Understanding Cognitive Variability in Alzheimer’s Disease

Alzheimer’s Disease (AD) is highly heterogenous, both clinically and biologically. This variability is exacerbated by the ways within which, the clinical presentation is assessed with cognitive measures. This inhibits clinical trial success and earlier diagnosis of individuals. Marrying the clinical presentation to the pathology of the disease has so far proved troublesome. This thesis will look at how cognitive measures can best capture the clinical presentation of AD and how these measures can link to the underlying pathology using machine learning methods.

This thesis studied this problem across four analyses and two cohorts. Each study looked at a different aspect of cognitive testing within AD. This was done with the overarching aim to interrogate the cognitive variability across the spectrum of AD.

Study 1 showed a novel discrepancy score is different to memory measures at screening for AD. It also showed it tracks with AD severity, in the same way memory recall does. Studies 2 & 3 uncovered broad psychometric variance within amnestic measurement of impairment due to AD. This was done in two different populations across two different constructs of amnestic measurement, story recall and verbal list learning. These tests are frequently used interchangeably. These two studies show they should not be. Finally, Study 4 built models from cognitive measures to predict AD pathology. The performance of these models was moderate showing that even with novel cognitive measures, further work is needed to link the clinical and amyloid related biological presentations of AD.

Bridging the gap between clinical presentation and pathology of AD using clinical and cognitive markers alone is not possible. Even when using a novel measure of discrepancy score. The discrepancy measure shows promise but was limited due to the inability of the MMSE to measure verbal ability. Conceptually a discrepancy score remains a promising avenue of research for screening, but broader language measures, as well as other AD biomarkers are needed to further test the construct validity of this measure.