Associations of Lipoprotein(a) Levels With Incident Atrial Fibrillation and Ischemic Stroke: The ARIC (Atherosclerosis Risk in Communities) Study

Study Cohort

The ARIC study is a mostly biracial, community‐based, prospective epidemiological trial of cardiovascular disease. A complete description of the study design, objectives, and sampling strategy has been previously described.25 Briefly, the ARIC study enrolled 15 792 participants from 1987 to 1989. Participants were aged 45 to 64 years and came from 4 communities in the United States (Washington County, MD; the suburbs of Minneapolis, MN; Jackson, MS; and Forsyth County, NC). After baseline assessment, participants took part in up to 4 additional examinations (1990–1992, 1993–1995, 1996–1998, and 2011–2013). In addition, ARIC study participants received annual follow‐up calls to screen for hospitalizations or major changes in their health status. The ARIC study protocols were approved by the institutional review boards at all ARIC study sites and the Coordinating Center. All participants provided written informed consent at each study visit. For this analysis, we used ARIC visit 4 (1996–1998), the time of Lp(a) measurement, as baseline. Visit 4 included 11 656 participants.

For our primary aim we included all ARIC study participants who were free of prevalent or incident AF occurring before visit 4. Exclusion criteria were: an AF event at or before visit 4 (n=293); missing Lp(a) levels (n=165); racial/ethnic group other than white or black (n=31); blacks from the MN and MD field centers (n=37; low numbers limited our ability to adjust for their race‐center groups); missing covariates (n=1221); or follow‐up that ended on or before visit 4 (n=1). After exclusions, 9908 participants were included in our primary analysis for incident AF (43.9% men, 21.6% blacks) (Figure 1).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1.

Participant flow diagram for aim 1 (incident atrial fibrillation [AF]). ARIC indicates Atherosclerosis Risk in Communities study; Lp(a), lipoprotein(a).

For our secondary aim, we included all ARIC study participants who were free of prevalent or incident stroke before visit 4. Exclusion criteria were: ischemic stroke at or before visit 4 (n=139); missing Lp(a) levels (n=167); racial/ethnic group other than white or black (n=31); blacks from the MN and MD field centers (n=37); missing covariates (n=1153); or follow‐up that ended on or before visit 4 (n=2). After exclusions, we included 10 127 participants (44.1% men, 21.1% blacks) in our secondary analysis for incident stroke (Figure 2).

• Download figure
• Open in new tab
• Download powerpoint

Figure 2.

Participant flow diagram for aim 2 (incident stroke, stratified by history of atrial fibrillation vs no history of atrial fibrillation. ARIC indicates Atherosclerosis Risk in Communities study; Lp(a), lipoprotein(a).

Lp(a) Measurement

Lp(a) levels were measured in EDTA plasma collected at ARIC study visit 4 using the Denka Seiken assay (Denka Seiken Co). This is a well‐validated assay, with good reproducibility (coefficient of variation, 1.26–2.22%), that yields robust results and is generally insensitive to apolipoprotein(a) size heterogeneity.26 As per prior ARIC study analyses,8 Lp(a) levels were examined: (1) as a continuous logarithmically transformed variable, and (2) as 5 prespecified Lp(a) categories defined by the arbitrary thresholds of ≤10, >10 to ≤20, >20 to ≤30, >30 to ≤50, and >50 mg/dL.

AF Ascertainment

Our primary outcome was first occurrence of AF (after visit 4) through December 31, 2011. We defined the incidence date of AF as the date for the first ECG showing AF, the first hospital discharge coded as AF, or when AF was listed as a cause of death, whichever occurred earlier. AF was ascertained by 3 different sources: ECGs performed during study visits, hospital discharge International Statistical Classification of Diseases (ICD) codes, and death certificates.19 At each ARIC study visit, a 10‐second 12‐lead ECG was obtained and transmitted to the ARIC ECG Reading Center for coding, interpretation, and storage. All abnormal ECGs were subsequently adjudicated by a trained cardiologist, and both atrial flutter and AF were coded as AF.27 Hospitalizations were detected with annual follow‐up telephone calls and review of local hospital discharges. AF events were identified by the presence of the ICD, 9th Revision, Clinical Modification code of 427.31 (AF) or 427.32 (atrial flutter), listed as a discharge diagnostic code at any position. We excluded AF events associated with cardiac surgery. AF events were also identified if ICD‐9 code 427.3 or ICD‐10 code I48 was listed as a cause of death. More than 98% of incident AF cases were identified from hospital discharge codes, and less than 1% of AF cases were identified from death certificates.28 The sensitivity and specificity of incident AF by hospital discharge was evaluated to be over 80% and 97%.19

Stroke Ascertainment

Our secondary outcome was incident ischemic stroke through December 31, 2013. Stroke events were identified by annual phone calls, hospital ICD‐9 codes 430 to 438 until 1997 and ICD‐9 codes 430 to 436 afterwards (listed as a discharge diagnostic code at any position), or by death certificates. Trained ARIC study personnel abstracted from hospital records, stroke signs and symptoms, and cerebrovascular imaging findings. On the basis of National Survey of Stroke criteria,29 a computer algorithm and a panel of ARIC study physicians (which included a neurologist) categorized all potential stroke events (hospitalized or out‐of‐hospital stroke deaths) as definite or possible strokes; an additional study reviewer adjudicated discrepancies. Strokes were subclassified as ischemic versus hemorrhagic. For this analysis, we included only ischemic strokes. In addition, the ARIC study physicians further classified ischemic strokes into thrombotic, lacunar, or cardioembolic subtypes, according to criteria adopted from the National Survey of Stroke subtype classification.29, 30 Given that the aim of this study was in regards to AF‐related stroke outcomes, we considered only cardioembolic strokes in a sensitivity analysis.

