Doctoral candidate in Biostatistics, Anne Eaton, will present:
“Non-Parametric Estimation of Probability in Disease States and Mean Cumulative Marker Process”
PhD Adviser: Xianghua Luo
Abstract: This talk focuses on endpoints for time to event variables which integrate additional outcomes such as non-fatal events or other measures of patients’ health state while alive. We introduce new statistical tools to estimate the event free survival probability and the mean cumulative marker process, an endpoint recently proposed in the literature as a benefit-risk summary measure. Many clinical studies focus on composite endpoints defined as the time to the earliest of either death or a non-fatal event (for example, cancer progression). However, if death is right censored and cancer progression is measured periodically at visits, the resulting composite displays a mix of interval censoring and right censoring known as component-wise censoring, and standard survival analysis methods cannot be applied. We propose novel estimators for the event free survival curve and the restricted mean event free survival time using component-wise censored data by combining existing non-parametric estimators to circumvent the component-wise censoring problem. We derive the large sample properties of the estimators and compare their performance to standard methods using a simulation study. These estimators can also be used in multistate modelling, and to estimate the mean cumulative marker process when the marker of interest is measured at visits. However, the estimators require independence between the visit process and the clinical outcomes of interest, which may be violated in some real datasets. We use inverse visit rate weights to develop estimators that work under a relaxed assumption, that the visit rate is conditionally independent of the clinical outcomes, given possibly unobserved baseline covariates. We explore the performance of our proposed estimators under various scenarios.