The Application Of Structural Equation Modeling For The Study Of Cognitive Aging And Associated Factors

【Objective】In the modern society, the growing aging problem, together with issues on economy, health, social security, etc, has caught people’s attention. Many diseases caused by degeneration of cognitive function, such as mild cognitive impairment and Alzheimer’s, seriously damage the quality of people’s later years, which have raised special concern in public. Therefore, exploring the evolution of cognitive aging, determinating the risk factors and protective factors in the process of aging and clarifying the interactions between cognitive functions play a significant role in delaying the cognitive aging, improving the life quality and optimizing the health evaluating system of the aged. At present, many studies have been done on cognitive aging; however, studies on cognitive functions of the aged in China based on large samples, on comprehensive exposition and systematic analysis about various factors’ influence on cognitive functions and on the guidance of the structural equation modeling in processing multigroup and multilevel data are lacking. This paper aims to accurately reveal the changes and trajectories of various cognitive measurement indexes, to comprehensively explain the evolution rules influenced by physiology, psychology, society and other aspects, to further refine the specific conditions of occurrence and change of effects with aging, and to enhance the accuracy of cognitive measurement and evaluation of the elderly. Structural equation modeling has been used more and more widely in the field of cognitive neuroscience. The study based on this method further builds a multi-group structural equation modeling and multilevel structural equation modeling, which forms a set of statistical analysis guidance for this kind of data and provides more methodological basis and supports for the application of structural equation modeling to the study of cognitive aging.【 Content 】 Based on the BABRI database, this research uses a piecewise linear regression to explore the trajectories of episodic memory, visual spatial, working memory and processing speed with aging based on related tests, to find out the critical age window for the changes of tests scores, to determine the sequence of cognitive decline, and to provide reference for the extraction of intermediary cognitive function. The research builds the SEM of interaction between factors and cognitive functions based on single field and multiple fields respectively, which determine the risk and protective factors in the process of aging, verify the intermediation of cognitive function, and find out the direct and indirect effects between factors and functions. The multigroup SEM is used to compare the different model parameters in order to discuss the effects of factors as well as the invariance of cognitive tests and measuring reliability between groups. For the hierarchical related data, the MSEM is used to describe the relationship between variables of all hierarchies and to explore the evolution of cognitive abilities under the combined effects of variables within and between groups.This study makes full use of various statistical analysis methods and professional analysis tools, such as SAS, LISREL and MPLUS. The research not only conducts a thorough discussion on related influence factors and mode of action with cognitive aging, but also provides a lot of methodological guidance in practical application of SEM.【Methods】Based on the BABRI database established by Beijing Normal University, the study collects demographical, behavioral, cognitive ability evaluation and clinical data for cognitive aging analysis. First, regression analysis is used to preliminarily screen candidate factors which may influence the cognitive abilities. Then, the piecewise linear regression is applied to study the trajectories of each cognitive test scores with aging, which aims to find out the key period of recession in each cognitive ability and provide reference for the hypothesis of intermediary cognitive ability. SEM in the single field is used to analyze the influence of the factors on cognitive abilities, with focus on direct effects between cognition and factors. SEM in multiple fields is used in further study on possible interaction between cognitive abilities, with focus on indirect effects between cognition and factors. To explore the differences of cognitive aging evolution in different groups affected by external or internal factors, multigroup SEM is used to compare the significance and effect size of path coefficient between models of different genders or different education levels, etc. Finally, the average values(or proportions) of individual level variables are computed as group level variables based on the original data of processing speed score. In the condition that observations within groups do not meet the independence, MSEM is used to explore the interaction of various factors and other cognitive abilities affected by differences within and between groups.