Clinical trial biostatistics and biopharmaceutical applications ( PDFDrive.com )

In the fourth section, the book describes multiple test problems with applications to adaptive designs, graphical approaches to multiple testing, the estimation of simultaneous confidenc

Clinical Trial Biostatistics and Biopharmaceutical Applications presents novel

biostatistical methodologies in clinical trials as well as up-to-date biostatistical

applications from the pharmaceutical industry Each chapter is self-contained with

references

Divided into five sections, the book begins with emerging issues in clinical trial

design and analysis The second section examines adaptive designs in drug

devel-opment, discusses the consequences of group-sequential and adaptive designs,

and illustrates group sequential design in R The third section focuses on oncology

clinical trials, covering competing risks, escalation with overdose control (EWOC)

dose finding, and interval-censored time-to-event data

In the fourth section, the book describes multiple test problems with applications

to adaptive designs, graphical approaches to multiple testing, the estimation of

simultaneous confidence intervals for multiple comparisons, and weighted

para-metric multiple testing methods The final section discusses the statistical analysis

of biomarkers from omics technologies, biomarker strategies applicable to clinical

development, and the statistical evaluation of surrogate endpoints

Features

• Examines numerous critical issues, such as dosing, assay sensitivity,

competing risks, multiplicity, and the use of biomarkers

• Explores major advances in adaptive clinical trial designs, including

group-sequential survival trials and extensions to Bayesian adaptive dose-finding

designs

• Explains how to implement the procedures using R and other software

• Discusses regulatory considerations and U.S FDA guidelines

• Illustrates the models and methods with many real examples and case

studies from drug development, cancer research, and more

This book clarifies important issues when designing and analyzing clinical trials,

in-cluding several misunderstood and unresolved challenges It will help you choose

the right method for your biostatistical application

K21709

w w w c r c p r e s s c o m

Clinical Trial Biostatistics and Biopharmaceutical Applications

and Biopharmaceutical Applications

and Biopharmaceutical Applications

Edited by

Walter R Young Ding-Geng (Din) Chen

© 2015 by Taylor & Francis Group, LLC

CRC Press is an imprint of Taylor & Francis Group, an Informa business

No claim to original U.S Government works

Version Date: 20141021

International Standard Book Number-13: 978-1-4822-1219-8 (eBook - PDF)

This book contains information obtained from authentic and highly regarded sources Reasonable efforts have been made to publish reliable data and information, but the author and publisher cannot assume responsibility for the validity of all materials or the consequences of their use The authors and publishers have attempted to trace the copyright holders of all material reproduced in this publication and apologize to copyright holders if permission to publish in this form has not been obtained If any copyright material has not been acknowledged please write and let us know so we may rectify in any future reprint.

Except as permitted under U.S Copyright Law, no part of this book may be reprinted, reproduced, ted, or utilized in any form by any electronic, mechanical, or other means, now known or hereafter invented, including photocopying, microfilming, and recording, or in any information storage or retrieval system, without written permission from the publishers.

transmit-For permission to photocopy or use material electronically from this work, please access www.copyright com (http://www.copyright.com/) or contact the Copyright Clearance Center, Inc (CCC), 222 Rosewood Drive, Danvers, MA 01923, 978-750-8400 CCC is a not-for-profit organization that provides licenses and registration for a variety of users For organizations that have been granted a photocopy license by the CCC,

a separate system of payment has been arranged.

Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are used

only for identification and explanation without intent to infringe.

Visit the Taylor & Francis Web site at

http://www.taylorandfrancis.com

and the CRC Press Web site at

http://www.crcpress.com

the conference.

Walter Young

To my parents and parents-in-law who value higher education and hardwork, and to my wife, Ke, my son, John D Chen, and my daughter, Jenny K.Chen, for their love and support This book is also dedicated to my passionfor the Deming Conference!

Ding-Geng (Din) Chen

List of Figures xi

List of Tables xv

Preface xix

Early History of the Deming Conference, by J Stuart Hunter xxiii

Some Nonstatistical Reminiscences of My 44 Years of Chairing the Deming Conference, by Walter R Young xxv

Contributors xxxiii

Section I Emerging Issues in Clinical Trial Design and Analysis 1 Emerging Challenges of Clinical Trial Methodologies in Regulatory Applications 3

H.M James Hung and Sue-Jane Wang 2 Review of Randomization Methods in Clinical Trials 41

Vance W Berger and William C Grant 3 First Dose Ranging Clinical Trial Design: More Doses? Or a Wider Range? 55

Guojun Yuan and Naitee Ting 4 Thorough QT/QTc Clinical Trials 75

Yi Tsong 5 Controversial (Unresolved) Issues in Noninferiority Trials 97

Brian L Wiens Section II Adaptive Clinical Trials 6 Adaptive Designs in Drug Development 117

