Dynamic Prediction Of Longitudinal Data For Alzheimer's Disease Based On A Joint Model

Longitudinal data and survival data are widely existed in the field of medical follow-up.When longitudinal data and survival data exist at the same time,if the relationship between the two types of data is ignored,modeling alone will often lead to deviations in the parameter estimates.The joint model uses the potential connection of the two to carry out joint modeling,which can reduce the deviation and simultaneously study the vertical development process and survival probability and outcome of individuals.The development of Alzheimer's disease(AD)is divided into three stages: the early stage without symptoms,the mild cognitive impairment(MCI)stage,and finally the diagnosis stage of AD.MCI is an important transitional stage of AD.Using a joint model to study the development process of MCI to AD,its vertical sub-model can study the changes of vertical markers and the survival sub-model can predict the conversion probability of MCI to AD,and can use the new vertical observation to realize dynamic prediction of the disease conversion probability.In this paper,the statistical analysis and processing of AD follow-up data are first carried out,using visualization to analyze the missing value generation mechanism and using the limit tree for multiple filling,using a random forest-based filtering variable selection algorithm to obtain variable screening results,and sort them according to the variable importance score.Then this paper builds a joint model that uses the linear random effect mixed model as the longitudinal sub-model and the extended Cox proportional hazard regression model as the survival sub-model.The conversion probability at the AD stage was dynamically estimated.According to the model constructed in this article,the following are the main conclusions of this article:(1)In the study of the development process of longitudinal markers MMSE,age,years of education and follow-up time all have a significant effect on the change of the longitudinal markers MMSE score;(2)In the process of predicting the conversion probability of MCI to AD,whether to carry Apo E4 is an important indicator.MCI subjects carrying Apo E4 have a higher tendency to suffer from AD;(3)With the increase of follow-up time,the continuous addition of follow-up data can realize the dynamic prediction of the individual's longitudinal markers and the probability of MCI to AD disease conversion,and obtain more accurate prediction results.