Blood Lipid Levels, Lipid-Lowering Medications, and the Incidence of Atrial Fibrillation: The Atherosclerosis Risk in Communities Study

The ARIC study is a prospective cohort study of cardiovascular disease and atherosclerosis risk factors. Participants at baseline (1987–1989) included 15 792 men and women, aged 45 to 64 years, recruited from 4 communities in the United States (Washington County, MD; the suburbs of Minneapolis, MN; Jackson, MS; and Forsyth County, NC). Participants were mostly white in the Washington County and Minneapolis centers, only African American in the Jackson center, and both races in Forsyth County. After the initial assessment, study participants were examined 3 additional times (1990–1992, 1993–1995, and 1996–1998). Response rates for surviving participants at each examination were 93%, 86%, and 80%, respectively. In addition, ARIC participants have received annual follow-up calls since baseline, and survivors have a response rate ≥90%. The study was approved by institutional review boards at each participating center, and all study participants provided written informed consent.

AF diagnoses were ascertained by 3 different sources in the ARIC study: ECGs performed at study visits, hospital discharge codes, and death certificates.³ At each ARIC study visit, a 10-second 12-lead ECG was performed using an MAC PC cardiograph (Marquette Electronics Inc; Milwaukee, WI) and transmitted to the ARIC ECG Reading Center for coding, interpretation, and storage. All ECGs automatically coded as AF were visually checked by a trained cardiologist to confirm AF diagnosis.²²

Annual follow-up calls and a review of local hospital discharges identified hospitalizations in ARIC participants through the end of 2007. The International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9) code of 427.31 or 427.32, listed as a discharge diagnostic code, identified AF cases. AF events associated with cardiac surgery were excluded in this study. Validity was sufficient because ≈90% of the cases were confirmed in a physician review of discharge summaries from 125 possible AF cases.³ AF cases were also identified if ICD-9 code 427.3 or International Statistical Classification of Diseases, 10th Revision (ICD-10) code I48 was listed as a cause of death. Most incident AF cases (>98%) in this analysis were identified from hospital discharge codes. In this analysis, the incidence date of AF was defined 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.

Blood samples were collected at each visit after a fast of at least 8 hours and sent to the ARIC Central Lipid Laboratory for processing. Detailed procedures are available elsewhere.^23,24 Briefly, total plasma cholesterol and triglycerides were determined by enzymatic methods, HDLc was measured after dextran-magnesium precipitation, and LDLc was calculated by the Friedewald equation²⁵ in those with triglyceride levels <4.52 mmol/L.

The use of lipid-lowering medications was self-reported and confirmed by checks on medications brought to each visit by the patient. Lipid medications were divided into 2 categories: statins and other lipid medications. Included in the “other” category were niacin (vitamin B3), bile sequestrants, fibrates, and other antihyperlipidemics. Participants taking both statins and another type of lipid-lowering medication were considered to be in the statin category.

At each study visit, participants underwent a physical examination and provided self-reported information. Detailed procedures on covariate measurements are available elsewhere.²³ At each visit, prevalent HF was defined as the reported use of HF medication in the previous 2 weeks, the presence of HF according the Gothenburg criteria (only at the baseline visit), or having developed incident HF from the previous visit.²⁶ Incident HF was defined as the presence of ICD-9 code 428 in any hospitalization during follow-up.²⁷ Prevalent CHD was defined as prior cardiovascular revascularization, physician-diagnosed myocardial infarction, or presence of a previous myocardial infarction by ECG. Incident CHD was ascertained by the ARIC Morbidity and Mortality Classification Committee using data from follow-up calls, hospitalization records, and death certificates.²⁸ Prevalent stroke was defined as the self-reported physician diagnosis of a stroke before visit 1, and after visit 1, it was adjudicated from diagnosis codes indicative of cerebrovascular disease.²⁹ Incident stroke was defined as the first probable or definite hospitalized stroke occurring in a participant free of a history of diagnosed stroke at baseline.

