Adaptive Design Lecture Series 2011
From Biopharmaceutical Network
Presentations given in the adaptive design lecture series in 2011. The series is chaired by chaired by José Pinheiro, Weili He and Judith Quinlan.
Back to Adaptive Design Lecture Series.
January 14, 2011
Andy Grieve (ClinResearch GmbH)
Dose Response in Early Phase Studies
There have been significant methodological advances in dose response studies in early phase over the last 15 years, but these advances have had limited impact in practice. In this talk I look at three examples -- First-in-Human studies, determining the MTD in oncology and Phase IIb dose selection studies. I look at the reasons that these advances have not had greater impact and suggest some consequences.
Download the slide set.
February 11, 2011
Björn Bornkamp (Novartis Pharma AG)
Response Adaptive Dose-Finding under Model Uncertainty using the DoseFinding R package
This talk will consist of two parts. In the first part we will discuss a methodology for designing response adaptive dose-finding studies, based on ideas from Bayesian statistics, optimal design theory and the MCPMod methodology. We will illustrate the method in a simulation study based on a case study in asthma. The second part will give an overview of the DoseFinding R package and will give hints on how to use it to implement the proposed design methodology.
Download the slide set.
March 11, 2011
Caroline Morgan (Cardinal Systems)
An adaptive confirmatory trial with interim treatment selection: practical experiences and unbalanced randomisation
This talk will focus on an ongoing international adaptive confirmatory trial. At the interim analysis of this two-stage trial, none, one or two active treatment regimens are selected for further study in the second stage. A combination test approach is used in this practical setting with an extension of the theory to unbalanced randomisation. We show that a combination test with suitable weights can still preserve the overall Type I error rate provided that the test statistic is chosen appropriately and the unspooled Z-test for proportions is not used. The accuracy of stagewise p-values is also discussed in a more general framework. Monte Carlo simulations confirm the validity of the approach retained and evaluate the necessary sample size. Additional issues addressed during the design of the trial are also examined such as multiplicity due to testing hypotheses on key secondary endpoints, a non-inferiority comparison to an active treatment and covariate adjusted analyses for various types of outcome.
Download the slide set.
April 08, 2011
Val Fedorov (Research Statistics Unit, GSK)
Dose-Finding Studies: To Treat or to Learn?
For decades statisticians have been contributing to creation and improvement of optimal experimental design methods in model fitting, hypothesis testing, parameter estimation and optimal control. In most of these areas these methods worked flawlessly. However in a few cases (mostly in optimal control) some of the intuitively attractive approaches failed. Clearly that in clinical trials the direct application or reinvention of similar methods may lead to unexpected and undesirable results. Medical ethics, huge expenses, and interactions with regulatory agencies call for more meticulous analysis of models and designs. Often the initial enthusiasm about seemingly ethically attractive or (mathematically) efficient designs quickly disappears after thorough mathematical analysis combined with Monte-Carlo simulations. I consider a few simple models to illuminate the above.
Download the slide set.
May 13, 2011
Christopher Coffey (University of Iowa)
Adaptive Design for Clinical Trials: Perspective From an NIH-Funded Workshop
Adaptive clinical trial designs provide the flexibility to make adjustments to aspects of the design of the trial based on data reviewed at interim stages. The potential for adaptive designs to improve clinical research has generated widespread interest in the biomedical research community, and a wide variety of adaptations have been proposed or implemented. Yet specific approaches have met with differing levels of support, and their use has been limited in publicly funded research. The Scientific Advances in Adaptive Clinical Trial Designs Workshop was organized to give participants from government agencies, industry, non-profit foundations, the patient advocacy community, and academia a forum to discuss the use of adaptive designs in publicly funded research. After addressing issues of adaptive designs that arise at the planning, design, and execution stages of clinical trials, participants set forth seven recommendations for guiding action to promote appropriate use of adaptive designs in this setting. In this presentation, I will summarize the recommendations from the workshop and discuss some of the logistical issues that must be addressed in order to achieve the promise of adaptive design in publicly funded research.
Download the slide set.
