Atrial Fibrillation and Risk of Incident Heart Failure with Reduced Versus Preserved Ejection Fraction

Abstract

Objective:

Associations among atrial fibrillation (AF) and heart failure (HF) have been established. We compared the extent to which AF is associated with each primary subtype of HF, with reduced (HFrEF) versus preserved ejection fraction (HFpEF).

Methods:

We included 25,787 participants free of baseline HF from the REasons for Geographic And Racial Differences in Stroke (REGARDS) cohort. Baseline AF was ascertained from electrocardiogram and self-reported history of physician diagnosis. Incident HF events were determined from physician-adjudicated review of hospitalization medical records and HF deaths. Based on ejection fraction (EF) at the time of HF event, HFrEF, HFpEF, and mid-range HF were defined as EF<40%, ≥50%, and 40–49%, respectively. Multivariable Cox proportional-hazards models examined the association between AF and HF. The Lunn-McNeil method was used to compare associations of AF with HFrEF versus HFpEF.

Results:

Over median 9 years follow-up, 1,109 HF events occurred (356 HFpEF, 388 HFrEF, 77 mid-range, and 288 unclassified). In a model adjusted for sociodemographics, cardiovascular risk factors, and incident coronary heart disease, AF was associated with increased risk of all HF events (hazard ratio [HR] 1.67, 95% confidence interval [CI] 1.38–2.01). The associations of AF with HFrEF versus HFpEF events did not differ significantly (HR [95% CI] 1.87 [1.38–2.54], and 1.65 [1.20–2.28], respectively; p-value for difference=0.581). These associations were consistent in sex and race subgroups.

Conclusions:

AF is associated with both HFrEF and HFpEF events, with no significant difference in the strength of association among these subtypes.

INTRODUCTION

Growth in the global burden of atrial fibrillation (AF)¹ represents a troubling population health concern. While clinical management of AF is largely driven by symptom relief and abatement of stroke risk, new AF diagnosis appears to carry an approximately doubled risk of heart failure (HF) relative to that of stroke.² As such, HF frequently develops in a population with AF, with which it has bidirectional associations³ driven through shared risk factors and pathophysiology.⁴ When concurrent, AF with HF carries a more than doubled mortality relative to either condition alone,⁵ so HF risk stratification in individuals with AF is an important research priority.⁶ Despite this, the extent to which AF is associated with each of the primary subtypes of HF, with reduced (HFrEF) and preserved (HFpEF) ejection fraction, remains uncertain.

Therefore, we aimed to compare the associations of AF with incident HF and HFrEF versus HFpEF events in the REasons for Geographic And Racial Differences in Stroke (REGARDS) Study, a racially and geographically diverse contemporary cohort. Furthermore, as appreciable sex and race-group differences are found in the epidemiology of AF⁷ and HF,⁸ we examined divergence in these differences across sex and race subgroups.

METHODS

Sample & Design

The REGARDS Study enrolled 30,239 Black or White participants aged ≥45 years living in the contiguous U.S. from 2003–2007.⁹ Potential participants were selected at random from publicly-available lists and enrolled by telephone and/or mail. Sampling design intentionally oversampled residents of the “Stroke Belt”, a region of excess stroke mortality in the southeastern United States,¹⁰ and Black Americans. Individuals were excluded from participation in the cohort for insufficient English proficiency, active treatment of a malignancy, residing in or waitlisting for a nursing home, any medical condition likely to preclude long-term follow-up, or interviewer suspicion of cognitive impairment.

