Multimodal latent composites are associated with cognition and Alzheimer’s disease dementia: a framework for systems-level brain health

Abstract Biomarkers from diverse methodological domains are increasingly important in the detection, diagnosis and tracking of neurological diseases and brain health, yet they are often evaluated in isolation. Statistical integration approaches, such as factor analysis, provide a means to combine complementary biomarkers and capture higher-order domains of brain health. Exploratory factor analysis has previously been employed to identify latent brain health constructs using multimodal MRI, fluid biomarkers and cardiovascular risk factors in a non-clinical older population. The current study aimed to validate this integrative framework using confirmatory factor analysis in an independent cohort and test construct associations with cognition and diagnosis of mild cognitive impairment (MCI) or Alzheimer’s disease (AD) dementia. Data were analysed from 197 participants in the Prospective Imaging Study of Ageing, including 157 cognitively normal controls (CN), 18 participants with MCI and 22 participants with early AD dementia. MRI, cardiovascular, and plasma biomarker processing closely replicated previous methods. Confirmatory factor analysis was conducted in CNs to validate the previously reported latent constructs. Weighted factor composites were then compared between each diagnostic group and tested for associations with cognitive performance (verbal reasoning, verbal memory, visual memory and language) and sensitivity to MCI and AD diagnosis. Three factors were reproducible across cohorts: 1) Brain & Vascular Health (hippocampal and ventricular volumes, cerebral blood flow); 2) White Matter (WM) Fluid Dysregulation (Free Water, WM enlarged perivascular spaces); and 3) Blood Biomarkers (GFAP, NfL, pTau181). Compared to the CN group, both MCI (β=−1.25, SE=0.19, p