Sue-Jane Wang and H.M James Hung

vii

7 Optimizing Group-Sequential Designs with Focus on

Adaptability: Implications of Nonproportional Hazards in

Clinical Trials 137

Edward Lakatos

8 Group Sequential Design in R 179

Keaven M Anderson

Section III Clinical Trials in Oncology

9 Issues in the Design and Analysis of Oncology

Clinical Trials 213

Stephanie Green

10 Competing Risks and Their Applications in Cancer

Clinical Trials 247

Chen Hu, James J Dignam, and Ding-Geng (Din) Chen

11 Dose Finding with Escalation with Overdose Control in Cancer Clinical Trials 273

André Rogatko and Mourad Tighiouart

12 Interval-Censored Time-to-Event Data and Their Applications

in Clinical Trials 307

Ling Ma, Yanqin Feng, Ding-Geng (Din) Chen, and Jianguo Sun

Section IV Multiple Comparisons in Clinical Trials

13 Introduction to Multiple Test Problems, with Applications to

Adaptive Designs 337

Jeff Maca, Frank Bretz, and Willi Maurer

14 Graphical Approaches to Multiple Testing 349

Frank Bretz, Willi Maurer, and Jeff Maca

15 Pairwise Comparisons with Binary Responses:

Multiplicity-AdjustedP-Values and Simultaneous

Confidence Intervals 395

Bernhard Klingenberg and Faraz Rahman

16 Comparative Study of Five Weighted Parametric Multiple

Testing Methods for Correlated Multiple Endpoints in

Clinical Trials 421

Changchun Xie, Xuewen Lu, and Ding-Geng (Din) Chen

Section V Clinical Trials in a Genomic Era

17 Statistical Analysis of Biomarkers from -Omics Technologies 435

Herbert Pang and Hongyu Zhao

18 Understanding Therapeutic Pathways via Biomarkers and

Other Uses of Biomarkers in Clinical Studies 453

Michael D Hale and Scott D Patterson

19 Statistical Evaluation of Surrogate Endpoints

in Clinical Studies 497

Geert Molenberghs, Ariel Alonso Abad, Wim Van der Elst,

Tomasz Burzykowski, and Marc Buyse

Index 537

1.1 Relationship between γ and λ as a function of RR = C/P 7

1.2 Probability that m of four regions show a nonpositive treatment effect where sample size allocation for the four regions: (0.20, 0.10, 0.30, 0.40) 33

3.1 Dose–response results from the three studies of osteoarthritis drug 57

3.2 Several possible dose response curves 58

3.3 Graphical summary of the evaluation on statistical power 68

3.4 Graphical summary of the evaluation on false positive rate 69

3.5 Graphical summary of the evaluation on statistical power with a total sample 70

3.6 Graphical summary of the evaluation on two-sided false positive rate with a total sample size of 420 71

4.1 ECG waves 76

4.2 General time course of Moxifloxacin effect of a double-blind randomized study 81

5.1 Tipping point plot for the example data in Table 5.1 103

5.2 Gatekeeping procedure for example in Table 5.2 110

6.1 Impact on Type I error rate one cannot rule out when the (interim) observed data are used to update the noninferiority margin 126

6.2 Select among treatment dose groups based on clinical endpoint, cutoff D = 0 129

6.3 Adaptive monitoring-logistics models proposed in regulatory submissions 130

7.1 Sketch of the survival experience of a statin arm of the PROVE-IT trial 140

7.2 Sketch of the survival curve of a statin arm of the PROVE-IT trial compared with two exponential curves fitted to the earlier and later portions of the statin curve 140

7.3 Estimating failure probabilities from graph 141

7.4 Piecewise constant estimate of the failure rate function 142

7.5 Piecewise smoothed version of failure rate function 142

7.6 Constructed sketch showing the general shape of survival curves when there is a threshold treatment lag 145

7.7 Constructed sketch showing the general shape of survival curves when the treatment effect diminishes over time 146

7.8 Smoothed piecewise constant hazard ratio function derived from the BHAT survival curves 147

xi

7.9 Comparing power based on exponential assumption with

power based on the actual shape of the observed

survival curves 148

7.10 Two derived hazard ratio curves from the CARE trial, one

assuming proportional hazards and the other using STOPP,

assuming piecewise exponential hazards 151

7.11 Hazard ratio curves from CARE and BHAT assuming

constant hazards 154

7.12 Time-dependent hazard ratio curves from BHAT and CARE

derived from Kaplan–Meier plots 154

7.13 Comparing the original O’Brien–Fleming and

Pocock boundaries 160

7.14 Comparing the original O’Brien–Fleming boundary with

the first variant of Pocock boundary 161

7.15 Comparing original O’Brien–Fleming boundary with the

first variant of O’Brien–Fleming boundary 162

7.16 Calculating the cumulative survival probabilities

using conditional survival probabilities for each

of the subintervals 166

8.1 Examples of (a) asymmetric and (b) symmetric group

sequential design bounds for trials with a single

interim analysis 186

8.2 Hwang–Shih–DeCani spending functions with γ= −8

(efficacy) and γ= 3 (futility) 190

9.1 Efficacy–toxicity contours defining sets of equally

desirable outcomes 217

9.2 Association of early outcome j with final outcome on

treatment arm A: change in event rate for final outcome

after occurrence of early event 231

10.1 Schematic illustration of competing risks in the multistate

model framework 250

10.2 Cumulative incidence of failures in one-sample setting for

LA-NSCLC patients in RTOG 9410 263

10.3 Comparing cumulative incidences of first failures for

LA-NSCLC patients in RTOG 9410 264

11.1 Example of the logistic tolerance distribution used to model

dose–toxicity relationship 279

11.2 Estimated mean length of HPD of the posterior distribution

of the MTD as a function of the number of patients accrued

to the trial for different target probabilities of DLT θ 286

11.3 Posterior density of the MTD when the number of treated

patients (from bottom to top) is 1, 5, 10, 15, 20, 25, 30, 33 288

11.4 Posterior density of the MTD after 33 patients

have been treated 288

11.5 Recommended dose of PNU as a function of anti-SEA

concentration at both the onset and the conclusion of the

phase I trial 292

11.6 EWOC recommended dose for selected values of the targeted probability of DLT in two distinct trials 297

11.7 Standard paradigm of clinical evaluation of new cancer therapies restricts the determination of dose to the initial phase of the process 298