An initial intake telephone interview was conducted to obtain verbal consent for participation and medical history. This was followed by an in-person assessment in the participant’s home, during which written consent, a resting electrocardiogram (ECG), biometric measurements, a medication inventory, and fasting blood and urine samples were obtained. ECGs were sent to the central electrocardiogram reading center (Epidemiological Cardiology Research Center, Wake Forest School of Medicine, Winston-Salem, North Carolina, USA) where they were interpreted and coded by clinicians blinded to other participant data. Glucose, total cholesterol, high-density lipoprotein, and triglyceride concentrations were measured in baseline blood samples via colorimetric reflectance spectrophotometry using the Orthos Vitros 950 IRC Clinical Analyzer (Johnson & Jonson Clinical Diagnostics, New Brunswick, New Jersey); the Friedewald equation¹¹ was used to derive low-density lipoprotein (LDL) concentration. Estimated glomerular filtration rate (eGFR) was calculated using the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) equation,¹² with plasma creatinine measured by isotope dilution mass spectrometry-traceable methods.

Resting blood pressure was measured by a trained examiner following a protocol using an aneroid sphygmomanometer over the brachial artery.¹³ Height and weight were measured without shoes using a metal tape measure and 300-pound-calibrated scale, respectively.

A cohort of participants free of suspected heart failure at baseline was assembled using information on medication use and medical history, as recently described.¹⁴ Detailed exclusions are shown in Figure 1.

Figure 1.

Exclusions. Abbreviations: ACEI, angiotensin converting enzyme inhibitor; AF, atrial fibrillation; ARB, angiotensin receptor blocker; HFrEF, heart failure with reduced ejection fraction; HFpEF, heart failure with reduced ejection fraction; HFmrEF, heart failure with midrange ejection fraction; HF, heart failure; ISMN, isosorbide mononitrate; ISDN, isosorbide dinitrate; REGARDS, REasons for Geographic And Racial Differences in Stroke;

The institutional review boards of all institutions involved in data ascertainment and/or processing approved the methods of the REGARDS study. The public were not directly involved in the design or conduct of REGARDS. The REGARDS Publications Committee reviewed and approved the data analysis plan for this manuscript and reviewed and approved the final manuscript and adherence to this plan.

Exposure & Outcomes

Prevalent AF was defined by detection of AF on baseline ECG and/or self-reported affirmative response to the telephone interview question “Has a physician or a health professional ever told you that you had atrial fibrillation?”¹⁵

Participants or their proxies were followed semiannually by telephone for monitoring of hospitalizations or deaths likely to involve HF and other outcomes such as stroke and coronary heart disease (CHD). Pertinent medical records for hospitalizations associated with suspected HF were obtained, and two clinicians independently adjudicated the diagnosis of heart failure using clinical documentation, left ventricular ejection fraction (LVEF) assessment from imaging studies (for example, echocardiography), and biomarkers such as natriuretic peptides. Discordance in adjudications was resolved by committee.

Incident HF events were defined as initial hospitalizations or deaths due to HF and were further subclassified into HF subtypes by LVEF. HFrEF was defined by documented LVEF <40% or qualitative report of reduced LVEF. HFpEF was defined by documented LVEF ≥50% or qualitatively normal LVEF. HF with mid-range ejection fraction (HFmrEF) was defined by documented LVEF ≥40% and <50%. Some HF events did not involve quantitative or qualitative evaluation of ventricular function and were therefore unclassified.

Variables

Baseline clinical and behavioral variables included age (years), sex, race, annual income (<$20,000, $20,000-$34,999, $35,000-$74,999, ≥$75,000, or declined to report), education level (< high school, high school graduate, some college, or ≥ college graduate), region (Stroke Belt, Stroke Buckle, or other¹⁰) and pack-year tobacco smoking history.

Prevalent medical conditions identified at baseline included diabetes mellitus (self-reported use of insulin or hypoglycemic medications, fasting glucose ≥6.99 mmol/L [126 mg/dL], or glucose ≥11.10 mmol/L [200 mg/dL] among those failing to fast), body mass index (BMI; calculated using height and weight obtained at baseline; continuous in kg/m² or in categories: underweight [<18.5 kg/m²], normal weight [18.5–24.9 kg/m²], overweight [25.0–29.9 kg/m²], or obesity [≥30.0 kg/m²]), left ventricular hypertrophy (LVH; defined by Sokolow-Lyon ECG criteria¹⁶), and current use of warfarin, aspirin, statins, or antihypertensive medications of any dose or brand from home medications inventory. A history of coronary heart disease (CHD) was indicated by self-reported history of myocardial infarction (MI), evidence of prior infarct on baseline ECG, or history of coronary artery bypass graft surgery or percutaneous coronary intervention with use of stents or angioplasty.

