Connectome‐based predictive modeling of cognitive reserve. (31st December 2021)

• Record Type: Journal Article
• Authors: Boyle, Rory; Connaughton, Michael; McGlinchey, Eimear; Knight, Silvin P.; De Looze, Celine; Carey, Daniel; Stern, Yaakov; Robertson, Ian H.; Kenny, Rose Anne; Whelan, Robert
• Is Part Of: Alzheimer's & dementia. Volume 17(2021)Supplement 5
• Page Start: n/a
• Page End: n/a
• Publication Date: 2021-12-31
• Subjects: Alzheimer's disease -- Periodicals; Alzheimer Disease -- Periodicals; Dementia -- Periodicals; Démence; Maladie d'Alzheimer; Périodique électronique (Descripteur de forme); Ressource Internet (Descripteur de forme); 616.83
• DOI: 10.1002/alz.057654
• Languages: English
• ISSNs: 1552-5260

Abstract

Abstract: Background: Cognitive reserve (CR) enables individuals to maintain cognitive function despite the presence of pathology or atrophy. CR may be most directly measured using functional neuroimaging. Connectome‐based predictive modeling can develop measures of cognitive phenotypes using task‐fMRI that can be applied to resting‐state fMRI, which is easier to collect and unaffected by task‐related confounds. We aimed to develop and validate a CR measure that could be applied to resting‐state fMRI data. Method: Paper Folding task‐fMRI scans from 220 participants of the Cognitive Reserve/Reference Ability Neural Network studies formed a training set. Resting‐state fMRI scans from 294 participants from The Irish Longitudinal Study on Ageing formed a test set. fMRI data were preprocessed, motion‐corrected and parcellated into 205 nodes using the Shen atlas, excluding 63 brainstem and cerebellar nodes due to incomplete coverage. Functional connectivity matrices were calculated via Fisher z‐normalised correlations between the mean time series of each node‐pair. Connectome‐based predictive modeling with leave‐one‐out cross‐validation was used to predict a CR residual from task‐based functional connectivity in the training set (Fig. 1). This generated three network‐strength predicted CR measures. We established the validity of these measures via a positive correlation with a CR proxy (NART scores) and global cognition, independent of brain structure. We then applied the model to the test set and assessed the validity of the resulting network‐strength measures. Result: The three network‐strength measures accurately predicted the CR residuals of unseen participants (Fig. 2) and demonstrated validity as they were positively correlated with NART scores (Fig. 3) and global cognition, independent of brain structure. However, application of the model to the resting‐state data in the test set did not generate accurate predictions of the CR residual (Fig. 4) and the resulting measures did not correlate with NART scores nor global cognition (Fig. 5). Conclusion: Validated measures of CR were generated using task‐based fMRI data but these measures did not generalise to resting‐state fMRI from an independent dataset. The inability to generalise may be because CR involves a reorganisation of connectivity in response to task demands or that the data here excluded the cerebellum.