Association of Atrial Fibrillation With Stroke and Dementia Accounting for Left Atrial Function and Size

Abstract

Background

Atrial fibrillation (AF) is associated with higher risks of ischemic stroke (IS) and dementia. Whether alterations in left atrial (LA) function or size—atrial myopathy—confound these associations remains unknown.

Objectives

The purpose of this study was to examine the association of prevalent and incident AF with ischemic stroke and dementia in the ARIC (Atherosclerosis Risk In Communities) study, adjusting for LA function and size.

Methods

Participants at visit 5 (2011-2013) with echocardiographic LA function (reservoir, conduit, contractile strain, and emptying fraction) and size (maximal, minimal volume index) data, and without prevalent stroke or dementia were followed through 2019. For analysis, we used time-varying Cox regression.

Results

Among 5,458 participants (1,193 with AF, mean age of 76 years) in the stroke analysis and 5,461 participants (1,205 with AF, mean age of 75 years) in the dementia analysis, 209 participants developed ischemic stroke, and 773 developed dementia over 7.1 years (median). In a demographic and risk factor-adjusted model, AF was significantly associated with ischemic stroke (HR, 1.63; 95% CI: 1.11-2.37) and dementia (HR: 1.38, 95% CI: 1.13-1.70). After additionally adjusting for LA reservoir strain, these associations were attenuated and no longer statistically significant (stroke [HR: 1.33, 95% CI: 0.88-2.00], dementia [HR: 1.15, 95% CI: 0.92-1.43]). Associations with ischemic stroke and dementia were also attenuated and not statistically significant after adjustment for LA contractile strain, emptying fraction, and minimal volume index.

Conclusions

AF-ischemic stroke and AF-dementia associations were not statistically significant after adjusting for measures of atrial myopathy. This proof-of-concept analysis does not support AF as an independent risk factor for ischemic stroke and dementia.

Central Illustration

Atrial fibrillation (AF) is a serious public health problem because it is the most commonly sustained arrhythmia, its prevalence is increasing in the aging population and it is associated with elevated risks of ischemic stroke and dementia.¹ Although the AF-stasis hypothesis has been the accepted mechanism of AF-related thromboembolism and morbidity,² recent compelling evidence has emerged to suggest an alternate hypothesis. Novel insights into the temporal dissociation between AF episodes and ischemic stroke events,^3,4 and data linking markers of abnormal atrial substrate with higher risk of ischemic stroke independent of AF,^5,6 suggest that another mechanism—other than the dysrhythmia of AF—may drive AF-related thromboembolism and morbidity.

Atrial myopathy (AM) (or atrial cardiomyopathy or atrial cardiopathy) defined as “any complex of structural, architectural, contractile, or electrophysiological changes affecting the atria with the potential to produce clinically relevant manifestations” manifests with lower (worse) left atrial (LA) function and greater LA size.^7,8 Emerging evidence suggests that thromboembolism can occur in the setting of AM even in the absence of AF.³ Therefore, AM may be a risk factor that explains the morbidity previously attributed to AF. Because no study has specifically examined the association of AF with ischemic stroke and dementia by taking into account underlying LA function and size, the aim of this study was to address this important knowledge gap.

Methods

Study population and design

The ARIC (Atherosclerosis Risk In Communities) study is an ongoing observational cohort study that started in 1987 to 1989 with 15,792 participants to investigate cardiovascular disease etiology and risk factors—including race-based differences—in the general population.⁹ In 2011 to 2013, 6,538 participants aged 65 to 90 years attended visit 5 (V5), which was chosen as the baseline for this study because echocardiograms were performed at this visit. Figure 1 shows the study flow diagram. The ARIC study has been approved by institutional review boards at all participating institutions and participants provided written informed consent at each study visit.

Figure 1.

Study Flowchart of Participants: The ARIC Study (1987-2019)

The flow chart of participants included in this analysis and the number of participants excluded for each exclusion criteria.

AF ascertainment

Prevalent and incident AF were ascertained from 4 sources: 1) 12-lead electrocardiograms (ECGs) from study visits, 2) hospital discharge records, 3) death certificates, and 4) leadless patch ambulatory monitoring at V6 (2016-2017). Data collection and AF diagnosis details are described in Supplemental Methods. Prevalent AF was defined as an AF diagnosis up to a participant’s V5 date (2011-2013); incident AF was defined as AF diagnosis after the V5 date and until the latest available data (2019).

Ischemic stroke ascertainment

Incident stroke was defined as fatal and nonfatal stroke based on hospital record review and diagnosed using criteria adapted from the National Survey of Stroke.¹⁰ Details of ischemic stroke ascertainment are in Supplemental Methods. All incident ischemic stroke events through 2019 were included in the present study. Stroke events defined as hemorrhagic or transient ischemic attacks were not counted as ischemic stroke outcomes.

Dementia adjudication

Incident dementia was adjudicated from in-person evaluations at study visits, telephone interviews, International Classification of Diseases (ICD) hospitalization codes or death certificate codes.¹¹ Details of dementia adjudication are in Supplemental Methods. In the present study, incident dementia was defined as diagnoses made after V5 and through 2019.

