Pre-conference workshops – September 2nd, 2015
Short course 1 and 2 are from 9:00 am – 12:30 pm
Short course 3 and 4 are from 1:30 am – 5:00 pm
Topics for the four half-day short courses:
- Fundamental Concepts & Techniques of Multiple Testing
(Frank Bretz & Vishwanath Iyer (Mahesh))
- Dose Finding and MCP-Mod Methodology
- Trial Designs with Multiple Treatments and Multiple Endpoints Using East®
(Lingyun Liu, Ajit Tamhane and Vidyadhar Phadke)
- Selective Inference in Genomics and Elsewhere
(Yoav Benjamini, Daniel Yekutieli and Saumyadipta Pyne)
Frank Bretz, Novartis Pharma AG, Basel, Switzerland
Vishwanath Iyer (Mahesh), Novartis Oncology, Hyderabad, India
Abstract: Multiple testing techniques include the traditional methods by Tukey and Dunnett, and modern approaches such as gate-keeping and graphical methods. Error rate concepts include the traditional familywise error rate of Tukey, and the more recent false discovery rate of Benjamini and Hochberg. This half-day course will connect fundamental concepts and techniques using the partitioning principle (PP), and demonstrate how to use graphical approaches (GA) to flexibly design and implement multiple tests in both simple and complex situations. Specifics to be covered are as follows.
- Dunnett’s, Holm’s, and Hochberg’s methods for simple scenarios (such as which doses are efficacious for a single endpoint) will be constructed based on PP
- We then indicate how to construct methods for complex scenarios (such as which doses are efficacious for multiple ordered endpoints) using PP, and show they can be readily and flexibly implemented using GA.
- Finally, while this course concentrates on familywise error rate control, we indicate how a decision-theoretic approach (assigning different losses to errors with different endpoints), and error rate calculations (conditional on evidence of efficacy in the primary endpoints, for example), are ideas worthy of consideration.
Bjoern Bornkam, Novartis Pharma AG
Abstract: Appropriate information on the dose-response relationship (efficacy and safety) at the end of Phase II in pharmaceutical development is key to determine the dose(s) used in Phase III clinical trials.
To enable informed decision making appropriate trials need to be conducted in Phase II, and statistical analysis methods should be aligned to the objectives in these studies.
Multiple testing and dose-response regression both have a complementary roles in this regard. Both type of approaches will be discussed in this short course as well as a hybrid approach MCP-Mod Multiple Comparisons & Modeling). The twist of MCP-Mod is that it acknowledges uncertainty about the dose response model both at design and analysis stage. The DoseFinding R package implementing the MCP-Mod approach will be presented and illustrated based on examples.
Lingyun Liu, Cytel Inc.
Ajit C. Tamhane, Northwestern University
Vidyadhar Phadke, Cytel, Pune
Abstract: Modern clinical trials are often designed to address multiple clinical questions which need multiplicity adjustments to ensure strong type I error control. Commonly encountered sources of multiplicities include multiple treatments, multiple endpoints, interim analyses and subgroup analyses. This workshop will cover the most common multiplicity problems in trials designed to compare multiple dose/treatment arms to a common control arm or to compare two treatment arms with respect to multiple endpoints. We will also cover adaptive multi-arm multi-stage group sequential designs with flexible interim analyses at which ineffective or unsafe arms can be dropped and sample sizes can be re-estimated to ensure a well powered study while strongly controlling the type I error rate at nominal level. Using clinical trial examples, we will demonstrate the use of the industry standard software package East® to illustrate a variety of multiple comparison procedures, including Dunnett, Bonferroni, Holm, Hochberg, Hommel, Fixed Sequence and Fallback tests, which are commonly used in clinical trials. After introducing these basic methods we will cover gatekeeping methods, which are used to address hierarchical multiplicity problems in more complex settings. We will also demonstrate how to design multi-arm multi-stage group sequential trials with the flexibilities to drop ineffective or unsafe arms and to re-estimate sample sizes at interim looks. The methods will be illustrated with the help of the East software for group sequential and adaptive designs
Yoav Benjamini, Dep. of Statistics and Operations Research, Tel Aviv University
Saumyadipta Pyne, C R Rao AIMSCS, Hyderabad, India
Abstract: Drawing inference on a selected subset of the parameters, a subset that is selected because the parameters within seem interesting after viewing the data, is the domain of selective inference. We shall illustrate with examples that selective inference is both common and unavoidable in genomics as well as in modern scientific research in general, yet it is an acute problem that if unattended hampers the replicability of discoveries. We shall discuss the ‘ average over the selected’ approach to this problem, embodied in the False Discovery Rate (FDR) and False Coverage-statement Rate (FCR) criteria, and address the why, when and how needed to use them. In particular we shall describe post selection testing, confidence intervals, and modelling in simple and complex inference challenges. We shall also discuss the Bayes and Empirical Bayes approaches to selective inference.