Heart failure, a strong risk factor for atrial fibrillation (AF), is often accompanied by elevated liver transaminases. The aim of this study was to test the hypothesis that elevated transaminases are associated with the risk for incident AF in the community. A total of 3,744 participants (mean age 65 ± 10 years, 56.8% women) from the Framingham Heart Study Original and Offspring cohorts, free of clinical heart failure, were studied. Cox proportional-hazards models adjusted for standard AF risk factors (age, gender, body mass index, systolic blood pressure, electrocardiographic PR interval, antihypertensive treatment, smoking, diabetes, valvular heart disease, and alcohol consumption) were examined to investigate associations between baseline serum transaminase levels (alanine transaminase and aspartate transaminase) and the incidence of AF over up to 10 years (29,099 person-years) of follow-up. During follow-up, 383 subjects developed AF. The 2 transaminases were significantly associated with greater risk for incident AF (hazard ratio expressed per SD of natural logarithmically transformed biomarker: alanine transaminase hazard ratio 1.19, 95% confidence interval 1.07 to 1.32, p = 0.002; aspartate transaminase hazard ratio 1.12, 95% confidence interval 1.01 to 1.24, p = 0.03). The associations between transaminases and AF remained consistent after the exclusion of participants with moderate to severe alcohol consumption. However, when added to known risk factors for AF, alanine transaminase and aspartate transaminase only subtly improved the prediction of AF. In conclusion, elevated transaminase concentrations are associated with increased AF incidence. The mechanisms by which higher mean transaminase concentrations are associated with incident AF remain to be determined.
Heart failure is 1 of the strongest risk factors for atrial fibrillation (AF) and commonly goes along with an elevation of liver transaminases. We hypothesized that elevations of liver transaminases may also be a marker of subclinical heart failure and thus an indirect marker of AF risk. In addition, the current guidelines for the management of patients with AF of the American College of Cardiology, the American Heart Association, and the European Society of Cardiology recommend the assessment of liver function for the evaluation of patients with initial, incident diagnoses of AF, at least if their heart rates are difficult to control. Because many drugs prescribed in the context of AF management are metabolized hepatically or affect liver function, presumably, the guidelines recommend the assessment of liver function for pharmacodynamic reasons. With the present study, we therefore sought to assess whether transaminases are associated with risk for new-onset AF in a community-based sample.
The Framingham Heart Study was founded in 1948 by enrolling 5,209 subjects, who were followed regularly every 2 years. For the present study, we examined 1,401 participants who attended the 20th examination cycle (1986 to 1990). Starting in 1971, the offspring of the Original cohort and their spouses were enrolled in the Framingham Offspring Study (n = 5,124) and were followed every 4 to 8 years. Offspring participants who attended the 7th examination cycle (1998 to 2001) were eligible (n = 3,539). We combined participants from the Original and the Offspring cohorts and excluded 1,196 participants for the following indications (detailed in Supplemental Table 1 ): age <45 years at examination (because the incidence of AF below this threshold is very low, and to make the age distribution of the 2 cohorts more comparable), did not attend the examination on site, incomplete or missing follow-up, incomplete covariate information, prevalent AF or heart failure, and transaminases >3 times the upper limit of normal (>120 U/L) (suggestive of prevalent liver disease; values ≤120 U/L are considered only mildly elevated ). The Boston University Medical Center Institutional Review Board approved the study protocols, and participants provided written informed consent at each examination.
We studied AF risk over up to 10 years of follow-up from the baseline examinations (cohort 20th [1986 to 1990] and Offspring 7th [1998 to 2001] examination). An AF diagnosis was based on AF or atrial flutter on electrocardiography or medical record information from routinely collected Framingham Heart Study examinations and inpatient or outpatient medical visits. Health history updates during examination visits and between examinations also contained a routine question regarding AF. If any cardiovascular, cancer, or orthopedic diagnosis or AF was indicated, all available medical records to substantiate diagnoses were reviewed. All incident AF electrocardiograms were individually reviewed by ≥2 Framingham cardiologists.
