with only one analysis. Just as the Central Limit Theorem shows that test statistics involving very different types of clinical trial outcomes are asymptotically normal, this book shows that the joint distribution of the test statistics at different analysis times is asymptotically multivariate normal with the correlation structure of Brownian motion (``the B-value") irrespective of the test statistic. The so-called B-value approach to monitoring allows us to use, for different types of trials, the same boundaries and the same simple formula for computing conditional power. Although Brownian motion may sound complicated, the authors make the approach easy by starting with a simple example and building on it, one piece at a time, ultimately showing that Brownian motion works for many different types of clinical trials.

The book will be very valuable to statisticians involved in clinical trials. The main body of the chapters is accessible to anyone with knowledge of a standard mathematical statistics text. More mathematically advanced readers will find rigorous developments in appendices at the end of chapters. Reading the book will develop insight into not only monitoring, but power, survival analysis, safety, and other statistical issues germane to clinical trials.

Michael Proschan, Gordon Lan, and Janet Wittes are elected Fellows of the American Statistical Association. All have spent formative years in the Biostatistics Research Branch of the National Heart, Lung, and Blood Institute (NHLBI/NIH). While there, they were intimately involved in the design and statistical monitoring of large-scale randomized clinical trials, developing methodology to aid in their monitoring. For example, Lan developed, with DeMets, the now widely-used spending function approach to group sequential designs, whose properties were furtherinvestigated by Proschan. The B-value approach used in the book was introduced in a very influential paper by Lan and Wittes. The statistical theory behind conditional power was developed by Lan, along with Simon and Halperin, and was the cornerstone for the conditional error approach to adaptive clinical trials introduced by Proschan and Hunsberger. All three authors have expertise in adaptive methodology for clinical trials.

Michael Proschan is a Mathematical Statistician at the National Institutes of Health; Gordon Lan is Senior Director of Biometrics at Johnson & Johnson Pharmaceutical Research & Development, L.L.C.; Janet Wittes is President of Statistics Collaborative, a statistical consulting company she founded in 1990.