Innovative Statistics in Regulatory Science

Shein-Chung Chow

• Publisher: Chapman and Hall/CRC
• Year: 2019
• Print ISBN: 9780367224769
• eBook ISBN: 9781000710816
• Edition: 1
• Topic: Matematica
• Subtopic: Farmacologia

Statistical methods that are commonly used in the review and approval process of regulatory submissions are usually referred to as statistics in regulatory science or regulatory statistics. In a broader sense, statistics in regulatory science can be defined as valid statistics that are employed in the review and approval process of regulatory submissions of pharmaceutical products. In addition, statistics in regulatory science are involved with the development of regulatory policy, guidance, and regulatory critical clinical initiatives related research. This book is devoted to the discussion of statistics in regulatory science for pharmaceutical development. It covers practical issues that are commonly encountered in regulatory science of pharmaceutical research and development including topics related to research activities, review of regulatory submissions, recent critical clinical initiatives, and policy/guidance development in regulatory science.

• 
  
  

• Devoted entirely to discussing statistics in regulatory science for pharmaceutical development.

• 
  
  

• Reviews critical issues (e.g., endpoint/margin selection and complex innovative design such as adaptive trial design) in the pharmaceutical development and regulatory approval process.

• 
  
  

• Clarifies controversial statistical issues (e.g., hypothesis testing versus confidence interval approach, missing data/estimands, multiplicity, and Bayesian design and approach) in review/approval of regulatory submissions.

• 
  
  

• Proposes innovative thinking regarding study designs and statistical methods (e.g., n-of-1 trial design, adaptive trial design, and probability monitoring procedure for sample size) for rare disease drug development.

• 
  
  

• Provides insight regarding current regulatory clinical initiatives (e.g., precision/personalized medicine, biomarker-driven target clinical trials, model informed drug development, big data analytics, and real world data/evidence).

This book provides key statistical concepts, innovative designs, and analysis methods that are useful in regulatory science. Also included are some practical, challenging, and controversial issues that are commonly seen in the review and approval process of regulatory submissions.

About the author

Shein-Chung Chow, Ph.D. is currently a Professor at Duke University School of Medicine, Durham, NC. He was previously the Associate Director at the Office of Biostatistics, Center for Drug Evaluation and Research, United States Food and Drug Administration (FDA). Dr. Chow has also held various positions in the pharmaceutical industry such as Vice President at Millennium, Cambridge, MA, Executive Director at Covance, Princeton, NJ, and Director and Department Head at Bristol-Myers

Squibb, Plainsboro, NJ. He was elected Fellow of the American Statistical Association and an elected member of the ISI (International Statistical Institute). Dr. Chow is Editor-in-Chief of the Journal of Biopharmaceutical Statistics and Biostatistics Book Series, Chapman and Hall/CRC Press, Taylor & Francis, New York. Dr. Chow is the author or co-author of over 300 methodology papers and 30 books.

1

Introduction

1.1 Introduction

For approval of pharmaceutical products (including drugs, biological products, and medical devices) in the United States the Food and Drug Administration (FDA) requires that substantial evidence regarding the safety and effectiveness of the test treatment under investigation be provided in the regulatory submission process Section 314 of 21 Codes of Federal Regulation (CFR). The substantial evidence regarding safety and effectiveness of the test treatment under investigation will then be evaluated by the reviewers (including statistical reviewers, medical reviewers, and reviewers from other relevant disciplines) in the review and approval process of the test treatment under investigation. Statistics plays an important role to ensure the accuracy, reliability, and reproducibility of the substantial evidence obtained from the studies conducted in the process of product development. Statistical methods and/or tools that are commonly used in the review and approval process of regulatory submissions are usually referred to as statistics in regulatory science or regulatory statistics. Thus, in a broader sense, regulatory statistics can be defined as valid statistics that may be used in the review and approval of regulatory submissions of pharmaceutical products. The purpose of regulatory statistics is to provide an objective, unbiased and reliable assessment of the test treatment under investigation.