LA function and size

ECGs at V5 were performed by trained sonographers and analyzed in a blinded core laboratory as previously described.¹² For the current study, the LA variables examined were: 1) reservoir, conduit, and contractile strain for strain measures of LA function; 2) LA emptying fraction as a volumetric measure of LA function; and 3) maximal and minimal volume index for LA size. Details of echocardiographic LA characterization are in Supplemental Methods.

Covariates

Participant covariate data were obtained from V5. Participants self-reported age, sex, race, and current cigarette smoking status. Additional data collected from participants included body mass index; prevalent hypertension, diabetes, and heart failure; total cholesterol; low density lipoprotein cholesterol; and APOE (apolipoprotein E) ε4 status. Definitions of covariates are in Supplemental Methods.

Statistical analysis

Time-varying Cox proportional hazards regression was used to examine the association of prevalent and incident AF with ischemic stroke or dementia. Schoenfeld residuals tests were used to verify the proportional hazards assumption, which was not violated in the fully adjusted models for the stroke and dementia analyses. Follow-up time began on the date of a participant’s V5 examination and continued through December 31, 2019. Participants with prevalent AF were defined as exposed at the beginning of follow-up, whereas participants with incident AF were changed to expose on the date of their AF diagnosis. Participants were censored at the date of death or the end of the study period, whichever came first. For the ischemic stroke analysis, model 1 adjusted for demographics, and model 1a to e additionally adjusted for LA function and size variables. Model 2 was model 1 additionally adjusted for body mass index, current cigarette smoking, hypertension, diabetes, and prevalent coronary heart disease and heart failure. Model 3 additionally adjusted for use of statins and antihypertensives. Model 4 additionally adjusted for use of anticoagulants and anti-platelet agents. Model 4a to f was model 4 additionally adjusted for individual LA function and size variables. Model covariates for the dementia analysis were similar to the stroke analysis with the following additions: model 1 additionally adjusted for APOE ɛ4 status and model 5 additionally adjusted for prevalent and incident ischemic stroke. LA variables were adjusted as continuous variables based on per 1-SD change. For LA conduit and contractile strain, the absolute value was used instead of the negative percentage to keep the direction of estimates consistent with LA reservoir strain estimates (smaller values indicate lower function). Bootstrapping with 1,000 iterations was used to calculate 95% CIs for cumulative incidence plots.

Several sensitivity analyses were conducted. First, participants taking anticoagulant medication at V5 were excluded. Second, since AF can potentially lead to lower LA function and greater LA size, we excluded participants with prevalent AF at V5 and considered only incident AF after V5 as an exposure. Third, we also excluded participants with incident AF, thus considering only prevalent AF as the exposure. Fourth, because stroke is a risk factor for dementia,¹³ participants with prevalent and incident ischemic stroke were excluded in the dementia analysis. Fifth, multiple imputation by chained equations was used to impute LA function data for participants who were missing LA function or size data. The proportion of missingness was 6% for LA function data and 16% for LA size data. Clinical characteristics and echocardiographic variables were used with the R package mice to create 10 datasets without missing values.

Finally, because past studies have identified stroke and dementia incidence disparities between Black and White individuals,^14,15 and the ARIC study was funded by the National Institutes of Health to study the etiology and variation of cardiovascular disease and its clinical sequelae,⁹ we conducted race-stratified analyses and assessed interactions by race. Additionally, we conducted sex-stratified analyses and assessed interactions by sex. Significance threshold was defined as a 2-sided P < 0.05. All analyses were performed with SAS (version 9.4) and R (version 4.1.1).

Results

Study population

Study participant characteristics, stratified by AF status, are presented in Table 1. In the stroke analysis, 5,458 participants were included after all exclusions: 478 (9%) had prevalent AF at V5 and 715 (13%) developed incident AF; 29 (2%) were diagnosed by study visit ECG, 74 (7%) were diagnosed by ambulatory monitoring, and 1,090 (91%) were diagnosed by hospitalization discharge codes. Over a median follow-up of 7.2 (IQR: 5.4-7.8) years, 209 (4%) participants developed ischemic stroke.

Table 1.