Incident coronary heart disease (CHD) events, defined as definite or probable nonfatal MI or CHD death,¹⁷ were also identified by semiannual telephone call and independently adjudicated by two clinicians with discordance resolved by committee.

Statistical Analysis

Baseline characteristics were compared between participants with and without baseline AF using Fisher’s exact tests (categorical variables), one-way analysis of variance (ANOVA; continuous variables), or Wilcoxon rank-sum test (pack-year smoking history; skewed distribution).

Survival analysis was conducted using Cox proportional-hazards models. Survival time began on the in-home visit date and ended on HF event date (failure) or censoring at date of last follow-up or December 31, 2016 (creation of the heart failure analytic cohort). First, Cox proportional-hazards models were fitted to risk of all HF events (aggregately, including HFrEF, HFpEF, HFmrEF, and unclassified HF). Sex and race differences in the association of AF with HF were evaluated with AF*sex and AF*race multiplicative interaction terms in the all-events models. Next, the data set was augmented for competing risks analysis according to Lunn and McNeil.¹⁸ HFmrEF and unclassified HF events were censored in the main analysis. We also planned a sensitivity analysis in which HFmrEF events were incorporated into the HFpEF event group, thus considering LVEF ≥40% as a threshold for preserved EF. P-values for the difference in survival function between HFrEF and HFpEF events are reported from interaction terms for HF subtype*AF in the augmented data set. Estimates were also reported separately within each race and sex subgroup.

Four sequential sets of covariates were used in all multivariable modeling. Model 1 included demographics: age, sex, race, income, education, and region. Model 2 included Model 1 covariates and added heart failure risk factors: smoking history, systolic blood pressure, diabetes mellitus, BMI (continuous), LDL, LVH, eGFR, antihypertensive medication use, and baseline CHD. Model 3 included Model 2 covariates and added baseline use of medications that modify cardiovascular disease risk: aspirin, warfarin, and statins. Model 4 included Model 3 covariates and added incident CHD as a time-varying covariate.

Unadjusted Kaplan-Meier failure curves stratified by baseline AF status were plotted for all HF events in the overall sample and for HFrEF and HFpEF events in the augmented Lunn-McNeil dataset. All statistical tests were two-sided, with p-values considered statistically significant when below 0.10 for multiplicative interaction terms and below 0.05 for all other tests. Participants missing data were excluded from analyses in which relevant information was missing. This analysis was performed with Stata, version 16.1 (StataCorp, College Station, Texas).

RESULTS

Sample Characteristics

This analysis included 25,787 participants who were HF-free at baseline. Prevalent AF at baseline was detected in 7.4% (n=1,896; 105 by ECG only, 1,637 by self-reported medical history only, and 154 by ECG and self-reported medical history). Table 1 compares baseline characteristics of included participants with and without AF at baseline. Participants with baseline AF were more likely to be older, less educated, with lower LDL, eGFR, and annual income, with more White participants, CHD history, diabetes mellitus, incident CHD events, use of warfarin, statins, or aspirin, and higher pack-year smoking history.

Table 1.

Baseline Characteristics of Included Participants by Baseline Atrial Fibrillation Status

