Multivariate analysis and prediction of inter-relationships between cognitive and non-cognitive measures in the CLSA comprehensive sample

There is a growing understanding that different facets of “healthy cognitive aging” are strongly influenced by a large number of non-cognitive variables such as diet and lifestyle, physical activity, nutrition, and other physical health indices. Moreover, evidence suggest that the influence of such variables form correlated components that, considered together as a cluster or latent factor, have greater explanatory and predictive power than any individual measure examined in isolation. However, to reliably identify these kinds of latent factors requires large studies that collect hundreds or even thousands of measurements on large groups of individuals. The Canadian Longitudinal Study in Aging (CLSA) offers exactly this kind of large-sample dataset that is needed to capture the complex multivariable relationships between cognition and other non-cognitive domains. Using statistical methods designed to capture commonalities across large sets of variables, we will capture the elements of lifestyle, physical health, and social participation that are most strongly related to cognitive performance in an aging population in Canada.

With this study we developed multivariate techniques for the analysis of multi-table datasets with variables grouped according to different domains (e.g. behavioural, physiological, cognitive). We developed new open source tools for latent variables analyses, including multiple factor analysis, multi-view discriminant analysis, and kernel manifold alignment. These methods allow one to map groups of variables to a common latent space to better understand and visualize relationships between sets of variables. It also allows for what is called “domain transfer”, that is, the ability to translate from one domain to another, using a linear or nonlinear transformation. The code developed for these analyses are available on Github at the following URLs:

• https://github.com/bbuchsbaum/multivarious
• https://github.com/bbuchsbaum/genpca
• https://github.com/bbuchsbaum/musca