12.1 Turnbull’s nonparametric estimator overlaid with intcox estimator 326

13.1 Effect of number of tests on type I error rate, under independence 338

13.2 Comparison of the rejection regions for the Bonferroni and Simes procedures 341

13.3 Schematic diagram of closed testing procedure 342

13.4 Pictorial representation of two-stage adaptive designs 344

14.1 Conventions for the graphical approach 355

14.2 Numerical example of the graphical Holm procedure with m= 2 hypotheses and α = 0.025 356

14.3 Example of a graphical multiple test procedure to illustrate Algorithm 14.1 with α = 0.025 and unadjusted p-values p1= 0.015, p2= 0.001, and p3= 0.1 358

14.4 Graphical multiple test procedure from Figure 14.3 revisited 362

14.5 Graphical visualization of a viable multiple test procedure for the diabetes case study 364

14.6 Test strategies for composite endpoints and their components 365

14.7 Graphical illustration of the test strategy for noninferiority and superiority in a combination trial with multiple endpoints 367

14.8 Numerical example for the diabetes case study with unadjusted p-values p1= 0.1, p2= 0.001, p3= 0.0001, and p4= 0.005 369

14.9 Screenshot of the GUI from the gMCP package 370

14.10 Visualization of the (a) hierarchical and (b) fallback procedures for m= 3 hypotheses 371

14.11 Two extensions of the original fallback procedure for m= 3 hypotheses 372

14.12 Graphical visualization of the parallel gatekeeper procedure with two familiesF1= {H1, H2}  F2= {H3, H4} 374

14.13 (a) Weighting strategy to test noninferiority and superiority for two doses (b) Updated weighting strategy after rejecting H1 . 379

14.14 Graphical test procedure without memory 385

14.15 Entangled graphs ( G1,G2)with memory 386

14.16 Graphical visualization of a k-out-of-m gatekeeper

procedure with m = 4 primary hypotheses, k = 3, and one

secondary hypothesis H5 . 38915.1 One-sided critical values of maxi T i1using the exact

distribution based on the binomial probabilities, the

approach based on the asymptotic multivariate normality

(MVN) of (T21, T31, T41)and the approach based on the

Sidak inequality when comparing three groups to a

control, for various sample sizes n1, , n4and true

binomial probability vector π 399

15.2 Two-sided critical values c of max (i,j) |T ij| using the exact

distribution based on the binomial probabilities, the

approach based on the asymptotic multivariate normality

(MVN) of (T21, T31, , T43), and the approach based on

the regular Bonferroni correction for all six pairwise

comparisons among K= 4 groups 400

15.3 Simultaneous confidence region for the case of two

comparisons to control (a) and three pairwise comparisons(b, projected into the two-dimensional space), using the

first three dose groups of the IBS trial as data 415

18.1 Biochemical coverage, pharmacodynamic assessment and

clinical outcome 460

18.2 Graphical representation of the attenuation of an

endogenous agonist by an antagonist 470

18.3 A few analysis possibilities for various combinations of

continuous and binary biomarkers and responses 483

18.4 Common performance characteristics of screening and

diagnostic assays 485

18.5 Illustration of a receiver operating characteristic

(ROC) curve 486

18.6 Hypothetical treatment X marker predictiveness plot 487

19.1 Psychiatric study Individual and mean profiles for each

scale by treatment group 502

1.1 Incidence of Adjudicated Events in RENAAL Trial 26

1.2 Treatment Effect (mmHg) of a Test Drug Relative to Placebo Based on Mean Change from Baseline in Blood Pressure at a Targeted Clinical Visit 28

3.1 Number of Patients in a Center from a Multicenter Trial 62

3.2 Simulation Schema 66

3.3 Assessment of EmaxModel Fit Performance 72

3.4 Summary Statistics of Estimated ED50s from EmaxModel Simulations 73

5.1 Example Data from a Clinical Trial with Missing Data 102

5.2 Treatment Groups in Example 109

6.1 Data Information of the Two Interim Analyses of WIZARD 127

7.1 Assumed Failure Rates Assuming Constant Hazards 143

7.2 Assumed Failure Rates Captured from PROVE-IT Kaplan–Meier Plots 143

7.3 Assumed Failure Rates Captured from PROVE-IT Plots and Assumed Treatment Lag 143

7.4 Summary of Sample Size Calculations 144

7.5 Two Trials with the Same Overall Proportions Failing in Each Group, but the Required Sample Size Adjustment for Group-Sequential Boundaries Go in Opposite Directions 150

7.6 Declining Hazards Derived from PROVE-IT Plots 156

7.7 Operating Characteristic for Trial with N= 3300 157

7.8 Operating Characteristic for Trial with N= 3475 157

7.9 Cumulative Failure and At-Risk Probabilities for Control and Experimental Arms of Trial 167

7.10 How Censoring Effects Cumulative At-Risk Probabilities of Trial in Table 7.9 172

7.11 Conditional 2× 2 Table for Log-Rank Statistic 175

8.1 Basic Summary of Bounds for Binomial Trial 195

8.2 Asymmetric Two-Sided Group Sequential Design with Nonbinding Futility Bound, 2 Analyses, Time-to-Event Outcome with Sample Size 260 and 176 Events Required, 85% power, 2.5% (1-Sided) Type I Error to Detect a Hazard Ratio of 0.6 196

8.3 Incremental Upper Boundary Crossing Probabilities at Each Analysis under the Null and Alternate Hypotheses 200

xv

9.1 Prognostic vs Predictive: Not Possible to Distinguish in a

Single Arm Study 220

10.1 Commonly Used Statistical Methods for Competing Risks 253

10.2 Comparison of Competing Risks Regression Models Examining Treatment and Important Covariates for Competing Outcomes in LA-NSCLC 266

11.1 Estimated Proportions of Patients Given Doses in an ε-Neighborhood of the True MTD under Designs1P and3P and Differences between These Proportions on the Average 284