June 10, 2011
Kenneth Liu (Merck Research Laboratories)
Adaptive Dose Finding Using the “Maximizing Procedure”: Case Study and Missing Data Simulation
This is a case study of a recent clinical trial that combines proof-of-concept with dose finding. Proof-of-concept in clinical trials usually focuses on identifying a maximum tolerated dose (MTD) and assumes that higher doses provide better efficacy. However adverse events associated with an MTD may blunt efficacy. We present an adaptive dose-finding strategy which concentrates assignments at and around the dose which has the best efficacy/tolerability profile based on a utility function. The strategy is applied to a three period crossover design to improve power and allow for inclusion of an active control. Operating characteristics over various efficacy/tolerability scenarios and missing data mechanisms (MCAR, MAR, NMAR, and a mixture of the three) were studied via simulation. Advantages and disadvantages of this adaptive design versus traditional trial designs will be discussed.
References
Anastasia Ivanova, Ken Liu, Ellen Snyder, Duane Snavely. An Adaptive Design for Identifying the Dose with the Best Efficacy/Tolerability Profile with Application to a Crossover Dose-Finding Study. Statistics in Medicine, 2009, 28:2941-2951.
Kenneth Liu, Richard Entsuah. Missing Data Mechanisms In A Dose Finding Adaptive Trial. Journal of Biopharmaceutical Statistics (in press).
Download the slide set.
July 08, 2011
William Mietlowski (Novartis Pharmaceuticals)
Bayesian Adaptive Phase I Studies in Oncology -- a Call for Flexibility
While accurate targeting of the maximum tolerated dose (MTD) and long-run operating characteristics ("group ethics") are important considerations, FDA Oncologists have been concerned about some individual trials that appear to enroll an excessive number of patients at excessively toxic doses using the continuous reassessment method (CRM). This talk will describe an alternative approach, which has been used as a standard design for Phase I Oncology trials at Novartis since 2004, that relies on Bayesian logistic regression using the escalation with overdose control principle (EWOC) espoused by Babb et al (1998). The model, based on 1st cycle DLTs, identifies doses which are potentially unsafe as input to a dose-escalation teleconference with investigators and Novartis oncologists. A clinical synthesis of all relevant safety (non-DLTs and later cycle data), PK/PD, tumor response, and biomarker data is used to determine which dose among the model-identified "safe" doses should be given to the next cohort. We’ll discuss the importance of flexible cohort sizes and that one should not use the same algorithm to assign the number of patients at each dose level, but enrich at safe and active doses to learn more about activity, PK (especially if erratic), safety (especially if chronic toxicity, e.g. hepatotoxicity is a concern), potential predictive biomarkers, etc. Currently, not enough attention is paid to choosing the correct dose(s)/regimens for dose expansion with a strictly algorithmic approach; perhaps a more flexible approach, like the one discussed in the talk, will lead to better dose/regimen/patient population selection for Phase II.
Download the slide set.
August 12, 2011
Josh Chen (Merck Research Laboratories)
Some Alternative Designs for A Confirmatory Study with Uncertain Dose
One of the key objectives of a Phase II clinical program is to identify appropriate dose(s) for subsequent confirmatory studies. Traditionally, one single dose is selected for a Phase III confirmatory study at the end of dose-ranging Phase IIB. In some situations, this traditional paradigm could be inefficient operationally (due to gap between Phase II and III) and/or statistically (due to exclusion of Phase II patients in the final analysis). In some other situations, the traditional paradigm could be risky if an appropriate dose cannot be differentiated at the end of Phase II. In this talk, I will discuss some designs that may provide alternatives to the traditional paradigm under certain scenarios. In particular,
- Combined Phase IIB/III designs, if it is expected that a single "best" (or at least a "good") dose can be selected at the end of Ph II. Special considerations are given to life-threatening diseases such as HIV/AIDS and cancers where studied doses are generally high and likely located at the plateau of the dose-response curve. A "MiniPool" approach, which is an application of the closed testing principle, is particularly useful in these scenarios.
- "Drop-the-loser" designs, if at the end of Phase IIB choices are narrowed down to 2-3 doses but the team cannot pick one single good dose. A few candidate doses enter the confirmatory study and ineffective and/or toxic doses compared to the control may be dropped at the interim analyses as the study continues ("drop-the-loser"). The study may be stopped once the accumulated data have demonstrated convincing efficacy and an acceptable safety profile for one or more doses. Several "drop-the-loser" designs and their characteristics will be discussed.
