1、Applied Bioinformatics and Biostatistics in Cancer ResearchSeries editors:Jeanne Kowalski,Steven PiantadosiFor further volumes:http:/ Tang Xin TuEditorsModern Clinical TrialAnalysis123EditorsWan TangDepartment of Biostatisticsand Computational BiologySchool of Medicine and DentistryUniversity of Roc
2、hesterElmwood Ave.601Rochester,New York,USAXin TuDepartment of Biostatisticsand Computational BiologySchool of Medicine and DentistryUniversity of RochesterElmwood Ave.601Rochester,New York,USAISBN 978-1-4614-4321-6ISBN 978-1-4614-4322-3(eBook)DOI 10.1007/978-1-4614-4322-3Springer New York Heidelber
3、g Dordrecht LondonLibrary of Congress Control Number:2012943945 Springer Science+Business Media New York 2013This work is subject to copyright.All rights are reserved by the Publisher,whether the whole or part ofthe material is concerned,specifically the rights of translation,reprinting,reuse of ill
4、ustrations,recitation,broadcasting,reproduction on microfilms or in any other physical way,and transmission or informationstorage and retrieval,electronic adaptation,computer software,or by similar or dissimilar methodologynow known or hereafter developed.Exempted from this legal reservation are bri
5、ef excerpts in connectionwith reviews or scholarly analysis or material supplied specifically for the purpose of being enteredand executed on a computer system,for exclusive use by the purchaser of the work.Duplication ofthis publication or parts thereof is permitted only under the provisions of the
6、 Copyright Law of thePublishers location,in itscurrent version,and permission for use mustalways be obtained from Springer.Permissions for use may be obtained through RightsLink at the Copyright Clearance Center.Violationsare liable to prosecution under the respective Copyright Law.Theuseofgeneral d
7、escriptive names,registered names,trademarks,service marks,etc.in this publicationdoes not imply,even in the absence of a specific statement,that such names are exempt from the relevantprotective laws and regulations and therefore free for general use.While the advice and information in this book ar
8、e believed to be true and accurate at the date ofpublication,neither the authors nor the editors nor the publisher can accept any legal responsibility forany errors or omissions that may be made.The publisher makes no warranty,express or implied,withrespect to the material contained herein.Printed o
9、n acid-free paperSpringer is part of Springer Science+Business Media()PrefaceWith advances in medical sciences,cancer research no longer focuses primarilyon effective treatments,but rather a host of other issues such as tools for earlydiagnosis,cost of treatmentand long-termcare,and qualityof life.T
10、hus,in additionto traditional survivalanalyses for comparingtreatment differences,modernclinicaltrials in cancer research are also designed to address these new emerging issues.This volume covers classic as well as cutting-edge topics on the analysis of clinicaltrial data in biomedical and psychosoc
11、ial research and discusses each topic inan expository and user-friendly fashion.The intent of the book is to provide anoverview of the primary statistical and data analytic issues associated with each ofthe selected topics,followed by a discussion of approaches for tackling such issuesand available
12、software packages for carrying out the analyses.Some of the topics covered are quite standard,such as survival data analysisand longitudinal data.Although in-depth discussions of such classic topics canbe found in various topic-specific texts,our coverage highlights their specific andimportantroles
13、in clinical trials.Further,by presentingthe topics in a self-containedfashion,the materials become more assessable by researchers in other disciplines,particularly clinicians and medical investigators who seek a“crash course”tounderstand the“nuts and bolts”of clinical trials.With new medical discove
14、ries,advances in technology and treatment delivery,rises in health-care cost,and emphasis on quality of life and patient-centeredcare,clinical trials have become increasingly complex in design to address allthese concerns.For example,in most randomized controlled trials,patients arerandomizedto a pa
15、rticulartreatment at baseline,regardlessof whetherthe treatmentis optimized for the patient.Although it is a necessary step to study the efficacy ofa treatment,this traditional approach does not fully reflect how patients are treatedin practice,when multiple treatments are generally used either sequ
16、entially or incombinationto effectivelytreat a patient.In fact,dynamicallyadjusting treatmentinaccordancewith the patientresponseto thepreviouslyassignedtreatmentis the onlyviable option for treating many mental disorders.Some of the chapters are devotedto addressing such cutting-edgeissues to refle
17、ct advances in statistical methodologyfor clinical trials.vviPrefaceAs a series to provide an overview of the core concepts of clinical trials anda guide to statistical methods for analyzing data from such studies,the chaptersare organized in an order following the logical considerations of the issu