11.2 Estimated Proportions of Patients Exhibiting DLTs, Treated above the MTD, MSE, and Bias of the MTD under Designs 1P and3P and Differences between These Proportions on the Average 285

11.3 Average Posterior Standard Deviation and Average Length of HPD of the Posterior Distribution of the MTD That Are Achieved for a Given Sample Size for θ= 0.3 286

11.4 Number of Patients Accrued to the Trial (N) to Achieve an Estimated Mean Length of 0.74 for the 95% HPD Interval of the Posterior Distribution of the MTD for Selected Target Probabilities of DLT(θ) 299

12.1 Interval-Censored Breast Cancer Data 325

15.1 Data for the IBS Trial and the Resulting Observed Differences in the Proportions of Relief from Abdominal Pain between Dose Groups i = 2, , 5 and Placebo 408

15.2 Multiplicity Adjusted P-values padjij Corresponding to the Family of Null Hypotheses{H0 ij} for All 10 Pairwise Comparisons between the 5 Dose Groups of the IBS Trial 410

15.3 Simultaneous 95% Lower Bounds for the Difference in the Probability of Relief from Abdominal Pain Comparing Dose Group i = 2, , 5 to Placebo (i = 1) for the IBS Data 416

15.4 Simultaneous 95% Confidence Intervals for All 10 Pairwise Comparisons among the 5 Dose Groups for the IBS Data 417

16.1 Two Endpoints: Simulated Family-Wise Type I Error Rate (%)Based on 1,000,000 Runs for WMTC, FFS, and 4A When Different Correlations Are Specified in Multiple Testing Adjustments 428

17.1 Confusion Matrix Example 440

17.2 Confusion Matrix Definitions 440

17.3 Confusion Matrix of Results 448

17.4 Top Pathways with p-Values ≤0.0002 and q-Values ≤0.005 449

19.1 Schizophrenia Study Results of the Trial-Level (R2trial) Surrogacy Analysis 511

19.2 Examples of Possible Surrogate Endpoints in

Various Diseases 513

19.3 Schizophrenia Study: Trial-Level and Individual-Level

Validation Measures (95% Confidence

Intervals)—Binary–Binary Case 524

The Annual Deming Conference on Applied Statistics (http://www.demingconference.com/) has long been deemed as an important event inthe statistics profession since 1945 This book is to honor the contributions

of the Deming conference toward clinical trials and the biopharmaceuticalindustries for the last decades by inviting prominent Deming conferencespeakers to contribute chapters on the cutting-edge developments in biosta-tistical methodologies in clinical trials and biopharmaceutical applications.The novel methodology development and up-to-date biopharmaceuticalapplication validate the importance of this book

The book starts with two historical notes: one from Stuart Hunter fromPrinceton University and the other from Walter Young, Deming conferencechair, which provide a historical overview of the Deming conference Follow-ing this, the book is divided into five sections containing the contributionsfrom prominent Deming conference speakers

Section I, Emerging Issues in Clinical Trial Design and Analysis, consists

of five chapters Chapter 1 covers topics ranging from emerging challenges

of clinical trial methodologies in regulatory applications to active controlledtrial design, adaptive design, multiple comparison considerations, roles ofmodeling and simulation, as well as multiregional clinical trials Chapter 2 is

a review of randomization methods in clinical trials and discusses the prosand cons of some randomization procedures Chapter 3 discusses the impor-tance of selecting a wide range of doses in designing Phase II dose-rangingtrials Many practitioners misunderstand this critical issue and tend to addmore study doses in a narrow dose range The chapter clarifies this point.Chapter 4 is about thorough QT/QTc clinical trials A thorough QT clinicalstudy is required for any nonantiarrhythmic drugs in early drug develop-ment to assess the potential of drug-induced prolongation of QT interval andproarrhythmic potential since 2005 This chapter discusses the approachesand issues in the design and analysis of a thorough QT clinical trial Chapter 5discusses controversial (unresolved) issues in noninferiority trials, especially

on assay sensitivity and the constancy assumption

Section II, Adaptive Clinical Trials, consists of three chapters Chapter 6discusses the roles of adaptive designs in drug development This chaptergives a brief overview of some of the major advances in adaptive clinicaltrial designs and presents several scenarios where adaptive design is worth-while Chapter 7 is on optimal design of group-sequential survival trialsincluding adaptability This chapter focuses on the survival trials that arenotorious for surprises, such as far fewer events, or smaller treatment effects,than anticipated Particular focus is on the consequences of group-sequentialand adaptive designs Chapter 8 presents the group sequential design in R

xix

This chapter demonstrates the gsDesign R package using examples of trialsplanned with a single futility analysis.

Section III, Clinical Trials in Oncology, contains four chapters Chapter 9gives an overview on issues in the design and analysis of oncology clinicaltrials, such as choice of treatment arms, primary endpoint, randomizationscheme, trial size, use of historical information, model assumptions, andtesting strategy, as well as issues such as those related to missing data, evalu-ability, subset analysis, competing risks, outcome-by-outcome analysis, andsurrogate endpoints, which may be critical to interpretation Chapter 10 isaimed at competing risks and their application in oncology clinical trials.Competing risks are frequently encountered in clinical cancer research, whereindividuals under study may experience one of two or more different types(causes) of failures, and the time to the first failure of any type is typicallyused for efficacy evaluation This chapter then pays special attention to com-peting risks since the occurrence of first failure either prevents the occurrence

of other types of failures or the disease natural history is altered due to thesubsequent first-line treatment Chapter 11 presents the escalation with over-dose control (EWOC) dose finding in cancer clinical trials, which is a Bayesianadaptive dose-finding design that produces consistent sequences of doseswhile controlling the probability that patients are overdosed This chapterpresents the original model and several extensions, such as the incorporation

of individual patient characteristics by continuous or categorical covariatesand introduction of intermediate grade toxicities in addition to dose-limitingtoxicities Chapter 12 is on interval-censored time-to-event data and theirapplications in cancer research Interval-censored time-to-event or failuretime data occur in many areas especially in oncology and biopharmaceutics.This chapter provides a review of the recent developments in the field with

a focus on the applications of existing methodologies and available softwarefor biopharmaceutical applications

