Chapter Objectives

In the previous chapter, we presented implementations of fundamental approaches to estimation of an optimal single decision treatment regime and its associated value. All of the methods previously discussed limit the search for an optimal regime to a restricted class that is specified either indirectly through posited models, e.g., outcome regression estimators, or directly and whose elements typically are characterized by a parameter of low dimension, e.g., value search estimators.

In this chapter, we present a weighted classification interpretation where regimes are likened to classifiers and thus can be potentially complex, involving high-dimensional parameterizations. The weighted classification estimators for $$d^{opt}$$ have been implemented in R package DynTxRegime. This package is freely available through the repository maintained by R, CRAN. We provide detailed examples of how to use DynTxRegime to estimate the optimal treatment regime.

Also covered in this chapter of the book are estimators that consider a restricted class that is deliberately chosen to comprise regimes characterized by decision rules that are easy to understand, interpret, and present to clinicians and patients. At this time, the implementation for decision lists has not been incorporated into DynTxRegime, but is planned for the near future. This site will be updated to include an example of this method at a later time.