Methodology And Applications Of Global Cognitive Function Trajectory Based On Latent Class Mixed Model

BACKGROUND:The prevalence of Alzheimer’s disease(AD)is increasing,causing a severe public health problem.The process of AD is slow and indetectable,and the effective treatments are lacked,so that primary preventions are the main strategies for preventing AD.Trajectory analyses based on the global cognitive function has become the important measurement to identify high-risk sub-group and explore risk factors for AD.Global cognitive function could be evaluated by Mini-Mental State Examination(MMSE)directly,or be constructed by multidimensional scale scores.Based on these two kinds of datasets,researchers have reported various cognitive function trajectory literatures.However,there are still two problems should be solved.Firstly,the assumption of latent class mixed model(LCMM)is that the response variable is a continuous variable obeyed normal distribution.However,the scale scores are categorical variables.The rationality of direct analogy of categorical variable to continuous variable is still unknown.Furthermore,previous researches have used multivariate latent class mixed model(MLCMM),total score method,or calculating the mean value of the standardized scores to establish trajectory model of multidimensional scale scores.Nevertheless,the selection of methods has still lacked theoretical guidance.Used simulation studies and applications,this research aims to explore the effect of the class number of the variable on the performance of LCMM and compare the performance of different joint trajectory method furtherly,and based on the result of simulation studies to identify the trajectory of MMSE,the joint trajectory of episodic memory and verbal fluency,and explored their risk factors.METHODS:In this research,simulated dataset was generated based on LCMM,and then the response variable was translated as categorical variable based on suitable cut-off values.Subsets were sampled from the total dataset,and the effects of the category number on the probability of correct classification and the performance of model-selection of LCMM in different sample sizes,missing and censored data were explored.Every simulation was repeated 1000 times.Based on these results,we compared the performance of MLCMM,total score method,and the method of calculating the mean value of standardized scores,furtherly.Based on the Chinese Longitudinal Healthy Longevity Survey(CLHLS),we used LCMM to identify the trajectories of MMSE and explored the relationship between rest heart rate(RHR)and trajectory groups.Using Survey of Health,Ageing and Retirement in Europe(SHARE),we chose the suitable measurement to identify the joint trajectory of episodic memory and verbal fluency,and estimated the effects of mental factors and lifestyle on trajectory groups.RESULTS:①when the class number of the variable was more than 10,the performance of LCMM was close to the counterpart of continuous variable,with great probability of correct classification and model selection efficiency;② the probability of correct classification of MLCMM,total score method,and the method of calculating the mean value of the standardized scores was similar,but in small sample(N<1000),higher probability of missing and censored data(π>0.3)circumstance,the accuracy of parameters of MLCMM was the lowest;③considering the probability of correct classification and the rate of convergence,MLCMM was not suitable to fit trajectory when the correlation was lower than 0.4;④when the category difference is greater than 10,it is not suitable to use the total score method for modeling.Based on CLHLS,three trajectory groups were identified during 75-100 years old:high-stable(n=1298),moderate-decreasing(n=1620),and rapid-decreasing(n=382).Both individuals with moderate-decreasing and rapid-decreasing changing patterns were defined as cognitive decline.Logistic regression models showed that compared with RHR between 60 bpm and 69 bpm,those with 70-79 bpm,80-89 bpm and more than 90 bpm all performed high risk of cognitive decline,with ORs(95%CI)1.27(1.03～1.56),1.30(1.01～1.67),and 1.62(1.07～2.47)respectively.Using SHARE,we found the category of immediate memory,delayed memory and verbal fluency were 10,10,and 100.The correlations of these three variables were 0.3-0.6,so that calculating the mean value of the standardized scores to establish the global cognition was the best strategy.We then discovered four distinct trajectory groups of those during 50-75 years old,and depression and alcohol drinking were significant influencing the cognition changing patterns.In the comparison between persist low level group(n=1100)and persist high level group(n=1265),the ORs(95%CIs)of 4-5 and 6-12 scores of depression were 1.74(1.35～2.25)and 3.10(2.14～4.45).While alcohol drinking was a protective factor whose ORs(95%CIs)were 0.79(0.69～0.91)0.60(0.52～0.70)and 0.44(0.36～0.53)for persist middle-high,persist middle-low,and persist low group,respectively.CONCLUSIONS:①LCMM could use to identify trajectory groups when category was more than 10;②the correlation and category difference of variates should be considered when the strategy of joint trajectory was selected;③there were three distinct trajectory groups during 75-100 years old in Chinese,and RHR more than 70 bpm was an important risk factor of cognitive decline;④in the European old individuals,four cognitive function trajectories was discovered,and were influenced by mental factors and lifestyle.