A1110
Title: Dynamic prediction for the joint model of a longitudinal biomarker and interval-censored failure time data
Authors: Yang-Jin Kim - Sookmyung Women University (Korea, South) [presenting]
Abstract: Joint modeling approaches have been widely used to reflect the effect of time-varying biomarkers on failure time. The aim is to propose a joint model for interval-censored failure time data in the presence of longitudinal biomarkers. To assess predictive accuracy, dynamic measures are introduced, including the time-dependent area under the ROC curve (AUC) and the Brier score. The inference procedure is based on the EM algorithm that accounts for the latent failure time and subject-specific random effects. A dynamic marker is defined as the conditional survival probability of being alive at time $t>s$, given survival up to time $s$. Various simulation scenarios are considered to validate the proposed joint model and compare its performance to the landmarking approach. Finally, the method is illustrated using the well-known Paquid dataset, which includes interval-censored dementia onset times and two longitudinal cognitive scores.