Characteristics of Study Participants: The ARIC Study (2011-2013)^a

	Analysis of Ischemic Stroke (n = 5,458)		Analysis of Dementia (n = 5,461)
	AFb (n = 1,193)	No AF (n = 4,265)	AFb (n = 1,205)	No AF (n = 4,256)
Exposure
Prevalent AF at visit 5 (2011-2013)	478 (40.1)	N/A	484 (40.2)	N/A
Incident AF after visit 5 through 2019	715 (59.9)	N/A	721 (59.8)	N/A
Incidence rate (95% CI), per 100 person-yearsc	2.56 (2.37-2.74)	N/A	2.58 (2.39-2.77)	N/A
Demographics
Age, years	77.0 (73.0-81.0)	74.0 (71.0-78.0)	77.0 (73.0-81.0)	74.0 (71.0-78.0)
Women	594 (49.8)	2,583 (60.6)	598 (49.6)	2,566 (60.3)
Men	599 (50.2)	1,682 (39.4)	607 (50.4)	1,690 (39.7)
Black raced	138 (11.6)	955 (22.4)	137 (11.4)	962 (22.6)
White raced	1,055 (88.4)	3310 (77.6)	1,068 (88.6)	3,294 (77.4)
Clinical characteristics
Body mass index, (kg/m2)	28.4 (25.3-31.9)	27.7 (24.8-31.2)	28.4 (25.4-32.0)	27.8 (24.8-31.3)
Systolic blood pressure, mm Hg	128.5 (116.5-140.9)	128.5 (118.0-140.5)	128.5 (116.5-140.5)	129.0 (117.6-140.5)
Prevalent coronary heart diseasee	319 (26.7)	491 (11.5)	330 (27.4)	503 (11.8)
Prevalent heart failuref	176 (14.8)	106 (2.5)	185 (15.4)	118 (2.8)
Prevalent diabetes	413 (34.6)	1,250 (29.3)	422 (35.0)	1,259 (29.6)
Prevalent stroke	N/A	N/A	51 (4.2)	103 (2.4)
Total cholesterol, mg/dLg	172.8 ± 41.9	183.8 ± 41.5	N/A	N/A
LDL cholesterol, mg/dLg	50.1 ± 13.0	52.9 ± 14.2	N/A	N/A
APOE ɛ4 carrierh	N/A	N/A	328 (27.2)	1,192 (28.0)
Medication use
Antihypertensive usei	1,009 (84.6)	3,025 (70.9)	1,022 (84.8)	3,032 (71.2)
Anticoagulant usej	298 (25.0)	73 (1.7)	310 (25.7)	78 (1.8)
Anti-platelet usek	878 (73.6)	2,894 (67.9)	N/A	N/A
Statin use	682 (57.2)	2,128 (49.9)	693 (57.5)	2,137 (50.2)
Echocardiographic measures
LA reservoir strain, %	24.6 ± 10.1	33.3 ± 7.4	24.6 ± 10.1	33.2 ± 7.4
LA conduit strain, %	12.4 ± 5.3	15.0 ± 5.7	12.4 ± 5.2	15.0 ± 5.7
LA contractile strain, %	12.5 ± 7.5	18.3 ± 5.5	12.5 ± 7.5	18.2 ± 5.5
LA maximal volume index, (ml/m2)j	38.9 (30.2-48.1)	31.6 (25.6-38.8)	39.0 (30.2-48.2)	31.7 (25.6-38.9)
LA minimal volume indexl, (ml/m2)j	18.1 (12.4-25.2)	12.7 (9.4-16.9)	18.1 (12.4-25.2)	12.7 (9.5-17.0)
LA emptying fraction, %j	51.7 ± 12.1	58.9 ± 9.0	51.9 ± 11.9	58.9 ± 9.0
LV ejection fraction, %	64.3 (59.3-68.2)	66.0 (62.5-69.4)	64.3 (59.6-68.1)	66.0 (62.5-69.4)
E/A ratio	0.9 ± 0.4	0.8 ± 0.2	1.0 ± 0.4	0.8 ± 0.3
Average E/e’ ratio	3.1 ± 1.3	2.7 ± 0.9	3.1 ± 1.3	2.7 ± 0.9
Moderate or severe mitral regurgitation	62 (5.2)	89 (2.1)	57 (4.7)	86 (2.0)

Values are n (%), median (IQR), or mean ± SD unless otherwise indicated.

AF = atrial fibrillation; LA = left atrial; LDL = low density lipoprotein; APOE = apolipoprotein E; N/A = not applicable.

^aAnalytic sample was restricted to participants who had an echocardiogram at visit 5 (2011-2013) and did not meet any exclusion criteria.

^bTime-varying AF, includes participants with prevalent AF at visit 5 and incident AF prior to stroke or dementia; participants with incident AF happening after stroke or dementia were not considered as AF cases.

^cIncludes only incident AF cases.

^dRace categories other than Black or White were excluded from this analysis. Race was self-selected by participants from fixed categories.

^eCoronary heart disease was defined as prior cardiovascular revascularization, physician-diagnosed myocardial infarction, presence of prior myocardial infarction by ECG, or ascertained by ARIC through follow-up phone calls and hospitalization records.

^fHeart failure was defined by Gothenburg criteria (visit 1 only), heart failure medication use or hospitalization ICD codes.

^gTo convert cholesterol to mmol/L, multiply values by 0.0259.

^hAPOE ɛ4 carriers were defined based on the number of Taqman assay detected ɛ4 alleles: 0, 1, or 2.

ⁱIncluded all medications indicated for hypertension, including diuretics, calcium channel blockers, ACE inhibitors, Angiotensin II receptor antagonist, adrenergic receptor antagonists, aldosterone receptor antagonists, and alpha-2 adrenergic receptor agonists.

