Statistical Approaches in Oncology Clinical Development

Current Paradigm and Methodological Advancement

Satrajit Roychoudhury, Soumi Lahiri, Satrajit Roychoudhury, Soumi Lahiri

• Publisher: Chapman and Hall/CRC
• Year: 2018
• ISBN:
  9781351650472
• Edition: 1
• Topic: Mathematik
• Subtopic: Wahrscheinlichkeitsrechnung & Statistiken

Statistical Approaches in Oncology Clinical Development: Current Paradigm and Methodological Advancement presents an overview of statistical considerations in oncology clinical trials, both early and late phase of development. It illustrates how novel statistical methods can enrich the design and analysis of modern oncology trials. The authors include many relevant real life examples from the pharmaceutical industry and academia based on their first-hand experience. Along with relevant references, the book highlights current regulatory views.

The book covers all aspects of cancer clinical trial starting from early phase development. The early part of the book covers novel phase I dose escalation design, exposure response analysis, and innovative phase II design. This includes early development strategy for cancer immunotherapy trials. The contributors also emphasized the role of biomarker and modern era of precision medicine. The second part focuses on the late stage development. This includes the application of adaptive design, safety analysis, and quality of life (QoL) data analysis. The final part discusses current regulatory perspective and challenges.

Features:

• 
  
  

• Covers a wide spectrum of topics related to real-life statistical challenges in oncology clinical trials.

• 
  
  

• Provides a comprehensive overview of novel statistical methods to improve trial design and statistical analysis.

• 
  
  

• Detailed case studies illustrate the real life applications.

Satrajit Roychoudhury is a Senior Director and a member of the Statistical Research and Innovation group in Pfizer Inc. Prior to joining; he was a member of Statistical Methodology and consulting group in Novartis. He has 11 years of extensive experience in working with different phases of clinical trial. His area of research includes early phase oncology trials, survival analysis, model informed drug development, and use of Bayesian methods in clinical trials. He is industry co-chair for the ASA Biopharmaceutical Section Regulatory-Industry Workshop and has provided statistical training in major conferences including the Joint Statistical Meetings, ASA Biopharmaceutical Section Regulatory-Industry Workshop, and ICSA Applied Statistics Symposium.

Soumi Lahiri has 12 years of extensive experience in working different therapeutic areas. She is the former Director of Biostatistics in Clinical Oncology, GlaxoSmithKline. She has also worked in the oncology division of Novartis Pharmaceutical Company for two years. She is an active member of the ASA Biopharmaceutical section and former chair of the membership committee.

1

 

 

Introduction

 

Ohad Amit

GlaxoSmithKline

In 2010, the I-SPY2 trial [1] was launched as an innovative collaboration across five pharmaceutical companies in a phase 2 breast cancer trial. The benefits of this collaborative approach, which significantly reduces the cost, time, and number of subjects required for efficiently bringing new drug therapies to patients, embodies the culmination of much of the innovation in statistical methodology that has characterized oncology drug development in recent years.

Unlike other disease areas, the statistical methodology supporting oncology drug development has long been specialized and has evolved independently of other clinical trial methodologies. The unmet need and urgency underpinning oncology drug development have put many new innovations in statistical methodology front and center.

Fundamentally, the endpoints used to evaluate new therapeutics in oncology have not changed over the last 30 years. Tumor shrinkage, disease progression, and overall survival (OS) remain the mainstays for evaluating the efficacy of new treatments in phase 2 and phase 3 trials. These endpoints confer a readily interpretable result for researchers in a great majority of trials and have been at the core of significant advances in treatment over the last few years. But there is fertile ground to move beyond these traditional modalities. The emergence of more sophisticated imaging modalities and new biomarkers resulting from a more granular understanding of cancer at the molecular level has created tremendous opportunities to further expedite the development of new cancer therapeutics. These new markers can be predictive, defining an enriched population more likely to benefit from treatment, or can be used to evaluate the efficacy of new treatments.

