Atrial fibrillation (AF) is likely secondary to multiple different pathophysiological mechanisms that are increasingly but incompletely understood. Motivated by the hypothesis that 3 previously described electrocardiographic predictors of AF identify distinct AF mechanisms, we sought to determine if these electrocardiographic findings independently predict incident disease. Among Cardiovascular Health Study participants without prevalent AF, we determined whether left anterior fascicular block (LAFB), a prolonged QT C , and atrial premature complexes (APCs) each predicted AF after adjusting for each other. We then calculated the attributable risk in the exposed for each electrocardiographic marker. LAFB and QT C intervals were assessed on baseline 12-lead electrocardiogram (n = 4,696). APC count was determined using 24-hour Holter recordings obtained in a random subsample (n = 1,234). After adjusting for potential confounders and each electrocardiographic marker, LAFB (hazard ratio [HR] 2.1, 95% confidence interval [CI] 1.1 to 3.9, p = 0.023), a prolonged QT C (HR 2.5, 95% CI 1.4 to 4.3, p = 0.002), and every doubling of APC count (HR 1.2, 95% CI 1.1 to 1.3, p <0.001) each remained independently predictive of incident AF. The attributable risk of AF in the exposed was 35% (95% CI 13% to 52%) for LAFB, 25% (95% CI 0.6% to 44%) for a prolonged QT C , and 34% (95% CI 26% to 42%) for APCs. In conclusion, in a community-based cohort, 3 previously established electrocardiogram-derived AF predictors were each independently associated with incident AF, suggesting that they may represent distinct mechanisms underlying the disease.
Although atrial fibrillation (AF) is the most common arrhythmia, there is no known means to prevent it. AF is generally considered “one” disease; however, it may actually represent the final pathway of multiple different pathophysiological mechanisms. An understanding of these mechanisms and the identification of accessible clinical predictors of AF may be the key to developing more custom-built prevention and therapeutic strategies. The electrocardiogram (ECG) is a noninvasive, readily available test. Our group has identified 3 electrocardiographic predictors of AF that may reflect distinct mechanistic phenotypes: left anterior fascicular block (LAFB), a prolonged QT interval, and atrial premature complexes (APCs). Based on theoretical reasoning found in the literature and the underlying biologic processes, these electrocardiographic markers may represent distinct underlying mechanisms of AF: LAFB may represent atrial fibrosis, the QT interval may be a marker of cardiomyocyte refractoriness, and APCs may represent triggers for AF. However, whether these different predictors overlap or identify distinct mechanistic phenotypes has not been investigated. Therefore, we sought to determine if there is overlap between these electrocardiographic predictors and whether they independently predicted AF risk.
Methods
The Cardiovascular Health Study is a prospective cohort study established in 1989 that enrolled adults aged ≥65 years. Detailed methods have been previously published. Briefly, 5,201 subjects were recruited from 4 US communities and, beginning in 1992, an additional 687 African-American participants were recruited. All participants underwent a comprehensive baseline examination, including a thorough medical history, physical examination, and 12-lead ECG at rest. A random subset of 1,429 participants (the “Holter cohort”) was assigned to 24-hour ambulatory ECG (Holter) monitoring at baseline. Participants were followed semiannually with alternating telephone calls and clinic visits for 10 years, after which semiannual telephone contact was continued. Study participants provided written informed consent, and the study protocol was approved by the institutional review board at each center.
Baseline demographics and medical conditions were ascertained by participant report and validated by components of the baseline examination, physician report, and the medical record (see Supplementary Table 1 ). During follow-up through June 30, 2008, incident AF was determined from clinic visit ECGs at rest, hospital discharge diagnosis codes, and inpatient Medicare claims data.