【Results】(1) The piecewise linear regression is done for each cognitive test score with aging in the study. The results show that three indicators of MEMORY(N5, N1N5 and RODELAY) contain single inflection point respectively. The inflection points in N5 and N1N5 are found at the age of 60 or so, while the RODELAY’s inflection point appears later, around the age of 73. In the SPACE field(ROCOPY, CLOCK), the slope of ROCOPY test which shows a relatively stable trend of change during the age window is not significant, while a critical period of decline is observed in the CLOCK test, which may appear at 58-66 years old. Two indicators of WKMEMORY(FORWARD, BACKWARD) both contain an inflection point. The slope of the FORWARD test is insignificant before or after the inflection point with a large fluctuation during the age range, and the inflection point of the BACKWARD test appears around 58 years old, with insignificant slope as FORWARD. Four indicators of PROCESS(SDMT, TMTA, STRA and STRB) are affected by age remarkably. SDMT and TMTA are monotonously changing without any inflection point; while the STRA and STRB tests both contain an inflection point with about 5 years gap of age.(2) SEM is constructed for individual cognitive function affected by factors, and multigroup SEM is used to compare model parameters of different groups. The results show that MEMORY declines with increasing age. Females, with high education level and regular participation in leisure activities, perform better in memory field. The MEMORY models of male and female share the same settings and path coefficients, and the same with the situation of carrying ε4 or not, but different levels of education will affect the interpretation of the MEMORY latent variable. Male, highly educated subjects perform better in visual-spatial ability test. ROCOPY test reflects the visual-spatial ability of lowly educated females with relatively higher reliability, while CLOCK test better describes males’ visual-spatial ability. In working memory field, highly educated males who live a regular life and often participate in leisure activities have better working memory performance. The ε4 allele may affect the interpretation of the WKMEMORY latent variable, and the regularity of life in ε4 non-carriers has significant effect on WKMEMORY. The comparisons between the models of different gender and educational levels have no difference. Diabetic subjects with low level of education, who do not often participate in activities, have slightly worse performance in processing speed field. The evaluation of processing speed ability using STRB test for men can achieve higher reliability. Whether they carry the ε4 allele or have high educational level does not affect the model fitting.(3) SEM in multiple fields is used to find out the existence of the interaction between various cognitive abilities, and to explore the indirect effects from factors to indicators. The results show that there are significant impacts derived from the processing speed to the other three cognitive abilities, and memory and working memory are also adjusted by visual-spatial ability. Factors such as age, leisure activities, and diabetes perform significant direct effects on processing speed and indirect effects on the other three abilities. Gender has both direct and indirect effects on memory. Direct effects are observed from education to all the four cognitive abilities, and the effect to visual-spatial and working memory abilities contain two parts. The law of life only affects working memory, and hypertension can affect episodic memory and working memory through visual-spatial function.(4) Based on processing speed scores, MSEM is used to analyze the changes of effects between factors and other cognitive abilities under similar and dissimilar degrees of processing speed performance. The results show that in the same group, the younger female individuals have better memory performance; however, in different groups, subjects with high average participation in leisure activities perform better in memory field. Highly educated subjects with high average educational years are better at memory tests. In the field of visual-spatial ability, male individuals are superior to females in the same group; as for education, it is exactly the same with memory. In regard to working memory, gender, education, leisure activities and regularity of life all have a significant impact on the individual level. The analysis of multiple domains show that the impact from visual-spatial ability to episodic memory is both present within and between groups, indicating that the individual level variables(individual gender and individual level of education) and group level variables(average level of education) can have significant indirect effects on memory through visual space simultaneously. However, the influence from visual space to working memory exists only within groups.【Conclusions】In this study, the cognitive measurements of the aged from eight communities in Beijing are thoroughly analyzed, which obtains satisfactory results. Our study suggests that the critical declining period of processing speed ability appears to be earlier than several other cognitive abilities, indicating that the processing speed ability is likely to be a key factor in overall cognitive aging process. The following multi-domain SEM verifies the significant unidirectional effect from processing speed to the other abilities. This conclusion will help us conduct interventions to delay the process of cognitive decline in the elderly more precisely and more specifically. The study also finds out that the reliability of ROCOPY, TMTA and other tests can be affected by individual differences, which has an important reference value for improving the completeness and accuracy of cognitive tests. At the same time, MSEM is applied to resolve the modeling problem of related data. Our study also summarizes and concludes the change under the combined effects of different hierarchies, aiming to provide guidance and basis for promoting the application of this statistical method.