Assessment of Other Covariates

We included demographic and clinical variables collected during ARIC study visit 4 as follows: age, sex, race, study site, systolic blood pressure (BP), diastolic BP, hypertension treatment, heart rate, height, body mass index, ECG left ventricular hypertrophy, PR interval, low‐density lipoprotein cholesterol, high‐density lipoprotein cholesterol, triglycerides, N‐terminal pro–B type natriuretic peptide, diabetes mellitus, smoking, personal history of CHD (history of myocardial infarction, myocardial infarction indicated on baseline ECG, or revascularization), personal history of stroke, personal history of systemic embolism, personal history of heart failure (HF), and personal history of peripheral artery disease. We calculated the CHA₂DS₂‐VASc score.31 Hypertension was defined as systolic BP ≥140 mm Hg and/or diastolic BP ≥90 mm Hg, or BP medication use in the past 2 weeks. Diabetes mellitus was considered present if one of the following criteria was met: (1) fasting glucose ≥126 mg/dL or nonfasting glucose ≥200 mg/dL, (2) self‐reported physician diagnosis of diabetes mellitus, or (3) current use of medication for diabetes mellitus. Prevalent HF was identified by the Gothenburg criteria during visit 132 or an incident HF‐related admission ascertained by ARIC study investigators between visit 1 and 4. Peripheral artery disease was defined as an ankle‐brachial index <0.9 assessed at visit 4 or peripheral artery disease–related hospitalization, amputation, leg revascularization, or intermittent claudication as assessed by annual questionnaire.33

Statistical Analysis

We presented baseline characteristics of study participants as mean±SD for continuous variables that were normally distributed, median (interquartile range) for continuous variables that were not normally distributed, and proportions for categorical variables, across the 5 Lp(a) categories. We used progressively adjusted multivariable Cox proportional hazards models to estimate the hazard ratios (95% confidence intervals) of incident AF and incident stroke across different Lp(a) levels, as further described below. We performed analyses using Lp(a) as a categorical variable with the 5 defined groups and as a logarithmically transformed continuous variable.

Baseline Characteristics

The mean age of the study population was 62.7 years (5.6 years). Baseline characteristics, stratified by Lp(a) levels, are shown in Table 1. Women had higher Lp(a) levels. Low‐density lipoprotein cholesterol levels and the percentage of participants taking lipid‐lowering medications increased with increasing Lp(a) levels. Prevalence of CHD was higher in participants with higher Lp(a) levels.

Incident AF

The median follow‐up was 13.9 years (interquartile range, 12.4–14.8 years). Among 9908 participants free of AF at baseline (visit 4), the incidence rate of AF was 8.8 events per 1000 person‐years overall, and was higher in whites compared with blacks (9.5 versus 6.1 events per 1000 person‐years) and men compared with women (10.8 versus 7.3 events per person‐years). There was no association between Lp(a) levels and incidence of AF (Table 2). The AF incidence rate was similar in participants with Lp(a) ≤10 mg/dL and those with >50 mg/dL (9.1 versus 8.4 events per 1000 person‐years). In the fully adjusted model (adjusted for demographics, prevalent CHD, other CVD risk factors, and lipids), the hazard ratio of AF for high Lp(a) levels ≥50 mg/dL compared with <10 mg/dL was 0.98 (95% confidence interval, 0.82–1.17). Figure 3 shows a restricted cubic spline of AF risk across increasing levels of Lp(a), using the 10th percentile of Lp(a) levels as a reference. There was no effect modification of the association between Lp(a) levels and AF risk by race (P interaction=0.39) or sex (P interaction=0.38).

• Download figure
• Open in new tab
• Download powerpoint

Figure 3.

Atrial fibrillation (AF) risk across lipoprotein(a) (Lp[a]) levels. Multivariable‐adjusted restricted cubic spline model showing the hazard ratios of AF (95% confidence intervals) by Lp(a) levels at visit 4. The solid line represents hazard ratios and the dashed lines represent 95% confidence intervals. Knots at 5th, 35th, 65th, and 95th percentiles (corresponding to 1.3, 7.7, 23.9, and 83.1 mg/L). The spline is centered at the 10th percentile. The histogram shows the distribution of Lp(a) levels. Restricted cubic spline is truncated at the 1st and 99th percentiles of Lp(a). The model is adjusted for age, sex, race‐center groups, smoking, systolic and diastolic blood pressure, treatment for hypertension, heart rate, height, body mass index, ECG left ventricular hypertrophy, PR interval, prevalent heart failure, coronary artery disease, diabetes mellitus, low‐density lipoprotein cholesterol, high‐density lipoprotein cholesterol, triglycerides, lipid‐lowering medication, and log‐transformed N‐terminal pro–B type natriuretic peptide.

