Brain structure and function predict adherence to an exercise intervention in older adults
Individualized and precision medicine approaches to exercise for cognitive and brain health in aging have the potential to improve intervention efficacy. Predicting adherence to an exercise intervention in older adults prior to its commencement will allow for adaptive and optimized approaches that could save time (no need to demonstrate failure before changing course) and money (cost of de-implementing approaches that do not work for certain individuals) which ultimately could improve health outcomes (e.g., preventative medicine approaches prior to the onset of symptoms). Individual differences in brain structure and function in older adults are potential proxies of brain and brain reserve or maintenance and may provide strong predictions of adherence. We hypothesized that brain-based measures would predict adherence to a six-month randomized controlled trial of exercise in older adults, alone and in combination with psychosocial, cognitive and health measures. In 131 older adults (aged 65.79 {+/-} 4.65 years, 63% female) we found, using regularized elastic net regression within a nested cross-validation framework, that brain structure (cortical thickness and cortical surface area) in somatosensory, inferior temporal, and inferior frontal regions and functional connectivity (degree count) in primary information processing (somatosensory, visual), executive control, default, and attentional networks, predicted exercise adherence (R2 = 0.15, p < 0.001). Traditional survey and clinical measures such as gait and walking self-efficacy, biological sex and perceived stress also predicted adherence (R2 = 0.06, p = 0.001) but a combined multimodal model achieved the highest predictive strength (R2 = 0.22, p <0.001). Neuroimaging features alone can predict adherence to a structured group-based exercise intervention in older adults which suggests there is substantial utility of these measures for future research into precision medicine approaches. The best performing model contained multimodal features suggesting that each modality provided independent relevant information in the prediction of exercise adherence.