Two Methods to Fit Longitudinal Sub-Model of Joint Model
Download as PDF
DOI: 10.25236/icess.2019.174
Corresponding Author
Rong Li
Abstract
The joint model is a new shared parameter model consisting of longitudinal and survival data. In general, we often use a linear mixed-effects model to estimate the longitudinal sub-model and a Cox proportional hazard model to estimate the surviving sub-model. However, the linear model requires a hypothesis of normal distribution. Consequently, when the assumption is not valid, the estimation result will produce large deviations. In this paper, we use machine learning to estimate the longitudinal model, and the survival model still uses the Cox model. We compare the effects of the linear mixed-effects model and the machine learning method. These two estimation methods are compared by simulated data. The results of the comparison show that the residuals diagnosis outcomes of the machine learning survival sub-model are more in line with the theoretical results, while the residuals of the longitudinal sub-model are more dispersed than the linear model. Additionally, the running time of the machine learning is shorter than that of the linear mixed effect model.
Keywords
Joint model, Longitudinal data, Survival sub-model, Linear mixed effect model, Cox proportional, Hazard model, Machine learning