Incident Stroke

The median follow‐up was 15.8 years (interquartile range, 13.6–16.7 years). Among 10 127 participants free of stroke at baseline (visit 4), the incidence of stroke was higher by 4‐fold in participants with AF compared with those without (10.8 versus 2.9 events per 1000 person‐years). Among participants without a history of prevalent AF, those with Lp(a) >50 mg/dL had 42% higher risk of developing a stroke compared with those with ≤10 mg/dL (hazard ratio, 1.42; 95% confidence interval, 1.07–1.90), even after adjustment for numerous cardiovascular disease risk factors, other lipid markers, and CHA₂DS₂‐VASC score (Table 3). On the other hand, we failed to demonstrate a statistically significant relative increase in ischemic stroke risk among participants with a history of prevalent AF (hazard ratio, 1.06; 95% confidence interval, 0.70–1.61); however, there was no statistically significant interaction by AF status (P value for interaction=0.25). The relative hazard of stroke was not increased in participants with or without AF for intermediate Lp(a) levels of >10 to 20, >20 to 30, and >30 to 50 mg/dL compared with those with ≤10 mg/dL (Table 3). Figure 4 shows restricted cubic splines of the risk of stroke across increasing levels of Lp(a) in participants with and without AF, using the 10th percentile of Lp(a) levels as a reference. There was no effect modification of the association between Lp(a) levels and stroke risk by race or sex; the 3‐way multiplicative interactions of Lp(a) and AF by race and sex were P=0.85 and P=0.45, respectively, for continuous Lp(a).

• Download figure
• Open in new tab
• Download powerpoint

Figure 4.

Ischemic stroke risk across lipoprotein(a) (Lp[a]) levels stratified by atrial fibrillation (AF) status. Multivariable‐adjusted restricted cubic spline models showing the hazard ratio of ischemic stroke (95% confidence intervals) by Lp(a) levels at visit 4. The solid line represents the hazard ratio and the dashed lines represent the 95% confidence intervals. Knots at 5th, 35th, 65th, and 95th percentiles (corresponding to 1.3, 7.7, 23.9, and 83.1 mg/L). The splines are centered at the 10th percentile. The histograms show the distribution of Lp(a) levels. Restricted cubic splines are truncated at the 1st and 99th percentiles of Lp(a). The models are adjusted for age, sex, race‐center groups, smoking, systolic and diastolic blood pressure, treatment for hypertension, heart rate, height, body mass index, ECG left ventricular hypertrophy, PR interval, prevalent heart failure, coronary artery disease, and diabetes mellitus. hx indicates history.

In a sensitivity analysis, examining the risk of incident cardioembolic stroke (n=129 of 541 ischemic stroke), there was no association between Lp(a) levels and cardioembolic stroke (Table 4).

Study Strengths

Our study has several strengths. First, we utilized a large, biracial cohort with adequate AF and stroke events to test our hypotheses. The ARIC study's design enabled assessment for statistical interaction by race and sex. The ARIC study employed standardized protocols that facilitated accurate collection of outcomes and covariates. Second, Lp(a) levels were measured with a well‐validated assay that is robust to different apolipoprotein(a) isoform sizes. Third, AF was an adjudicated outcome in data collected from ARIC study visits, although AF events detected using ICD codes were not adjudicated. However, the use of ICD codes to ascertain AF in the ARIC study is a valid method, as ≈90% of AF cases were confirmed in a physician review of discharge summaries from 125 possible AF cases.19 Last, stroke events and their subtypes were adjudicated by trained study physicians including neurologists.

Study Limitations

Our study has several limitations. First, this was a community‐based sample unselected for lipid disorders, and extreme phenotypes of high Lp(a) levels (ie, >100 mg/dL) were rare and could not be examined. Thus, whether extremely high Lp(a) levels are associated with AF risk or stroke in patients with AF remains unknown. Second, we included only whites and blacks and thus these results may not generalize to other racial/ethnic groups. Third, we only had a single measure of Lp(a) at baseline. However, since Lp(a) levels are genetically determined,37 levels are generally stable in men and postmenopausal women over time. Although Lp(a) levels can be modifiable with lipid therapy, we did adjust for use of lipid‐lowering medications in our analyses. Fourth, we did not include data from the Centers for Medicare and Medicaid Services that were linked to ARIC study participants. However, more than half of ARIC participants were younger than 65 years (mean age, 63 years), which would not be captured by this Centers for Medicare and Medicaid Services data set, and it has previously been shown that adding Medicare data does not have a meaningful impact in most associations between AF and risk factors in the ARIC study cohort.38 Last, this is an observational study and we cannot exclude residual confounding, despite adjusting for potential covariates.