Identifying optimal dynamic strategies for prostate cancer control

项目总结/摘要 大数据有可能彻底改变癌症研究和护理，但提取它所掌握的信息 癌症控制的最佳策略将需要数据科学的尖端工具。的最优策略 癌症控制将是动态的策略，随着时间的推移， 临床病史不幸的是，传统的统计方法不能适当地比较动态 策略，所以我们需要专门为此任务设计的方法：g方法。G方法有助于 在许多领域形成临床护理，但它们尚未系统地应用于癌症研究。此外，本发明还 虽然g方法可以让我们有效地估计预先指定的策略的效果，但这些策略可能不是最佳的 战略布局我的首要目标是应用并进一步发展分析方法，以学习最佳策略 从复杂的纵向数据中获取癌症控制的信息，并生成用户友好的公开软件， 将这些方法提供给癌症研究界。 我将应用这些方法来回答前列腺癌控制连续体中的关键临床问题：1） 预防侵袭性前列腺癌的最佳饮食和生活方式策略，2）最佳筛查策略 在基线PSA检测后，最大限度地检测侵袭性疾病，同时最大限度地减少 惰性肿瘤，和3）最佳的他汀类药物治疗策略，以最大限度地提高非转移性肿瘤患者的生存率。 前列腺癌该项目将利用一项大型前瞻性队列研究的数据和一个新的平台， 电子健康记录与基因数据相关联。我将首先应用g方法来估计 推荐的癌症控制策略，随机试验将有有限的可行性进行评估。我 然后将研究从数据中学习最佳策略的新方法是否会导致 改进的、有针对性的建议，在正确的时间向正确的人提供正确的干预措施。 这一创新项目将推进癌症护理的比较有效性研究， 数据科学基于我在癌症方面的专业知识， 流行病学和因果推理; 2）由全球 3）无与伦比的研究环境，支持我的职业发展。 通过这项工作，我将扩展我在新领域的专业知识，包括机器学习。拟议 研究和培训将帮助我实现成为一名独立调查员的长期职业目标 并领导一个跨学科的研究项目，该项目将因果推理和机器学习相结合， 癌症控制的最佳策略。利用丰富的现有数据，该提案代表了一个重要的 有机会开发，应用和传播大临床数据的强大方法，以加速 癌症研究和护理。

PROJECT SUMMARY/ABSTRACT Big Data has the potential to revolutionize cancer research and care, but extracting the information it holds on the optimal strategies for cancer control will require cutting-edge tools in data science. The optimal strategies for cancer control will be dynamic strategies that adapt clinical decisions over time to a patient’s evolving clinical history. Unfortunately, conventional statistical methods cannot appropriately compare dynamic strategies, so we need methods specifically designed for this task: g-methods. G-methods have helped to shape clinical care in many areas, but they have not been systematically applied to cancer research. Further, while g-methods let us validly estimate the effect of pre-specified strategies, these may not be the optimal strategies. My overarching goal is to apply and further develop analytic methods to learn the optimal strategies for cancer control from complex longitudinal data and generate user-friendly, publicly-available software to make these methods available to the cancer research community. I will apply these methods to answer key clinical questions across the prostate cancer control continuum: 1) the optimal dietary and lifestyle strategies to prevent aggressive prostate cancer, 2) the optimal screening strategy following a baseline PSA test to maximize detection of aggressive disease while minimizing detection of indolent tumors, and 3) the optimal statin therapy strategy to maximize survival among men with nonmetastatic prostate cancer. This project will leverage data from a large prospective cohort study and a novel platform of electronic health records linked with genetic data. I will first apply g-methods to estimate the effects of recommended strategies for cancer control that a randomized trial would have limited feasibility to evaluate. I will then investigate whether novel methods that learn the optimal strategies from the data may lead to improved, targeted recommendations that get the right interventions to the right people at the right time. This innovative project will advance comparative effectiveness research for cancer care at the cutting edge of data science. I am optimally positioned to undertake this research based on my 1) expertise in cancer, epidemiology, and causal inference; 2) exceptional multidisciplinary mentoring team comprised of global leaders in their respective fields; and 3) unparalleled research environment to support my career development. Through this work, I will expand my expertise in new areas, including machine learning. The proposed research and training will help me achieve my long-term career goal to become an independent investigator and lead a transdisciplinary research program that integrates causal inference and machine learning to identify optimal strategies for cancer control. Leveraging rich, existing data, this proposal represents a significant opportunity to develop, apply, and disseminate powerful methods for big clinical data to accelerate progress in cancer research and care.