|
Abstract:
|
Future events prediction is a major objective in survival analysis. In this scenario, events data and longitudinal observations of the predictors associated with the subjects are commonly available. For events predictions, existing approaches typically rely on jointly modeling the events process and the longitudinal observables. Thus difficulties associated with the model specification may increasingly arise especially when handling modern large and complex data sets. In this study, we propose a new predictive approach for events prediction constructed based on modeling the forward events intensity function. Our approach is advantageous in that it does not require a separated model for the longitudinal observables. To handle the practical issues with missing and irregularly spaced observations, we propose a kernel smoothing approach to more efficiently utilize the data information. We justify the validity of our approach by establishing the estimation consistency and asymptotic normality. We also demonstrate the promising performance of the proposed new method by extensive simulations and data analysis.
|
