Multimodal Neuromarkers Predict Successful Physical Activity Engagement in Older Adults with Vascular Disease

Abstract

AbstractBackgroundPhysical inactivity is a significant risk factor for cognitive decline, particularly in vulnerable populations including those with cardiovascular risk‐factors1. To better understand the mechanisms that support positive physical activity engagement, the current study aimed to evaluate whether neuromarkers can predict future change in physical activity among older adults with a newly diagnosed cardiovascular risk‐factor.MethodWe analyzed baseline resting‐state functional and structural magnetic resonance imaging (MRI) data from the UK‐Biobank, a large population longitudinal cohort (n = 365; mean age = 62.10 years ± 6.5; cognitively normal). Brain imaging was obtained at baseline, and physical activity data was obtained at baseline and follow‐up after 5 years. Inclusion criteria were a new diagnosis of a cardiovascular risk factor (hypertension, type‐II diabetes‐mellitus, dyslipidemia, cardiac angina or myocardial infarction) between baseline and follow‐up; and did not meet the World Health Organization recommended 150 minutes/week of moderate or 75 minutes/week of vigorous physical activity at baseline. Demographic variables including age, sex, and education were included as covariates of non‐interest.To assess whether baseline resting‐state brain imaging predicts future change in physical activity behaviour, we performed a kernel‐ridge‐regression model with 5‐fold nested‐cross‐validation. Preprocessed neuroimaging data was used as input for the analysis, and the machine‐learning pipeline included feature reduction using grid‐search, hyperparameter selection, and model‐building: the model predicts whether test subject successfully increased physical activity engagement at follow up (reported above 150 min/week moderate or 75 min/week vigorous physical activity). Permutation‐testing (100 times) was performed before evaluating prediction metrics using accuracy and Receiver Operating Characteristic curves.ResultOur prediction model delivered an accurate performance with average accuracy of 0.861 ± 3.8, and 0.883 ± 4.2 area under the curve (AUC) in predicting future behaviour change from physically inactive at baseline to physically active in follow up.ConclusionThese results show baseline brain imaging can accurately predict physical activity behavior change in older adults with new cardiovascular disease. Leveraging machine‐learning methods to predict future lifestyle engagement will help characterize the neural mechanisms that support successful lifestyle change after a new cardiovascular diagnosis.1. Gallen et al. Trends in Cog Sci. 2019