Skip to content

Advertisement

  • Poster presentation
  • Open Access

Revisiting the multi-armed bandit model for the optimal design of clinical trials: benefits and drawbacks

  • 1, 2,
  • 1 and
  • 1
Trials201314 (Suppl 1) :P36

https://doi.org/10.1186/1745-6215-14-S1-P36

  • Published:

Keywords

  • Experimental Treatment
  • Randomise Clinical Trial
  • Significant Treatment Effect
  • Average Proportion
  • Entire Duration

In a traditional randomised clinical trial, patients are allocated to an experimental treatment or the standard therapy arm - with equal probability - for its entire duration. However, using data from past patients to change allocation probabilities for future patients can increase the average number of patients who receive the best treatment. A drawback of doing this is that the power of the trial can be considerably reduced. The Bernoulli Multiarmed bandit problem (MABP) is an idealised model that illustrates such a conflict. For such model, Gittins & Jones (1974) provided a rule that maximises the expected patient benefit in a large trial.

Bandit models for clinical trials have been extensively studied in theory, yet the resulting schemes have been rarely used in practice. There are many reasons for this, however the power limitations to detect a significant treatment effect is a major drawback. In this presentation, we discuss allocating patients between several experimental treatments and a shared control group in a multi-arm trial, using the Gittins index. We modify the MABP design so that the shared control group is allocated separately to protect the power of the trial. The design provides considerable gains not only in efficiency over separate randomised trials but also in the average proportion of patients allocated to the best experimental treatment. This design can be advantageously extended for trials in rare diseases, where it becomes particularly relevant to optimize the choice of treatment both for patients in the trial and for those to treat after it.

Authors’ Affiliations

(1)
MRC Biostatistics Unit, Cambridge, UK
(2)
Mathematics and Statistics Department, Lancaster University, Lancaster, UK

Copyright

© Villar et al; licensee BioMed Central Ltd. 2013

This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Comments

By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate. Please note that comments may be removed without notice if they are flagged by another user or do not comply with our community guidelines.

Advertisement