Shourt Courses




top

Gatekeeping procedures in clinical trials

Alex Dmitrienko (Eli Lilly and Company)
Ajit Tamhane (Northwestern University)

This half-day course will focus on issues arising in clinical trials with ordered multiple objectives (e.g., primary and secondary endpoints, primary and secondary patient populations, etc.) using gatekeeping procedures. The course will provide a detailed overview of novel statistical approaches developed over the past five years. There will be a well-balanced coverage of theory and applications, regulatory considerations and software implementation of gatekeeping procedures in SAS and R. Examples from clinical trials will be used throughout the discussion to illustrate the statistical approaches discussed in the course.


top

Graphical approaches to multiple test problems

Frank Bretz
Ekkehard Glimm
Willi Maurer (Novartis)

Methods for addressing multiplicity are becoming increasingly more important in clinical trials. In the recent past several multiple test procedures have been developed that allow one to map the relative importance of different study objectives as well as their relation onto an appropriately tailored multiple test procedure, such as fixed-sequence, fallback, and gatekeeping procedures. In this course we focus on graphical approaches that can be applied to common multiple test problems, such as comparing several treatments with a control, assessing the benefit of a new drug for more than one endpoint, and combined non-inferiority and superiority testing. Using graphical approaches, one can easily construct and explore different test strategies and thus tailor the test procedure to the given study objectives. The resulting multiple test procedures are represented by directed, weighted graphs, where each node corresponds to an elementary hypothesis, together with a simple algorithm to generate such graphs while sequentially testing the individual hypotheses. The class of procedures covered in this course include weighted Bonferroni tests, weighted parametric tests accounting for the correlation between the test statistics, and weighted Simes' tests. The approach is illustrated with the visualization of several common gatekeeping strategies. We also present several case studies to illustrate how the approach can be used in cinical practice. In addition, we briefly consider power and sample size calculation to optimize a multiple test procedure for given study objectives. The presented methods will be illustrated using the graphical user interface from the gMCP package in R, which is freely available on CRAN.


top

Multiple Comparisons in Complex Clinical Trial Designs

H.M. James Hung
Sue-Jane Wang (FDA)

In the last decade, the methodology for clinical development became increasingly complex. For example, active control designs are increasingly used. In a clinical program for assessing cardiovascular risks, multiple trials may be combined to assess a mortality endpoint while each trial is planned to assess a different endpoint. In many of such cases, particularly under regulatory application settings, the statistical framework of inference and the conventional notion of experimentwise type I error are often unclear. This short course will be devoted to emerging multiplicity issues in a number of complex design settings including flexible design strategies and suggest a number of approaches. Case examples will be presented to facilitate discussion. The topics to be covered include:
1. Family of combinable clinical trials
Multiple doses
Multiple endpoints
Composite endpoint
2. Active-controlled trials
With or without a placebo arm
Non-inferiority and superiority analyses
3. Adaptive or flexible design trials
Early phase trials
Pivotal trials



top

Simultaneous confidence bands in regression

Wei Liu, University of Southampton

Using simultaneous confidence bands to bound an unknown function or the differences between unknown functions is a direct generalization of using confidence intervals to bound an unknown parameter or the differences between unknown parameters. Simultaneous confidence bands are intuitive, informative and can make useful statistical inferences (such as one-sided comparison of two models) that are beyond the standard statistical methods. This course will focus on the construction and applications of simultaneous confidence bands for various inferential purposes in parametric regression analysis, including linear and generalized linear regression models. It aims to provide an overview of the methods available, and demonstrates with examples the implementation of the methods using computer software MATLAB.


top

Adaptive Designs for Clinical Trials

Martin Posch (European Medicines Agency, London), Franz Koenig (Medical University of Vienna)

Adaptive designs allow for mid-course design adaptations based on interim data without compromising the overall type I error rate. Examples of design adaptations are the adjustment of sample sizes or the number and timing of interim analyses. These design parameters may be adapted depending on interim estimates of the variance, the treatment effect and safety parameters. An important field of application of the adaptive design methodology are clinical trials with several treatment arms, where promising treatments can be selected at an interim analysis. Using adaptive multiple test procedures the type I error rate can be controlled even if the selection rule or the number of selected treatments is not prefixed. Adaptive multiple testing procedures can also be used in adaptive designs with the option of population enrichment. In such designs a sub population may be selected in an interim analysis and further recruitment of patients is restricted to the selected subgroup.
The course provides an overview of methods from the published literature including the most recent developments. Special emphasis is put on sample size adjustment and multiple hypotheses testing with adaptive designs. Furthermore, regulatory issues will be discussed.


top

Concepts and Techniques of Multiple Testing in Clinical and Biomarker Studies

Jason C. Hsu, The Ohio State University

This course covers fundamental concepts and techniques of multiple testing.
Error rates:
  • FWER (Familywise Error Rate)
  • gFWER (generalized Familywise Error Rate)
  • FDR, Fdr (False Discovery Rate)
How incorrect decision-making rate is impacted by issues of true, average, or worst case scenario control, number or proportion of incorrect rejections control, conditional or unconditional error rate, tail probability or expectation error rate, will be discussed.
Techniques:
  • Closed Testing
  • Partition Testing
Holm's and Hochberg's tests, both simple examples of partition testing, ignore joint distribution in computing critical values. Using testing for efficacy in genomic subgroup as an example, it will be shown that when joint distribution is taken into account, step-down methods are often more powerful than Hochberg's step-up method.
Applications:
  • Multiple endpoints
  • Bioinformatics
Partition Testing and Closed (Gatekeeping) Testing will be compared in the setting of Multiple Endpoints, in terms of simplicity/complexity and power. In the setting of association studies between biomarkers and drug response, we will discuss assumption needed on the joint distributions of gene expression levels or SNP alleles for permutation tests to be valid.