Adaptive design methods in clinical trials. Shein-Chung Chow and Mark Chang, Chapman & Hall/CRC,...

2
BOOK REVIEWS 4611 set from the logistic regression trick and the Cox model trick for fitting the model. They attribute this to a difference between maximum likelihood and partial likelihood—an explanation that cannot possibly be correct because both approaches should lead to the same conditional likelihood. In fact, the Cox model trick is using the wrong set of match- ing variables, and when this is corrected the results are identical. A good feature of this book is the presentation of actual code and output. This is valuable in show- ing the user where the useful numbers lurk in the masses of printout. Equally importantly, it should also guarantee that the code really runs and gives the answers shown. There is one important caveat to this: the R code usually does not include the library() calls needed to load any necessary packages, and the text does not always say which packages are needed. The code is included on a CD in the back of the book (rather than on a web site), and unfortunately is collated into a single document in Microsoft Word format. The data sets are all constructed in-line in the code rather than read from files, which is convenient for presentation purposes but is not how students should be encour- aged to manage their data. Perhaps for this reason most of the data sets are small and several are ‘hy- pothetical’. There are few or no competing books that at- tempt such a broad coverage of advanced applied statistics for epidemiology. For most purposes, however, it would be more useful to have a book that focuses in much more detail on a subset of the methodology and/or a subset of the computing. THOMAS LUMLEY Department of Biostatistics School of Public Health and Community Medicine University of Washington Seattle, U.S.A. (DOI: 10.1002/sim.3331) 2. ADAPTIVE DESIGN METHODS IN CLINICAL TRIALS. Shein-Chung Chow and Mark Chang, Chapman & Hall/CRC, Boca Raton, FL, 2007. No. of pages: 277. Price: $89.95. ISBN10 1-58488- 776-1, ISBN13 978-1-58488-776-8 Adaptive clinical trials are currently a hot topic in pharmaceutical research, and a large number of articles have been published in journals during re- cent years. This book targeting the topic is there- fore highly needed, and was published at an oppor- tune time. The authors are to be commended for the endeavour to provide a monograph about this rapidly evolving area in—to our knowledge—the first book specifically devoted to adaptive design methods. Adaptive design is a big and heterogeneous re- search field. As a consequence, chapters on quite differing areas are provided. The book starts with protocol amendments, adaptive randomization, and adaptive hypotheses, where the latter focuses on the choice of non-inferiority margin for the infer- ence in non-inferiority studies. Chapters for special adaptive design follow: dose-escalation trials, group sequential designs, sample size adjustment, seam- less phase II/III designs, and treatment switch- ing. Additional chapters are devoted to a Bayesian approach, clinical trial simulation, and case studies. Finally, a comprehensive and very valuable bibli- ography about relevant literature is included. The choice of topics included in the book conveys an impression that the authors had core content around adaptive design and then added other interesting topics. For example, issues around protocol amendments, non-inferiority mar- gins, clinical trial simulations, treatment switching based on individual responses are unexpected in this book with an intended focus on adaptive de- sign, but they are certainly of value to the clinical statistician. On the other hand, the book does not cover the area of adaptive dose finding in Phase II. This is remarkable since we think that this is an area with large potential for adaptive designs, as evident from recent publications, see, for example [1, 2]. As the authors mention, it is very important in applications to focus not only on significance testing but also on estimates and confidence intervals. The book, however, does not describe methods for this in detail. By having it as the first chapter, the book highlights the role of protocol amendments. This choice is notable since changes due to protocol amendments are not even included in most def- initions of adaptive designs. In adaptive designs changes are typically done based on an analysis of accumulated data. Protocol amendments, on the other hand, are typically ad hoc changes due to facts overseen in the planning stage, or unfore- seeable events occurring during the study. The main focus in this chapter is on a change in target Copyright 2008 John Wiley & Sons, Ltd. Statist. Med. 2008; 27:4610–4613

Transcript of Adaptive design methods in clinical trials. Shein-Chung Chow and Mark Chang, Chapman & Hall/CRC,...