^jIncluded warfarin, dabigatran, apixaban, edoxaban, or rivaroxaban.

^kIncluded aspirin, clopidogrel, ticagrelor, or prasugrel.

^lAmong participants with LA volume and emptying fraction measures (n = 4,603 for stroke analysis, n = 4,595 for dementia analysis).

In the dementia analysis, 5,461 participants were included after exclusions: 484 (9%) had prevalent AF at V5 and 721 (13%) developed incident AF; 29 (3%) were diagnosed by study visit ECG, 70 (6%) were diagnosed by ambulatory monitoring, and 1,106 (92%) were diagnosed by hospitalization discharge codes. Over a median follow-up of 7.1 (IQR: 5.4-7.8) years, 773 (14%) participants developed dementia; 35% were diagnosed via in-person cognitive assessments at study visits, 46% were diagnosed from telephone or informant interview, and 19% were diagnosed by ICD hospitalization or death certificate codes.

In both study populations, participants with AF had higher prevalence of coronary heart disease, heart failure, anticoagulant use, lower LA function, and greater LA size, compared to participants without AF (Table 1). There were also significant associations between prevalent AF and LA function and size (Supplemental Table 1). Clinical characteristics of participants who did not attend V5, who were missing LA function, and who were missing echocardiographic data are presented in Supplemental Tables 2-6, respectively.

Outcome: ischemic stroke

AF was significantly associated with a higher risk of ischemic stroke in demographic-adjusted model 1 (Table 2) (HR: 1.77, 95% CI: 1.27-2.47). However, after additional adjustment for LA reservoir strain, the association was attenuated and no longer statistically significant (model 1a; HR: 1.22, 95% CI: 0.81-1.85). Next, AF was significantly associated with ischemic stroke after adjusting for demographics, risk factors for stroke, and medications (Table 2) (model 4, HR: 1.63, 95% CI: 1.11-2.37). However, after additional adjustment for LA reservoir strain, contractile strain, maximal volume index, minimal volume index, and emptying fraction (but not LA conduit strain), the association of AF with stroke was attenuated and no longer statistically significant (models 4a to 4f). Figure 2A shows the cumulative incidence of ischemic stroke by AF status. The demographic-adjusted cumulative incidence of ischemic stroke in participants with AF was significantly higher than in those without AF (8.0% [95% CI: 5.3%-10.7%] vs 4.6% [95% CI: 3.5%-5.7%] at 8 years; P = 0.001). After additional adjustment for LA reservoir strain, there was no longer a significant difference in the cumulative incidence of ischemic stroke in participants with AF vs without AF (6.2% [95% CI: 3.9%-8.6%] vs 5.1% [95% CI: 3.8%-6.5%] at 8 years; P = 0.44).

Table 2.

Association of Time-Varying Atrial Fibrillation^a With Incident Ischemic Stroke: The ARIC Study (2011-2019)

	AF (n = 1,193)	No AF (n = 4,265)
Ischemic stroke	93 (7.8)	116 (2.7)
Incidence rate (95% CI), per 100 PY	1.10 (0.79-1.41)	0.52 (0.44-0.60)
Follow-up
Years since visit 5	6.8 (4.7-7.7)	7.2 (5.8-7.8)
Years since incident AF after visit 5	2.5 (0.7-4.1)	N/A

Models	HR (95% CI)	P value
Model 1 (adjusted for age, sex, and race/center)	1.77 (1.27-2.47)	<0.001
Model 1a (model 1 + LA reservoir strain [%])	1.22 (0.81-1.85)	0.34
Model 1b (model 1 + LA conduit strain [%])	1.63 (1.16-2.28)	0.005
Model 1c (model 1 + LA contractile strain [%])	1.44 (0.98-2.11)	0.06
Model 1d (model 1 + LA maximal volume index [mL/m2])	1.44 (0.94-2.23)	0.10
Model 1e (model 1 + LA minimal volume index [mL/m2])	1.34 (0.85-2.11)	0.20
Model 1f (model 1 + LA emptying fraction [%])	1.24 (0.78-1.98)	0.36
Model 2 (adjusted for model 1 covariates plus BMI, smoking status, diabetes, hypertension, coronary heart disease, and heart failure)	1.55 (1.11-2.18)	0.01
Model 3 (adjusted for model 2 covariates plus use of statins, and use of antihypertensives)	1.55 (1.1-2.18)	0.01
Model 4 (adjusted for model 3 covariates plus use of anticoagulants and anti-platelet agents)	1.63 (1.11-2.37)	0.01
Model 4a (model 4 + LA reservoir strain [%])	1.33 (0.88-2.00)	0.17
Model 4b (model 4 + LA conduit strain [%])	1.56 (1.07-2.29)	0.02
Model 4c (model 4 + LA contractile strain [%])	1.43 (0.96-2.13)	0.08
Model 4d (model 4 + LA maximal volume index [mL/m2])	1.31 (0.83-2.08)	0.25
Model 4e (model 4 + LA minimal volume index [mL/m2])	1.26 (0.79-2.02)	0.34
Model 4f (model 4 + LA emptying fraction [%])	1.20 (0.74-1.93)	0.46

Values are n (%) or median (IQR) unless otherwise indicated.