Clinical covariates were routinely ascertained during the Framingham Heart Study examination visits. Physicians performed interviews and physical examinations and collected data on self-reported medication use (e.g., hypertension and statins), smoking status, and alcohol consumption. Participants were considered current smokers if they reported smoking cigarettes during the preceding year. Alcohol consumption was categorized as light or moderate to heavy drinking for men, if they consumed 1 to 14 or >14 drinks per week, respectively, and for women if they consumed 1 to 7 or >7 drinks per week, respectively. Systolic blood pressure was assessed as the average of 2 seated measurements. A systolic murmur of grade ≥3 on a scale of 6, or any diastolic murmur, was considered a clinically significant heart murmur. A dedicated end point committee adjudicated all diagnoses of heart failure on the basis of criteria published elsewhere.
Fasting blood samples were drawn and processed immediately for storage at −70°C during the examination visits. Liver function tests assessed at the baseline examination included alanine transaminase (ALT; previously referred to as alanine aminotransferase or serum glutamic pyruvic transaminase) and aspartate transaminase (AST; previously referred to as aspartate aminotransferase or serum glutamic oxaloacetic transaminase). In the Original cohort, ALT and AST were measured using a Cobas Mira Analyzer using Roche Diagnostics reagents (Roche Diagnostics Corporation, Indianapolis, Indiana). In the Offspring cohort, ALT and AST were determined enzymatically using a Roche Hitachi 911 analyzer (Roche Diagnostics Corporation). For ALT, the intra- and interassay coefficients of variation were 3.8% and 4.4%, respectively. For AST, the intra- and interassay coefficients of variation were 3.1% and 4.5%, respectively. Details on the distribution of ALT and AST are provided in Supplemental Table 2 and Supplemental Figure 1 . The 2 assays had comparable detection ranges. C-reactive protein (CRP) was available only in the Offspring cohort and was determined using the Dade Behring BN 100 high-sensitivity CRP reagent kit (Dade Behring, Deerfield, Illinois). Intra- and interassay coefficients of variation were 3.2% and 5.3%, respectively.
All discrete variables are expressed as frequencies and percentages. Untransformed biomarkers are expressed as medians with 25th and 75th percentiles; all other continuous variables, including natural logarithmically transformed ALT, AST, and CRP, are summarized as mean ± SD. Log-transformed ALT and AST were standardized to a mean of 0 and an SD of 1.
Interaction testing did not suggest effect modification by gender, so Cox models are gender pooled. We used Cox proportional-hazards models to assess the relations between ALT and AST and the incidence of AF. The follow-up time was up to 10 years; participants were followed from their baseline examinations until the development of AF. Censoring occurred at the time of death or at the end of follow-up. All models adjusted for age, gender, and cohort. Further multivariable models additionally adjusted for baseline risk factors included in the Framingham AF risk prediction model: body mass index, systolic blood pressure, electrocardiographic PR interval, antihypertensive treatment, smoking, diabetes, and valvular heart disease. To account for its potential involvement in liver pathology, we also incorporated alcohol consumption into the model. We assessed the assumption of proportional hazards by calculating a supremum test on the basis of the cumulative sums of Martingale-based residuals. Correlation between the clinically related transaminases was assessed using Pearson’s correlation; models including the 2 transaminases were not assessed, because of potential collinearity. Instead, we forced the inclusion of previously established AF risk factors and used a stepwise, automated selection process with p = 0.05 to determine which transaminase would be selected. Effect estimates and confidence intervals for ALT and AST are shown for a 1-SD increase in the log-transformed biomarker values. For graphical presentation, we used cumulative hazard plots and spline plots with 3 knots around the untransformed median biomarker values of 18.0 U/L for ALT and 21.0 U/L for AST.
In secondary analyses, we assessed whether inflammation as measured by CRP might partially mediate the risk associated with elevated transaminases. As sensitivity analyses, we restricted our association analysis to subjects at the baseline examination with transaminases within the reference range (transaminase values ≤40 U/L), and we excluded participants with moderate or heavy alcohol consumption. We also adjusted for the competing risk for death during follow-up. Further analyses additionally adjusted for the interim occurrence of heart failure and myocardial infarction, respectively.