BOOK REVIEWS 4611

set from the logistic regression trick and the Coxmodel trick for fitting the model. They attributethis to a difference between maximum likelihoodand partial likelihood—an explanation that cannotpossibly be correct because both approaches shouldlead to the same conditional likelihood. In fact, theCox model trick is using the wrong set of match-ing variables, and when this is corrected the resultsare identical.

A good feature of this book is the presentationof actual code and output. This is valuable in show-ing the user where the useful numbers lurk in themasses of printout. Equally importantly, it shouldalso guarantee that the code really runs and givesthe answers shown. There is one important caveatto this: the R code usually does not include thelibrary() calls needed to load any necessarypackages, and the text does not always say whichpackages are needed. The code is included on aCD in the back of the book (rather than on a website), and unfortunately is collated into a single

document in Microsoft Word format. The data setsare all constructed in-line in the code rather thanread from files, which is convenient for presentationpurposes but is not how students should be encour-aged to manage their data. Perhaps for this reasonmost of the data sets are small and several are ‘hy-pothetical’.

There are few or no competing books that at-tempt such a broad coverage of advanced appliedstatistics for epidemiology. For most purposes,however, it would be more useful to have a bookthat focuses in much more detail on a subset of themethodology and/or a subset of the computing.

THOMAS LUMLEYDepartment of Biostatistics

School of Public Health and Community MedicineUniversity of Washington

Seattle, U.S.A.

(DOI: 10.1002/sim.3331)

2. ADAPTIVE DESIGN METHODS IN CLINICALTRIALS. Shein-Chung Chow and Mark Chang,Chapman & Hall/CRC, Boca Raton, FL, 2007. No.of pages: 277. Price: $89.95. ISBN10 1-58488-776-1, ISBN13 978-1-58488-776-8

Adaptive clinical trials are currently a hot topicin pharmaceutical research, and a large number ofarticles have been published in journals during re-cent years. This book targeting the topic is there-fore highly needed, and was published at an oppor-tune time. The authors are to be commended forthe endeavour to provide a monograph about thisrapidly evolving area in—to our knowledge—thefirst book specifically devoted to adaptive designmethods.

Adaptive design is a big and heterogeneous re-search field. As a consequence, chapters on quitediffering areas are provided. The book starts withprotocol amendments, adaptive randomization, andadaptive hypotheses, where the latter focuses onthe choice of non-inferiority margin for the infer-ence in non-inferiority studies. Chapters for specialadaptive design follow: dose-escalation trials, groupsequential designs, sample size adjustment, seam-less phase II/III designs, and treatment switch-ing. Additional chapters are devoted to a Bayesianapproach, clinical trial simulation, and case studies.Finally, a comprehensive and very valuable bibli-ography about relevant literature is included.

The choice of topics included in the bookconveys an impression that the authors had corecontent around adaptive design and then addedother interesting topics. For example, issuesaround protocol amendments, non-inferiority mar-gins, clinical trial simulations, treatment switchingbased on individual responses are unexpected inthis book with an intended focus on adaptive de-sign, but they are certainly of value to the clinicalstatistician. On the other hand, the book doesnot cover the area of adaptive dose finding inPhase II. This is remarkable since we think thatthis is an area with large potential for adaptivedesigns, as evident from recent publications, see,for example [1, 2]. As the authors mention, it isvery important in applications to focus not onlyon significance testing but also on estimates andconfidence intervals. The book, however, does notdescribe methods for this in detail.

By having it as the first chapter, the bookhighlights the role of protocol amendments. Thischoice is notable since changes due to protocolamendments are not even included in most def-initions of adaptive designs. In adaptive designschanges are typically done based on an analysisof accumulated data. Protocol amendments, on theother hand, are typically ad hoc changes due tofacts overseen in the planning stage, or unfore-seeable events occurring during the study. Themain focus in this chapter is on a change in target

Copyright q 2008 John Wiley & Sons, Ltd. Statist. Med. 2008; 27:4610–4613

4612 BOOK REVIEWS

population due to modified inclusion/exclusioncriteria.