AF = atrial fibrillation; BMI = body mass index; LA = left atrial; PY = person-years.

^aFor time-varying AF, prevalent AF was defined as exposed at the beginning of follow-up.

Figure 2.

Cumulative Incidence of the Primary Study Outcomes

(A) Cumulative incidence of ischemic stroke stratified by AF status with and without adjustment for LA reservoir strain. (B) Cumulative incidence of dementia stratified by AF status with and without adjustment for LA reservoir strain. AF = atrial fibrillation; LA = left atrial.

In sensitivity analyses, participants using anticoagulants at V5 were excluded (Supplemental Table 7), multiple imputation accounted for missing LA function data (Supplemental Table 8), and participants with incident AF were excluded (Supplemental Table 9). Significant AF-stroke associations in demographic-adjusted model 1 were attenuated and no longer significant after additional adjustment for LA reservoir strain (model 1a in Supplemental Tables 7 to 9). After excluding participants with prevalent AF (Supplemental Table 10), time-varying incident AF was not significantly associated with ischemic stroke (HR: 1.48, 95% CI: 0.87-2.52). Nonetheless, as in the main analysis, the point estimate for the association between incident AF and ischemic stroke was attenuated after adjustment for LA reservoir strain (HR: 1.18, 95% CI: 0.68-2.05).

There were no significant interactions by race or sex. Race- and sex-stratified analyses are described in Supplemental Results and presented in Supplemental Table 11.

Outcome: dementia

AF was significantly associated with dementia after demographic and APOE ɛ4 status adjustment (model 1 [HR: 1.59, 95% CI: 1.34-1.90]), but additional adjustment for LA reservoir strain (model 1a) attenuated the HR to 1.17 (95% CI: 0.95-1.44 [Table 3]). Next, AF was significantly associated with dementia after adjusting for demographics, APOE ɛ4 carrier status, risk factors for dementia, medications, and prevalent and incident ischemic stroke (model 5 [HR:1.38, 95% CI: 1.13-1.70]). However, after additional adjustment for LA reservoir strain, contractile strain, minimal volume index, and emptying fraction (but not conduit strain or maximal volume index), the association of AF with dementia was attenuated and was no longer statistically significant (models 5a, 5c, 5e, and 5f). Figure 2B shows the cumulative incidence of dementia by AF status. The demographic-adjusted cumulative incidence of dementia in participants with AF was significantly higher than in those without AF (25.5 % [95% CI: 22.5%-28.5%] vs 17.2 % [95% CI: 15.9%-18.7%] at 8 years; P < 0.001). After additional adjustment for LA reservoir strain, there was no longer a significant difference in the cumulative incidence of dementia in participants with AF vs without AF (21.5% [95% CI: 18.5%-24.5%] vs 18.8 % [95% CI: 17.3%-20.2%] at 8 years; P = 0.13).

Table 3.

Association of Time-Varying Atrial Fibrillation^a With Incident Dementia: The ARIC Study (2011-2019)

	AF (n = 1,205)	No AF (n = 4,256)
Dementia	254 (21.1)	519 (12.2)
Incidence rate (95% CI), per 100 PY	4.33 (3.72-4.95)	1.91 (1.76-2.07)
Follow-up
Years since visit 5	6.7 (4.5-7.7)	7.2 (5.6-7.8)
Years since incident AF after visit 5	2.7 (1.1-4.1)	N/A

Models	HR (95% CI)	P Value
Model 1 (adjusted for age, sex, race/center, education, and APOE ɛ4)	1.59 (1.34-1.90)	<0.001
Model 1a (model 1 + LA reservoir strain [%])	1.17 (0.95-1.44)	0.14
Model 1b (model 1 + LA conduit strain [%])	1.52 (1.27-1.81)	<0.001
Model 1c (model 1 + LA contractile strain [%])	1.26 (1.03-1.55)	0.02
Model 1d (model 1 + LA maximal volume index [mL/m2])	1.40 (1.13-1.73)	0.002
Model 1e (model 1 + LA minimal volume index [mL/m2])	1.28 (1.03-1.60)	0.03
Model 1f (model 1 + LA emptying fraction [%])	1.23 (0.99-1.54)	0.07
Model 2 (adjusted for model 1 covariates plus BMI, smoking status, diabetes, hypertension, coronary heart disease, and heart failure)	1.51 (1.25-1.83)	<0.001
Model 3 (adjusted for model 2 covariates plus use of statins, and use of antihypertensives)	1.52 (1.25-1.83)	<0.001
Model 4 (adjusted for model 3 covariates plus use of anticoagulants)	1.46 (1.19-1.79)	<0.001
Model 5 (adjusted for model 4 covariates plus prevalent and incident ischemic stroke)	1.38 (1.13-1.70)	0.002
Model 5a (model 5 + LA reservoir strain [%])	1.15 (0.92-1.43)	0.23
Model 5b (model 5 + LA conduit strain [%])	1.35 (1.10-1.66)	0.005
Model 5c (model 5 + LA contractile strain [%])	1.19 (0.96-1.49)	0.12
Model 5e (model 5 + LA maximal volume index [mL/m2])	1.29 (1.02-1.64)	0.04
Model 5e (model 5 + LA minimal volume index [mL/m2])	1.22 (0.96-1.55)	0.11
Model 5f (model 5 + LA emptying fraction [%])	1.18 (0.92-1.50)	0.19

Values are n (%) or median (IQR) unless otherwise indicated.

