Longitudinal Changes in Brain Diffusion Characteristics Associated With Cognition and Vascular Risk Factors: The ARIC-NCS Study

Abstract

Background and objectives: It remains unclear whether longitudinal changes in brain microstructural integrity, measured by diffusion MRI, relate to cognition and vascular risks. We investigated whether annualized changes in fractional anisotropy (FA) and mean diffusivity (MD) are associated with changes in cognitive domains among nondemented older adults, and how these patterns differ by vascular risk factors.

Methods: Data were longitudinally collected from the Atherosclerosis Risk in Communities Neurocognitive Study, conducted across 4 US sites with 6,471 participants attending the baseline assessment. Individuals who underwent diffusion MRI and neurocognitive evaluations at least twice between 2011 and 2019 were included; those with baseline dementia were excluded. Linear mixed-effects models assessed associations between FA and MD values in 140 brain regions and domain-specific cognitive scores (executive function, language, and memory). Annualized changes in FA and MD values were compared between individuals with and without vascular risk factors.

Results: 592 participants (mean age: 75.8 ± 4.6 years; 56% women) were followed for 6 years on average. A 1-SD decrease in FA values in the left cingulum bundle was associated with a 0.166-SD reduction in executive function (95% CI 0.065-0.267; p = 0.025) and a 0.158-SD reduction in language (95% CI 0.054-0.263; p = 0.031). In addition, a 1-SD increase in MD values in the left hippocampus was associated with a 0.191-SD reduction in memory (95% CI -0.306 to -0.076; p = 0.009). Significant differences in annualized changes in these FA and MD values were found depending on the presence or absence of type 2 diabetes and smoking.

Discussion: Deterioration in microstructural integrity-reflected by lower FA in the left cingulum bundle and higher MD in the left hippocampus-was associated with declines in executive function, language, and memory. Furthermore, longitudinal changes in these FA and MD values had differences based on the presence of type 2 diabetes and smoking. These findings provide a foundational basis for future research to determine whether managing vascular risk factors can delay cognitive decline by affecting the microstructural integrity of the brain.

Figure 1.. Flow chart of the participant selection

Of the 6,471 participants who completed the ARIC-NCS in-person assessment, 1,978 participants underwent visit 5 brain MRI scans. In the present study, we selected the participants who fulfilled the following inclusion criteria: 1) MRI scans with DTI sequences undergone at visit 5 and followed at visit 6 or 7; 2) detailed cognitive performance and vascular risk factor surveillance performed through visits 5 to 7. Then, we excluded participants who had prevalent dementia at visit 5 (N = 110) and those whose neuropsychological assessments ended on or prior to visit 5 (N = 71). From a total of 615 eligible participants, those with incomplete/poor-quality MRI exams (N = 15) and missing cognitive function measures (N = 8) were further excluded, resulting in the study participants of N = 592.

Figure 2.. Correlations between longitudinal changes in DTI metrics and cognition

An overview of correlation analyses between annualized changes (Δ) in FA and MD values and those in cognitive function scores is provided. Scatterplot and linear regression with 95% CI are shown in each panel. P indicates Bonferroni-corrected P value; r, Spearman’s rank correlation coefficients.

Figure 3.. Differences in annualized changes in DTI metrics by vascular risk factors

Annualized changes (Δ) in FA and MD values within regions surrounding the limbic system were presented, stratified by the presence or absence of hypertension, dyslipidemia, type 2 diabetes, smoking, drinking, and obesity. Asterisks indicate significant differences in these values between participants with and without each vascular risk factor (*Bonferroni-corrected P <0.05; **Bonferroni-corrected P <0.01).