Multiplicity Issues in Clinical Trials JSM 2010

From MultXpert

Jump to: navigation, search

Selected presentations given at Joint Statistical Meetings in Vancouver, Canada on August 1-5, 2010.

Topic-contributed presentations

Choice of multiple comparison procedure in two pivotal clinical trials for approval of a new pharmaceutical product: Power and aesthetics

Presented by Brian L. Wiens, Director, Biostatistics and Data Management, Alcon Laboratories (jointly with Alex Dmitrienko, Eli Lilly and Company).

We consider analysis of two identical pivotal trials with correlated multiple endpoints evaluated by the fixed sequence, weighted Holm or fallback procedure. For approval, at least one endpoint must be significant in both studies. Evaluation of the procedures as closed tests distinguishes which has better power in different situations. The fixed sequence procedure has the highest power to reject the first hypothesis and to reject all three hypotheses. The fixed sequence often has the highest chance of obtaining inconsistent results, which makes it less aesthetically pleasing. The weighted Holm and fallback procedures have very similar properties. Under some plausible assumptions, we find weights that maximize or minimize power differences between the fallback and weighted Holm procedure, and provide practical guidance on which is preferable.

Download slides.

A gatekeeping multiple comparison procedure based on the Hommel test for clinical trials with hierarchically ordered objectives

Presented by Thomas Brechenmacher, Dainippon Sumitomo Pharma (jointly with Jane Xu, Sepracor; Alex Dmitrienko, Eli Lilly and Company; Ajit C. Tamhane, Northwestern University)

When conducting clinical trials with hierarchically ordered objectives composed of multiple treatment comparisons, it is essential to use multiplicity adjustment methods to control the familywise error rate in the strong sense while taking into account the hierarchical structure of the hypotheses. To address this issue, a gatekeeping procedure based on the Hommel test is proposed and is shown to yield more statistical power compared to similar methods already available in the literature (Dmitrienko Tamhane 2010). A general description of the procedure is given and details are presented on how it can be applied to complex clinical trial designs by setting up highly flexible logical restrictions. Two clinical trial examples are given to illustrate the methods.

Download slides.

On optimal grouping strategies using parallel gatekeeping procedures

Presented by Haiyuan Zhu, Associate Director, Biostatistics, Forest Research Institute.

Multiple hypotheses are common in clinical trials, and it is often required to control the Type I error rate across hypotheses. When the hypotheses are tiered to several families based on clinical importance (primary, secondary etc), serial or parallel gatekeeping procedures can used for multiplicity adjustment. When several hypotheses share the similar clinical importance, and therefore belong to the same family, one can use an appropriate multiple testing procedures (e.g. Hochberg). Concerns arise, however, when the number of the hypotheses in the family increase and some promising hypotheses may loose the chance of being tested. In this research, we will explore an alternative approach that divides hypotheses into ordered groups and tests them using parallel gatekeeping procedures. The power performances of various grouping strategies will be evaluated.

Download slides.

Testing a primary and a secondary endpoint in a group sequential design

Presented by Ajit C. Tamhane, Senior Associate Dean, McCormick School of Engineering, Professor of IEMS and Statistics, Northwestern University (jointly with Cyrus R. Mehta, Cytel; Lingyun Liu, Cytel).

We consider a clinical trial where the primary endpoint is a gatekeeper for the secondary endpoint. A two-stage group sequential test is used. The FWER of false significance on either endpoint is to be controlled at level a. The type I error rate for the primary endpoint is controlled by choosing any alpha-level boundary, e.g., the O'Brien-Fleming (OF) or the Pocock (PO) boundary. Given any alpha-level boundary for the primary endpoint, the boundary for the secondary endpoint to control the FWER is determined. We show that the FWER is maximized when the correlation coefficient rho between the two endpoints equals 1. The critical constants required to control the FWER are computed for different combinations of OF and PO boundaries and for different rhos. When rho is unknown, the upper confidence limit on rho from the first stage data is used to compute boundaries. A clinical trial example is provided.

Download slides.

Multiple testing problems with general logical restrictions in clinical trials

Presented by Alex Dmitrienko, Research Advisor, Global Statistical Sciences, Eli Lilly and Company (jointly with Ajit C. Tamhane, Northwestern University).

This presentation discusses multiple testing problems with general logical restrictions arising in clinical trials. The mixture-based framework can be used to construct gatekeeping procedures that account for complex logical relationships among multiple null hypotheses. These gatekeeping procedures control the familywise error rate in the strong sense and enable trial sponsors to set up powerful procedures based on p-value-based tests (eg, Hochberg-type or Hommel-type tests) or parametric tests (eg, regular and step-down Dunnett tests). The general methodology is illustrated using clinical trials with multiple dose-placebo comparisons and patient populations as well as dose-placebo comparisons and noninferiority/superiority objectives.

Download slides.

