Dynamic Treatment Regimes: Statistical Methods for Precision Medicine
Dynamic Treatment Regimes: Statistical Methods for Precision Medicine
复制标题
动态治疗方案：精准医学的统计方法
DOI：
--
复制
发表时间：
2022
影响因子：
3.7
通讯作者：
Ying‐Qi Zhao
中科院分区：
文献类型：
--
作者：
Ying‐Qi Zhao
动态治疗方案（DTR）用于管理慢性疾病，并很好地适应更大的精准医学范式。人们越来越关注动态治疗方案的方法。动态治疗方案：《精准医学的统计方法》是这一领域的一本优秀书籍，它既涉及基础问题，也涉及更高级的问题。本书共10章。第1-7章和第9章涵盖了与DTR相关的研究设计和分析的基础材料，这些材料可以用于这一主题的博士水平入门课程。第8、10和11章介绍了更高级和专业的主题。本书在第1章中对DTR进行了一般性介绍。在本章中，提供了几个激励性的例子来阐述DTR的概念。基本框架和定义定义如下的例子。最后一节介绍了本书的大纲，其中详细说明了如何组织其余章节的全面发展。在第2章中，作者提供了一个在书中使用的汇编和工具的一般概述。特别是，它们涵盖了因果推断和潜在结果中的关键概念，这是建立DTR统计框架的基础。他们还审查标准的统计建模技术和相关的渐近性质。第3章和第4章主要集中在一个决策点的设置上。本章首先回顾了治疗方案的相关概念。通过潜在结果框架，他们定义了一个制度的价值，这代表了预期的结果，如果人口中的所有个人都接受治疗，根据这一制度。然后，讨论了一个固定制度的价值估计。在下文中，定义了最佳方案，并且非常详细地描述了用于估计最佳治疗方案的方法。介绍了常用的估计方法，包括基于回归的估计、A-学习和值搜索估计。本章的内容通俗易懂，可以作为一个入门讲座，听众可以立即了解单一决策处理机制的全貌。第4章讨论了其他方法，包括从分类的角度估计最佳制度。第5-7章全面回顾了多决策动态治疗方案的方法。第5章介绍了一些概念和例子，技术细节较少，第6章和第7章提供了一个严格的多决策机制的统计框架。第6章从潜在结果框架开始，并证明与固定制度相关的潜在结果的分布可以从观察到的数据中确定。然后介绍了各种估计固定制度价值的方法。第7章致力于最佳动态治疗方案的表征和估计。我推荐第8、10和11章给了解动态治疗方案基础知识的读者，并有兴趣学习更高级的分析主题。至事件发生时间结局通常是慢性病研究中关注的主要终点。第8章讨论了在数据受到删失的情况下制定治疗方案的更先进方法。第10章讨论了治疗方案的统计推断，它可以回答诸如哪种治疗方案表现更好，以及关键的定制变量是什么等问题。由于渐近分析中的标准平滑假设可能会被违反，因此量化治疗方案的不确定性是相当具有挑战性的。为了掌握这一领域的基本思想，读者可以回顾10.3.1节，10.3.2节和10.4.1节的介绍。本章的其余部分提供更多的技术结果。第11章介绍了一些额外的主题，包括基于多个竞争性结局的方案，变量选择，连续治疗等。序贯多分配随机试验（SMART）是开发多阶段治疗方案的金标准研究设计，它允许在初始随机化后对替代治疗方案进行多次随机化，以评估治疗序列的总体效果。第9章首先提供了几个SMART示例来说明一般概念。设计考虑因素，包括功率和样本量计算的SMART，进行了讨论。最后，本书中介绍的方法的计算工具可以通过一个全面的R软件包DynTxRegime获得，该软件包由作者之一（霍洛威）开发。学生和研究人员谁有统计和相关的定量领域强烈建议阅读这本书的接触动态治疗制度，其中沿着其软件包可以作为有用的参考。
Dynamic treatment regimes (DTRs) are used for managing chronic disease, and fit nicely into the larger paradigm of precision medicine. There is an increasing focus on methodology for dynamic treatment regimes. Dynamic Treatment Regimes: Statistical Methods for Precision Medicine is an excellent book in this area, which addresses both foundational and more advanced topics. The book consists of 10 chapters. Chapters 1–7 and 9 cover foundational material on study design and analysis related to DTRs, which could be used for an introductory Ph.D.level course on this topic. Chapters 8, 10, and 11 present more advanced and specialized topics. The book starts with a general introduction of DTRs in Chapter 1. In this chapter, several motivating examples are provided to elaborate the concept of DTRs. The basic framework and definition are defined following the examples. The last section presents the outline of the book, which details how the remaining chapters will be organized for full developments. In Chapter 2, the authors provide a general overview of the preliminaries and tools to be used in the books. In particular, they cover the key concept in causal inference and potential outcomes, which is fundamental for establishing the statistical framework of DTRs. They also review standard statistical modeling techniques and associated asymptotic properties. Chapters 3 and 4 primarily focus on the setting of a single decision point. This chapter first reviews concept related to treatment regime. Through the potential outcome framework, they define the value of a regime, which represents the expected outcome if all individuals in the population were to receive the treatment according to this regime. Then, the estimation of the value of a fixed regime is discussed. In the following, the optimal regime is defined and methods for estimating an optimal treatment regime are described in great details. The commonly used methods, including regressionbased estimation, A-learning and value search estimation, are illustrated. The content in this chapter is accessible, and easy-tounderstand, which can serve as an introductory lecture that an audience can immediately learn the big picture of single decision treatment regimes. Chapter 4 discusses additional approaches, including estimation of optimal regimes from a classification perspective. Chapters 5–7 comprehensively review the methodology of multiple decision dynamic treatment regimes. Chapter 5 introduces the concepts and examples with less technical details, and Chapters 6 and 7 provide a rigorous statistical framework for multiple decision regime. Chapter 6 starts with the potential outcomes framework, and demonstrate that the distribution of potential outcomes associated with a fixed regime can be identified from observed data. A variety of methods on estimation of the value of a fixed regime is then introduced. Chapter 7 is devoted to characterization and estimation of an optimal dynamic treatment regime. I would recommend Chapters 8, 10, and 11 for readers who know the basics of dynamic treatment regimes, and are interested in learning more advanced topics for the analysis. Time to event outcome is commonly the primary endpoint of interest in chronic disease research. Chapter 8 discusses more advanced approaches on developing treatment regimes when data are subject to censoring. Chapter 10 discusses statistical inference for the treatment regimes, which can answer questions such as which treatment regime performs better, and what are the key tailoring variables. Since the standard smoothness assumptions in asymptotic analyses could be violated, quantifying uncertainty for treatment regimes is quite challenging. To grasp a basic idea in this area, readers can review Section 10.3.1, introduction to Sections 10.3.2 and 10.4.1. The remaining sections in this chapter provide more technical results. Chapter 11 introduces a few additional topics, including regimes based on multiple competing outcomes, variable selection, continuous treatments, etc. The sequential multiple assignment randomized trial (SMART) is the gold standard study design for developing multistage treatment regimes, which enables multiple randomizations to alternative treatment options after the initial randomization to evaluate the overall effect of the treatment sequences. Chapter 9 first provides several SMART examples to illustrate the general concept. Design considerations, including power and sample size calculation for a SMART, are discussed. Finally, computational tools for the methods introduced in this book are available through a comprehensive R package, DynTxRegime, developed by one of the authors (Holloway). Students and researchers who have statistics and related quantitative fields are highly recommended to read the book for an exposure to dynamic treatment regimes, which along with its software package could serve as useful references.