Valid Inferences Based on Proportional Hazards Model for High-Dimensional, Interval-Censored Data

• Activity Number: 506 - Robust and Efficient Analysis of Complex Time-to-Event Data
• Type: Topic Contributed
• Date/Time: Thursday, August 11, 2022 : 8:30 AM to 10:20 AM
• Sponsor: ENAR
• Author(s): Zhe Fei* and Gang Li and Rui Yang and Hua Zhou
• Companies: UCLA and University of California, Los Angeles and UCLA and UCLA
• Keywords: Interval-Censored; Dimension Reduction; P-value; Joint Modeling

Abstract

Interval-censored failure time data commonly arise in clinical trials and longitudinal studies, which refer to a censoring mechanism where an event time cannot be directly observed but is only known to lie between two adjacent examination times in a sequence of visits. In addition, the high dimensionality refers to data with large numbers of predictors commonly seen in modern biomedical studies. Appropriate modeling of high dimensional interval-censored data requires accurate estimation of the model coefficients, and reliable inference of their significance, which leads to a sparse final model. Numerous efforts have been put into high dimensional inference for various types of regression models and data, yet that for proportional hazards models and interval censored data possesses unique challenges and remains largely unsolved. We propose a novel approach that utilizes dimension reduction together with re-sampling and aggregation to provide estimation and inference for the high dimensional regression parameters in this context. Our method is applied to the UK Biobank SNP data and offers novel discoveries in a cohort of diabetes patients with chronic adverse events.