The book often provides a listing of variousmethods, but a critical discussion of the methodsand a comparison between them are either veryshort or totally missing. An example is groupsequential methods, where the Wang and Tsi-atis boundary is introduced but the reader is notadvised about which � to choose in the boundaryfamily when she/he wants to apply it. One wouldexpect to get some guidance as to the appropriatechoices of � for various situations.

In the chapter about sample size adjustment,the authors provide an overview of the area andintroduce several methods when sample size isadjusted based on an interim observed treatmentdifference. A useful further step would be to com-pare the different methods and to discuss rela-tionships between, for example, the inverse-normalmethod and Cui-Hung-Wang’s method.

Although the book ends with a chapter on casestudies, we think that the small examples usedthroughout the book to illustrate the introducedmethods are more valuable to the reader.

It is our general impression that many impor-tant problems and issues with adaptive design arenot sufficiently dealt with. There is a lot of text,including quite a lot of repetition, but an actualaccount of the problem is not given. This can beillustrated by a couple of issues related to seamlessphase II/III designs. In this instance, if the time toachieve data for relevant endpoints is long in rela-tion to recruitment time, the benefits of an adaptivedesign may vanish. Similarly, there may be no gainin time to market if a required second phase IIItrial triggers the new drug application. Such issuesthat may entirely eliminate the case for an adaptivedesign deserve thorough discussion, but in thisbook they are merely mentioned in passing in afew sentences. The motivation for seamless designsgiven in the book is not convincing, but we agree

with the authors that these designs can often beefficient.

The book contains a lot of typographical mis-takes, lacks consistent notation, and is missingsome definitions. There are also duplications,apparently by mistake, such as a section aboutthe combination of independent p-values, whichis included both in Chapters 6 and 7. This mightbe a didactical trick but is more likely just a mis-take. We hope that an effort will be made to docorrections in the next edition.

The authors claim to cover ‘all of the statisticalissues that may occur at various stages of adaptivedesign and analysis of clinical data’. It is almostinevitable that such an ambitious claim cannot bemet. Replacing the claim with a more realistic goalfor the book, we think that it provides a valuablecontribution to the area of adaptive design.

REFERENCES

1. Bornkamp B, Bretz F, Dmitrienko A, Enas G,Gaydos B, Hsu CH, Konig F, Krams M, Liu Q,Neuenschwander B, Parke T, Pinheiro J, Roy A,Sax R, Shen F. Innovative approaches for designingand analyzing adaptive dose-ranging trials. Journalof Biopharmaceutical Statistics 2007; 17:965–995.

2. Gaydos B, Krams M, Perevozskaya I, Bretz F,Liu Q, Gallo P, Berry D, Chuang-Stein C, Pinheiro J,Bedding A. Adaptive dose–response studies. DrugInformation Journal 2006; 40:451–461.

FRANK MILLER AND STIG JOHAN WIKLUNDClinical Information Science, Biostatistics

AstraZenecaSodertalje

Sweden

(DOI: 10.1002/sim.3297)

3. USING AND UNDERSTANDING MEDICALSTATISTICS (4th edn). David E. Matthews andVernon T. Farewell, Karger, Basel, 2007. No. ofpages: XX+322. Price: CHF 49.00, EUR 35.00,$44.00. ISBN: 978-3-8055-8189-9

It is a pleasure to read this book about statis-tics aimed at medical researchers, with the mainpurpose to provide them sufficient understanding

to enable them to read statistical methods anddiscussions of papers published in medical jour-nals. The authors are great scholars, and this bookdemonstrates that they are great teachers as well.The 23 chapters are all well accessible for medi-cal researchers: the chapters are compact, they arepresented in logical order, and do not contain toomuch technical detail. Moreover, each chapter con-tains many medical examples that are clearly ex-plained and analyzed using the statistical technique

Copyright q 2008 John Wiley & Sons, Ltd. Statist. Med. 2008; 27:4610–4613