	No Atrial Fibrillation (n=23,889)	Atrial Fibrillation (n=1,896)	p
Age (years; mean [95% CI])	64.2 (64.1, 64.4)	66.8 (66.3, 67.2)	<0.001
Male Sex (%)	45.0	44.8	0.829
Black Race (%)	40.6	35.8	<0.001
Annual Income (%)
<$20,000	16.6	20.1
$20,000–$34,999	23.8	25.3
$35,000–$74,999	30.6	28.0	<0.001
≥$75,000	17.2	13.1
Refused	11.9	13.5
Education (%)
< High School	11.4	12.0
High School Graduate	25.3	27.6	0.030
Some College	26.9	27.0
≥ College	36.4	33.3
Region (%)
Stroke Belt	34.4	33.8	0.019
Stroke Buckle	20.6	23.3
Other	45.0	42.9
Estimated glomerular filtration rate (mL/min/1.73m2; mean [95% CI])	86.6 (86.4, 86.8)	82.5 (81.8, 83.2)	<0.001
Smoking History (pack-years; median [IQR])	0.3 [0, 19]	1.5 [0, 22]	<0.001
Systolic Blood Pressure (mmHg, mean [95% CI])	127.2 (127.0, 127.4)	127.9 (127.2, 128.7)	0.07
Diabetes Mellitus (%)	18.7	21.0	0.014
Body Mass Index (%)
Underweight (<18.5 kg/m2)	1.1	1.2
Normal Weight (18.5–24.9 kg/m2)	24.8	26.3	0.401
Overweight (25–29.9 kg/m2)	38.0	36.5
Obesity (≥30 kg/m2)	36.1	36.0
Warfarin Use (%)	1.4	17.9	<0.001
Statin Use (%)	28.8	35.6	<0.001
Aspirin Use (%)	41.2	47.5	<0.001
Low-Density Lipoprotein (mmol/L; mean [95% CI])	3.00 (2.98, 3.01)	2.84 (2.80, 2.88)	<0.001
Left Ventricular Hypertrophy (%)	9.2	9.0	0.867
Antihypertensive Drug Use (%)	51.8	64.1	<0.001
Baseline Coronary Heart Disease (%)	13.4	30.9	<0.001
Coronary Heart Disease Events in Follow-Up (%)	5.1	8.8	<0.001

Abbreviations: CI, confidence interval; mmHg, millimeters of mercury; mmol/L, millimoles per liter; IQR, interquartile range

A total of 1,109 incident heart failure events were identified over mean follow-up of 9.0 years (standard deviation 3.6 years), including 356 HFpEF, 388 HFrEF, 77 HFmrEF, and 288 unclassified HF events.

Association of Atrial Fibrillation with All Heart Failure Events

Table 2 presents the association of AF with all incident HF events. Figure 1 depicts Kaplan-Meier curves for all HF events, stratified by baseline AF status. Baseline AF increased risk of all incident HF events in all models; this association was moderately attenuated across subsequent models. Associations of AF with all HF events were consistent across race and sex.

Table 2.

Distributions of Heart Failure Event Subtypes by Baseline Atrial Fibrillation Status

HF Event Subtype	No Atrial Fibrillation	Atrial Fibrillation	Total
All HF	915	194	1,109
HFpEF	295	61	356
HFmrEF	69	8	77
HFrEF	316	72	388
Unclassified HF	235	53	288
No HF	22,976	1,702	24,678
Total	23,891	1,896	25,787

Abbreviations: HF, heart failure; HFmrEF, heart failure with mid-range ejection fraction; HFpEF, heart failure with reduced ejection fraction; HFrEF, heart failure with reduced ejection fraction

Association of Atrial Fibrillation with HFrEF & HFpEF Events

Figure 2 depicts separate Kaplan-Meier curves for incident HFpEF and HFrEF events in the augmented dataset, stratified by baseline AF status. The separate associations of AF with incident HFrEF and HFpEF events in the augmented dataset are presented in Table 2. The cause-specific association of AF with HFpEF differed by sex (Model 4 sex difference interaction p = 0.024). In men, AF was associated with HFpEF in the demographic model, but this was attenuated and was no longer statistically significant when considering other risk factors.

Figure 2.

Unadjusted Kaplan-Meier failure curves for all incident heart failure events stratified by baseline atrial fibrillation status. Failure includes heart failure with reduced, preserved, midrange, and unclassified ejection fraction. Abbreviations: AF, atrial fibrillation; HF, heart failure.