Section IV, Multiple Comparisons in Clinical Trials, contains four chapters.Chapter 13 introduces multiple test problems with applications to adaptivedesigns This chapter focuses on the general concepts behind multiplic-ity and some of the basic techniques for correcting for multiplicity withapplications to adaptive trial design Chapter 14 focuses on the graphicalapproaches to multiple testing that can be applied to common multiple testproblems, such as comparing several treatments with a control, assessingthe benefit of a new treatment for more than one endpoint, and combinednoninferiority and superiority testing along with the graphical user inter-face from the gMCP package in R Chapter 15 presents pairwise comparisonswith binary responses and focuses on estimating simultaneous confidenceintervals for multiple comparisons to a control or all pairwise comparisons,using the risk difference as the effect measure Chapter 16 discusses theweighted parametric multiple testing methods for correlated multiple end-points To incorporate the correlations among the test statistics, severalweighted parametric multiple testing methods have been proposed, such

as Huque and Alosh’s flexible fixed-sequence (FFS) testing method, Li andMehrotra’s adaptive α allocation approach (4A), Xie’s weighted multipletesting correction (WMTC), Bretz et al.’s graphical approaches, and Millenand Dmitrienko’s chain procedures This chapter provides the relationshipbetween these weighted parametric multiple testing methods and discussesthe effect of misspecified correlations in these methods Tentative guidelines

to help choose an appropriate method are provided

Section V, Clinical Trials in a Genomic Era, contains three chapters

Chapter 17 is for the statistical analysis of biomarkers from -omics

technolo-gies This chapter presents an overview of classification, validation, andsurvival prediction methodologies, and real data examples for analyzing

-omics data As new data types evolve, there is ample opportunity for

statis-ticians and bioinformastatis-ticians to develop novel methodologies for solvingthese big data challenges Chapter 18 presents understanding of therapeu-tic pathways via biomarkers and other uses of biomarkers in clinical studies.Specifically, this chapter gives an overview of biomarker strategies applica-ble to clinical development, covering phases I–III, with an emphasis on earlydevelopment studies, which include examples of successes and challengesassociated with these approaches; study designs and analysis appropriatefor different uses of biomarkers; practical considerations related to acquiringsamples and developing or selecting a suitable assay; basic models useful for

a quantitative approach to pharmacology, including the Hill and Gaddummodels; regulatory considerations and recent FDA guidance; and conceptsaround biomarker validation Chapter 19 is on statistical evaluation of sur-rogate endpoints in clinical studies, which emphasizes the meta-analyticapproach and its information-theoretic versions The methods are illustratedwith the help of case studies

Each chapter is self-contained with references provided at the end of everychapter, for the readers’ convenience

We would like to express our gratitude to many individuals First, thanks

to David Grubbs, from Taylor & Francis Group, for his interest in the book,and Shashi Kumar, from ITC, for assistance in LATEX Special thanks to theauthors of the chapters, to the Deming Conference Committee, and to Xinyan(Abby) Zhang, a PhD candidate from the Department of Biostatistics, Univer-sity of Alabama at Birmingham, and to Jenny K Chen from Pittsford MendonHigh School, for assistance in LATEX and formatting, which helped to speed

up the production of this book

We welcome any comments and suggestions on typos, errors, andfuture improvements to this book Please contact Ding-Geng (Din) Chen atDG.Chen@gmail.com

Walter Young

Wayne, Pennsylvania

Ding-Geng (Din) Chen

Rochester, New York

The conference organizer was Princeton professor Sam S Wilks (1906–1964) Sam Wilks was then a member of the National Defense ResearchApplied Mathematics panel and the organizer of the Statistical TechniquesResearch Group (STRG) at both Princeton and Columbia universities Wilkswas widely recognized as a leader in the adaptations of statistics to indus-trial problems His 1941 paper “Determination of Sample Sizes for Setting

Tolerance Limits” (Ann Math Stat 12, 1, 91–96) and his close

associa-tion with Walter Shewhart as founding editors of the John Wiley & Sons’

“Statistical Series” serve as examples of his commitment The PrincetonApplied Statistics seminars became an annual event from 1946 Recall, too,that the American Society for Quality Control (ASQC) was established from

1946, and the Shewhart Medal in 1949 The Princeton conference continued

to attract academics and research workers throughout its early years

A significant change of conference environment occurred in the early 1950swith the appearance of Box and Wilson’s 1951 paper “Response Surface

Methodology” ( J Royal Stat Soc., B 13, 1–45) This paper stimulated the

invitation to George Box to become a visiting professor at North CarolinaState University in 1953 and ultimately led to his appointment as director

of a newly established STRG at Princeton in 1955 Sam Wilks’ influencethroughout is obvious

Interest in experimental design applied to industrial problems soonexploded The Metropolitan Section of ASQC held its first annual conference

as part of the Princeton Conference in 1954 In 1956, the Chemical Division ofthe ASQC was established to offer myriad short courses on statistical appli-

cations In 1957, George Box’s paper on evolutionary operation (Appl Stat 6,

2–23) appeared, adding renewed emphasis on the importance of statistics in

the industrial environment And in 1959 the journal Technometrics appeared.

