Handbook of Methods for Designing, Monitoring, and Analyzing Dose-Finding Trials

John O'Quigley, Alexia Iasonos, Björn Bornkamp, John O'Quigley, Alexia Iasonos, Björn Bornkamp

• Publisher: Chapman and Hall/CRC
• Year: 2017
• ISBN:
  9781351648028
• Edition: 1
• Topic: Matematica
• Subtopic: Probabilità e statistica

Handbook of Methods for Designing, Monitoring, and Analyzing Dose-Finding Trials gives a thorough presentation of state-of-the-art methods for early phase clinical trials. The methodology of clinical trials has advanced greatly over the last 20 years and, arguably, nowhere greater than that of early phase studies. The need to accelerate drug development in a rapidly evolving context of targeted therapies, immunotherapy, combination treatments and complex group structures has provided the stimulus to these advances. Typically, we deal with very small samples, sequential methods that need to be efficient, while, at the same time adhering to ethical principles due to the involvement of human subjects.

Statistical inference is difficult since the standard techniques of maximum likelihood do not usually apply as a result of model misspecification and parameter estimates lying on the boundary of the parameter space. Bayesian methods play an important part in overcoming these difficulties, but nonetheless, require special consideration in this particular context. The purpose of this handbook is to provide an expanded summary of the field as it stands and also, through discussion, provide insights into the thinking of leaders in the field as to the potential developments of the years ahead. With this goal in mind we present:

• 
  
  

• An introduction to the field for graduate students and novices

• 
  
  

• A basis for more established researchers from which to build

• 
  
  

• A collection of material for an advanced course in early phase clinical trials

• 
  
  

• A comprehensive guide to available methodology for practicing statisticians on the design and analysis of dose-finding experiments

• 
  
  

• An extensive guide for the multiple comparison and modeling (MCP-Mod) dose-finding approach, adaptive two-stage designs for dose finding, as well as dose–time–response models and multiple testing in the context of confirmatory dose-finding studies.

John O'Quigley is a professor of mathematics and research director at the French National Institute for Health and Medical Research based at the Faculty of Mathematics, University Pierre and Marie Curie in Paris, France. He is author of Proportional Hazards Regression and has published extensively in the field of dose finding.

Alexia Iasonos is an associate attending biostatistician at the Memorial Sloan Kettering Cancer Center in New York. She has over one hundred publications in the leading statistical and clinical journals on the methodology and design of early phase clinical trials. Dr. Iasonos has wide experience in the actual implementation of model based early phase trials and has given courses in scientific meetings internationally.

Björn Bornkamp is a statistical methodologist at Novartis in Basel, Switzerland, researching and implementing dose-finding designs in Phase II clinical trials. He is one of the co-developers of the MCP-Mod methodology for dose finding and main author of the DoseFinding R package. He has published numerous papers on dose finding, nonlinear models and Bayesian statistics, and in 2013 won the Royal Statistical Society award for statistical excellence in the pharmaceutical industry.

Part II

More Advanced Phase I and Phase I/II Methodology

5
Phase I/II Dose-Finding Designs with Efficacy and Safety Endpoints

Oleksandr Sverdlov

Novartis Institutes for Biomedical Research

Lei Gao

Sanofi US

CONTENTS

1. 5.1 Introduction
2. 5.2 Statistical Background
   1. 5.2.1 Basic concepts
   2. 5.2.2 A Bayesian formulation
   3. 5.2.3 Phase I/II trial objectives
   4. 5.2.4 Adaptive designs
3. 5.3 Phase I/II Designs with Bivariate Binary Outcomes
   1. 5.3.1 Nonparametric designs
   2. 5.3.2 Parametric model-based designs
      1. 5.3.2.1 Extensions of the CRM
      2. 5.3.2.2 Designs based on efficacy{toxicity tradeoff
      3. 5.3.2.3 Bayesian decision-theoretic designs
      4. 5.3.2.4 Penalized optimal adaptive designs
4. 5.4 More Complex Settings
   1. 5.4.1 Binary toxicity and continuous efficacy
   2. 5.4.2 Handling late-onset outcomes
      1. 5.4.2.1 Using surrogate efficacy outcomes
      2. 5.4.2.2 Bayesian data augmentation
      3. 5.4.2.3 Efficacy as a time-to-event outcome
   3. 5.4.3 Incorporating covariates
5. 5.5 Discussion
6. References