In supplemental analyses, we assessed the ability of the transaminases to improve risk prediction of AF. To ensure complete follow-up for all subjects, we calculated all prediction analyses restricted to 8-year risk for AF. We calculated C-statistics and the difference in the C-statistic between the models with and without the respective transaminase. To assess calibration, we calculated a Hosmer-Lemeshow statistic adapted for survival analyses. In addition, we investigated the integrated discrimination improvement and the relative integrated discrimination improvement for each transaminase. Reclassification of AF cases on the basis of the risk score including transaminases versus the score without the biomarkers was determined using a continuous and a user-defined net reclassification improvement analysis. Eight-year risk categories for net reclassification improvement were selected as 5%, 5% to 10%, and 10%. Confidence intervals for prediction analyses were calculated using bootstrapping and 1,000-times resampling.
Our overall study consisted of 3,744 participants, 875 of whom were derived from the Original cohort and 2,869 from the Offspring cohort. Clinical characteristics and the distributions of biomarkers are provided in Table 1 . The mean follow-up duration was 7.77 ± 2 years (minimum 0.04, maximum 10.00) in 29,099 person-years of observation. During follow-up, 383 participants developed AF ( Table 2 ). The mean age at the time of AF diagnosis was 76 years. At baseline, participants who subsequently developed AF were on average 6.8 years older than those who remained AF free.
|Age (yrs)||65 ± 10|
|Current smokers||454 (12.1%)|
|Light alcohol drinkers||1,769 (47.3%)|
|Moderate to heavy alcohol drinkers||636 (17.0%)|
|Body mass index (kg/m 2 )||27.8 ± 5.2|
|Systolic blood pressure (mm Hg)||132 ± 21|
|Diastolic blood pressure (mm Hg)||75 ± 10|
|Heart rate (beats/min)||66 ± 11|
|PR interval (ms)||168 ± 28|
|Diabetes mellitus||360 (9.6%)|
|Antihypertensive treatment||1,408 (37.6%)|
|Statin medication use||532 (14.2%)|
|Significant cardiac murmur||120 (3.2%)|
|ALT (U/L)||19 (14–25)|
|Log e ALT||3.0 ± 0.4|
|AST (U/L)||21 (18–25)|
|Log e AST||3.1 ± 0.3|
|CRP (mg/L) ∗||2.2 (1.0–5.2)|
|Log e CRP ∗||0.8 ± 1.1|
|n Event||n Total||Person-Years||AF Incidence/1,000 Person-Years|
|≤40 U/L for both markers||348||3,492||27,204||12.8|
|>40 U/L for either marker||35||252||1,894||18.5|
In an age-, gender-, and cohort-adjusted model, ALT was significantly associated with incident AF; AST slightly missed statistical significance. In the multivariable-adjusted model, ALT and AST were significantly associated with incident AF ( Table 3 ). Cumulative hazard curves illustrate the incidence of AF for subjects with baseline transaminase levels within reference limits (≤40 U/L) compared to those with elevated levels (>40 U/L) ( Figure 1 ). Spline plots illustrate the increased risk for AF by increased transaminase values ( Supplemental Figure 2 ). For ALT, a hazard ratio of 2 was reached at untransformed values of about 80 U/L, a hazard ratio of 3 at about 120 U/L, and a hazard ratio of 4 at about 150 U/L. ALT and AST were highly correlated (r = 0.73). In the Offspring cohort, neither of the transaminases was significantly correlated with CRP (ALT r = −0.07, and AST r = 0.05). By stepwise regression, ALT but not AST reached statistical significance for risk for AF once established risk factors were considered.
|HR (95% CI)||p Value||HR (95% CI)||p Value|
|Age and gender||1.21 (1.09–1.34)||<0.001||1.11 (1.00–1.23)||0.05|
|Multivariable||1.19 (1.07–1.32)||0.002||1.12 (1.01–1.24)||0.03|
|Multivariable model with additional adjustments|
|CRP||1.18 (1.04–1.35)||0.01||1.10 (0.98–1.24)||0.12|
|Competing risk for death during follow-up||1.21 (1.08–1.35)||<0.001||1.12 (1.01–1.25)||0.03|
|Interim heart failure||1.19 (1.06–1.32)||0.002||1.12 (1.01–1.24)||0.03|
|Interim myocardial infarction||1.20 (1.07–1.34)||0.002||1.10 (0.99–1.22)||0.08|
|Restricting to transaminases ≤40U/L||1.14 (0.99–1.32)||0.07||1.06 (0.93–1.22)||0.39|
|Excluding moderate to heavy alcohol consumption||1.28 (1.10–1.39)||<0.001||1.18 (1.05–1.32)||0.005|