Bayesian approaches for missingness and causality in cancer and behavior studies

描述(由申请人提供)：这项建议将开发新的贝叶斯方法来处理遗漏，并对生物医学研究中与癌症和行为研究特别相关的重要问题进行因果推断。数据缺失是临床研究中的一个主要问题。最近，人们花了更多的时间试图将遗漏的数量降到最低，但这仍然是一个问题。我们将解决临床环境中不完全数据分析中的几个紧迫的复杂问题，如最近美国国家科学院的一份报告所述，包括对观察数据的模型t进行评估，为辅助协变量开发贝叶斯方法，以及对不可忽略的缺失进行非参数建模。治疗效果的机制在行为试验中特别令人感兴趣。具体地说，不同的过程如何调节干预的效果？这可以促进 构建未来的干预措施。然而，很难确定这些“调解人”对结果的因果影响。我们将制定新的办法，在有多个调解人和几乎没有做过什么工作的纵向调解人的复杂情况下确定这些影响。另一个重要的问题是，在设定半竞争性风险的情况下，如何确定干预措施对结果的因果影响。半竞争性风险发生在研究中，如果进展终点可能因死亡而先发制人，或因失去随访或研究终止而被审查。经历进展事件的受试者也会被跟踪以求生存，这可能会被审查。这种形式的数据被称为半竞争性风险数据。这一模式与某些脑癌试验特别相关，在这些试验中，死亡和小脑进展是半竞争性的风险。对于所有这些设置，贝叶斯方法是理想的，因为它允许人们适当地表征关于不变假设的不确定性(这些假设存在于所有这些问题中)。这里开发的方法将有助于回答许多重要的临床问题，包括体重管理和戒烟方面的行为变化机制，通过适当评估调解的能力，以及治疗对脑癌死亡时间和小脑进展的联合因果影响。我们将(通过PI的网页)发布这些方法的代码，以确保这些方法可供研究人员在自己的研究中随时使用。PI与个别临床研究的PI和统计学家联合调查员合作的历史将帮助团队产生最好的科学，并通过主题出版物和在相关会议上的演示将我们的临床结果和新方法传播给适当的受众。

DESCRIPTION (provided by applicant): This proposal will develop novel Bayesian approaches to handle missingness and conduct causal inference for important problems in biomedical research with particular relevance to cancer and behavioral studies. Missing data is a major problem in clinical studies. Of late, more e ort is spent to try to minimize the amount of missingness, but it remains a problem. We will address several pressing complications in the analysis of incomplete data in clinical settings as documented in a recent National Academies of Science report, including assessing model t to the observed data, developing Bayesian approaches for auxiliary covariates, and nonparametric modeling of nonignorable missingness. The mechanisms of treatment effectiveness are of particular interest in behavioral trials. Specifically, how do different processes mediate the effect of an intervention? This can facilitate constructing future interventions. However, determining the causal effect of such 'mediators' on the outcomes is difficult. We will develop new approaches to identify these effects in complex settings with multiple mediators and longitudinal mediators for which little work has been done. Another important question is how to de ne and identify causal effects of interventions on outcomes in the setting of semi-competing risks. Semi-competing risks occur in studies where a progression endpoint may be pre-empted by death or censored due to loss to follow-up or study termination. Subjects who experience a progression event are also followed for survival, which may be censored. Data of this form has been termed semi-competing risks data. This paradigm is particularly relevant to certain brain cancer trials, where the semi-competing risks are death and cerebellar progression. For all these settings, a Bayesian approach is ideal as it allows one to appropriately characterize uncertainty about invariable assumptions (which are present in all these problems). The methods developed here will help answer numerous important clinical questions including the mechanisms of behavior change, both in weight management and smoking cessation, via the ability to appropriately assess mediation, and the joint causal effect of treatment on time to death and cerebellar progression in brain cancer. We will disseminate code for these methods (via the PI's webpage) to ensure the methods will be readily usable by investigators in their own studies. The history of the PI's collaboration with the PI's of the individual clinical studies and the statistician co- investigators will help the team produce the best science and facilitate dissemination of our clinical findings and new methods to the appropriate audience via both subject matter publications and presentations at relevant conferences.