STATISTICAL ANALYSIS.

Baseline characteristics of the study population by FGCRS tertiles were compared using chi-square tests for categorical variables and 1-way analysis of variance or Wilcoxon rank sum tests for continuous variables.

Linear mixed-effects models were used to estimate β-coefficients and 95% confidence intervals (CIs) for the associations between cardiovascular risk burden (i.e., continuous and categorical FGCRS) and annual change in global cognitive function and 5 cognitive domains, with follow-up time (in years) as the time scale. The fixed effect included cardiovascular risk burden, follow-up time, and their interaction. The random effect included random intercept and slope, allowing the individual differences at baseline and across follow-up. Age, sex, education, BMI, stroke, heart disease, alcohol consumption, physical activity engagement, and APOE ε4 were adjusted for as potential confounders in the multiadjusted model. To further explore the role of APOE ε4 in the association between FGCRS and cognitive function, an interaction term between FGCRS categories and APOE ε4 status was also included in the models first; then, stratified analysis by APOE ε4 was performed.

In MRI analyses, linear regression was used to estimate β-coefficients and 95% CIs for the relationship between cardiovascular risk burden using continuous FGCRS and brain volumes. Mixed-effects models were used to estimate the association between brain volumes and cognitive functions. All models were adjusted for age, sex, education, BMI, stroke, heart disease, physical activity engagement, alcohol consumption, global cognitive function, and APOE ε4 as potential confounders. In the sensitivity analysis, we excluded 385 individuals with MCI at baseline and reran the linear mixed-effects model. All statistical analyses were performed using Stata SE 15.0 for Windows (StataCorp, College Station, Texas). To broadly investigate the association between cardiovascular burden and cognitive decline, multiple comparisons were not mathematically corrected for to reduce the chance of type II error. A 2-tailed p value <0.05 was considered to be statistically significant for all tests.