Practical Bayesian design and analysis for drug and device clinical trials

Brian P. Hobbs, Brad Carlin

Research output: Contribution to journalArticlepeer-review

44 Scopus citations


Perhaps the most valuable contribution of Bayesian methods to health care evaluation involves study design. Drug and medical device clinical trialists are increasingly confronted with data that feature complex correlation structures, and are costly and difficult to obtain. In such settings, Bayesian trial designs are attractive since they can incorporate historical data or information from published literature, thus saving time and expense and minimizing the number of subjects exposed to an inferior treatment. Bayesian designs can also adapt to unexpected changes in the protocol, and allow the investigator to explore the plausibility of various outcome scenarios before any patients are enrolled in the trial. Recently, the FDA Center for Devices has encouraged hierarchical Bayesian statistical approaches which allow for the incorporation of such valuable historical data into the design and analysis of new device trials. The practical application of these methods has only become feasible in the last decade due to advances in computing via Markov chain Monte Carlo (MCMC) methods, especially as implemented in the popular BUGS software package. In this paper we illustrate Bayesian analysis and sample size calculations using BRugs, a function for calling BUGS from R. We provide illustrations in two applied settings where incorporation of available historical information is crucial, one concerning an AIDS drug trial and the other a comparison of left ventricular assist devices (LVADs).

Original languageEnglish (US)
Pages (from-to)54-80
Number of pages27
JournalJournal of Biopharmaceutical Statistics
Issue number1
StatePublished - Jan 2008

Bibliographical note

Funding Information:
Brian P. Hobbs is Graduate Assistant and Bradley P. Carlin is Professor of Biostatistics and Mayo Professor in Public Health at the Division of Biostatistics, School of Public Health, 420 Delaware St. S.E., University of Minnesota, Minneapolis, MN, 55455. The work of both authors was supported in part by NIH grant 1-R01-CA95955. The authors thank Prof. Jim Neaton and Ms. Grace Peng for discussions and assistance that were essential to the completion of this project.


  • BRugs software
  • Historical controls
  • Interim monitoring
  • Markov chain Monte Carlo (MCMC)
  • Sample size calculations
  • WinBUGS software


Dive into the research topics of 'Practical Bayesian design and analysis for drug and device clinical trials'. Together they form a unique fingerprint.

Cite this