No comments about Epidemiology in Health Care (3rd Edition).

2 comments about The Use of Restricted Significance Tests in Clinical Trials (Statistics for Biology and Health).

1. David Salsburg is a statistician with a good knowledge of the foundations of statistics as well as being a practical statistician with a great deal of clinical trials experience. In Chapter 1 he discusses randomized controlled clinical trials. Chapters 2-5 deal with foundational issues in probability and statistics. It leads to the conclusion that hypothesis testing is appropriate for clinical trials and statistical power against alternatives is an important concept for choosing appropriate test statistics but (1)for most practical problems uniformly most powerful tests do not exist and (2)powerful tests can only be found if one is able to meaningfully restrict the class of alternative hypotheses being considered.

   In Chapter 6, Salsburg makes the case for restricted tests by providing a real clinical example. Special methods are then covered in the remaining chapters. Chapter 8 deals with resampling approaches including permutation methods and the bootstrap. In Chapter 10, Neyman's often neglected theory of restricted chi-squared tests, is presented.

   This is a well written and unusual book that covers methodology not seen in very many biostatistics books. However, these techniques are very relevant to the clinical trials commonly conducted at pharmaceutical companies. It is an important reference source for biostatisticians.

   Those interested in statistical theory and its foundational issues will find clear and concise coverage in the first 5 chapters. However if you want more, take a look at Salsburg's new book "The Lady Tasting Tea" which just came out in 2001. In that book he raises all the same issues and more in the context of discussing the great statisticians of the 20th Century.




2. David Salsburg is a statistician with a good knowledge of the foundations of statistics as well as being a practical statistician with a great deal of clinical trials experience. In Chapter 1 he discusses randomized controlled clinical trials. Chapters 2-5 deal with foundational issues in probability and statistics. It leads to the conclusion that hypothesis testing is appropriate for clinical trials and statistical power against alternatives is an important concept for choosing appropriate test statistics but (1)for most practical problems uniformly most powerful tests do not exist and (2)powerful tests can only be found if one is able to meaningfully restrict the class of alternative hypotheses being considered.
   In Chapter 6, Salsburg makes the case for restricted tests by providing a real clinical example. Special methods are then covered in the remaining chapters. Chapter 8 deals with resampling approaches including permutation methods and the bootstrap. In Chapter 10, Neyman's often neglected theory of restricted chi-squared tests, is presented.
   
   This is a well written and unusual book that covers methodology not seen in very many biostatistics books. However, these techniques are very relevant to the clinical trials commonly conducted at pharmaceutical companies. It is an important reference source for biostatisticians.
   
   Those interested in statistical theory and its foundational issues will find clear and concise coverage in the first 5 chapters. However if you want more, take a look at Salsburg's new book "The Lady Tasting Tea" which just came out in 2001. In that book he raises all the same issues and more in the context of discussing the great statisticians of the 20th Century.

2 comments about Statistical Monitoring of Clinical Trials: A Unified Approach (Statistics for Biology and Health).

1. This is one of two excellent books on group sequential methods and adaptive designs. All three authors are ASA Fellows. Wittes and Proschan have worked at the NIH and Proschan formerly worked at the FDA. Gordan Lan has published widely on group sequential methods and has developed software with David deMets that can be downloaded for free from deMets' website at the University of Wisconsin. Lan and deMets developed the theory of alpha spending functions that are commonly used in software such a EaSt (Cytel Corporation) to help determine an appropriate shape to the stopping boundary. Two group sequential methods with markedly different spending functions are the Pocock design anf the O'Brien-Fleming design. I have written a detailed book review for Technometrics, that also compares the book to Jennison and Turnbull's text. Both of these books will be classics. my review will appear in the May 2007 issue of Technometrics.



2. The topics listed in the table of contents of the book are really interesting. On the other hand, the notations used in the book make it so much more difficult to read. For example, the letters chosen for many variables used have no relation at all with their meaning, so that constantly, throughout the book, you need to go back to previous chapters. Also, various statistical results are just used, without any proof or at least hints for proof. Therefore, one needs to read this book while having another graduate stats inference book at hand.

1 comments about Randomization Tests, Fourth Edition (Statistics: a Series of Textbooks and Monographs).

