Bayesian Adaptive Methods for Clinical Trials

Transcription

1 CENTER FOR QUANTITATIVE METHODS Department of Biostatistics COURSE: Bayesian Adaptive Methods for Clinical Trials Bradley P. Carlin (Division of Biostatistics, University of Minnesota) and Laura A. Hatfield (Dept. of Health Care Policy, Harvard Medical School) 8-9 October 2012 Location: TBA Room: TBA Deadline for registration: 15 September For registration and further practical information, please contact Eline Van Gent Department of Biostatistics, Erasmus MC, Dr. Molewaterplein 50, 3015 GE Rotterdam, Ee 2124 (21 st floor), tel:

2 Abstract Thanks in large part to the rapid development of Markov chain Monte Carlo (MCMC) methods and software for their implementation, Bayesian methods have become ubiquitous in modern biostatistical analysis. In submissions to regulatory agencies where data on new drugs or medical devices are often scanty but researchers typically have access to large historical databases, Bayesian methods have emerged as particularly helpful in combining disparate sources of information while maintaining traditional frequents protections regarding Type I error and power. Biostatisticians in earlier phases (especially in oncology trials) have long appreciated Bayes ability to get good answers quickly. Finally, an increasing desire for adaptability in clinical trials (to react to trial knowledge as it accumulates) has also led to heightened interest in Bayesian methods. This two-day course begins with a brief introduction to Bayesian methods, computing, and software, and then goes on to elucidate their use in safety and efficacy trials of drugs and medical devices. In particular, we will illustrate how simulation may be used to calibrate a Bayesian procedure to guarantee good long-run frequents performance (i.e., low Type I and II error rates), a subject of keen ongoing interest to regulators. We also discuss Bayesian meta-analysis, and the closely related area of appropriate use of historical information in clinical trials. In terms of software, we will focus on BUGS and its variants (like Open BUGS and JAGS) as well as methods for calling it from R (such as R2WinBUGS, BRugs, and rjags), but also mention more recent developments, including both MCMC-based (e.g., PyMC, STAN/NUTS) and non-mcmc-based (INLA, ABC) alternatives. Students will be expected to bring their own laptop computers to the session, and to have recent versions of BUGS and R already installed on these computers. All necessary computer code will be provided during the two afternoon computer lab sessions, and the presence of two instructors will ensure adequate assistance for those who become lost during the presentation.

3 Required textbook: Berry, S.M., Carlin, B.P., Lee, J.J., and Muller, P. (2011). Bayesian Adaptive Methods for Clinical Trials, Boca Raton, FL: Chapman and Hall/CRC Press. (ISBN ) Options for obtaining the book include: 1. Buy it from the publisher using the promotion code 194CM on the CRC website to gain a 20% discount: The priced listed here is $94.95, but with the 20% discount this drops to $ Buy it direct from amazon.com: Cost using this method is currently $72.73, slightly cheaper (though this could change). Of course, other vendors and options (say, purchasing a used copy) may be even less expensive, but please be aware that we strongly suggest that each participant obtain his or her own copy of the book. Necessary background for the course Short course participants should be familiar with standard probability distributions (including normal, gamma, binomial, Poisson, beta, and multivariate normal) and statistical techniques (including linear regression, logistic regression, and the basics of generalized linear models). We will not assume significant previous experience with Bayesian data analysis or computing, although students with basic knowledge of these areas will certainly face a gentler learning curve. Chapter 2 of the textbook provides a quick review of Bayesian methods and computing for those who require it.

4 Target Audience Professional statisticians working in applied environments where hierarchical modelling and clinical trials are key issues, where flexible and adaptive methods are important, and where well justified approaches are needed for making informed inferences; this would include biostatisticians working in the pharmaceutical industry, regulatory agencies, or academic centers running large clinical trials Researchers and students from epidemiology, biostatistics, statistics, health services research, and related fields who are interested in gaining exposure to Bayesian adaptive methodology, either as a solution to their own challenging applied problems, or as an area of research in its own right