Despite the promises held by these new endpoints, single-arm trials with endpoints based on tumor shrinkage remain the mainstay of phase 2 development in oncology. Such trials rely on the use of historical controls to draw inferences based on well-established response criteria, such as Response evaluation criteria in solid tumors (RECIST) [2]. Such trials are well-placed to incorporate Bayesian methods, and methods have been developed to continuously monitor the data emerging from these trials in a Bayesian decision framework. In many situations, single-arm trials have provided an efficient framework for speeding the approval of new agents. This is particularly true of agents with transformational efficacy where response rates are observed significantly in excess of what one might expect, with last line therapy where few treatment options are available. However, in many other settings, single-arm trials remain difficult to interpret. For agents whose activity is more cytostatic in nature, single-arm trials have been designed with the aim of evaluating efficacy based on the comparison of a time-to-event endpoint relative to historical control. Unlike tumor shrinkage, with time-to-event endpoints in such a trial, there is inherent difficulty in differentiating between the natural history of disease and treatment efficacy. Similarly, with combination agents studied in a single-arm trial, there are inherent challenges in differentiating monotherapy activity from combination efficacy in the absence of a control arm. In both these situations, randomized trials are typically needed. Randomized trials are longer, larger, and more expensive than single-arm trials. There is an opportunity and a clear need to develop new, more efficient designs in these situations.

Unlike in phase 2, progression free survival (PFS) and overall survival (OS) remain the mainstays in phase 3 trials. As more and more effective therapies become available across many tumor types, OS is becoming an increasingly difficult endpoint to study. Long OS times and confounding of results from effective post-progression therapies have greatly complicated the ability to design trials with OS as a primary objective. However, in later line settings or histologies where survival times are shorter, there is a stronger rationale for using OS as a primary endpoint. A key challenge in designing trials with an OS endpoint in later lines for later lines of treatment centers around whether subjects randomized to the control arm should be allowed to crossover to the experimental arm at the time of progression. Many cogent ethical arguments have been made for and against allowing such crossover. What remains indisputable, however, is that allowing crossover will complicate and confound the ability to estimate an unbiased treatment effect with respect to OS. Several statistical methods have been developed over the last few years to provide for more accurate estimates of the treatment effect on OS in the presence of potentially confounding post-progression therapies.

With complications in the evaluation of OS, PFS has rapidly gained traction in many settings as a key endpoint in phase 3 evaluation. Much has been written in the literature over the last few years about the methodological challenges in both the assessment and statistical analysis of PFS. At the core of this discussion is a debate around whether a treatment effect on PFS represents clinical benefit in and of itself or whether PFS is a surrogate for OS. Demonstrating the surrogacy of PFS has been accomplished in some histologies but in general has remained challenging. There has been growing acceptance of the PFS as an endpoint which measures the clinical benefit directly. A large treatment effect in terms of PFS strengthens the argument of clinical benefit. However, many methodologic challenges remain. These include optimizing the interval for assessment and scanning frequency, handling dropouts and missing data, and the need and value of a blinded central independent review. The recent introduction of new immuno-oncology therapies has also created new challenges in the evaluation of PFS.

The endpoints and associated trial designs in oncology have been responsible for many successful development programs in oncology, but often times registration has not been with the optimal dose regimen. Dose selection remains a critical challenge in the development of new oncology therapeutics. Unfortunately, the traditional paradigm developed for cytotoxic drugs of identifying the maximum tolerated dose (MTD) and subsequent p2 and 3 doses based on dose-limiting toxicities remains entrenched as a key feature of early development in many of the oncology programs. Over the last few years, this approach has led to several undesirable outcomes, including multiple sponsors having to initiate post-marketing commitments to further refine the dosing regimen. The oncology literature is rich with various dose escalation schemes for identifying the MTD. Many of these methods incorporate statistical modeling and formal use of historical data via Bayesian methods. There is hopefully little controversy in the notion that statistical modeling and Bayesian methods should be at the core of any dose-selection strategy. The current opportunity is to parlay these methods into more robust p2 trial designs that allow for a more informed evaluation and differentiation of doses based on risk and benefit.

In 2015, the United States government launched the precision medicine initiative, validating many of the novel concepts incorporated into the I-SPY2 trial 5 years earlier. The mission of this initiative is “To enable a new era of medicine through research, technology, and policies that empower patients, researchers, and providers to work together toward development of individualized care.” For many years now oncology has been at the forefront of precision medicine, with many examples now of targeted therapies approved based on the individual molecular and genetic profile of the tumor. The mission of the precision medicine initiative has the potential to open tremendous opportunities for statistical innovation in oncology. The evaluation of medicines in the context of precision medicines requires a new approach to trial design in early and late phases of drug development. It also necessitates consideration of companion diagnostics, and there are many other nontraditional statistical methods that could be considered. Such initiatives leave oncology and its researchers poised to break down the doors and enter a new and exciting era of drug development, an era that will undoubtedly be characterized by many bold statistical innovations.