5.1 Introduction

The primary objective of a conventional phase I oncology trial of a cytotoxic agent is to identify the maximum tolerated dose (MTD) for testing in subsequent studies. The major assumption is that both the probability of toxicity and the probability of therapeutic response are monotonically increasing with dose; therefore, by determining the MTD, one indirectly determines a dose with maximum potential for therapeutic effect. Many phase I trial designs for determining the MTD with a prespecified level of toxicity rate have been proposed in the literature [53].

With recent developments in personalized medicine, there has been an increasing interest in clinical trials of cytostatic agents—the therapies that act on specific molecular targets expressed in cancer cells and, due to their specific mechanism of action, can inhibit tumor growth or prevent proliferation of cancer cells at doses that are not necessarily very toxic. Clinical trial designs of cytostatic agents generally require special considerations, different from those of cytotoxic agents [31, 71]. In particular, for a cytostatic agent, the risk of toxicity may not necessarily increase with increased doses. In addition, the dose–efficacy curve may have a peak or may reach a plateau at some dose levels below the MTD, and therefore, higher doses may at best result in only marginal improvements in clinical benefit. Hence, a simultaneous assessment of early signals of efficacy along with toxicity can be of great importance for a successful development of a cytostatic therapy. Recently, there has been significant methodological research on statistical designs for phase I clinical trials that incorporate both efficacy and toxicity outcomes in dose-finding objectives (see Refs [10, 52, 54, 76] for recent reviews). Such designs are also referred to as seamless phase I/II designs [10] because they integrate the objectives of traditional phases I and II—namely, identifying a dose or doses that exhibit promising signals of efficacy (phase II goal) with acceptable levels of toxicity (phase I goal)—in a single study.

An advantage of pursuing a seamless phase I/II trial is that doses with desirable benefit–risk ratio can be identified faster and more efficiently than in a conventional sequence of separate phase I and II trials. Clinical development programs can be potentially accelerated because a phase I/II trial avoids an administrative wait between phase I and II protocol activation. Gains in statistical efficiency can be achieved because a phase I/II trial would typically be larger and collect more data than would a single phase I trial. Since therapeutic response and toxicity are frequently correlated, one can perform joint modeling of the dose–efficacy–toxicity relationship—this can yield many insightful findings on the benefit–risk ratio of the compound.

The goal of this chapter is to provide an overview of state-of-the-art seamless phase I/II dose escalation oncology designs that utilize both efficacy and toxicity outcomes in dose assignment decisions. In Section 5.2, we present statistical background, define experimental objectives of a phase I/II trial, and discuss adaptive designs to achieve these objectives in practice. In Section 5.3, we discuss several important types of phase I/II designs for trials where efficacy and toxicity are correlated binary random variables. In Section 5.4, we describe phase I/II designs for more complex settings, namely when the toxicity is binary and efficacy is continuous, when efficacy (and toxicity) outcomes are delayed, and when study patients have heterogeneous prognostic profiles. Section 5.5 provides a discussion.

Note that in this chapter, we exclude phase I/II “cohort expansion” designs where phase I toxicity-based dose escalation is followed by a phase II single-arm or multiarm randomized selection design using efficacy response. Examples of “cohort expansion” designs can be found elsewhere [29, 32]. Also, our review is focused on phase I/II dose-escalation studies of a single compound. More complex phase I/II drug combination and dose-schedule-finding studies are discussed in Chapters 6 and 7 of the present volume. For a recent book-length discussion on Bayesian phase I/II trial designs, re...