AF = atrial fibrillation; BMI = body mass index; LA = left atrial; PY = person-years.

^aFor time-varying AF, prevalent AF was defined as exposed at the beginning of follow-up.

In sensitivity analyses: 1) participants using anticoagulants at V5 were excluded (Supplemental Table 12); 2) participants with prevalent or incident ischemic stroke were excluded (Supplemental Table 13); 3) participants with prevalent AF were excluded (Supplemental Table 14); 4) participants with incident AF were excluded (Supplemental Table 15); and 5) multiple imputation accounted for missing LA function data (Supplemental Table 16). In all sensitivity analyses, significant AF-dementia associations in demographic-adjusted model 1 were attenuated and no longer significant after additional adjustment for LA reservoir strain (model 1a in Supplemental Tables 12 to 16).

There were no significant interactions by race or sex. Race- and sex-stratified analyses are described in Supplemental Results and presented in Supplemental Table 17.

Discussion

In this analysis of a U.S. community-based cohort of older adults over a median follow-up of 7.1 years, AF was associated with ischemic stroke and dementia after adjusting for demographic factors and risk factors for stroke and dementia. However, these associations were substantially attenuated and were no longer statistically significant after adjusting for measures of AM, such as LA reservoir strain, contractile strain, emptying fraction, and minimal volume index. These findings suggest that AF is not independently associated with ischemic stroke and dementia.

The earliest evidence to support an association between AF and stroke came from studies of rheumatic mitral valve disease-associated AF in the 1930s.¹⁶ Wolf and Kannel subsequently demonstrated the same risk of stroke in Framingham Heart Study participants with nonvalvular AF.^17,18 Further corroboration by multiple cohort studies entrenched the AF-stroke relationship as conventional wisdom.19, 20, 21 In recent years, this paradigm has been challenged. In the LOOP study, anticoagulation patients without clinical AF but with continuous ambulatory monitoring-detected AF did not significantly reduce stroke risk.²² Based on analysis of data from cardiac implantable electronic devices, the ASSERT (Asymptomatic Atrial Fibrillation and Stroke Evaluation in Pacemaker Patients and the Atrial Fibrillation Reduction Atrial Pacing Trial) and IMPACT (Improving Plasma Collection) trials found no temporal relationship between AF episodes and stroke.^3,4 This temporal dissociation suggests that the AF-stroke association may be explained by another factor other than the dysrhythmia.²³ One possible factor is AM: 1) greater LA size was significantly associated with higher risk of ischemic stroke, even in participants without AF or adjusting for AF^24,25; 2) lower LA function has been linked to ischemic stroke and brain infarcts on MRI, independent of AF26, 27, 28; and 3) ECG markers of AM were significantly associated with increased risk of ischemic stroke and vascular brain injury.^5,6,29 However, no study has reexamined the association of AF with ischemic stroke while adjusting for AM, a critical knowledge gap.

In addition to stroke, more recent evidence has linked AF to increased risk of dementia. Ott et al first identified an association of AF with dementia in the Rotterdam study in 1997.³⁰ Subsequent studies and meta-analyses have now confirmed these original observations.31, 32, 33 A proposed mechanism underlying this association is subclinical cardioembolic brain infarcts, cerebral imaging of patients with AF in the absence of clinical stroke have found evidence suggestive of old asymptomatic infarcts.34, 35, 36, 37 However, none of these studies adjusted for the presence of AM. Recent studies have also implicated AM as a risk factor for dementia, independent of AF: 1) ECG markers of AM were significantly associated with incident dementia^38,39; 2) LA enlargement was significantly associated with lower cognitive function and cognitive decline^40,41; and 3) lower LA function was significantly associated with an increased risk of subsequent dementia.⁴² Nevertheless, studies have not reexamined the AF-dementia association by adjusting for AM; thus, whether AM confounds the AF-dementia association remains another critical knowledge gap.

