[R-pkgs] New package ‘bcrm’ to implement Bayesian continuous reassessment method designs
michael.sweeting at mrc-bsu.cam.ac.uk
Fri Jan 13 18:23:08 CET 2012
Dear R users,
I am pleased to announce the release of a new packaged called `bcrm’
(version 0.1), now available on CRAN.
The package implements a wide range of Bayesian continuous reassessment
method (CRM) designs to be used in Phase I dose-escalation trials. The
package is fully documented and highlights include
• A choice of 1-parameter working models or the 2-parameter logistic
• Bayesian updating performed after outcomes from each cohort of
patients becomes available, using exact computation or MCMC methods (via
either BRugs or R2WinBUGS).
• Full functionality allowing a range prior distributions for the
model parameter(s) as well as calculation of standardised doses from
prior estimates of toxicity at each dose level
• The ability to choose the summary estimate of the posterior
distribution used to select the next dose (options are posterior mean or
plug-in mean toxicity or a quantile of the maximum tolerated dose (MTD)
distribution, as in an EWOC design).
• Specification of a modified (constrained) or unmodified CRM design.
• Varying the cohort size
• Plots of the data and estimated dose-toxicity curve after each
cohort has been recruited.
• Fully integrated simulation code, which returns operating
characteristics of the design as output. Alternatively, the user can
interactively run the design, specifying doses and outcomes as each new
cohort is recruited, as if conducting a trial.
In a future release I hope to allow full loss function based decisions
to be made, using toxicity intervals (areas of the posterior distribution).
As this package is new I would appreciate any feedback users have.
Dr Michael Sweeting
MRC Biostatistics Unit
Institute of Public Health
Tel: 01223 768257
Fax: 01223 760729
More information about the R-packages