Exploring the link between biologically defined resilience, brain pathology, and cognitive functioning in old age and the contribution of biological sex

Introduction: Exposure to dementia risk factors throughout life can lead to brain atrophy and older- appearing brains on neuroimaging. Resilience mechanisms can help sustain the brain structure (brain maintenance, BM) and/or compensate for the neuropathological damage (cognitive reserve, CR), preserving cognition. Traditional proxy-based approaches (e.g., education) face challenges in measuring these mechanisms as they cannot capture the core biological dimension. However, with deep learning is possible to develop algorithms predicting the biological age of the brain from raw brain images. This study investigated whether differences between predicted brain age and chronological age (PBA-CA) can be used as a marker of BM and/or CR following the NIH-funded Collaboratory on Reserve and Resilience framework. Methods: The study population included 719 dementia-free septuagenarians from the Gothenburg H70-1944 MRI cohort. We applied the deep learning model—developed in-house using minimally processed T1-w MRI from ∼17000 neurologically intact individuals from UK Biobank, ADNI, AIBL, and GENIC and validated via hold-out and cross-validation approaches—to H70 participants’ MRI to predict their brain age, and computed PBA-CA. MRI markers of brain pathology included cortical thickness (overall and Alzheimer's disease-related), cerebral small vessel disease (SVD; individual markers and score), and white-matter microstructural alterations (DTI's fractional anisotropy). Global and domain-specific cognitive function was based on composite scores from ten tests. Data analysis included regression models and stratification by sex. Results: In the brain, decreasing differences between PBA and CA (reflecting younger-appearing brains) were associated with a thicker brain (overall and AD signature areas), lower SVD score—particularly lower white matter hyperintensities volume, lacunes, large infarcts—, and higher fractional anisotropy (more integrity). Decreasing differences were also related to better cognitive performance, globally and in attention/speed, executive function, and visuospatial abilities. In stratified analysis by sex, such associations were evident in men but not in women, except fractional anisotropy [robust regressions’ β-coefficients for women -4.89 (95%CI-8.87,-0.90) and men -23.7 (95%CI -32.9,-14.4)]. Discussion: Negative differences PBA