Estimating treatment effects using real world data when there are competing risks

随机对照试验(RCT)是确定药物治疗和其他干预效果的证据的黄金标准。然而，来自真实世界数据(RWD)的治疗效果证据的价值越来越被认识到，因为它为研究大量和多样化的患者群体的治疗效果提供了机会。该项目将侧重于使用RWD评估治疗对事件发生时间结局的影响，例如癌症死亡。在估计治疗对事件发生时间结果的影响时，重要的是要考虑竞争风险，这意味着个人受到其他事件的影响。竞争风险的重要例子是前列腺癌治疗研究，其中许多患者不是死于癌症，而是死于其他原因，以及关于使用他汀类药物对痴呆症患者心血管死亡率的影响的研究，后者也有很高的死于其他原因的风险。在这种情况下量化治疗效果是具有挑战性的，需要考虑正在调查的治疗对其他死亡原因的可能影响。当存在相互竞争的风险时，通常使用的治疗效果指标包括病因特定风险比和次分布风险比，但最近的统计文献表明，这些指标没有因果解释，并推荐了替代的治疗效果指标，称为ESTMAND。这项最近的工作重点放在RCT上，需要扩展以在RWD分析中使用推荐的ESTMAND，并结合处理与时间相关的混淆所需的专门方法。因此，该博士项目将评估和开发适当处理竞争风险的方法，以估计使用RWD治疗的因果影响。这些方法将使用来自癌症登记、医院和死亡记录的相关联的国家数据在前列腺癌治疗的例子中进行说明。该项目的预期结果是在存在竞争性风险时使用RWD估计多水平治疗对生存的影响的统计方法，以及与前列腺癌最佳治疗相关的实质性发现。这将为RWD在监管框架中的地位提供重要证据，并为药物开发提供信息，特别是为第三阶段试验的设计提供信息，这对医疗保健决策至关重要。实现这些影响的战略将包括在统计/流行病学和临床期刊上发表论文。我还将在会议上展示我的工作，出席会议和演讲技能已纳入我的培训计划。通过与项目联合资助者阿斯利康的合作，我将有机会展示我的工作，并与为新研究和政策建议提供信息的专家接触。通过我的学生身份，我从LSHTM：博士可转移技能计划的内部课程和外部课程：伦敦大学学院扩展课程、InkPath、RSS、SOAS、流行病学研究学会和统计学博士培训学院获得了主题培训。这些课程包括研究伦理、有效的文献搜索技能、数据管理、统计计算、高级编程、高维统计学、计算机密集统计学、统计机器学习、癌症存活率、竞争风险，以及使用模拟研究来评估统计方法。这些课程使我掌握了如何创建这个项目所需的快速和准确的统计程序的技能。2022年，我在青年统计学家会议上介绍了我们的工作，我们被评为最佳演讲，并被赞助在2022年RSS会议上介绍我们的工作。2023年，我在SGUL统计研讨会和ISCB44会议上介绍了我们的工作。关键词：竞争风险、治疗效果、真实数据、预测、因果效应、事件发生时间结果

Randomized controlled trials (RCTs) are the gold standard for establishing evidence for the effects of medical treatments and other interventions. However, the value of evidence on treatment effects from real world data (RWD), such as electronic health records, is increasingly being recognized as it presents opportunities to study treatment effects in large and diverse patient populations. This project will focus on estimating effects of treatments on time-to-event outcomes, such as death due to cancer, using RWD. When estimating treatment effects on time-to-event outcomes it is important to consider competing risks, meaning other events that individuals are subject to. Important examples of competing risks are studies of treatments for prostate cancer where many patients will die not from their cancer but due to other causes, and studies of the impact of statin use on cardiovascular mortality in dementia patients, who also have a high risk of death from other causes. Quantifying treatment effects in this setting is challenging and needs to account for possible effects of the treatments being investigated on other causes of death. Commonly used treatment effect measures when there are competing risks include cause-specific and subdistribution hazard ratios, but recent statistical literature has shown that these do not have a causal interpretation and has recommended alternative treatment effect measures known as estimands.This recent work has focused on RCTs and extensions are required to use the recommended estimands in the analysis of RWD, in conjunction with specialized methods required to address time-dependent confounding. This PhD project will thus evaluate and develop methods for the appropriate handling of competing risks to estimate causal effects of treatments using RWD. The methods will be illustrated in an example of treatments for prostate cancer using linked, national data from cancer registries, hospital and death records.The anticipated outcomes of this project are statistical methods for estimating impacts of multi-level treatments on survival when there are competing risks using RWD, and substantive findings relating to optimal treatments for prostate cancer. This will contribute important evidence for the place of RWD in the regulatory framework and to inform drug development, particularly to inform design of phase III trials which are critical for healthcare decision-making. Strategies to achieve these impacts will include publication of papers in both statistical/epidemiological and clinical journals. I will also present my work at conferences, and conference attendance and presentation skills have been incorporated into my training plan. Through collaboration with the project co-funders, AstraZeneca, I will have opportunities to present my work and engage with experts who inform new studies and policy recommendations.Throughout my studentship, I have obtained subject-matter training from internal courses at LSHTM: Doctoral Transferable Skills Programme, and from external courses: UCL Extend, Inkpath, RSS, SOAS, Society for Epidemiologic Research and Academy for PhD Training in Statistics. The courses include research ethics, effective literature searching skills, data management, statistical computing, advanced programming, high dimensional statistics, computer intensive statistics, statistical machine learning, cancer survival, competing risks, and using simulation studies to evaluate statistical methods. These courses have equipped me with skills on how to create fast and numerically accurate statistical programs needed for this project. In 2022, I presented our work in the Young Statisticians Meeting where we were voted as the best talk and sponsored to present our work at the 2022 RSS conference. In 2023, I presented our work at SGUL statistical seminar and the ISCB44 conference.Keywords: Competing risks, treatment effects, real world data, estimands, causal effect, time-to-event outcomes