Difference in Associations of AF with HFrEF vs. HFpEF

No significant differences in the associations of AF with HFrEF vs. HFpEF events were observed in the overall group, although a subjectively larger-magnitude association of AF with HFrEF was consistently observed across models.

No significant differences in the associations of AF with HFrEF vs. HFpEF events were observed in the Black or White subgroups. Despite the lack of a significant association of AF with HFpEF in men in models, 2–4, the associations of AF with HFrEF vs. HFpEF events did not statistically differ in either sex subgroup.

Sensitivity Analysis

Results of a planned sensitivity analysis redefining HFpEF as LVEF ≥40% are reported in Table 3 and Figure 3. Results did not differ substantially from those of the primary analysis, except for a subjectively lower magnitude of associations of AF with HFpEF events when defining HFpEF as LVEF ≥40% vs. LVEF ≥50%.

Table 3.

Association of Atrial Fibrillation With Incident Heart Failure Events

		n events / n at risk	Model 1	Model 2	Model 3	Model 4
	HF Group*		HR (95% CI)	HR (95% CI)	HR (95% CI)	HR (95% CI)	χ2 p For Subtype Difference
Overall	All HF	1,109/25,787	2.61 (2.24, 3.05)	2.11 (1.79, 2.50)	1.75 (1.46, 2.11)	1.67 (1.38, 2.01)
	HFrEF	388/25,787	2.88 (2.22, 3.72)	2.36 (1.79, 3.12)	2.02 (1.48, 2.74)	1.87 (1.38, 2.54)	0.581
	HFpEF	356/25,787	2.54 (1.92, 3.34)	1.99 (1.48, 2.68)	1.74 (1.26, 2.39)	1.65 (1.20, 2.28)
Women †	All HF	493/14,176	2.40 (1.89, 3.04)	1.92 (1.47, 2.50)	1.75 (1.33, 2.30)	1.76 (1.34, 2.32)
	HFrEF	141/14,176	2.89 (1.90, 4.40)	2.23 (1.39, 3.57)	2.16 (1.33, 3.52)	2.21 (1.36, 3.59)	0.145
	HFpEF	198/14,176	2.82 (1.98, 4.02)	2.49 (1.70, 3.65)	2.28 (1.54, 3.40)	2.26 (1.52, 3.36)
Men †	All HF	616/11,611	2.73 (2.22, 3.36)	2.27 (1.82, 2.83)	1.74 (1.34, 2.24)	1.59 (1.23, 2.06)
	HFrEF	245/11,611	2.84 (2.05, 3.94)	2.51 (1.77, 3.55)	1.94 (1.30, 2.90)	1.75 (1.17, 2.60)	0.943
	HFpEF	158/11,611	2.07 (1.33, 3.24)	1.46 (0.90, 2.36)	1.15 (0.67, 2.00)	1.05 (0.61, 1.83)
Black ‡	All HF	445/10,310	2.46 (1.88, 3.23)	1.81 (1.34, 2.44)	1.64 (1.20, 2.23)	1.60 (1.17, 2.18)
	HFrEF	146/10,310	2.92 (1.87, 4.57)	1.91 (1.16, 3.14)	1.77 (1.06, 2.96)	1.80 (1.08, 3.02)	0.434
	HFpEF	144/10,310	2.15 (1.31, 3.53)	1.89 (1.12, 3.20)	1.79 (1.05, 3.05)	1.74 (1.02, 2.96)
White ‡	All HF	664/15,477	2.60 (2.15, 3.15)	2.24 (1.82, 2.75)	1.84 (1.45, 2.32)	1.75 (1.38, 2.22)
	HFrEF	242/15,477	2.74 (2.00, 3.77)	2.54 (1.81, 3.55)	2.14 (1.46, 3.16)	2.01 (1.37, 2.97)	0.918
	HFpEF	212/15,477	2.66 (1.89, 3.72)	2.00 (1.39, 2.87)	1.70 (1.13, 2.54)	1.61 (1.07, 2.42)

Abbreviations: AF, atrial fibrillation; HF, heart failure; HFrEF, heart failure with reduced ejection fraction; HFpEF, heart failure with preserved ejection fraction; HR, hazard ratio; CI, confidence interval

Covariates in Multivariable Models:

Model 1: age, sex, race, income, education, & region

Model 2: Model 1 covariates + smoking (pack-years), systolic blood pressure, diabetes mellitus, body mass index (continuous), low-density lipoprotein, left ventricular hypertrophy, estimated glomerular filtration rate, antihypertensive drug use, and coronary heart disease history.

