REGARDS is a national cohort study consisting of 30,239 community-dwelling Black and White adults from the contiguous United States. It was designed to examine causes for excess stroke mortality in Black compared to White adults, and in those residing in the Stroke Belt of the southeastern United States (where stroke mortality is greater than other areas). Eligibility criteria included age ≥45 and self-reported Black or White race, with exclusion criteria consisting of either self‐reported medical conditions that would prevent long‐term participation (such as active cancer) or being on a waiting list for a nursing home. The study features intentional oversampling (56%) from the Stroke Belt (North Carolina, South Carolina, Georgia, Tennessee, Alabama, Mississippi, Arkansas, and Louisiana) and Stroke Buckle (coastal plains of North Carolina, South Carolina, and Georgia). In this report, we consider race a sociopolitical, not biological, variable.

Participants were enrolled from 2003 to 2007 and completed a computer-assisted telephone interview (CATI) assessing demographic information, medical history, and stroke history, followed by an in-home visit 3 to 4 weeks later to collect blood pressure measurements, height and weight, resting electrocardiogram (ECG) recordings, blood and urine samples, and an inventory of current medications. Biosamples were collected with uniform methods and shipped overnight to a central laboratory at the University of Vermont. Participants are contacted every 6 months via telephone to complete cognitive assessments and report information about stroke events, hospitalizations, and death.

An external observational study monitoring board and institutional review boards of all participating institutions approved the study methods and procedures. All participants provided written informed consent.

AF was determined by self-reported physician diagnosis or the presence of AF on ECG. Self‐reported AF had a similar relationship to stroke risk as ECG‐detected AF. ECGs were centrally read by analysts masked to other REGARDS data, and AF was defined using the Standard Minnesota ECG Classification, followed by physician verification.

A 6-Item Screener (SIS) was introduced during the baseline exam 11 months after enrollment began and is administered annually via CATI. Derived from the Mini-Mental State Exam (MMSE), the SIS is a validated instrument that assesses global cognitive function using 3-item word recall and 3-item temporal orientation components, with a scored range of 0 to 6. Baseline cognitive impairment was defined as SIS ≤4.

An Extended Cognitive Battery (ECB) based on the Montreal Cognitive Assessment (MoCA) and consisting of the Consortium to Establish a Registry for Alzheimer Disease (CERAD) Word List Learning (range of 0 to 30; learning a list of 10 semantically unrelated words), Word List Delayed Recall (range of 0 to 10; a delayed free recall trial of the 10 words), Animal Fluency (assessing the number of animals that a participant could list over 1 minute), and Letter F (assessing the number of words starting with the Letter F that a participant could list over 1 minute) was introduced in 2006 (2008 for Letter F) and conducted at 2-year intervals. ICI was defined by robust norms among 6,264 participants who did not display cognitive impairment between the baseline assessment and a second in-home assessment ten years later. Baseline scores of this normative sample on the ECB were used to calculate estimated scores and standard deviations by age and sex for each cognitive test using linear regression equations. These equations were used to derive T-scores for each test across the REGARDS cohort, with scores lower than 35 (1.5 standard deviations below the expected score in the normative sample) marking impairment. At each follow-up visit, participants were considered to have ICI if at least 3 out of the 4 cognitive tests were impaired. For participants who did not complete all 4 tests on a single call, ICI was defined as scoring <4 on the SIS once and <5 twice during follow-up. For the latter definition, the 2 or 3 low SIS measures were not necessarily consecutive, and time to ICI in this case was defined at the time of the most recent low SIS.

Death from dementia was defined using the National Death Index (NDI), with death from dementia in any location on the NDI corresponding to ICD-10 codes F00-F03, G30, G31.0-G31.1, and R54.

Data on age, race, sex, region of residence, prevalent stroke, smoking status, alcohol use, education, and income were self-reported. Anticoagulant use was assessed by medication inventory. Hypertension was defined by self-reported physician/nurse diagnosis, self-reported use of antihypertensive medications, or systolic blood pressure ≥140 mmHg or diastolic blood pressure ≥90 mmHg. Dyslipidemia was defined as total cholesterol ≥240 mg/dL, LDL ≥160 mg/dL, HDL ≤40 mg/dL, or self-reported lipid-lowering medication use (drawn from the Third Report of the Expert Panel on Detection, Evaluation, and Treatment of High Blood Cholesterol in Adults (Adult Treatment Panel III), in use at the time that the REGARDS cohort was established). Diabetes was determined by fasting glucose ≥126 mg/dL, nonfasting glucose ≥200 mg/dL, or self-reported use of diabetes medications. C-reactive protein (CRP), cystatin C, albumin, and leukocyte count were measured in baseline blood samples as previously described. ^,,

We included all participants with AF information, complete covariate data, and ascertainment of ICI and death from dementia. Participants with baseline cognitive impairment and those missing AF information, cognitive assessments during follow-up, and missing covariate data were excluded from the analyses.

Baseline characteristics were tabulated by baseline AF status. Cox proportional hazards models were used to calculate hazard ratios (HRs) of time to ICI by baseline AF, with censoring at death or last follow-up. Model 1 was adjusted for age, sex, race, and region. Model 2 was adjusted for Model 1 variables plus education, income, prevalent stroke, diabetes, hyperlipidemia, hypertension, systolic blood pressure, smoking, and alcohol use. Model 3 was adjusted for Model 2 variables plus baseline oral anticoagulant use. Models were the same for the secondary outcome of death from dementia.

We used interaction terms (variable x AF) to test for 2-way interactions between AF and prevalent stroke in Models 2 and 3 and race group and baseline oral anticoagulant use in Model 3. The significance level to declare an interaction was p interaction < 0.10. For illustration, regardless of the interaction, we stratified on prevalent stroke, race, and oral anticoagulant use to calculate separate hazard ratios for AF and risk of ICI by each factor.

To understand the impact of inflammation on the association between baseline AF and ICI, we tested for attenuation of the association by CRP, leukocyte count, albumin, and cystatin C via individual addition of each biomarker to Model 2. These analyses were limited to participants with prevalent stroke, as this was the only group with an association of baseline AF with ICI.

We assessed the proportional hazards assumption using Kaplan-Meier curves and Schoenfeld residual tests. There was no violation for AF. For proportional hazards assumption violations for any covariate, we fit stratified Cox models specifying different baseline hazard functions for that covariate.

Since we did not observe an overall association of baseline AF with ICI, 3 sensitivity analyses were performed to investigate the possibility of a missed association. First, for cases where ICI was determined by 2 or 3 low SIS measurements, we replaced follow-up time in the Cox proportional hazards models with time to first abnormal SIS rather than time to last abnormal SIS. We did this to check if incorrect handling of follow-up time (i.e., waiting for multiple measurements to be completed before concluding impairment) was underestimating ICI. Second, disregarding the ECB, we used SIS measurements only with time to first occurrence of SIS ≤4 as the outcome. Third, disregarding the ECB, we used SIS measurements only with time to first occurrence of 2 consecutive SIS ≤4 as the primary outcome.

Cox modeling was used to provide an analysis of the independent predictors of ICI in the stroke and nonstroke cohorts.

Statistical analysis was performed using RStudio 2024.04.2 Build 764.