A case study will be used to illustrate the considerations when choosing an appropriate study design for the confirmatory study.
Download the slide set.
October 14, 2011
Carl-Fredrik Burman (AstraZeneca R&D) and Fredrik Öhrn (AstraZeneca R&D)
The Adaptive Programs Work Stream: Joint Optimization of Phase II and Phase III
Large efforts have been made to find innovative designs of single clinical trials. A natural extension is to use decision analysis to design a development program, including multiple trials. Therefore, the Adaptive Design Scientific Working Group (ADSWG) has initiated a work stream on Adaptive Programs (AP). The AP work stream has focused on both generic methodology as well as case studies in different therapeutic areas: neuropathic pain, oncology and diabetes.
In this presentation, we will in some detail consider the joint optimization of Phase II and Phase III, assuming one active dose and one comparator in both phases. The main purpose of Phase II is then to inform the Phase III go/no go decision and guide the Phase III sample size. Since the long-term Phase III endpoint often cannot be measured in Phase II, we use a model that quantifies the uncertainty about the Phase III endpoint after having observed a biomarker in a Phase II trial. Using this model we can for a particular biomarker assess how much information should be collected in Phase II to maximize the expected utility. We also consider how properties change when Phase III is made group sequential and when multiple doses are used in Phase II.
Download the slide set.
November 11, 2011
Bo Huang (Oncology Biostatistics, Pfizer)
An Adaptive Oncology Dose Finding Trial Using the Time-to-Event Continual Reassessment Method Incorporating Cycle Information
The standard 3+3 design in oncology phase I trials has a less rigorous statistical basis than model-based designs described in the more recent literature, such as the continual reassessment method (CRM) and the method of escalation with overdose control (EWOC). Both types of designs, however, are subject to stop and go in patient accrual due to staggered patient entry to the trial, and as such discourage enrollment and can result in prolonged trial duration. In addition, data of patients with early drop-out not due to safety reasons are excluded from analysis and decision making.
An oncology phase I dose-finding study is presented and discussed, where the standard of care is combined with a novel regimen. After evaluating the biologic mechanism and data generated from toxicology studies, the dose limiting toxicity (DLT) observation window was set as 9 weeks to estimate the Maximum Tolerated Dose (MTD). To address the issues with long observation time and patient drop-out, we employ the time-to-event continual reassessment method (TITE-CRM) (Cheung and Chappell, 2000), a Bayesian dose-finding design incorporating information not only from patients observed for the entire observation period but also from patients observed for less than the full observation period. TITE-CRM uses a weighted binomial likelihood with weights assigned to observations by the unknown time to toxicity distribution, and is open to accrual continually. To avoid dosing at overly toxic levels while retaining accuracy and efficiency, we propose an alternative adaptive weight function by incorporating cyclical data with parameters updated continually. This provides a reasonable estimate for the time to toxicity distribution by accounting for inter-cycle variability and maintains the same statistical property of consistency and coherence as the method described by Cheung and Chappell.
Design calibrations for the clinical and statistical parameters are conducted to ensure good operating characteristics. Simulation results show the proposed TITE-CRM designs with adaptive weight function are significantly shorter, maintain advantages of the CRM relative to the 3+3 design, and do not expose patients to significant additional risk. We will also share some early experience in implementing the design by coordinating with other functional lines in the study team.
Download the slide set.
December 09, 2011
Cong Chen (Merck Research Laboratories)
Optimal GNG decision rules for an adaptive seamless Phase II/III oncology trial
The Go-No Go (GNG) decision from Phase II to Phase III in a seamless design often has to be made with limited clinical endpoint data. The difficulty to pre-specify such a decision rule is arguably the single most important reason why seamless designs are less used in practice than expected. In this presentation, we'd like to address the following issues: 1) how to effectively incorporate surrogate biomarker data into the decision matrix; 2) how to derive objective GNG bars from a benefit-cost ratio perspective; 3) how to fully realize the potential of a seamless desgn with proper risk mitigation. Our work is based on a real example in the oncology therapeutic area. However, the general approach is equally applicable to various other areas.
Joint work with Linda Sun (Merck Research Laboratories ).
Download the slide set.