Model 3: Model 2 covariates + aspirin, warfarin, and statin use.

Model 4: Model 3 covariates + incident coronary heart disease events (time-varying)

^*The entire sample, including non-cases and HFrEF, HFpEF, HFmrEF, and unclassified events, was included in analysis for “all HF”. Unclassified and HFmrEF cases were censored in the Lunn-McNeil analysis for “HFpEF” and “HFrEF”

^†Model 4 AF*sex interaction p-values (cause-specific): All HF: 0.697; HFrEF: 0.692; HFpEF: 0.024; HFmrEF: 0.908

^‡Model 4 AF*race interaction p-values (cause-specific): All HF: 0.945; HFrEF: 0.995; HFpEF: 0.917; HFmrEF: 0.744

Figure 3.

Unadjusted Kaplan-Meier failure curves for HFrEF and HFpEF events in the augmented dataset for Lunn-McNeil analysis, with each HF subtype stratified by baseline atrial fibrillation status. Abbreviations: AF, atrial fibrillation; HF, heart failure; HFrEF, heart failure with reduced ejection fraction; HFpEF, heart failure with preserved ejection fraction.

DISCUSSION

In this prospective analysis of the contemporary and biracial REGARDS cohort, we showed that the associations of baseline AF with incident HFrEF vs. HFpEF events did not significantly differ over 9 years of follow-up, independent of risk factors, medication use, and incident CHD events. This suggests that AF is a similarly important risk factor for both primary subtypes of HF. Specifically, participants with vs. without prevalent AF had a 65% increased risk of all heart failure events, 86% increased risk of HFrEF events, and 64% increased risk of HFpEF events in the maximally adjusted model.

Our finding of an overall association of AF with incident HF events is consistent with multiple prior studies of various populations.^{3 19–21} However, our finding of no significant difference in the associations of AF with HFrEF vs. HFpEF events contrasts with one study of Framingham Heart Study participants reporting an association of AF with HFpEF, but not with HFrEF, and a resulting difference in the association between subtypes.²² This discordance is likely due to the smaller sample size, enrollment in an earlier era, and limited inclusion of non-White participants in that cohort.

The reasons underlying these associations remain uncertain. While individuals with AF may represent a population at higher risk for HF, little research has focused on how the pathophysiology of AF could also contribute to separate associations between AF and HFrEF and HFpEF. A causal relationship of AF with HF has been purported to occur through compromised diastolic filling and cardiac output as a result of elevated ventricular rate, irregular cycle length, and loss of atrial systole, as well as concurrent neurohormonal changes and molecular alterations.⁵ It is possible that different pathophysiologic characteristics of AF are associated with each subtype in similarly important ways. For example, loss of the physiologic increase in inotropy with increased contraction rate could play a more important role in the association of AF with HFrEF,²³ while decoupling of myocardial relaxation and contraction functions as a result of irregular cycle length⁵ could contribute more to the association of AF with HFpEF. However, the observed associations of AF with HF and its subtypes may be better attributed to a unifying disease process or overlapping or synergistic pathophysiology, although our findings and modeling approach suggest these separate associations are not entirely due to shared traditional risk factors between these phenotypes. Thus, further basic and translational investigation into the pathophysiology underlying associations of AF with HFrEF and HFpEF is necessary.