Hence, this study significantly advances the field by reexamining the long-held AF-stroke association and the AF-dementia association, by adjusting for AM using a comprehensive array of echocardiographic LA function and size measures. Among the LA measures, the most sensitive and robust marker of AM is LA reservoir strain.⁴³ First, we calculated the Akaike information criterion for each LA strain measure in our Cox models and confirmed that LA reservoir strain was the best-fit model (Supplemental Table 15). Second, when added to the CHA₂DS₂-VASc score, LA reservoir strain improves stroke risk prediction the most between LA function and size measures.⁴⁴ Third, LA reservoir strain quantifies the reservoir function of the LA during atrial filling and ventricular systole and reflects the cumulative impact of atrial remodeling from ischemia or elevated filling pressures.45, 46, 47 Fourth, LA reservoir strain can be measured regardless of cardiac rhythm; whereas, LA contractile strain can no longer be measured when atrial contraction is lost during AF.⁴⁸ Importantly, in patients with left ventricular diastolic dysfunction, lower LA reservoir strain is detected before larger LA size.⁴⁹ Therefore, the attenuation of AF-stroke and AF-dementia associations with models adjusting for LA reservoir strain provide the strongest support for our study conclusions.

Additional strengths of the study include the multiple sources for detection of AF and adjudication of stroke and dementia events. Nevertheless, approximately 90% of AF cases were ascertained from hospitalization discharge codes. The combination of the former, study visit ECGs and ambulatory monitoring (up to 4 weeks in some participants) resulted in an AF incidence rate of 2.6 per 100 person-years (PY). This is nearly identical to the control group of the LOOP trial at 2.5 per 100 PY²² and comparable to AF incidence rates in other observational studies for this age group.^50,51 Although some cases of subclinical AF may have been missed and the heterogenous methods to ascertain AF could bias our analysis, the overall incidence of AF in this study is consistent with other contemporary studies, and is thus clinically relevant.

A possible explanation for the study findings is that decreased LA compliance in AM can disrupt blood flow, thus causing stasis that results in cardioembolism. LA reservoir strain is measured as the peak global longitudinal change in length of the LA.⁵² It captures the maximal distention of the LA during the cardiac cycle, representing a sensitive marker of LA compliance.⁵³ Indeed, patients with cardiac amyloidosis—an infiltrative restrictive cardiomyopathy—have LA dysfunction and are predisposed to cardioembolism and stroke even during sinus rhythm.54, 55, 56 Conversely, shorter periods of subclinical AF (such as the 6 minute threshold utilized in the LOOP study)²² may not have concomitant AM, and therefore, may not be sufficient for thrombogenesis. Collectively, the findings of this study suggest that decreased LA compliance in AM, as indicated by decreased LA reservoir strain, may be an important mechanism of cardioembolism that results in ischemic stroke and dementia, independent of AF (Central Illustration). Of note, since patients with clinical AF are likely to have concomitant AM, the findings of this study do not challenge the benefit of anticoagulation to prevent ischemic stroke in patients with AF because anticoagulation would also prevent ischemic stroke due to underlying AM in such patients. Importantly, because current evidence is insufficient to support systematic screening for subclinical AF to prevent ischemic stroke, an alternative stroke prevention strategy may be needed.⁵⁷ In this regard, the findings of this paper support future research to evaluate the benefit of anticoagulation as primary prevention for ischemic stroke and dementia in the presence of AM but in the absence of AF.⁵⁸

Central Illustration.

Proposed Conceptual Model for Atrial Myopathy Vis-à-Vis Atrial Fibrillation as Risk Factors for Stroke or Dementia

Atrial myopathy—manifested as lower (worse) left atrial (LA) function and greater LA size—can promote hemostasis in the heart and lead to thrombus formation, such as in the LA appendage. embolization of this thrombus to the brain can cause infarction and ischemic stroke. clinical stroke or chronic subclinical brain infarcts can then lead to dementia. the association of atrial fibrillation with brain infarcts/ischemic stroke and subsequent dementia is confounded by atrial myopathy.

Study Limitations

First, we did not examine stroke subtypes in this study. This is because the classification of cardioembolic stroke in ARIC is contingent on the presence of AF, which introduces bias into the analysis. Some cases of ischemic stroke may be secondary to intracranial atherosclerosis rather than cardioembolism.²⁶ Second, the analytic sample consisted of older adults (mean age: 75 years) and results may not be generalizable to younger individuals. However, the findings from this analysis remain clinically relevant as the public health burden of AM, ischemic stroke, and dementia is greatest in older adults. Third, volunteer and survivorship bias may have limited the generalizability of our observations. However, differences between ARIC participants and eligible nonparticipants had been previously reported to be modest.⁵⁹ Furthermore, participants who attended V5 were generally healthier than those who did not; however, this is common and not unexpected among community-based cohort studies with 20 to 30 years of follow-up. Fourth, the proportion of missingness for LA size data was higher than LA strain data (16% vs 6%, respectively), but we accounted for this missingness by performing sensitivity analyses using multiple imputation by chained equations (Supplemental Tables 8 and 16) and our findings for models with LA reservoir strain were unchanged. Fifth, the point estimates for the AF-ischemic stroke association in the primary analysis were lower than those reported in older studies,¹⁸ and this likely reflects lower overall stroke incidence rates in more recent decades due to contemporary medical treatment.⁶⁰ Moreover, the ischemic stroke incidence rate in participants with AF of this study (1.10 per 100 PY) was comparable to those in the usual care group of the EAST-AFNET-4 (Early Treatment of Atrial Fibrillation for Stroke Prevention Trial) (0.9 per 100 PY)⁶¹ and those in the Optum database (1.03 per 100 PY).⁶² Fourth, 25% of participants with AF were taking anticoagulants. However, the use of anticoagulants was adjusted for and a sensitivity analysis excluding participants taking anticoagulants demonstrated that the results remained unchanged. Sixth, the ascertainment of dementia utilized different methods at the baseline visit and during follow-up. Nevertheless, ascertainment was performed by the same central committee via an algorithmic approach detailed previously.⁶³ Seventh, different modalities of imaging were utilized in the evaluation of ischemic stroke which may impact the sensitivity to detect minor strokes, particularly in the posterior circulation. However, the incidence rate of ischemic stroke in this study was comparable to other contemporary studies.^61,62 Eighth, our primary analysis was not replicated in a separate cohort study. Nevertheless, our study utilized the largest known LA strain dataset currently available and future research to replicate the findings of our study is warranted. Ninth, the generalizability of this study to all U.S. individuals may be limited given that the study sample consisted of Black and White individuals from 4 U.S. communities. Tenth, because of the observational nature of this study, residual confounding cannot be eliminated.

