A0957
Title: Cox regression model with auxiliary endpoints accounting for left truncation, complex censoring, and missing data
Authors: Sharon Xie - University of Pennsylvania (United States) [presenting]
Yidan Shi - New York University Grossman School of Medicine (United States)
Abstract: The time-to-event analysis is a widely employed approach for modelling disease progression data. However, obtaining the true survival endpoint, such as the age at which cerebrospinal fluid biomarkers for Alzheimer's disease become abnormal, can be costly, invasive, and contingent on participants' cognitive status. As a result, it is often accessible only to a small subset of participants, which can affect estimation accuracy and efficiency. An auxiliary event, which is less costly or invasive but may measure the true event with error, is often available. Additionally, delayed entry in observational studies can result in left truncation, which further complicates analyses. Since the auxiliary event is not the primary focus, it often suffers from left censoring due to study design. A likelihood-based method and an E-M algorithm for Cox regression models that incorporate the auxiliary endpoint while accounting for left truncation, complex censoring scenarios, and auxiliary-event-dependent missingness are proposed. The method improves efficiency and corrects for bias compared to complete case analysis. It is demonstrated that the proposed regression coefficient estimator is consistent and asymptotically normally distributed. The performance of the method is assessed in finite sample scenarios through simulation studies. Finally, the proposed method is illustrated using the Alzheimer's disease neuroimaging initiative (ADNI) study.