That no association of AF with HFpEF events was observed in men in Models 2–4 is consistent with established sex differences in the epidemiology of HFpEF.²⁴ The racial epidemiology of AF appears paradoxical,²⁵ whereby Black Americans have a higher prevalence of established AF risk factors, but White Americans appear to have higher risk for AF. Conversely, risk for HF appears to be greater in Black vs. White Americans.²⁶ Despite these racial differences in the separate epidemiology of AF and HF, no significant differences between Black and White participants were observed in the association of AF with all HF, HFrEF, or HFpEF. This importantly suggests that AF holds similar relevance to HF, regardless of race or HF subtype.

Strengths & Limitations

Several limitations of this study must be considered. Firstly, findings from studies of the REGARDS cohort may have limited generalizability to race groups other than Black or White. As AF is more realistically considered as a continuum of frequency rather than a dichotomy (present or absent),²⁷ we are unable to account for AF burden or frequency-based categories (paroxysmal, persistent, or long-standing persistent AF). AF burden likely has an impact on HF outcomes, as participants with permanent vs. paroxysmal AF had higher risk of incident HF in a recent study.²⁸ Nevertheless, studies evaluating the association of AF burden with risk of HF are sparse; further research on the association of AF burden with subtypes of HF is needed. Baseline assessment of cardiac function parameters (i.e. LVEF) was not available in REGARDS, so the possibility exists that some participants had subclinical HF at baseline. Participants developing HF not resulting in hospitalization or death, such as that managed in the outpatient care setting, did not meet the criteria for incident HF because of difficulty in detecting this across in a large cohort. Lastly, although we enhanced AF detection at baseline with electrocardiography, ¹⁵ the proportion of participants with AF observed on study-scheduled ECG was relatively low. The prevalence of subclinical or undetected AF is likely 2.5 to 4%^{29 30} and we cannot exclude that some participants developed AF after the initial visit. However, misclassification of participants with AF as not having AF would be expected to bias findings towards the null.

This study has several noteworthy strengths. We evaluated the difference in the associations of AF with HFpEF vs. HFrEF events in a contemporary and diverse cohort with ongoing follow-up. Events were rigorously and conservatively adjudicated, and we used a conservative threshold in defining preserved LVEF in the primary analysis. REGARDS has a similar number of HFpEF and HFrEF events. Importantly, in comparing the survival functions for HFrEF and HFpEF across baseline AF status, we used a Lunn & McNeil augmented dataset approach rather than cause-specific hazard functions (in which other relevant failure types are censored). This was critical to direct comparison of competing risks and the integrity of resulting Kaplan-Meier plots, given that the risks of HFrEF and HFpEF are unlikely to be independent of one another.

In conclusion, over median 9 years’ follow-up of the REGARDS study, a cohort of contemporary Black and White Americans, AF was associated with all HF, HFrEF, and HFpEF events, and there was no significant difference in the associations of AF with incident HFrEF vs. HFpEF events. This suggests that AF increases risk for each of these primary subtypes to a similar magnitude. No differences in the associations of AF with HFrEF vs. HFpEF events were observed in any sex or race group, and findings were corroborated in sensitivity analysis using a less conservative definition of HFpEF. Further basic and translational research is needed to differentiate the mechanisms underlying the separate associations of AF with HFrEF and HFpEF.

Figure 4.

Unadjusted Kaplan-Meier failure curves in sensitivity analysis for heart failure events with left ventricular ejection fraction <40% and ≥40% in the augmented dataset for Lunn-McNeil analysis, each stratified by baseline atrial fibrillation status. Abbreviations: AF, atrial fibrillation; HF, heart failure; LVEF, left ventricular ejection fraction.

Table 4.

