Comparison of Predictive Performance for Different Longitudinal Cognitive Markers in the Reversion of Mild Cognitive Impairment based on the Shared Random-Effect Model
Objective The shared random-effect model(SREM)was applied to study the reversion of mild cognitive impairment(MCI)to normal cognition(NC),to compare the predictive performance of different longitudinal cognitive markers for the reversion,and to evaluate the covariate effects of predictors.Methods The SREM model consisted of two sub-models:the longitudinal sub-model adopted linear mixed model to describe different longitudinal cognitive markers trajectories;and the survival sub-model used proportional risk model to fit the survival process.Goodness of fit was based on the log-likelihood value,and information criterion.AUC(area under the curve)was used to evaluate the predictive performance of different longitudinal cognitive markers,including MMSE,CDRSB,FAQ,ADAS11,ADAS13 and ADASQ4.Results There were 72(8.54%)of 843 patients with MCI reversed to NC at the end of follow-up.The log-likelihood value of the SREM model based on spline-PH-GH parameter distribution was the largest,and AIC and BIC were the smallest;the SREM model based on CDRSB score as longitudinal cognitive marker had the best fitting performance,with high AUC values at different times ranging from 0.797 to 0.852,and the lowest prediction error,ranging from 0.0427 to 0.0429.Age,gender,education level,marital status and APOEε4 gene affected the cognitive function and daily activity function.Six different longitudinal cognitive markers influenced the reversion.Conclusion CDR score was the best predictor of cognitive function and the reversion in MCI patients.Cognitive function and daily activity function can be used as dynamic monitoring indicators of MCI reversion.
The shared random-effect modelJoint modelLongitudinal dataSurvival dataMild cognitive impairmentReversion
NETL
NSTL
万方数据