18、es arisingfrom the design to the execution of clinical trials.Thus,we start with two chaptersfocusing on the classic topics of survival and longitudinal data analysis.Treatmentevaluations in clinical trials generally center on two types of outcome.If the studyinvolvespatientswith terminalillnesses s
19、uchas advancedcancers,thedurationfroma certain time point such as the initiation or termination of the treatment is often ofprimaryinterest.Survivalanalysis modelsare uniquelysuited to comparingsurvivaltimes between different treatment groups.In most clinical trials,treatment differences are evaluat
20、ed by comparing changesin an outcome of interest over time such as tumor volume in cancer studiesbetween different treatment groups.Longitudinal models are applied to facilitatecomparisons of such temporal changes.These models extend classic methodsfor cross-sectional data to address the within-subj
21、ect correlations in the repeatedassessments of the individual and missing data due primarily to premature dropoutby study subjects.Chapter 2 discusses these distinctive features of longitudinal dataand associated models.Valid inference not only replies on correct statistical models,but on quality an
22、dfidelity of outcomes of clinical trials as well.Although clinical trials typically usemeasures with established fidelity and reliability,it is important to have some levelof understanding of the inner workings of the process to develop and validate suchmeasures,especially for outcomes derived from
23、measures of latent constructs suchas quality of life.Chapter 3 provides an overview of such measurement error issuesand methods to address them.Upon settling down on the measures of treatment effect,the next step is to decideon the length of the study.In particular,we may want to know whether we can
24、expedite the trial as soon as evidence of treatment efficacy emerges,especiallyfor treatments with adverse reactions and side effects and studies with seriousdeterioration of health and fatality outcomes.Chapter 4 discusses dynamic decisionrules to stop a trial as soon as there is indication of trea
25、tment difference.As notedearlier,the standard protocolin clinical trials is to randomizepatients atbaseline,which,althougha necessarystepto studytheefficacyofa singletreatment,does not reflect real clinical practice.Also,with new discoveries on genetic linkagein disease predisposition and treatment
26、response and recent emphasis on patient-centered outcome research,this standard treatment protocol does not meet theneeds of the new patient-specific treatment and care model.Chapter 5 focuses onthis new person-centered treatment approach by dynamically adjusting treatmentin accordance with the pati
27、ent response to the previously assigned treatment.Thedynamic treatment regime,which continuously adjusts treatment type and dosage,is particularly effective for effectiveness research,because of the diverse range ofpatients conditions and disease progression in such studies.As the cost for research
28、in developing and delivering new treatment becomesincreasingly high,health care has become quite expensive,especially in the USA.In recent years,more and more clinical trials have included a costeffectivenessPrefaceviicomponentto also examinethe cost forthe new intervention.Thecosteffectivenessanaly
29、sis allows one to see if the added benefit is worth the increased cost and howto maximize such benefit-to-cost margins for a population of interest.Chapter 6 isdevoted to addressing these issues.In most clinical trials,we are interested in establishing the superiority of thenew intervention over exi
30、sting or conventional treatment.But,in some cases,we may be interested in equivalence between a new and conventional treatment,such as in replacing costly name brand medications with less expensive genericalternatives,and a simplified instrument for diagnosis of disease.More importantly,such equival
31、ence tests are employed in early drug development to assess thepotential of drug induced,prolonged duration of ventricular depolarization andsubsequentrepolarization,orQTinterval,as thedurationis derivedfromtheintervalof ECG tracing from the beginning of Q wave to the end of T wave.For somedrugs,sig
32、nificant prolongation of the absolute QT interval has been associatedwith the precipitation of a potentially fatal cardiac arrhythmia and can degenerateinto ventricular fibrillation,leading to sudden cardiac death.To ensure drug safety,thorough QT(TQT)trials are recommended by FDA drug regulatory re
33、quirementsto assess the treatment response of a new drug and ensure that it does not induceprolonged QT intervals.Thus,the primary object of the TQT is to demonstrateequivalence,rather than superiority as in most clinical trials.In Chap.7,we first discuss the fundamental issues arising from the para
