Latent Classes For Mild Cognitive Impairment Conversion To Alzheimer’s Disease Based On Joint Modelling

Objective:Based on the characteristics of covariates and outcome variables in the different latent classes of the elderly population,the aims of this study were to(1)explore influencing factors of cognitive function trajectories to identify key population of cognitive interventions and to provide theoretical evidence for prevention or intervention of cognitive impairment and health policy formulation among the elderly;(2)show the advantages of joint model in longitudinal data analysis in practical application and provide methodological reference for data with multiple longitudinal markers and multiple endpoint events.Methods:Study data were from the Alzheimer’s Disease Neuroimaging Initiative(ADNI),in which 245 patients with mild cognitive impairment(MCI)at baseline and MCI or Alzheimer’s disease(AD)as outcome events were screened.A joint latent class model(JLCM)was fitted.The joint modelling included a liner mixed model for the longitudinal marker trajectory,a proportional hazards model for the time-to-event analysis,and a multinomial logistic regression for latent classes process.We used Bayesian Information Criterion(BIC),posterior classification table,fitted values compared with observed values and conditional independence assumption check to assess the goodness-of-fit of the fitted model.Results:JLCM defined three latent classes for MCI conversion to AD: low-risk group,moderate-risk group and high-risk group.In survival model,risk factors for AD were ApoE ε4 carriers(hazard ratio[HR]: 2.96;95% confidence interval [CI]: 1.80,4.88),lower daily functional activity(HR: 1.12;95%CI: 1.06,1.18)for low-risk group;only lower daily functional activity(HR: 1.50;95%CI: 1.30,1.73)was a significant risk factor for moderate-risk group;while for high-risk group,female(HR: 40.51;95%CI: 5.05,324.94),lower educational level(HR: 9.05;95%CI: 1.97,41.62),ApoE ε4 carriers(HR: 4.70;95%CI: 1.86,11.85)and lower daily functional activity(HR: 1.16;95%CI: 1.04,1.29)were all significant predictors for AD.In longitudinal model adjusting for cognition,lower educational level(:-0.43;95%CI:-0.80,-0.06),ApoE ε4 carriers(:-0.90;95%CI:-1.11,-0.68)and lower daily functional activity(:-0.17;95%CI:-0.19,-0.15)would lead to cognitive decline for low-risk group;for moderate-risk group,the lower educational level(:-1.95;95% CI:-2.76,-1.13),the lower daily functional activity(:-0.38;95% CI:-0.42,-0.35),the greater cognitive impairment;for high-risk group,lower educational level(:-3.06;95% CI:-4.54,-1.58)and lower daily functional activity(:-0.69;95% CI:-0.73,-0.64)may lead to cognitive decline while married status(: 4.31;95% CI: 1.88,6.75)may increase cognitive level compared to single status.Conclusions:1.For MCI conversion to AD,there were three different latent classes in which traditional risk factors played different roles.The common risk factors for cognitive decline and risk of AD were ApoE ε4 and daily functional activity for low-risk group,daily functional activity for moderate-risk group,educational level and daily functional activity for high risk group,respectively.Cognitive stimulation and physical activity may delay the cognitive deterioration and dementia among the elderly.2.JLCM can be used to explore the relationship between longitudinal marker and outcome indicators.It will help to identify individuals at high risk for cognitive decline and to provide reliable epidemiological evaluation or strong statistical decision support for individualized prevention and early intervention for cognitive impairment among the elderly.