Censoring Robust Estimation of Covariate Effects on Discrete Survival Endpoints

描述（由申请人提供）：离散生存终点可以出现在观察性研究中，其中有周期性随访或当时间被离散测量时。这些数据通常出现的例子包括通过常规检查和监测作出疾病的亚临床诊断的情况。其中包括系统地使用结肠镜检查来检测结直肠癌中的息肉生长，以及使用常规乳房x光检查来检测乳腺癌中的肿瘤肿块。在这种情况下，确切的事件时间是间隔审查和未观察到的。当生存终点被确认为离散或间隔截短时，用于分析观察到的失效时间的常用半参数方法包括离散时间比例风险模型和比例几率模型。当半参数假设不成立时，使用半参数生存模型的推理依赖于观察到的审查分布。这使得对结果的解释在科学上没有意义，因为目标人群定义不清。最近，一些作者提出使用加权估计来消除当半参数假设不成立时对兴趣估计的审查的影响。虽然这些建议的估计器在模型错误规范下提供一致和可重复的结果，但它们仅适用于连续测量生存时间的设置，并且仅适用于k -样本比较的情况。本文提出的研究目标是建立一类滤波鲁棒离散生存估计量，并建立它们的渐近行为。此外，我们将通过开发这些方法用于可能涉及多个调整协变量的一般回归策略，扩展先前对连续生存数据的稳健估计器的审查工作。

DESCRIPTION (provided by applicant): Discrete survival endpoints can occur in observational studies where there is periodic follow-up or when time is measured discretely. Examples where these data commonly arise include situations where subclinical diagnosis of disease is made through routine checkups and surveillance. These include the systematic use of colonoscopy for detecting polyp growth in colorectal cancer and the use of regular mammographies for the detection of tumor mass in the case of breast cancer. In such settings, the exact event times are interval censored and unobserved. When the survival endpoint is acknowledged to be discrete or interval censored, common semiparametric methods for the analysis of the observed failure times include the discrete-time proportional hazards model and the proportional odds model. Inference using semiparametric survival models is dependent on the observed censoring distribution when the semiparametric assumption fails to hold. This renders the interpretation of the results scientifically unmeaningful since the target population i ill-defined. Recently, some authors have proposed the use of weighted estimators to remove the effect of censoring on the estimand of interest when semiparametric assumptions fail to hold. While these proposed estimators provide consistent and reproducible results under model misspecification, they have only been developed for settings where survival times are measured continuously and only in the case of K-sample comparisons. The goal of the research proposed here is to develop a class of censoring robust discrete survival estimators and establish their asymptotic behavior. In addition, we will extend the previous work censoring robust estimators for continuous survival data by developing these approaches for general regression strategies that may involve multiple adjustment covariates.