Dynamic Modeling and Risk Prediction with Complex Observational Semi-Competing Risks Data

使用次要数据来源的观察性研究，如登记、索赔和电子健康记录，是在现实世界环境中评估治疗效果和预测疾病结果的主要研究工具。然而，观测数据往往呈现出许多复杂性，需要大量的方法开发才能获得有效的结果。具体地，当晚期事件(例如，死亡)可以阻止观察到非晚期事件(例如，癌症复发)时，出现半竞争风险数据，但反之亦然。对这些数据的分析进一步复杂化，包括结果的集群化、治疗效果的时变性、混乱性和动态预测。该项目将利用复杂的观测半竞争性风险数据开发新的动态建模和风险预测方法。该项目是由癌症研究推动的。开发的方法还将广泛适用于其他健康状况、可靠性研究和社会科学，这些领域通常会出现此类数据。研究人员将通过培训研究生、设计高级专题课程以及吸引代表不足的少数民族学生来整合研究和教育。研究人员还将在R中开发开源、用户友好的软件包来传播结果。该项目有三个研究目标。第一个目标是建立一个基于Copula的、时变系数、随机效应的多水平半竞争风险数据模型。第二个目标是开发一种基于倾向得分匹配的方法来控制多水平观测半竞争风险数据中的混杂。将研究省略未测量的混杂因素的影响。第三个目标是利用这些数据开发一种新的针对非终端和终端事件的动态风险预测工具。与传统的预测模型不同，开发的模型将利用患者动态疾病进展和特征的数据。研究人员将得出新估计器的大样本属性，进行模拟评估，并应用这些方法分析真实世界的数据。这一奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Observational studies using secondary data sources, such as registry, claims, and electronic health records, are primary research tools to assess treatment effects and predict disease outcomes in real-world settings. Observational data, however, often present many complexities, for which substantial methods development is needed to obtain valid results. Particularly, semi-competing risks data arise when a terminal event (e.g., death) can prevent the observation of a non-terminal event (e.g., cancer recurrence), but not vice versa. The analysis of such data is further complicated with clustered outcomes, time-varying treatment effects, confounding, and dynamic prediction. This project will develop novel dynamic modeling and risk prediction methods with complex observational semi-competing risks data. The project is motivated by cancer studies. The developed methods will also be broadly applicable in other health conditions, reliability studies, and social science, where such data commonly arise. The investigator will integrate research and education by training graduate students, designing advanced topic courses, and engaging underrepresented minority students. The investigator will also develop open-source, user-friendly software packages in R to disseminate the results.The project has three research aims. The first aim is to develop a copula-based, time-varying coefficient, random-effect model for multilevel semi-competing risks data. The second aim is to develop a propensity score matching based method to control for confounding in multilevel observational semi-competing risks data. The impact of omitting unmeasured confounders will be studied. The third aim is to develop a novel dynamic risk prediction tool for non-terminal and terminal events with such data. Unlike traditional prediction models, the developed model will utilize data on patients’ dynamic disease progression and characteristics. The investigator will derive large sample properties of the new estimators, conduct simulations for evaluation, and apply the methods to analyze real-world data.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.