1. From Back Cover:
   
   "'Randomization Tests' is an experimenter's guide for the valid application of statistical tests to data form nonrandom samples. An excellent textbook on applied statistics as well, 'Randomization Tests' surpasses other textbooks in the field by demonstrating a general procedure for determining significance that can be used instead of the conventional parametric and nonparametric statistical tables. In addition, it supplements the other textbooks by showing experimenters how to take tests described elsewhere and extend their scope of application by freeing them from their parametric assumptions.
   
   This book provides all the necessary theory and practical guidelines, such as instructions for writing computer programs, to permit experimenters to transform any statistical test into a distribution-free test; draw valid statistical inferences about experimental treatment effects in the absence of random sampling; perform valid statistical tests on data from only one subject; and develop new statistical tests that are custom-made to the experimenter's research needs.
   
   Experimenters in all fields where statistical tests are commonly used, especially in the fields of psychology, education, biology, and medicine, will find that this book can play a vital role in experimental design and data analysis."

3 comments about Epidemiologic Research: Principles and Quantitative Methods (Industrial Health & Safety).

1. Most epidemiology texts I have seen do not show a proof of the commonly stated issue that misclassification bias always produces a bias toward the null. This text provides such a proof and does so eloquently



2. This is a fine textbook, but I am compelled to respond to the first review. Nondifferential misclassification unfortunately does not always produce a bias towards the null. See more recent texts for details.



3. Once upon a time this book may have been useful, but now this text is more than 20 years old and badly in need of a revision. You would be better off to read a more up-to-date work instead, such as the books by Rothman and Greenland, Clayton and Hills, or MacMahon and Trichopolos.

No comments about Managerial Epidemiology: Practice, Methods and Concepts.

No comments about Analyzing Medical Data Using S-PLUS (Statistics for Biology and Health).

1 comments about Applied Mixed Models in Medicine (Statistics in Practice).

1. One of my collegues recommended this book to me. It surprised me that it has a very broader coverage of topics in medicine where mixed model is appropriate. This book talked many pratical issues such as "why the mixed model failed to converge" etc. The mathematical background is not demanding.

   Another advantage of this book is that there are so many examples in the book.

1 comments about Clinical Trials in Oncology, Second Edition.

1. I should begin by admitting that I had the opportunity to review this little masterpiece in manuscript. Good then, it's even better now.

   It's good because it informs the reader, in sober prose, how to determine what works and what doesn't in medical practice, and what's safe and what isn't. It's good because it reveals what can go wrong when anecdotes ("it worked for me!") substitute for sound research as the basis for clinical practice. And it's good because it shows how serious are the consequences of even subtle failures to observe protocols in designing and carrying out clinical trials.

   It is reassuring to read of the care and precautions advocated for government-sponsored research; it is accordingly unsettling to contemplate the pressure that commercial interests (drug companies, for-profit hospitals, equipment manufacturers) might bring on researchers to cut a few corners.

   After reading "Clinical Trials" I came to appreciate that case studies, longitudinal studies, and retrospective questionnaires, so frequently hyped in the press and on television, are no substitute for actual well-designed and well-executed experiments. Because you and I are different, certainly genetically and probably in other essential ways, what helps you may well harm me. Only the proper application of statistics in designing clinical trials and in analyzing data from them can distinguish what's generally valuable from what's useless (however plausible and authoritatively touted it may be). Although the authors had the good taste to reject the aphorism, usually attributed to a nameless statistician, that "if experimentation be the queen of science, then statistics stands as the guardian of the royal virtue", its pithiness may give the reader the crucial insight into why alternative modes of research are untrustworthy.

   Some readers may feel disheartened to learn the truth that many, probably most, promising therapies prove, when adequately tested, worthless, and some may feel in some fuzzy way that to accept this reality is cruelly to deny hope to those who need it badly. On the contrary, this book makes it clear that to offer false hope is the ultimate cruelty, for without experimentation there can be no knowledge, and without knowledge there can be no real hope.

   Notwithstanding the slightly technical nature of this book (yes, there IS a chapter with mathematics), I recommend it highly for the general reader who is interested in such topics as personal health care, alternative medicine, managed care cost containment, and the like. Buy a copy for yourself, and, if you feel philanthropic, you might consider donating a copy to your health care provider. The world would be better if doctors' waiting rooms (like hotel rooms with their Gideon Bibles) all had a copy of "Clinical Trials in Oncology" available for patients' perusal.

No comments about Statistical Monitoring of Clinical Trials: Fundamentals for Investigators.