CMStatistics 2020: Start Registration
View Submission - CMStatistics
Title: Dynamic predictions via functional joint models for sparse longitudinal and time-to-event data Authors:  Toshihiro Misumi - Yokohama City University (Japan) [presenting]
Yuriko Takeda - Yokohama City University (Japan)
Abstract: Recently, joint modeling techniques of longitudinal and survival data have been frequently applied in the medical research area. In the framework of joint modeling, the model enables us to compute individual dynamic predictions of survival probabilities. The accuracy of prediction may be worse when longitudinal data are sparsely observed at irregular time-points since ordinal linear mixed-effects models are widely used in existing joint models. To overcome this issue, we propose a novel functional joint model for sparse longitudinal and survival data. We employ a reduced rank model for the longitudinal submodel to capture the trajectory of sparse longitudinal process accurately. A Cox proportional hazards model is applied to the survival submodel. Unknown parameters included in the model are estimated by a Bayesian approach. Some numerical examples are presented to demonstrate the effectiveness of our proposed modeling strategy.