During a median follow-up of 18.7 years, we identified 1433 incident AF cases. After adjustment for potential confounders and intermediates, there was no significant association between baseline levels of HDLc and triglycerides with incident AF (Table 2). Higher levels of LDLc and total cholesterol at baseline were associated with a lower risk of incident AF. Compared to those with LDLc levels <2.59 mmol/L, the hazard ratios (HRs) and 95% CIs of AF in individuals with an LDLc of 2.59–4.13 mmol/L was 0.84 (0.72–0.97); and in those with an LDLc ≥4.14 mmol/L, 0.85 (0.72–1.01) (P=0.06 for trend). Compared with those with total cholesterol levels <5.18 mmol/L, the HR (95% CI) of AF was 0.88 (0.76–1.01) in those with levels ≥6.22 mmol/L (P=0.03 for trend). Models not adjusting for prevalent stroke, HF, and CHD and also models additionally adjusting for incident stroke, HF, and CHD produced similar results.

Results of the association between incident AF and each lipid level measurement at visits 2 to 4 were similar and are shown in Figure 1. A significant inverse association was seen between LDLc and total cholesterol and incident AF in most visits, with stronger associations in later visits. Results were comparable when we considered lipid levels as time-dependent exposures, also shown in Figure 1. The HRs (95% CIs) of AF by 1-SD increment in time-varying HDLc and triglyceride measurements were 0.97 (0.91–1.04) and 1.00 (0.96–1.04), respectively. The HR (95% CI) in time-varying LDLc measurements was 0.90 (0.85–0.96); and in total cholesterol, 0.89 (0.84–0.95). In analyses including only AF events identified in study ECGs (n=117), the HR (95% CI) of AF per 1 SD increment was 0.79 (0.63–0.98) for LDLc and 0.75 (0.60–0.93) for total cholesterol. Adjustment for exercise in the models did not change the significance of the association. An additional analysis was performed examining the change in LDLc from visit 1 to visit 4. After excluding those taking cholesterol medications, and after adjustment for visit 1 and visit 4 covariates and baseline LDLc levels, the HR (95% CI) for a 1-mmol/L (38.6 mg/dL) change of LDLc was 0.79 (0.69–0.90). These results indicate that the larger the decrease in LDLc from visit 1 to visit 4, the lower the risk of AF. LDLc was not associated with overall risk of hospitalizations (HR, 0.98 [95% CI, 0.96–1.00] for a 1-SD increment), and adjustment for hospitalizations as a time-dependent covariate did not change the association between LDLc and AF risk (HR, 0.91 [95% CI, 0.86–0.97] for a 1-SD LDLc increment as a time-dependent covariate).

AF-free survival curves from visits 1 and 4 by LDLc levels are shown in Figure 2, with higher survival in those with higher LDLc. There was no significant age or race interaction between HDLc, LDLc, and total cholesterol levels. However, the association of triglyceride levels with AF risk varied by race (P=0.02), with a stronger association in African Americans compared with whites (HR [95% CI], 1.14 [1.03–1.26] versus 0.99 [0.94–1.03], respectively).

We assessed the association of use of cholesterol medications with AF risk, considering visit 2 as baseline, while adjusting for covariates from the previous visit. Among the 13 044 eligible participants, 2102 were taking some cholesterol-lowering medication at 1 or more visits, with 68% of those taking statins at some point (39% at visit 2, 59% at visit 3, and 80% at visit 4). There were 1311 incident AF cases identified, and of those occurring in cholesterol medication users (278 events), 68% occurred in those taking statins. After multivariable adjustment, there was no association between time-varying cholesterol medication use across visits and the incidence of AF, with an HR (95% CI) of 0.96 (0.82–1.13), compared with those not taking cholesterol-lowering medications (Table 3). There was also no association with AF risk between those taking statins and in those taking other cholesterol medications: HR (95% CI), 0.91 (0.66–1.25). Models not adjusting for prevalent stroke, heart disease, and CHD, and models additionally adjusting for incident stroke, heart disease, and CHD, produced similar results.

A secondary analysis was performed using propensity scores calculated at visit 2. In a standard Cox model, the propensity score–adjusted HR (95% CI) was 1.09 (0.88–1.34) when comparing lipid-lowering medication use with nonuse. When matched 4:1 (not treated/treated) by propensity score, results were similar: HR (95% CI), 1.03 (0.89–1.20).

