Development and validation of objective measures of brain maintenance and cognitive reserve

Abstract

Age-related cognitive decline is an increasingly important societal issue, given projected increases in the proportion of older adults in the coming decades. Early identification of cognitive decline will enable earlier interventions which have a greater likelihood of slowing decline, maintaining quality of life, and reducing the burden on caregivers and society. The application of machine learning to neuroimaging data is a promising strategy to detect age-related cognitive decline. However, there has been little emphasis on the development of measures of two related, yet separable, constructs brain maintenance and cognitive reserve both of which support cognitive function as we age. Accurate measurement of these two constructs may improve our ability to detect age-related cognitive decline. Chapter 2 applied a machine learning method to structural MRI data in order to predict chronological age. Brain-predicted age difference scores were then created by subtracting chronological ages from the predicted age. A penalised regression with cross-validation was applied to generate the model using open-access structural MRI data. This model was then applied to structural MRI data in three independent datasets. Across these independent datasets, brain-predicted age differences were negatively correlated with measures of general cognitive status; semantic verbal fluency; executive function; and executive function (without processing speed). These results provided firm evidence of a robust relationship between increased brain-predicted age differences and reduced cognitive function in specific domains. As such, the findings established the validity of brain-predicted age difference scores as an operational measure of brain maintenance. Chapter 3 applied a data-driven framework in order to establish the validity of different socio-behavioural variables as proxy measures of cognitive reserve in a cross-sectional study of cognitively healthy older adults. To demonstrate face validity as a measure of cognitive reserve, candidate neuroimaging measures must be shown to correlate with a socio-behavioural proxy. Furthermore, socio-behavioural proxies are the most commonly used measures of cognitive reserve. However, there is little empirical evidence demonstrating the validity, and guiding the choice, of proxy variables as measures of cognitive reserve. The validity of five common proxies and all possible combinations of their composites were assessed across two community-dwelling older adult cohorts. Verbal intelligence was found to be the most robust socio-behavioural proxy measure of cognitive reserve. Chapter 4 applied a novel machine learning method, connectome-based predictive modelling, to functional connectivity data in order to develop and validate an objective measure of cognitive reserve. This measure was developed using task-based functional connectivity data from one dataset and then applied to resting-state fMRI data from an independent dataset. Face validity of the measure was assessed by establishing its association with the most robust socio-behavioural proxy measure of CR, verbal intelligence, as identified in Chapter 3. The protective effects of the measure was assessed by establishing its association with cognitive function, independent of brain structure. The measure accurately predicted CR in the training set and was validated as a measure of CR as it demonstrated face validity and protective effects on cognition. However, the measure was not validated in the independent dataset when generated using resting-state data. Overall, the findings demonstrate the value of machine learning for the development of robust and objective measures of brain maintenance and cognitive reserve using neuroimaging data. Chapter 2 established that brain-predicted age difference scores can serve as a valid measure of brain maintenance across cohorts, and may prove to be useful biomarkers of cognitive ageing. Chapter 3 identified verbal intelligence as the most robust socio-behavioural proxy of cognitive reserve and therefore recommended that researchers should use this variable when assessing the face validity of potential cognitive reserve neuroimaging measures. Chapter 4 developed a functional neuroimaging measure of cognitive reserve based on task-based fMRI data but further research is needed to validate this measure using resting-state data. Further innovations to the models outlined in Chapters 2 and 4 may provide important insights into the development and enhancement of brain maintenance and cognitive reserve and will further improve our understanding of these constructs. These validated measures of brain maintenance and cognitive reserve could be used to improve the early identification of cognitive decline and to directly assess the efficacy of preventative interventions targeted at the enhancement of brain maintenance and/or cognitive reserve.