As an example of a conference, consider the 18th Princeton Conferencecosponsored by the Metropolitan Section ASQC and held December 6–7,

xxiii

1962 The meeting announcement states “The Princeton Conference has beenthe traditional meeting place for practitioners of statistical methods andthose concerned with their theoretical development.” Conference Commit-tee: J S Hunter Chr A Bloomberg, E Sector, and L Pasteelnick Registration:(luncheon, beer, and cider party) $6.00 Registration only: $1.50 The program:Acheson Duncan on CuSum charts, Cuthbert Daniel on Biased Observa-tions, Frank Anscomb on Outliers, and Barrie Wetherill on Quantal ResponseCurves Almost 100 attended.

The conference became too large to be held at the Princeton campus,and responsibility for the conference gradually changed In 1967, the Bio-Pharmaceutical Section of the American Statistical Association (ASA) wasformed and it soon joined the Metropolitan Section of ASQ in cosponsorship.The two national societies, the ASQC and the ASA were thus cosponsor-ing the conference Walter Young began his service as conference chairmanduring this period and the conference was named the Deming Conference

in 1994

44 Years of Chairing the Deming Conference

Walter R Young

Deming Conference Chair

In 1964, I was group leader, Process Analysis, at Lederle Laboratories: ican Cyanamid in Pearl River, New York I worked closely with CharlieDunnett, whose department owned the computer and mechanical calcula-tors my group had to use In December, he was speaking at the PrincetonConference and he invited me to go along The conference started at 2:00 p.m

Amer-on a Friday and ran through lunch Amer-on Saturday It included a beer and ciderparty on Friday evening and a Saturday lunch for a total registration fee of $6.The topics ran the gamut of the highly theoretical from the Princeton Statis-tics Department to extremely simple QC talks I enjoyed the atmosphere andcamaraderie so I went every year for the next six years I usually stayed in acheap motel on Route 1 but occasionally splurged for the Nassau Inn

I had joined the ASQC in 1962 and by 1969 was quite active in theMetropolitan Section’s executive committee and thus was acquainted withthe members of the conference’s organizing committee Harry Howardwas the official conference chair but the de facto chair was Art Bobis, whoheaded the statistics group in Cyanamid’s Bound Brook, New Jersey facility

At that point I had to go to Bound Brook every few weeks to use their log computer to model chemical reactions After the conference, I stayed at

ana-my usual motel and went to Bound Brook on Monday morning via publictransportation I ranted with Art about the conference on such points thatmost attendees were grandfathered out of paying the $10.50 registration, theFriday afternoon to Saturday schedule was weird, and there was no unifiedtheme to the talks After listening to me grumble for about five minutes, hetold me that he was leaving Cyanamid to become vice president of his fam-ily’s folding chair business and how would I like to be chair It sounded likefun, so I accepted

I enlisted Khushroo Shaikh, Charlie Dunnett, and Don Behnkin fromCyanamid and my old professor from New York University, John Kao, onthe committee and retained Art Bloomberg who did a tremendous job

on arrangements and registration until his death 25 years later John stayed

on the committee for the next 25 years and in 2012 at the age 96, we gave him

a distinguished service award (Figure 1) (John died on June 8, 2014 at theage of 97.)

xxv

FIGURE 1

Young and Kao, 2013.

Stu Hunter was also extremely helpful with arrangements for the next nineyears until the conference left Princeton I got together with him once or twice

a year for a luncheon at the faculty club where I gained a reputation for ing only the desserts In 2013, we gave Stu the distinguished service awarddespite the fact that he was only a youngster of 90 (Figure 2)

eat-Immediately upon becoming chairman, I changed the format to all dayFriday with four 3-hour sessions of three 1-hour related talks, while retain-ing the luncheon, beer, and cider party (I claim credit for enforcing the term

related, as prior to this anyone who wanted to talk could In a similar vein,

I added the word Applied to the title of the conference in 1972.) We were forced

to raise the registration fee to $15 but we also raised attendance to about 100and tried to let absolutely no one in for free The Metropolitan Section waspleasantly surprised that the conference actually made money, and with theexception of an experimental outlier year of 2001 when statistics was radicallydeemphasized, the conference never lost money again

In 1971, the ASA’s Biopharmaceutical Subsection assumed cosponsorship

In 1973, Bill Wooding became their first official representative and the Sectionhas been continually active in the conference Ivan Chan, who joined thecommittee in 1996, has been their longest serving agent In 1979, the Statis-tics Division of the ASQC assumed cosponsorship Bill Strawderman joinedthe committee in 1980 for 14 years and Frank Alt in 1983 for 11 years astheir official representatives They did an excellent job in evenly balancing

FIGURE 2

Hunter, Laroia, and Young, 2013.

biopharmaceutical topics with QC and applied topics in other industries.Once they left the committee, the conference gradually shifted to being com-pletely biopharmaceutical and the Statistics Division became a sponsor inname only Biopharmaceutical made sense due to the heavy preponderance

of pharmaceutical firms in New Jersey, although in recent years more than10% of our participants have been international The conference steadily grew

in attendance and length Book sales and tutorials based on recently lished texts started in 1976 The conference had a peak attendance of 476 in

pub-1977 when it overflowed the Woodrow Wilson School For the first time wedecided to check badges The registration fee had at this point been raised

to $40 but this included meals Literally dozens of people hadn’t paid Oneindividual threatened to sue as he said he’d been coming for 5 years withoutpaying and it was illegal for us to check badges without advance notice Anauthor of a highly acclaimed text who flew in from California also said thatshe had never paid in the past Two individuals from a major drug company

in NYC were stopped at the door of the meeting room When they were nextcaught trying to slip through a back door, they paid, but came back 15 min-utes later to ask for a refund Their boss told them that if they couldn’t sneak

in, they couldn’t attend A member of the program committee, Kris Arora,

FIGURE 3

Young and Deming, 1974.