5 About the course instructors Brad Carlin is Mayo Professor in Public Health and Professor and Head of the Division of Biostatistics at the University of Minnesota. He has published more than 130 papers in refereed books and journals, and supervised or co-supervised 19 PhD dissertations to completion. He is co-author of three popular textbooks in Bayesian methods, the most recent of which ("Bayesian Adaptive Methods for Clinical Trials" with Scott Berry, J. Jack Lee, and Peter Muller) forms the primary text for this course. He is a winner of the Mortimer Spiegelman Award from the APHA, and from served as editor-in-chief of Bayesian Analysis, the official journal of the International Society for Bayesian Analysis (ISBA). Prof. Carlin has extensive experience teaching short courses and tutorials, and has won teaching awards from both the faculty and the graduate students at the University of Minnesota, as well as from the Joint Statistical Meetings CE program. During his spare time, Brad is a musician and bandleader, providing keyboards and vocals in a variety of venues, some of the more interesting of which are visible by typing the phrase "Bayesian cabaret" into the search window at YouTube. Laura Hatfield is an Assistant Professor of Health Care Policy, with a specialty in biostatistics, in the Department of Health Care Policy at Harvard Medical School. Dr. Hatfield's primary research interest is developing hierarchical Bayesian methods for analysis of complex health data including longitudinal patient-reported outcomes and survival in oncology clinical trials, consumer ratings of healthcare quality, and post-market surveillance of medical devices. Laura earned her BS in Genetics from Iowa State University and her MS and PhD in Biostatistics from the University of Minnesota. She was primary author of a complete solutions manual for Carlin & Louis' "Bayesian Methods for Data Analysis" (3rd ed), won several research and teaching awards at Minnesota, and co-taught or assisted short courses in Bayesian methods for the Biometric Society (ENAR), the New England Statistics Symposium, the Centers for Disease Control, and Yale University.

6 COURSE OUTLINE Day 1 9:00 10:45 Introduction: Motivation and overview, potential advantages, key differences with traditional methods (Ch 1, BCLM text) 10:45-11:00 Coffee break Review of Bayesian inference: prior determination, point and interval estimation, hypothesis testing, prediction, model choice (Sec BCLM) 11:00-12:30 Review of Bayesian computation: Markov chain Monte Carlo (MCMC) methods, Gibbs sampling, extensions (Sec 2.3 BCLM) 12:30-13:30 Lunch Basics of Bayesian clinical trial design: range of equivalence, community of priors, available software (Sec 2.5 BCLM) 13:30-15:00 Hierarchical modeling case studies: Hierarchical modeling and meta-analysis; survival analysis; joint longitudinal-survival modeling MCMC software options: R2WinBUGS, BRugs, mcmc, JAGS, and PyMC, extending WinBUGS to new distributions, simulation studies

7 15:00-15:15 Coffee break 15:15-17:00 Computer Lab session I: Demo and hands-on experimentation with R and WinBUGS for various models (simple failure rate; linear, nonlinear, and logistic regression, metaanalysis, survival modeling, joint longitudinal and survival modeling, etc.)

8 Day 2 9:00 11:00 Bayesian adaptive methods for safety studies: Rule- (3+3) and model-based (CRM, EWOC, TITE-CRM) designs for determining the maximum tolerated dose (MTD), efficacy and toxicity, combination therapy (Ch 3 BCLM) 11:00-11:30 Coffee break 11:30 12:30 Bayesian adaptive methods for efficacy studies: Standard designs, predictive probability-based methods, sequential stopping for futility, efficacy, and/or toxicity, adaptive randomization, biomarker-based adaptive designs (Ch 4 BCLM) 12:30-13:30 Lunch 13:30-15:00 Special topics: Methods for device trials, use of historical data in clinical trials, false discovery rate (FDR) in safety studies, incorporation of surrogate endpoints (Ch 6 BCLM) 15:00-15:15 Coffee break 15:15-17:00 Computer Lab session II: Practice with advanced, relevant R, WinBUGS, and BRugs software, opportunity for participants to install and try the programs themselves in realdata settings with one-on-one instructor guidance, wrap-up discussion, floor discussion, Q&A period

9 Administrative information Coffee breaks 2 coffee breaks (one in morning and one in afternoon) are included in the registration costs. Lunch is not included. Course materials Copies of the slides used in the course are included in the registration costs. Also a website is available with some computer code. Costs Erasmus University: 250,-, other universities/governmental: 500,- commercial organizations: 1500,- Registration is only effective upon receipt of payment. To register please fill the registration form and send it, in word-format, to: Secretariat.Biostatistics@ErasmusMC.nl