An R Toolkit for Dynamic Treatment Regimes
Shannon Holloway
Shannon
Holloway

IMPACT is excited to announce the first public release of its dynamic treatment regime toolkit, DynTxRegime. Developed for the R Statistical Computing Environment, the package includes several of the dynamic treatment regime analysis tools developed by IMPACT investigators. Currently, DynTxRegime provides implementations of Q-Learning, Interactive Q-Learning, and value-search estimators of optimal treatment regimes from a missing data perspective, from a classification perspective, and from a coarsened data perspective. Several additional methods including Outcome Weight Learning, Backwards Outcome Weighted Learning, and Residual Weighted Learning are actively under development and will be released in version 2.0, which is expected out in early 2016.

It is our goal that DynTxRegime become the foundation of a comprehensive software package dedicated to the study of dynamic treatment regimes. As such, great effort has been made to develop the toolkit using a simple interface and infrastructure with few artificial limitations imposed by developer choices. For example, all of the statistical methods currently implemented in DynTxRegime rely on at least one postulated regression model; these regression models can be linear or non-linear. To maximize flexibility, we made use of R package modelObj; this affords users the freedom to completely define each regression step of an analysis.

Of the methods implemented in DynTxRegime, only Q-Learning and Interactive Q-Learning have publicly available software. For example, these methods are available through R packages qLearn and iqLearn, respectively. The DynTxRegime implementation of Q-Learning is more general than qLearn in that regression methods are not limited to linear models and more than two decision points can be analyzed. The capabilities of iqLearn and the DynTxRegime implementation of Interactive Q-Learning are very similar, differing only in the allowed model structures for the regression analyses; iqLearn considers only linear models.

Other capabilities of DynTxRegime include the ability to: specify responder/non-responder status or feasible treatment sets; separate outcome models into main effect and contrast components, where an individual component can be linear or non-linear and the choice for one component does not dictate the choice for the other; specify models based on patient subsets; and select Augmented Inverse Probability Weighted estimators or Inverse Probability Weighted estimators. Because all methods use a common vocabulary and similar input structure, users can transition from one method to another with few changes and results can be compared easily.

DynTxRegime v1.0 is available from the R CRAN repository. We hope that you give it a try, and we welcome your feedback and suggestions!


Comments/Discussion
Submit a question or comment.
Welcome to the IMPACT Blog!

Each month, one of our program investigators will introduce him/herself and will discuss their research, new research directions, or advancements made toward our goal of improving the clinical trial process. Readers are encouraged to send questions or comments. In addition, we will announce new software releases, publications, and upcoming events.

Subscribe

Archive
blog icon
By: Xiaofei Wang and Jianwen Cai
Date: February 15, 2016
blog icon
By: Michael Kosorok
Date: December 16, 2015
blog icon
By: Shannon Holloway
Date: November 24, 2015
blog icon
By: Donglin Zeng
Date: October 15, 2015
blog icon
By: Alison Motsinger-Reif
Date: September 17, 2015
blog icon
By: Shannon Holloway
Date: August 24, 2015
blog icon
By: Michael R. Kosorok, Marie Davidian, and Kouros Owzar
Date: June 4, 2015