Regulatory statistics generally follow several principles to ensure the validity of the statistics used in the review and approval process of regulatory submissions. The first principle is to provide unbiased and reliable assessment of the substantial evidence regarding the safety and effectiveness of the test treatment under investigation. The second principle is to ensure quality, validity and integrity of the data collected for supporting the substantial evidence required for regulatory approval. The third principle is to make sure that the observed substantial evidence is not by chance alone and it is reproducible if the same studies were conducted under similar experimental conditions. To ensure the validity of regulatory statistics, it is suggested that statistical principles for Good Statistics Practice (GSP) that outlined in the ICH (International Conference Harmonization) E9 guideline should be followed (ICH, 2018).

The general statistical principles (or key statistical concepts) are the foundation of GSP in regulatory science, which not only ensure the quality, validity and integrity of the intended clinical research during the process of pharmaceutical development, but also provide unbiased and reliable assessment of the test treatment under investigation. Key statistical concepts include, but are not limited to, confounding and interaction; hypotheses testing and p-values; one-sided hypotheses versus two-sided hypotheses; clinical significance/equivalence; and reproducibility and generalizability. In practice, some challenging and controversial issues in the review and approval process of regulatory submissions may arise. These issues include totality-of-the-evidence versus substantial evidence; confusion between the use of $(1-\alpha )\times 100\%$ confidence interval (CI) approach for evaluation of new drugs versus the use of $(1-2\alpha )\times 100\%$ CI approach for assessment of generics/biosimilars; endpoint selection; selecting the proper criteria for decision-making at interim; non-inferiority or equivalence/similarity margin selection; treatment of missing data, the issue of multiplicity; sample size requirement; consistency test in multi-regional trials; extrapolation; drug products with multiple components; and the role of Advisory Committees (e.g., Oncologic Drug Advisory Committee). In addition, there are several critical clinical initiatives recently established by the FDA. These critical clinical initiatives concern precision and/or personalized (individualized) medicine; biomarker-driven clinical research; complex innovative design (CID); model-informed drug development (MIDD); rare diseases drug development; big data analytics; real-world data and real-world evidence, and machine learning for mobile individualized medicine (MIM) and imaging medicine (IM).

In Section 1.2, some key statistical concepts are briefly introduced. Section 1.3 describes some complex innovative designs and corresponding statistical methods. These complex innovative designs include adaptive trial designs, complete n-of-1 trial design, master protocols, and Bayesian approach. Challenging and controversial issues that are commonly encountered in the review and approval process of regulatory submissions are outlined in Section 1.4. Also included in this section are introduction of FDA recent critical clinical initiatives. Section 1.5 provides the aim and scope of the book.

1.2 Key Statistical Concepts

1.2.1 Confounding and Interaction

In pharmaceutical/clinical research and development, confounding and interaction effects are probably the most common distortions in the evaluation of a test treatment under investigation. Confounding effects are contributed by various factors such as race and gender that cannot be separated by the design under study, while an interaction effect between factors is a joint effect with one or more contributing factors (Chow and Liu, 2013). Confounding and interaction effects are important considerations in pharmaceutical/clinical development. For example, when confounding effects are observed, we cannot assess the treatment effect because it has been contaminated. On the other hand, when interactions among factors are observed, the treatment must be carefully evaluated to isolate those effects.

1.2.1.1 Confounding

In clinical trials, there are many sources of variation that have an impact on the primary clinical endpoints for evaluation relating to a certain new regimen or intervention. If some of these variations are not identified and properly controlled, they can become mixed in with the treatment effect that the trial is designed to demonstrate. Then the treatment effect is said to be confounded by effects due to these variations. To gain a better understanding, consider the following example. Suppose that last winter Dr. Smith noticed that the temperature in the emergency room of a hospital was relatively low and caused some discomfort among medical personnel and patients. Dr. Smith suspected that the heating system might not be functioning properly and called on a technician to improve it. As a result, the temperature of the emergency room was at a comfortable level this winter. However, this winter is not as cold as last winter. Therefore, it is not clear whether the improvement (temperature control) in the emergency room was due to the improvement in the heating system or the effect of a warmer winter. In fact, the effect due to the improvement of the heating system and that due to a warmer winter are confounded and cannot be separated from each other. In clinical trials, there are many subtle, unrecognizable, and seemingly innocent confounding factors that can cause ruinous results of clinical trials.