owned an excellent local Indian restaurant and he catered a very successfulIndian buffet in the basement of the Statistics Building for a number of years.During the meal, four students were caught exiting from a closet where theyhad earlier secreted themselves

In 1979, the conference established the W Edwards Deming Medal in tistical Excellence I was undeservedly awarded the first silver medal and forthe next 16 years the medal was presented at the conference to distinguishedstatisticians It is now awarded by the national ASQ to QC practitioners

Sta-Ed (in Figure 3) was the keynote speaker every year and presented the medaluntil his death in 1993, shortly after the conference The following year, withhis daughter’s permission, we renamed the conference in his honor I wasprivileged to get to know Ed during this period and was a guest in bothhis Washington home and Manhattan apartment on a number of occasions

Ed was one of the participants in the first conference and always had a softspot in his heart for it He spoke at the conference several times prior tothe medal’s establishment One amusing incident occurred in 1974 Ed wasscheduled to speak for two hours followed by a famous pollster headquar-tered in Princeton The pollster still had not shown up 30 minutes prior tohis talk I called his office, and heard in the background, “Tell them I’m nothere!” Ed graciously agreed to continue to talk and did not stop until wellafter three hours

The conference switched to its current format of 3 days of tutorials andtwo 2-day courses in 1980 One course was made the responsibility of theASA’s Biopharmaceutical Section and their agents have done an excellent job

of organizing this for the last 33 years The other one was given to the ASQC’sStatistics Division but after they terminated their active sponsorship in 1999,

I moderated at our calamitous year after which Fred Balch assumed controland has done a fine job for the last 13 years

I believe that my major claim to fame was coming up with the idea of the3-hour tutorial (preferably based on recently published texts) as opposed tothree 1-hour talks on related subjects hosted by a moderator Tutorials madetheir appearance in 1975 and the last 1-hour talk was essentially given in 1982.Coincident with the assumption of the current format, the conference moved

to the Holiday Inn at Newark Airport in 1980, where it remained for the nextfive years Until my wife compelled me to move in 1990, I was a lifelongManhattan resident who twice failed the driving test This obscure point isrelevant, because I insisted on including information on how to get to theHoliday Inn by public transportation My instructions in the program were

to take a nonstop bus to Newark Airport, walk about a quarter of a mile along

a busy exit ramp, try to safely cross it, and then go through a hole in a chainlink fence When I got there the day before the conference, I found that thehole had been repaired I had to take a cab to a Newark hardware store, buy

a pair of wire cutters, and regenerate the hole

After the Holiday Inn, we moved to Atlantic City in 1985, where with twonot too successful exceptions and one disaster, we have remained ever since

I got married in 1984 and my three kids virtually grew up in Atlantic City

My two-month-old son got thrown out of the Osmond Family production

of Fiddler on the Roof at the Claridge for crying I’ve never been paid for my

tenure at the conference but in 1986 I won a drawing (attended by MickeyMantle no less) at the Claridge for meeting planners where the grand prizewas a trip on the QE2 to England and back on the Concord 11 days later

We had a great time but my wife suffered some odd complications such asphysically losing her green card, being allowed on the ship without a Britishvisa and thus having a rather difficult time getting off, and losing her brownPhilippine passport two hours prior to our scheduled return, which causedher to somewhat believe in being cursed The passport was the same color asour hotel room carpet and our son had apparently grabbed it from his crib anddumped it These incidents somewhat assuaged my guilt for not rebookingthe Claridge for 1986

Besides attending the conference in their formative years, my oldest sonredesigned our program into its current format, my youngest son (who iscurrently in the Air Force in Qatar) served as registrar for a year, and mydaughter was awarded the conference’s Walter Young scholarship

In 1984, Karl Peace joined the program committee and for 11 years he did asuperlative job in getting speakers and successful short courses In 2010, Karlgraciously accepted our invitation to present a tutorial on his latest text and

we recognized him for his contributions to our conference His coauthor, and

my coeditor, Din Chen has done an equally outstanding job in replacing himfor the past three years (Figure 4)

FIGURE 4

Young and Chen, 2013.

After five quite successful and enjoyable years at the Sands, we were ished as the Sands replaced their meeting rooms with a gambling floor Wehad a multiple-year contract and the Sands negotiated a very favorable con-tract at Resorts as part of their settlement Resorts has a very strange meetingfacility on the 13th floor consisting of a large auditorium and two very longand narrow meeting rooms There is a parapet surrounding these rooms andone can walk along the outside of the building with a rather spectacular view

ban-of Atlantic City However, considering the December weather, not many istrants elected to exercise this option We remained there successfully forseven years, mainly because the contract included a cocktail hour with excel-lent food and an open bar This proved to be a mixed blessing as one of ourmoderators had to stay in the local hospital with an inebriated registrant for

reg-a number of hours until reg-a freg-amily member creg-ame reg-and picked him up