Conclusions

In this proof-of-concept analysis of a U.S. community-based cohort, the associations of AF with ischemic stroke and dementia were no longer statistically significant after adjusting for measures of AM. These findings do not support AF as an independent risk factor for ischemic stroke and dementia.

PERSPECTIVES.

COMPETENCY IN PATIENT CARE AND PROCEDURAL SKILLS: Although AF is associated with brain infarcts/ischemic stroke and subsequent dementia, this relationship is confounded by AM—manifested as lower (worse) left atrial function and greater left atrial size. Beyond the dysrhythmia of AF, AM is a possible treatment target for the prevention of stroke and dementia.

TRANSLATIONAL OUTLOOK: Future studies and clinical trials are necessary to determine the benefit of anticoagulation as primary prevention for ischemic stroke and dementia in the presence of AM but in the absence of AF.

Funding support and author disclosures

The ARIC study has been funded in whole or in part with Federal funds from the National Heart, Lung, and Blood Institute, National Institutes of Health, Department of Health and Human Services, under Contract nos. (HHSN268201700001I, HHSN268201700002I, HHSN268201700003I, HHSN268201700005I, and HHSN268201700004I). Neurocognitive data are collected by U01 2U01HL096812, 2U01HL096814, 2U01HL096899, 2U01HL096902, and 2U01HL096917 from the NIH (NHLBI, NINDS, NIA, and NIDCD), and with previous brain magnetic resonance imaging examinations funded by R01HL70825 from the NHLBI. This study is funded by R01HL141288. Dr Zhang is supported by F32HL152523. Dr Alonso is supported by R01HL137338, K24HL148521, and P30AG066511. Dr Johansen is supported by K23NS112459. Dr Gottesman is supported by the NINDS Intramural research program. Dr Shah is supported by R01HL135008, R01HL143224, R01HL150342, R01HL148218, and K24HL152008. Dr Chen is supported by R01HL141288, R01HL126637, RF1 NS127266, R01 HL158022, R01AG075883, and K24HL155813. Dr Zhang reported funding from the National Institutes of Health (NIH). Dr Johansen reported funding from the National Institute of Neurological Disorders and Stroke (NINDS). Dr Alonso has received funding from the NIH and the American Heart Association. Dr Mosley has received funding from the NIH. Dr Gottesman has received prior funding grants from the NIH. Dr Shah has received funding from the NIH/National Heart, Lung, and Blood Institute (NHLBI), Philips Ultrasound grant to BWH; personal fees from Philips Ultrasound advisory board; grants from Novartis grant to BWH; personal fees from Edwards LifeSciences; is a consultant; and has received personal fees from Janssen advisory board outside the submitted work. Dr Solomon has received funding from Actelion, Alnylam, Amgen, AstraZeneca, Bellerophon, Bayer, BMS, Celladon, Cytokinetics, Eidos, Gilead, GlaxoSmithKline, Ionis, Lilly, Mesoblast, MyoKardia, NIH/NHLBI, Neurotronik, Novartis, NovoNordisk, Respicardia, Sanofi Pasteur, Theracos, Us2.ai grant to institution, outside scope of manuscript; and has received personal fees from Abbott, Action, Akros, Alnylam, Amgen, Arena, AstraZeneca, Bayer, Boeringer-Ingelheim, BMS, Cardior, Cardurion, Corvia, Cytokinetics, Daiichi-Sankyo, GlaxoSmithKline, Lilly, Merck, Myokardia, Novartis, Roche, Theracos, Quantum Genomics, Cardurion, Janssen, Cardiac Dimensions, Tenaya, Sanofi-Pasteur, Dinaqor, Tremeau, CellProThera, Moderna, American Regent, Sarepta, Lexicon, Anacardio, Akros, PureHealth consulting, and outside scope of manuscript outside the submitted work. Dr Chen has received funding from the NIH. All other authors have reported that they have no relationships relevant to the contents of this paper to disclose.