Moses (1985) discussed an example of the devastating result with the confounder being the personal choice of a patient. The example concerns a polio-vaccine trial that was conducted on two million children worldwide to investigate the effect of Salk poliomyelitis vaccine. This trial reported that the incidence rate of polio was lower in the children whose parents refused injection than whose who received placebo after their parent gave permission (Meier, 1989). After an exhaustive examination of the data, it was found that susceptibility to poliomyelitis was related to the differences between the families who gave the permission and those who did not.

In many cases, confounding factors are inherent in the design of the studies. For example, dose-titration studies in escalating levels are often used to investigate the dose-response relationship of the antihypertensive agents during phase II stage of clinical development. For a typical dose titration study, after a washout period during which previous medication stops and the placebo is prescribed, N subjects start at the lowest dose for a prespecified time interval. At the end of the interval, each patient is evaluated as a responder to the treatment or a non-responder according to some criteria prespecified in the protocol. In a titration study, a subject will continue to receive the next higher dose if he or she fails, at the current level, to meet some objective physiological criteria such as reduction of diastolic blood pressure by a specific amount and has not experienced any unacceptable adverse experience. Figure 1.1 provides a graphical presentation of a typical titration study (Shih et al., 1989). Dose titration studies are quite popular among clinicians because they mimic real clinical practice in the care of patients. The major problem with this typical design for a dose-titration study is that the dose-response relationship is often confounded with time course and the unavoidable carryover effects from the previous dose levels which cannot be estimated and eliminated. One can always argue that the relationship found in a dose titration study is not due to the dose but to the time. Statistical methods for binary data from dose-titration studies have been suggested under some assumptions (e.g., see Chuang, 1987; Shih et al., 1989). Because the dose level is confounded with time, estimation of the dose-response relationship based on continuous data has not yet been resolved in general.

Another type of design that can induce confounding problems when it is conducted inappropriately is the crossover design. For a standard 2 × 2 crossover design, each subject is randomly assigned to one of the two sequences. In sequence 1, subjects receive the reference (or control) treatment at the first dosing period and the test treatment at the second dosing period after a washout period of sufficient length. The order of treatments is reversed for the subjects in sequence 2. The issues in analysis of the data from a 2 × 2 crossover design are twofold. First, unbiased estimates of treatment effect cannot be obtained from the data of both periods in the presence of a nonzero carryover effect. The second problem is that the carryover effect is confounded with sequence effect and treatment-by-period interaction. In the absence of a significant sequence effect, however, an unbiased estimate of the treatment effect can be estimated from the data of both periods. In practice, it is not clear whether an observed statistically significant sequence effect (or carryover effect) is a true sequence effect (or carryover effect). As a result, this remains a major drawback of the standard 2 × 2 crossover de...

Table of contents

• Cover
• Half Title
• Title Page
• Copyright Page
• Table of Contents
• Preface
• Author
• 1. Introduction
• 2. Totality-of-the-Evidence
• 3. Hypotheses Testing versus Confidence Interval
• 4. Endpoint Selection
• 5. Non-inferiority/Equivalence Margin
• 6. Missing Data
• 7. Multiplicity
• 8. Sample Size
• 9. Reproducible Research
• 10. Extrapolation
• 11. Consistency Evaluation
• 12. Drug Products with Multiple Components—Development of TCM
• 13. Adaptive Trial Design
• 14. Criteria for Dose Selection
• 15. Generics and Biosimilars
• 16. Precision Medicine
• 17. Big Data Analytics
• 18. Rare Diseases Drug Development
• Bibliography
• Index