34、digmshift from superiority to equivalence and methods for addressing them under thisalternative inference paradigm for equivalence.We then turn our attention to thedesignandanalysis ofTQTtrials byapplyingthemodelsforequivalenceinChap.8.Randomized controlled clinical trials rely on randomization,the
35、hallmark ofmodern clinical research,to deliver valid conclusions regarding treatment differ-ences.In some studies,it may not be possible to conduct such a trial.For example,it is clearlyunethicalto contemplatea randomizedcontrolledtrial to study the effectof smoking on lung cancer by randomizing sub
36、jects to a smoking group.In someother studies,decision to treat may also depend on the health condition of a subject,in which case treatment assignment is no longer random and treatment differencescannot be evaluated as in randomized trials.Chapter 9 discusses issues in assessingtreatment difference
37、 in such nonrandomizedtrials and methods for addressing themto enable valid inference.We conclude this series with a chapter discussing the opportunities and chal-lenges that lie ahead in developing on person-centered treatment regimens.Theadvances in cancer biology and the genetics of cancer have r
38、apidly providedus witha better fundamental understanding of cancer.These new developments require anew generation of clinical trials that modernize the processes and methods used toexamine the safety and efficacy of novel,gene-based therapies without sacrificinghigh standards.Chapter 10 revisits som
39、e of the basic components of clinical trialdesign within the context of timely areas of vaccine trials,cancer stem cell trials,and trials of epigenetic targeted therapies.This book is intended for biostatisticians with special interest in cancer researchor medical researchers with some background in
40、 biostatistics such as a workingviiiPrefaceknowledge of clinical trial designs and regression analysis.Since the authors forall the chapters are experienced in modern clinical trial data analysis and are at theforefront of their respective areas,this book should enable them to quickly applythesemeth
41、odstotheirownstudies,especiallyconsideringthefactthatmostchapterscontain illustrative real study data and associated software.We would like to express our appreciation to all who have contributed to thisbook.We are also thankful to editors Fiona Sarne and Rachel Warren for theirpatience and continui
42、ng support,despite multiple delays on the project on our part.Contents1Survival Analysis.1Changyong Feng,Yu Han,Pan Wu,and Xin Tu2Longitudinal Data Analysis.25Wan Tang,Naiji Lu,Rui Chen,and Hui Zhang3Assessment of Diagnostic Tests and Instruments.55Hua He,Douglas Gunzler,Yan Ma,and Yinglin Xia4Analy
43、sis of Sequential Clinical Trials.81W.Jackson Hall5Dynamic Treatment Regimes.127Min Qian,Inbal Nahum-Shani,and Susan A.Murphy6Statistical Issues with Trial Data and Economic Modelingfor Cost-Effectiveness Evaluation.149Kathleen A.Boyd,Andrew H.Briggs,Pierre Ducournau,Marlene Gyldmark,Frederic de Rey
44、det,and Jim Cassidy7Active-Controlled Clinical Trials.167Yi Tsong and Joanne Zhang8Thorough QT/QTc Clinical Trials.183Yi Tsong and Jinglin Zhong9Causal Inference in Cancer Clinical Trials.203Babette A.Brumback and Wendy B.London10Changing Paradigms in Cancer Clinical Trials.227Jeanne Kowalski and Wi
45、lliam MatsuiIndex.247ixChapter 1Survival AnalysisChangyong Feng,Yu Han,Pan Wu,and Xin Tu1.1IntroductionSurvival analysis concerns the time from a well-defined origin to some end event,such as the time from surgery to death of a cancer patient,the time from weddingto divorce,time from layoff to findi
46、ng a new job,and time between the first andsecond suicide attempts.Although originated in and driven much by research onlifetime,or survival of an object such as light bulbs and other electric devices inthe early days,modern applications of survival analysis include many non-survivalevents such as t
47、he aforementioned examples.Thus,survival analysis may be moreappropriatelycalledthetime-to-eventanalysis.However,inthis chapterwe continueto use the classic term survival analysis in our discussion of this statistical modeland its applications.Like all other statistical methods,survival analysis dea
48、ls with random eventssuch as death,divorce,and layoff.Thus,the time to event is a nonnegative randomvariable and usually right skewed because most events tend to occur in closeproximity to each other after being observed for a while.In survival analysis weare primarily interested in:(i)the distribut
49、ion of survival time,and(ii)the influenceon such a distribution by some explanatory variables such as age and gender.Insome applications,subjects may be clustered because of certain common featuressuch as genetic traits and shared environmentalfactors.Clustered survival time dataalso arise from anal
50、ysis involving multiple events from the same individual suchas repeated suicide events.Note that although estimates of model parameters,orpoint estimates,obtained from the standard likelihood or partial likelihood-basedC.Feng(?)Y.Han P.Wu X.TuDepartment of Biostatistics and Computational Biology,Uni
