Education and other cognitively stimulating activities (CSA) are potentially modifiable factors which may improve cognitive maintenance in later life. As these social exposures are not possible to effectively randomise, inferences from observational data are particularly important. Existing research findings differ regarding whether CSA are associated with cognitive maintenance. This may be explained in part by limitations and implicit assumptions in the most commonly used methods to analyse these associations. This thesis asks how modifiable social exposures affect cognitive maintenance and examines some of the assumptions underlying standard methods such as growth modelling. The analysis uses the English Longitudinal Study of Ageing (ELSA), a nationally representative longitudinal cohort study of adults aged over 50 living in England. A series of assumptions in standard regression approaches and their implications for the association between education or CSA and cognitive maintenance are examined. Firstly, ELSAâs scoring method for memory and executive function is examined using factor analysis. The memory score performs well, but the executive function score does not reflect the data. This leads to incorrect estimation of the association between cognitive maintenance and some important predictors such as age. I then tested for longitudinal measurement invariance (MI) in the cognitive factors and found this did not hold for memory in ELSA using Bayesian approximate MI. This is an advance on conventional tests of MI which had found equivocal results. The assumption that the ELSA sample is drawn from one homogenous population, and that the effect of education on cognitive maintenance is the same across sub-populations, were then tested using growth mixture modelling. A small beneficial effect of higher educational attainment on cognitive maintenance was found in a stable cognition latent class but no association was seen in latent classes with declining cognition. If CSA participation improves cognitive maintenance, and better cognition increases the likelihood of participation in CSA, this generates time varying confounding affected by prior exposure. Standard growth curves must assume this to be absent. Using inverse probability of treatment weighted marginal structural models to relax this assumption, volunteering and internet use activities were still found to reduce the risk of dementia or cognitive impairment. This research contributes methodologically to the existing literature by demonstrating how some of the assumptions underpinning the regression models most commonly used to estimate the association between CSA and cognitive maintenance can influence the substantive conclusions drawn. Specifically, it finds that ELSAâs executive function index does not represent the data well and Bayesian approximate MI can be used to clarify equivocal conventional tests of longitudinal measurement invariance of the cognitive test factors. Substantively, I find that the effect of education on cognitive maintenance varies somewhat depending on underlying trajectory, and that the association of volunteering and internet use activities with improved cognitive maintenance is robust after time varying confounding is accounted for.
Date of Award | 1 Aug 2020 |
---|
Original language | English |
---|
Awarding Institution | - The University of Manchester
|
---|
Supervisor | Neil Pendleton (Supervisor) & Tarani Chandola (Supervisor) |
---|
- cognition
- marginal structural model
- growth mixture model
- factor analysis
- internet use
- volunteering
- measurement invariance
- education
- dementia
- cognitive impairment
- cognitive reserve
- cognitively stimulating activities
Measuring cognitive maintenance in older adults and its association with education and other cognitively stimulating activities
Williams, B. (Author). 1 Aug 2020
Student thesis: Phd