New Methodology and Theory for Optimal Treatment Regimes with Applications to Precision Medicine

基于个体特征寻找最优治疗方案或一系列序贯治疗方案的问题在精准医学、政府政策和积极的劳动力市场干预等领域有着重要的应用。根据应用的不同，治疗可以代表一种药物、一种设备、一项计划、一项政策、一种干预或一种策略。在基因组学和医学成像等领域的进步的刺激下，过去十年见证了个性化医学令人兴奋和显著的进步，从乳腺癌的治疗到严重抑郁障碍的治疗。精确医学的成功取决于开发准确可靠的统计和机器学习工具，以便根据从随机实验或观察性研究收集的数据来估计最佳治疗方案。该项目将开发新的方法、理论和算法，有可能显著提高最优治疗方案的统计估计和推断的最新水平。拟议的研究将显著提高静态或动态最优治疗方案估计的统计方法和理论的可用性。将制定一个系统框架，利用新的分位数标准评估各种情况下的最佳治疗方案。这项研究将重点放在一阶段(静态)治疗方案和动态治疗方案上，后者允许治疗随时间变化。此外，这项研究将涉及完全观察到的反应和随机审查的反应(例如，生存时间)、随机试验和观察性研究，以及双重稳健估计。该框架还将扩展到其他标准，如基尼平均差。该项目将极大地推进最优治疗方案的一大类稳健估计器的理论基础。此外，它解决了开发新的方法和算法来识别高维环境下最佳治疗方案估计的重要变量这一具有挑战性和重要的问题。调查员将开发软件包，并向研究界免费提供。将特别鼓励来自少数群体的学生参与拟议的项目。

The problem of finding the optimal treatment regime, or a series of sequential treatment regimes, based on individual characteristics has important applications in areas such as precision medicine, government policies, and active labor market interventions. Depending on the application, a treatment can represent a drug, a device, a program, a policy, an intervention, or a strategy. Stimulated by the advancements in fields such as genomics and medical imaging, the last decade has witnessed exciting and remarkable progress in personalized medicine, ranging from treatments for breast cancer to treatments for major depressive disorders. The success of precision medicine depends on the development of accurate and reliable statistical and machine learning tools for estimating the optimal treatment regime given the data collected from randomized experiments or observational studies. This project will develop novel methodology, theory, and algorithms with the potential to significantly advance the state of the art in statistical estimation and inference for optimal treatment regimes. The proposed research will significantly enhance the availability of statistical methodology and theory for static or dynamic optimal treatment regimes estimation. A systematic framework for estimating optimal treatment regimes using a new quantile criterion for a variety of scenarios will be developed. The research will focus on both one-stage (static) treatment regimes and dynamic treatment regimes, the latter allowing for treatments to vary with time. In addition, the research will address completely observed responses and randomly censored responses (e.g., survival times), randomized trials and observational studies, and doubly robust estimation. The framework will also be extended to alternative criteria such as Gini's mean difference. This project will significantly advance the theoretical foundations of a large class of robust estimators of optimal treatment regimes. Furthermore, it addresses the challenging and important problem of developing new methodology and algorithms to identity important variables for optimal treatment regime estimation in the high-dimensional setting. The investigator will develop software packages and make them freely available to the research community. Students from minority groups will be especially encouraged to participate in the proposed project.