The following chapters present a comprehensive review of many of the important statistical aspects of oncology drug development, many of which directly address the challenges described earlier. The first five chapters focus on early development including phase 1 and phase 2 trials, model-based approaches, and biomarker-based approaches. The remaining chapters focus on later phase development including phase 3 trials, quality of life, risk benefit, and regulatory challenges.

We hope this book proves useful and provides a comprehensive treatment for all relevant statistical issues in oncology.

 

 

References

1. Barker, A., Sigman, C., Kelloff, G., Hylton, N., Berry, D., and Esserman, L. (2009), I‐SPY 2: An adaptive breast cancer trial design in the setting of neoadjuvant chemotherapy. Clinical Pharmacology & Therapeutics, 86: 97–100.

2. Eisenhauer, E.A., Therasse, P., Bogaerts, J., Schwartz, L.H., Sargent, D., Ford, R., Dancey, J., Arbuck, S., Gwyther, S., Mooney, M., Rubinstein, L., Shankar, L., Dodd, L., Kaplan, R., Lacombe, D., Verweij, J. (2009), New response evaluation criteria in solid tumours: Revised RECIST guideline (version 1.1). European Journal of Cancer, 45(2): 228–247.

2

Statistical Considerations in Phase I Oncology Trials

Simon Wandel

Novartis Pharma AG

Satrajit Roychoudhury

Pfizer Inc.

CONTENTS

2.1 Introduction

2.2 Dose Escalation in Phase I Studies

2.2.1 Operational Aspects

2.2.2 Statistical Aspects

2.2.3 Logistic Model for Single-Agent Escalation

2.2.4 Meta-Analytic-Combined Model for Single-Agent Escalation with Co-Data

2.2.5 Assessing Effective Sample Size

2.2.6 Case Study: Ceritinib for Non–Small Cell Lung Cancer

2.2.6.1 Western Dose Escalation

2.2.6.2 Japanese Dose Escalation

2.3 Dose Expansion in Phase I Studies

2.3.1 Dose Expansion for Signal Seeking

2.3.1.1 Safety Endpoints

2.3.1.2 Efficacy and Biomarker Endpoints

2.3.2 Dose Expansion with a Formal Success Criterion for Efficacy

2.3.2.1 Double Criterion Design

2.3.2.2 Single-Arm Design with Indirect Comparison to Comparator

2.3.3 Advanced Designs: Fully Hierarchical Models

2.4 Conclusion

Acknowledgments

Appendix

References

2.1 Introduction

Phase I trials in oncology aim at identifying a maximum tolerated dose (MTD) or a recommended phase II dose (RP2D) of a new anticancer therapy. An important difference between phase I trials in oncology and first-in-human (FIH) studies in other therapeutic areas is the study population—outside oncology, healthy volunteers are studied, whereas in oncology, terminally ill patients are enrolled. This difference originates from the fact that most oncology drugs can cause (eventually severe) side effects and it would be unethical to expose healthy volunteers to these drugs. However, they can be the last option for patients with end-stage, non-treatable cancer, who may be willing to accept some level of toxicity. The dilemma when conducting oncology phase I studies is, therefore, the following—escalation should happen quickly to reach a potentially efficacious dose, while overly aggressive dose increments could expose patients to unacceptable toxicities. A good phase I study design needs to address this dilemma prospectively, and we will discuss the corresponding challenges and potential solutions in Section 2.2.

Another challenge during dose escalation is the limited number of patients who are evaluated at the most interesting dose, that is, the MTD/RP2D. Typically, only one or two cohorts (around 6 to 12 patients) are studied at this dose. Furthermore, the patient population is often heterogeneous, which makes it difficult to interpret potential efficacy signals. Therefore, most phase I studies enter an expansion phase once the MTD/RP2D is declared, during which more patients are investigated. In recent years, the expansion phase has attracted particular interest in the phase I community (Manji et al., 2013), and it has prov...