Dynamic Treatment Regimes

Statistical Methods for Precision Medicine

Anastasios A. Tsiatis, Marie Davidian, Shannon T. Holloway, and Eric B. Laber

This website is designed to provide detailed demonstrations of the application of selected methods presented in "Dynamic Treatment Regimes: Statistical Methods for Precision Medicine" by Anastasios A. Tsiatis, Marie Davidian, Shannon T. Holloway, and Eric B. Laber. These applications are meant to assist a reader who has studied the methods in detail with their implementation and with gaining proficiency with R package DynTxRegime, a comprehensive toolkit for the analysis of dynamic treatment regimes. We intend for this website to be "dynamic," with periodic updates and modifications as the package and methods evolve. We will post supplemental materials and other resources as they become available. We encourage readers to check this website often.


Last updated 05/15/2022 1515 ⓘ

Questions? Comments? Typos? Website Issues?

Please contact Shannon Holloway (shannon dot t dot holloway at gmail dot com)

Please check the Additional Resources section often for important updates, clarifications, and corrections
(last updated June 10, 2020).

Table of Contents

• Preparatory Material
  In this section, site conventions for formatting and R coding are are specified. In addition, a brief tutorial on R's modelObj package is provided. This package is the framework on which all regression based implementations provided in through the website are build. Readers should be comfortable with the basic tools of this package before proceeding.


• Chapter 2 Preliminaries
  In this chapter, methods for the estimation of the average causal treatment effect are implemented. Specifically, the naive, outcome regression, stratification, inverse propensity weighted, and doubly robust estimators are discussed as well as their standard errors.


• Chapter 3 Single Decision Treatment Regimes: Fundamentals
  In this chapter, methods for the estimation of the value of a fixed treatment regime are implemented. Specifically, the outcome regression, inverse propensity weighted, and augmented inverse propensity weighted estimators are discussed as well as their standard errors. In addition, the DynTxRegime implementations for estimating an optimal treatment regime using the outcome regression and value search methods are discussed.


• Chapter 4 Single Decision Treatment Regimes: Additional Methods
  In this chapter, the classification, outcome weighted learning, and residual weighted learning implementations provided in DynTxRegime are discussed.


• Chapter 5 Multiple Decision Treatment Regimes: Overview
  In this chapter, the inverse propensity weighted estimator for the value of a fixed treatment regime in the setting where there is more than one decision point is implemented. In addition, the DynTxRegime implementations for estimating an optimal treatment regime using Q-learning, value search, and backward outcome weighted learning are discussed.


• Chapter 6 Multiple Decision Treatment Regimes: Formal Framework
  In this chapter, methods for the estimation of the value of a fixed treatment regime in the setting where there is more than one decision point is implemented. Specifically, the outcome regression, inverse propensity weighted, and augmented inverse propensity weighted estimators are discussed.


• Chapter 7 Optimal Multiple Decision Treatment Regimes
  In this chapter, the DynTxRegime implementations for estimating an optimal treatment regime using Q-learning, value search, classification, and backward outcome weighted learning are discussed.


• Chapter 8 Regimes Based on Time-to-Event Outcomes
  This chapter is under construction. The DynTxRegime implementations for the methods discussed in the book for estimating an optimal treatment regime for time-to-event outcomes are under active development. Once this update is released, examples of their application will be provided.


• Additional Resources
  This space houses clarifications, updates, corrections, new papers, websites, etc. that the authors feel might be useful to readers.