Can Critical Behaviour Change Windows Successfully Predict Physical Activity Engagement in Aging?

1473 

Abstract Submission 

Nagashree Thovinakere¹, Sue-Jin Lin², Robert Baumeister², Yasser Iturria Medina³, Maiya Geddes²

¹McGill University, Westmount, Quebec, ²McGill University, Montreal, QC, ³McGill University, Montreal, Quebec

Nagashree Thovinakere  
McGill University
Westmount, Quebec

Sue-Jin Lin  
McGill University
Montreal, QC

Robert Baumeister  
McGill University
Montreal, QC

Yasser Iturria Medina, YIM  
McGill University
Montreal, Quebec

Maiya Geddes  
McGill University
Montreal, QC

Physical activity is a crucial modifiable factor in preventing cognitive decline and dementia1. Lifestyle changes following a new cardiovascular disease diagnosis provide an opportunity for improved health outcomes2. However, successful behaviour change remains challenging, necessitating a deeper understanding of neurobiological mechanisms. This study aims to identify functional brain features predicting successful behaviour change (increased physical activity) among older adults with new diagnosis of a cardiovascular risk factor.

Methods: We analyzed baseline resting-state functional magnetic resonance imaging (rs-fMRI) data in a subsample from the UK Biobank, a large population longitudinal cohort (n=295; mean age = 63.13 years ± 7.5; cognitively normal). Brain imaging was obtained at baseline, and physical activity data was obtained for two time-points: Baseline and follow up after 5 years. Participants met the following inclusion criteria: 1) reported a new diagnosis of a cardiovascular risk factor (i.e., hypertension, type II diabetes, dyslipidemia, cardiac angina or myocardial infarction) between baseline and follow-up, and 2) did not meet the World Health Organization recommended 150 minutes/week of moderate-to-vigorous physical activity (MVPA) at baseline, 3) age >= 60, cognitively unimpaired. Self-reported MVPA were recorded using the Lifetime Total Physical Activity Questionnaire. Demographic variables including age, sex, years of education, and socioeconomic status were included as covariates of non-interest.

To assess whether baseline rs-fMRI connectivity predicts future change in physical activity behaviour, we used a Random Forest Classifier. Preprocessed rs-fMRI data from the UK Biobank was used as the input for these analyses. The Schaefer 2018 atlas was used to parcellate the brain, with 400 regions of interest (ROIs) and 7 rs fMRI networks. The analysis followed a nested cross-validation approach with 5 inner loop resampling and 10fold outer loop cross-validation to prevent overfitting. Our machine learning pipeline involved: 1) feature reduction using grid search, 2) hyperparameter selection and tuning within the inner loop, and 3) model building with the best cost value for each fold in outer loop cross-validation. Model performance, averaged across all 10folds in the outer loop, was assessed using two metrics-accuracy and Receiver Operating Characteristic (ROC) curves. The significance of accuracy was determined through a permutation test, repeating 1000 times to establish the null distribution. The post-hoc analysis involved extracting weights for all features from the best estimator, ranking the absolute values of coefficients, and selecting the top 30 features.

Results: Our prediction model delivered a highly accurate performance with mean accuracy (0.80 ± 0.05) and Area Under Curve (0.78 ± 0.08) in predicting future behavior change (≥150 min/week MVPA at follow-up). The frontal operculum/insula node in the Salience Ventral Attention Network emerged as the most critical predictor. Enhanced between-network connectivity, particularly between visual and somatomotor networks and transmodal brain networks (Default Mode and Salience) was associated with successful behaviour change.

The finding that functional network differentiation supports successful physical activity behaviour aligns with recent findings suggesting a shift in overall network activity balance during external versus internally guided decision-making. Notably, localized activity within unimodal networks supports cognition reliant on immediate perceptual input, while greater segregation of unimodal from transmodal networks supports internally oriented processing, including self-referential processes5. Speculatively, this suggests that functional segregation may sustain long-term engagement in physical activity.

Higher Cognitive Functions Other

Aging ²

Classification and Predictive Modeling ¹

Connectivity (eg. functional, effective, structural)

fMRI Connectivity and Network Modeling

Aging

FUNCTIONAL MRI

Machine Learning

Modeling