Latent Cognitive Decline Process Of The Elderly With Mild Cognitive Impairment Based On Multivariate Longitudinal Data

Objective:This paper aims to explore the cognitive decline process providing a theoretical basis for clinical workers to implement prevention strategies and to provide methodological references for longitudinal analysis of chronic diseases in other fields:(1)To explore the cognitive decline process from mild cognitive impairment(MCI)to Alzheimer’s Disease(AD)and to assess the covariates effects on this process;(2)to analyze the differences of influencing factors on the cognitive assessment scales and to show their application characteristics.Methods:Using the longitudinal study,233 patients with MCI who progressed to AD by waves of follow-up in the ADNI public database were selected.The cognitive function measures included the Alzheimer’s Disease Assessment Subscale-cognitive 11 item(ADAS-11),Clinical Dementia Rating-Sum of the Boxes(CDR-SB),Functional Activities Questionnaire(FAQ)and Mini-mental State Examination,(MMSE).Socio-demographic information and genetic information were also included.Based on the assumption that multiple scales collectively represent the same potential process,we constructed a latent process model using multivariate longitudinal data.A parametric nonlinear transformation is introduced to link multiple longitudinal indicators with latent processes.A parametric nonlinear latent process model is used to describe the continuous-time latent evolution of MCI to AD,and to evaluate the direct impact of covariates on it.A fixed contrast is made to distinguish the association with latent processes and the differences between various psychological tests.The model is screened according to Akaike information criterion(AIC)and Bayesian information criterion(BIC)and the number of parameters.The parameters were estimated by maximum likelihood method,and the model was evaluated based on residual analysis and the percentage of variance of regression interpretation in the common latent process model calculated by four tests at a given time.Results:(1)Gender(P=0.6210),education level(P=0.6009),ApoEε4(P=0.0609)and marital status(P=0.3529)were not significantly associated with potential cognitive processes,follow-up time(P<0.0001))was significantly associated with potential cognitive processes.The interaction between ApoEε4 and time was significantly correlated with the latent cognitive process(P=0.0277),and the interactions of other variables with time had no statistically significant effect on this process.(2)Gender(P=0.0139)and education level(P=0.0094)had different effects on different tests.Gender had effects on the scores of the FAQ and the CDR-SB,while educational level had effects on the scores of the MMSE and the FAQ.Effects of ApoEε4 carrires and marital status on each test were not statistically significant.(3)The MMSE and the CDR-SB all showed no significant difference in higher and lower level of cognition,while FAQ showed no significant difference in higher and lower level of function.(4)Residual analysis showed that the model fitted well,and with the extension of follow-up,the proportion of variances latent process interpreted by four tests increased from 16.42% to 93.94%,which also showed that the model fitted well.Conclusions:(1)The cognitive function of MCI with ApoEε4 carriers decreased faster than that of MCI without ApoEε4 carriers,and the longer the follow-up time,the greater the gap between the two groups.It reminds health care workers to pay more attention to the MCI population with ApoEε4 carriers,so as to prevent and slow down the cognitive decline process.(2)The influence of gender and education level on each cognitive test is not the same.The deviation caused by this influence on the results of cognitive test should be considered comprehensively in the cognitive measurement process.Cognitive assessment should be co-assessed using multiple scales that can test different cognitive functions and be able to distinguish between different levels of cognitive impairment to eliminate bias caused by a single scale.(3)Latent process model fits better and provides more information than single test analysis.It can provide methodologies for the analysis of multivariate mixed longitudinal data.