Sensitivity Analysis Defining HFpEF as LVEF≥40%: Association of Atrial Fibrillation With Incident Heart Failure Events

			Model 1	Model 2	Model 3	Model 4
	HF Group*	n events / n at risk	HR (95% CI)	HR (95% CI)	HR (95% CI)	HR (95% CI)	χ2 p For Subtype Difference
Overall	LVEF <40%	388/ 25,787	2.87 (2.22, 3.72)	2.35 (1.78, 3.11)	2.01 (1.48, 2.72)	1.85 (1.36, 2.51)	0.366
	LVEF ≥40%	433/ 25,787	2.33 (1.80, 3.02)	1.84 (1.39, 2.42)	1.61 (1.20, 2.17)	1.52 (1.13, 2.05)
Women †	LVEF <40%	141/14,176	2.89 (1.90, 4.40)	2.22 (1.39, 3.55)	2.16 (1.33, 3.51)	2.20 (1.35, 3.57)	0.114
	LVEF ≥40%	224/14,176	2.57 (1.83, 3.63)	2.23 (1.54, 3.23)	2.05 (1.40, 3.01)	2.03 (1.39, 2.98)
Men †	LVEF <40%	247/11,611	2.84 (2.05, 3.94)	2.50 (1.77, 3.54)	1.94 (1.30, 2.89)	1.72 (1.16, 2.56)	0.803
	LVEF ≥40%	209/11,611	1.96 (1.32, 2.92)	1.43 (0.93, 2.17)	1.15 (0.72, 1.85)	1.04 (0.65, 1.68)
Black ‡	LVEF <40%	146/10,310	2.89 (1.85, 4.52)	1.88 (1.14, 3.10)	1.75 (1.05, 2.92)	1.78 (1.07, 2.98)	0.334
	LVEF ≥40%	173/10,310	1.99 (1.25, 3.18)	1.73 (1.06, 2.83)	1.64 (1.00, 2.70)	1.58 (0.96, 2.61)
White ‡	LVEF <40%	242/15,477	2.74 (2.00, 3.76)	2.51 (1.79, 3.52)	2.12 (1.44, 3.13)	1.97 (1.34, 2.91)	0.745
	LVEF ≥40%	260/15,477	2.44 (1.79, 3.34)	1.86 (1.33, 2.61)	1.61 (1.11, 2.33)	1.51 (1.04, 2.20)

Abbreviations: AF, atrial fibrillation; HF, heart failure; HFrEF, heart failure with reduced ejection fraction; HFpEF, heart failure with preserved ejection fraction; EF, ejection fraction; HR, hazard ratio; CI, confidence interval

Covariates in Multivariable Models:

Model 1: age, sex, race, income, education, & region

Model 2: Model 1 covariates + smoking (pack-years), systolic blood pressure, diabetes mellitus, body mass index (continuous), low-density lipoprotein, left ventricular hypertrophy, estimated glomerular filtration rate, antihypertensive drug use, and coronary heart disease history.

Model 3: Model 2 covariates + aspirin, warfarin, and statin use.

Model 4: Model 3 covariates + incident coronary heart disease events (time-varying)

^*Unclassified cases were censored in the Lunn-McNeil analysis.

^†Model 4 AF*sex interaction p-values (cause-specific): LVEF <40%: 0.692; LVEF ≥40%: 0.022

^‡Model 4 AF*race interaction p-values (cause-specific): LVEF <40%: 0.991; LVEF ≥40%: 0.971

KEY QUESTIONS.

What is already known about this subject?

• Atrial fibrillation and heart failure have bidirectional associations that appear driven through shared risk factors and pathophysiology.

What does this study add?

• We confirmed an association of atrial fibrillation with all incident heart failure events exists in a contemporary and biracial cohort.

• We found that the associations of atrial fibrillation with heart failure with reduced versus preserved ejection fraction events do not differ significantly.

• We showed that our finding of no significant difference in the associations of atrial fibrillation with heart failure subtypes is consistent across sex and race subgroups.

How might this impact on clinical practice?

• Clinicians should be aware that their patients with atrial fibrillation appear to be at similar risk for both primary subtypes of heart failure.

• Further clinical and translational research examining the pathophysiology through which atrial fibrillation and each subtype of heart failure are associated may allow for strategies to prevent these often-concurrent diseases.

DATA AVAILABILITY STATEMENT

The data used in these analyses include potentially identifying participant information and therefore are not publicly available due to legal and ethical restrictions. Qualified investigators may request access from the University of Alabama at Birmingham to obtain de-identified data (regardsadmin@uab.edu).