The relationship between lipid levels and AF risk has not been extensively addressed. Previous publications reported that low HDLc levels were associated with a 20% to 40% higher risk of AF^7–9; however, residual confounding could have been responsible for that association. In our analysis, an initial association between HDLc and AF risk disappeared after adjustment for a wide selection of potential confounders. Similar results were observed for triglycerides.

The inverse association of LDLc and total cholesterol with AF risk is intriguing, and has been previously seen in an analysis of the Cardiovascular Health Study, which included individuals aged ≥65 years.¹¹ Similar results were found in 2 Japanese populations, in which individuals with hypercholesterolemia had a 25% lower odds of developing AF and an 8% reduction in AF risk with each 10% increase in LDLc.^9,32 Notably, this association was stronger in older individuals, consistent with our results of stronger inverse associations in later ARIC visits. However, no clear mechanism exists to explain this association. Subclinical hyperthyroidism might be potentially responsible, because hyperthyroidism reduces levels of total cholesterol and LDLc,³³ and is associated with increased risk of AF.³⁴ Unfortunately, we do not have information on thyroid profiles in ARIC participants. Blood lipids could also affect the composition of cell membranes, which is a major determinant of cell excitability. In vitro studies have shown that cholesterol modulates the distribution and function of some ion channels potentially involved in the occurrence of AF, such as the Kv1.5 potassium channel.^35,36 Whether this mechanism explains the observed association is hypothetical and requires further research.

Published reports addressing the relationship between statins and AF have provided inconsistent results, and the conclusions tend to be different based on clinical trials (no association) or observational studies (reduced AF risk).²¹ This study provided results consistent with those seen in most randomized clinical trials and indicated no relationship between statin use or any lipid-lowering medication use and the risk of AF. During the years the ARIC visits were conducted (1987–1998), statins were gaining in popularity, and that, combined with the participants getting older, meant most statin users were not taking them until around visit 4 (1996–1998). There have been no study visits since 1998, so we were unable to determine how many of the participants with incident AF were receiving statins or lipid-lowering medications in the years between visit 4 and the year in which their AF occurred.

Some other study limitations should be noted. Asymptomatic AF and AF managed exclusively in an outpatient setting could not be identified, because most of our incident AF cases were ascertained from hospitalization discharge records. However, the incidence rates of AF in the ARIC study are consistent with other population-based studies, and the validity of AF ascertainments using hospitalizations is acceptable.³ In a subanalysis in this study comparing AF ascertained from hospital records with AF from study ECGs, there was an even stronger inverse association seen between LDLc (HR, 0.79 for ECGs and 0.90 for hospital records) and total cholesterol (HR, 0.75 for ECGs and 0.89 for hospital records) and incident AF in the ECG group, thus supporting evidence that AF ascertained from hospitalizations is acceptable. Also, LDLc levels were not associated with the risk of hospitalization, and adjustment for incident hospitalizations before AF incidence or censoring did not change the association between LDLc and AF. There is also the possibility that those with dyslipidemia have more paroxysmal AF that was not captured by our AF ascertainment process. Other limitations include the possible misclassification of exposures, both lipid levels and lipid-lowering medications, because of unmeasured changes between visits. Along with no data on thyroid hormones, which might confound our results, the ARIC study also does not contain information on the dose of statins or other lipid medications. Higher doses have a stronger effect on cholesterol levels and other processes (eg, inflammation in the case of statins) and, therefore, different doses may differentially affect the incidence of AF. Also, we cannot determine the impact of lifestyle changes a patient may incorporate to become healthier after discovering he or she has high cholesterol. Despite these limitations, our study has important strengths, including a large sample size, a long follow-up, an elevated number of AF events, a biracial sample, and an extensive list of measured covariates.

In conclusion, we found that higher levels of LDLc and total cholesterol were associated with a lower incidence of AF independently of other risk factors. HDLc and triglycerides, however, were not independently associated with AF incidence after multivariable adjustment. Future research should replicate these results and study potential mechanisms. In addition, no association was found between lipid-lowering medication, including statins, and the incidence of AF. At present, there are still insufficient data to recommend the use of statins solely for AF prevention.