Adaptive enrichment designs based on the feedback test

Presented by Yan D. Zhao, Associate Professor, University of Texas Southwestern Medical Center (jointly with Alex Dmitrienko, Eli Lilly and Company; Martin Posch, Medical University of Vienna; Stephen Ruberg, Eli Lilly and Company)

In tailored therapies it is essential to select sensitive subgroups of patients who may benefit more from the study treatment than the overall population. In the design phase of the study several potential sensitive subgroups are pre-defined based on previously conducted studies. The objective of the study under consideration is to confirm or disregard these subgroups at pre-specified interim analyses or end of the study. In this talk we define adaptive enrichment designs based on the feedback test. The feedback test controls the Type I error level for multiple subgroup analyses and is more powerful than existing multiple tests. We illustrate this approach using a clinical trial example.

Download slides.

A multiplicity strategy incorporating fixed sequential, Hochberg and fallback procedures

Presented by Duane Snavely, Director, Biostatistics, Merck Research Laboratories (jointly with Kenneth Liu, Merck Research Laboratories).

A new drug may be simultaneously developed for two different indications within a patient population. Two sets of hypotheses (one for each indication) within a single study may include evaluation of multiple variables, time points or doses which contribute to the desired claim structure(s) for the new product. If strong type I error control is desired across these indications and factors, a number of multiplicity strategies may be conceived. We consider a combination of Bonferroni, fixed sequential, and Hochberg procedures with incorporation of a fallback procedure to increase the power for the second set of hypotheses that may be related to the indication of less importance or lower power. Simulations are used to demonstrate the type I error control and power implications of such a strategy compared to other viable alternatives.

Download slides.

Invited presentations

Presentations given at the invited session on key multiplicity issues in clinical trials organized by Alex Dmitrienko (Eli Lilly and Company) and chaired by Yan D. Zhao (UT Southwestern Medical Center).

Regulatory considerations for addressing multiplicity problems of clinical trials with multiple endpoints

Presented by Mohammad Huque, Director, Division of Biometrics IV, Office of Biostatistics, Office of the Translational Sciences, Center for Drug Evaluation and Research (CDER), U.S. Food and Drug Administration.

Confirmatory trials with multiple endpoints can raise multiplicity issues which might impact Type I error. Statistical approaches such as gatekeeping and tree-structured strategies are now available that can solve variety of multiplicity problems for multiple endpoints. However, applications of these approaches in the regulatory setting are often not straightforward as appropriate design and analysis should take into account the peculiarities of different diseases. Therefore, a regulatory guidance is needed. Such guidance would elaborate on basic multiplicity concepts and considerations that can be helpful in making sound regulatory decisions, and are useful in planning of trials and interpretation of results, and also in labeling and promotion of new beneficial treatments. This presentation will provide a list of key topics that can be included in such guidance with some discussions.

Download slides.

Industry perspective on the FDA guidance on multiplicity issues

Presented by Walter Offen, Senior Research Fellow, Global Statistical Sciences, Eli Lilly and Company.

There are numerous multiplicity issues in clinical research that need to be addressed at the protocol design stage. These issues relate to both efficacy and safety analyses. They pertain, for example, to interim analyses, multiple dose groups, multiple endpoints (primary or secondary), and subgroup analyses leading to tailored therapy indications and claims. Multiplicity issues are also very prevalent in so-called "substantial evidence" required by FDA DDMAC in order to allow promotion based on the findings. This talk will present an overview of all of these issues with proposed solutions, and will focus the majority of the discussion on key areas of the FDA Guidance that are potentially controversial or complex. The FDA draft guidance is not expected to become available until shortly before this session. Hence the specific key multiplicity topics will be identified later.

Variations of and types of statistical control for multiplicity in clinical trials: The academic perspective

Presented by Joseph Massaro, Associate Professor, Department of Biostatistics, Boston University School of Public Health (jointly with Ralph D'Agostino, Sr., Boston University School of Public Health).

Dealing with multiplicity and maintaining statistical control remains a major area of concern in randomized clinical trials. The Food and Drug Administration is in the process of publishing a guidance document. In the present talk we consider the many issues (e.g., primary outcome, co-primary outcomes, composite outcomes, subgroup analysis, interaction analyses and missing data issue). In each case we state the issue, why statistical control is essential, various approaches to achieve control and recommendations for dealing appropriately with the problem. Our objective is to focus on the issues as seen from the academics involved in the clinical trial and to anticipate how the new FDA guideline document may consider the issue.

Download slides.

Multiplicity in clinical trials: A European perspective

Presented by Norbert Benda, Group Head of Biostatistics, Federal Institute for Drugs and Medical Devices (BfArM), Germany.

Soon after the ICH E9 Biostatistical guideline has been released end of 1998, European regulators required further and more detailed information on several key methodological issues that were discussed. Among these issues were "role of meta-analysis in the licensing process", "non-inferiority", and "multiplicity issues". It was the general feeling that the guidance provided by the ICH E9 document should be supplemented in order to facilitate a more effective regulatory review of new submissions. Statisticians employed by European agencies have been rare even today. The supplementary guidances therefore also aimed at strengthening the methodological basis within the 16 national agencies. I will touch briefly the main topics from the CPMP Points-to-Consider document "Multiplicity Issues in Clinical Trials".

Download slides.