In 2000, the Metropolitan Section fired me as chairman as they felt the ference should be devoted to Deming’s QC principles despite the fact that

con-he was a statistician for tcon-he greater part of his career I did not contest this

as I felt a 31-year tenure was adequate A few weeks later they rehired me

as cochair as they couldn’t find anyone else to handle the statistics portion

of their revamped program All programs for the past 15 years can be found

at www.demingconference.com, and the Spring 2001 program was radicallydifferent from anything seen before or since It was held in the spring at theNewark Airport Holiday Inn because it could not attract sufficient registrants

in December It still lost quite a bit of money in the spring but as so much

money had been invested in the conference (e.g., paying a professional designfirm to compose the program), it was decided not to cancel The statisticstracks did marginally better than the QC tracks but both did poorly The meet-ing included meals but there were more speakers and committee members atthe meals than participants The statistics course that I moderated came close

to making money but the QC course was held with a single registrant andthe instructor was paid appreciably more than the registration fee The onepositive change we made that year was that Ed Warner joined the commit-tee as transactions chair and all registrants have been given a bound copyand a CD of all of the speaker’s slides ever since I regained my sole hold

on the chair and Satish Laroia (in Figure 2) was ensconced as arrangementschair and has done an outstanding job in improving and strengthening theconference’s amenities

Registration has always been a problem In our record year at the WoodrowWilson Center, I personally typed the 476 punched cards and wrote the list-ing program in Fortran We had a number of registrars but they did not staythat long due to the hassle of opening the mail and interpreting the poorhandwriting of the registration forms Fred Balch, our currently longest serv-ing committee member joined in 1993 as registrar and served for six yearsbefore transferring to the program committee Our biggest technical innova-tion came when Wenjin Wang joined the committee in 2005 as Bibliolater In

2006, he replaced Kalyan Ghosh (who only used the web to publicize ourprogram) as Webmaster and designed a website that allowed for online reg-istration, removing the need for the registrar to type all of the demographicinformation We still allow for mail registration but have not received a formfor several years

For the past 11 years we’ve been at the Tropicana with a very high rate

of recidivism as our attendees like the Havana Tower From past experience,we’re reluctant to change what has proved to be a relatively successful for-mat However, the conference is still evolving Since 2008, we’ve awarded a

$4,000 college scholarship to the offspring of a participant For the past fiveyears, Nandita Biswas has run a program where we support the attendance

of two to four local biostatistics PhD students who present their work in aposter In 2012, Manoj Patel, our current registrar, initiated a program where

we encourage our attendees to also present posters We’ve been lucky in thatwe’ve never lost a speaker due to weather Obviously, I can’t continue foreverbut I’ve already confirmed Din Chen as my speaker for 2014, my 45th year

as chair

Ariel Alonso Abad

Department of Methodology and

Statistics

Maastricht University

Maastricht, the Netherlands

Keaven M Anderson

Late Development Statistics

Merck Research Laboratories

North Wales, Pennsylvania

Vance W Berger

Biometry Research Group

National Cancer Institute

Interuniversity Institute for

Biostatistics and Statistical

Interuniversity Institute forBiostatistics and StatisticalBioinformatics (I-BioStat)Hasselt University

Hasselt, Belgium

Ding-Geng (Din) Chen

Center for Researchand

School of Nursingand

Department of Biostatistics andComputational BiologySchool of MedicineUniversity of RochesterRochester, New York

James J Dignam

Department of Health StudiesThe University of ChicagoChicago, Illinois

xxxiii

Office of Translational Sciences

Center for Drug Evaluation and

Research

U.S Food and Drug Administration

Silver Spring, Maryland

Geert Molenberghs

Interuniversity Institute forBiostatistics and StatisticalBioinformatics (I-BioStat)Hasselt University

Hasselt, Belgiumand

University of LeuvenLeuven, Belgium

Herbert Pang

Department of Biostatistics andBioinformatics

School of MedicineDuke UniversityDurham, North Carolinaand

School of Public HealthThe University of Hong KongHong Kong, People’s Republic

Cedars-Sinai Medical Center

Samuel Oschin Comprehensive

Cedars-Sinai Medical Center

Samuel Oschin Comprehensive

Office of Translational Sciences

Center for Drug Evaluation and

Research

U.S Food and Drug Administration

Silver Spring, Maryland

Wim Van der Elst

Interuniversity Institute forBiostatistics and StatisticalBioinformatics (I-BioStat)Hasselt University

Hasselt, Belgium

Sue-Jane Wang

Office of BiostatisticsOffice of Translational SciencesCenter for Drug Evaluation andResearch

U.S Food and Drug AdministrationSilver Spring, Maryland

Brian L Wiens

Portola PharmaceuticalsSan Francisco, California

Changchun Xie

Division of Epidemiology andBiostatistics

Department of EnvironmentalHealth

University of CincinnatiCincinnati, Ohio

Walter R Young (Retired)

Emerging Issues in Clinical Trial Design and Analysis

Emerging Challenges of Clinical Trial

Methodologies in Regulatory Applications

H.M James Hung and Sue-Jane Wang

CONTENTS

1.1 Active Controlled Trial Designs 41.1.1 Noninferiority Trial with Placebo Arm 51.1.2 Noninferiority Trial without Placebo Arm 81.1.3 Margin Determination in Noninferiority Trial

without Placebo 91.1.4 Intent-to-Treat versus Per-Protocol versus On-Treatment

Analysis 101.1.5 Testing for Superiority and Noninferiority 111.2 Adaptive Design 121.2.1 Adaptation of Statistical Information 131.2.2 Adaptive Selection 161.2.3 Trial Conduct and Logistics 191.2.4 Issues of Statistical Efficiency 201.3 Multiple Comparison Considerations 201.3.1 Primary Endpoint versus Secondary Endpoint 211.3.1.1 Testing Multiple Endpoints in Fixed Design Trial 221.3.1.2 Testing Multiple Endpoints in Group Sequential

Design Trial 231.3.1.3 Testing Coprimary Endpoints 241.3.2 Composite Endpoint 251.4 Multiregional Clinical Trials 271.4.1 Values of Multiregional Clinical Trial Strategy 281.4.2 Challenges of Multiregional Clinical Trial Strategy 301.4.3 Design Considerations 31Acknowledgment 34Disclaimer 34References 34

3