Clinical Trials: A Practical Guide to Design, Analysis, and Reporting
This is a general introduction to clinical trials co-edited by Duolao Wang, a lecturer in statistics at the London School
of Hygiene, and Ameet Bakhai, a cardiologist at Barnet General & Royal Free Hospitals in London. On the whole, I have a poor
opinion of edited books, which often have patchy coverage and an uneven style. This book succeeds against the odds.
I speculate that the first reason is that many of the contributors—despite the fact that many institutions are represented—come
from southeast England, whether from the local London medical schools or GlaxoSmithKline in Harlow. This has no doubt made
it easier for the editors to keep in contact with their contributing authors. The second reason is that at least one of the
editors is a co-author on every chapter.
Whatever the explanation, the result is an easy-to-read text with good coverage of topics and fairly uniform presentation.
I think that it would be fair to describe the book as "broad but not deep," in contrast to the John Matthews book1 I reviewed in ACT in 2000,2 which might be described as "profound but narrow." (Wang and Bakhai's work is also targeted at the nonstatistician, whereas
Matthews' intended readership was statisticians.) However, as Stuart Pocok points out in the preface, "The book contains over
300 references, facilitating a more in-depth pursuit of each topic if desired." And in fact some of the chapters do point
to deeper issues.
The book is split into 38 chapters divided into five major sections: Fundamentals of Trial Design, Alternative Trial Designs,
Basics of Statistical Analysis, Special Trial Issues in Data Analysis, and Reporting of Trials.
Inevitably in a book with such extensive coverage there are some things with which I disagree. For example, chapter 18 ("Significance
Test and Confidence Intervals") gives the following baffling example:
"Suppose that it is necessary to measure the average systolic blood pressure (SBP) level of all males aged 16 years in the
UK in 2005...we can conduct a survey (or sample) of 500 males within the population..." (p. 186)
This example is not only irrelevant (a health survey used to illustrate statistical inference for clinical trials), it is
likely to mislead. Clinical trials are not about representative inference but about comparative inference, being experiments.
This error is compounded by what follows in chapter 19 ("Comparison of Means"). In order to illustrate various statistical
tests, a trial in 24 patients with chronic airways limitation is considered and the following three questions are posed:
- At the start and end of the study are the mean FEV1 values seen in both treatment groups as expected from the overall population
of similar subjects?
- Is there evidence of a significant change in mean FEV1 from baseline in either treatment group?
- Is the posttreatment mean FEV1 or mean FEV1 change within the active treatment group significantly different from that in
the control group?
Of these three questions, the first is illogical. If the values are different the patients are not similar. This may seem
like a mere quibble, but once one thinks about it carefully one realizes that it points to a misunderstanding about clinical
Question two is pointless. What useful inference could one draw about the effect of treatment from the answers? (Unfortunately
many trialists try nonetheless to do so.)
Question three is the one that matters. To be fair to the authors, we do find on page 207:
"In a comparative clinical trial, the primary objective is to evaluate the efficacy and safety of a drug compared to a
And the authors do proceed to answer their question.