Baseline and annual 12-lead ECGs at rest were recorded using MAC PC ECG Machines (Marquette Electronics, Milwaukee, Wisconsin) and processed automatically after visual inspection for technical errors and quality. Baseline Holter data were analyzed at the Washington University School of Medicine Heart Rate Variability Laboratory using a MARS 8000 Holter scanner (GE Healthcare, Buckinghamshire, United Kingdom) and manually reviewed for accuracy. Electrocardiographic variables were defined in the same manner as previous publications establishing a relation with AF ( Supplementary Table 2 ). For the QT interval analyses, we excluded participants with QT intervals >600 or <200 ms, QRS duration ≥120 ms, left ventricular hypertrophy, ventricular preexcitation, Vaughan-Williams class I or III antiarrhythmic drug use, or ventricular pacing at baseline. We corrected the QT interval using the Framingham, Hodge, Fridericia, and Bazett formulas and used the Framingham formula for the analyses in the full cohort. In the Holter cohort, more participants met the definition for prolonged QT using Hodge’s formula compared with the other correction formulas and only QT corrected by Hodge-predicted AF ( Supplementary Table 3 ). Because the purpose of the present study was to examine distinct associations between established electrocardiographic predictors and AF (a prolonged QT interval has already been established as a predictor in 4 cohorts), we used QT corrected by the Hodge formula as the primary predictor in the smaller Holter cohort. Our APC analyses were in the Holter cohort only and excluded participants with poor-quality Holter data, atrial pacing, or wandering atrial pacemaker.
We excluded participants with prevalent AF. Normally distributed continuous variables were compared using t tests and are presented as means ± SD. Non-normally distributed continuous variables were compared using the Wilcoxon rank-sum test and are presented as medians and interquartile ranges (IQRs). Categorical variables were compared using chi-square and Fisher’s exact tests.
The relation between electrocardiographic markers was analyzed using logistic regression before and after adjusting for potential confounders and are reported as odds ratios with 95% confidence intervals (CIs). The relation between each electrocardiographic predictor and incident AF was analyzed using multivariate Cox proportional hazards models: model 1 included 1 electrocardiographic marker adjusted for potential confounders; model 2 included the addition of 1 other electrocardiographic predictor, and model 3 added both other electrocardiographic predictors. Interaction testing between each potential pair of electrocardiographic predictors as they related to the outcome of incident AF was conducted; the results of statistically significant interactions are reported and included in relevant multivariable models. Covariates were determined a priori based on biologic plausibility and included age, race, gender, body mass index, hypertension, diabetes, coronary heart disease, myocardial infarction, congestive heart failure, and study center. Because APC counts were skewed, we used log base 2 transformation in regression and Cox models to meet model linearity assumptions after adding 0.01 to the counts to retain participants with 0 counts.
The attributable risk in the exposed at 15 years of follow-up for each of the 3 predictors was analyzed in the Holter cohort using a counterfactual approach : for each participant with an exposure of interest, we estimated the fitted AF risk at 15 years under the Cox model, accounting for all observed risk factors, and a counterfactual fitted risk with exposure reset to a “safe” reference level. As reference levels, we used absence of LAFB and prolonged QT C and the lower quartile of APCs within the cohort. The attributable risk was then calculated as the average excess divided by the average observed risk. This ratio is interpretable as the proportion of disease among the exposed that is attributable to the exposure. Bootstrap resampling with 500 repetitions was used to obtain 95% CIs.
Data analysis was completed using Stata 14 (StataCorp, College Station, Texas). We considered a 2-tailed p value <0.05 statistically significant.
Results
After exclusion criteria were applied, data from 4,696 participants were available for the LAFB and QT C comparison ( Table 1 ). At baseline, 4 participants (0.09%) had both LAFB and a prolonged QT C . No relation was found between LAFB and a prolonged QT C (corrected by Framingham; see Methods ) both before and after multivariable adjustment ( Figure 1 ). This was consistent using the other QT interval correction formulas.
| Characteristic | Full Cohort | Holter Cohort | |||||||
|---|---|---|---|---|---|---|---|---|---|
| LAFB Status (n = 4,696) | P Value † | QT C Status (n = 4,696) ∗ | P value † | APC Status (n = 1,234) | P value † | ||||
| No LAFB (n = 4,579) | LAFB (n = 117) | No Prolonged QT C (n = 4,557) | Prolonged QT C (n = 139) | ≤Median Percent APCs (n = 634) | >Median Percent APCs (n = 600) | ||||
| Median age (years) (IQR) | 71 (68-76) | 73 (69-79) | 0.001 | 71 (68-76) | 73 (69-77) | 0.022 | 70 (67-73) | 72 (69-76) | <0.001 |
| Men | 1,829 (40%) | 87 (74%) | <0.001 | 1,845 (40%) | 71 (51%) | 0.012 | 257 (41%) | 293 (49%) | 0.003 |
| White | 3,855 (84%) | 100 (85%) | 0.71 | 3,843 (84%) | 112 (81%) | 0.23 | 603 (95%) | 572 (95%) | 0.85 |
| Mean body mass index (kg/m 2 ) | 27 ± 4.7 | 27 ± 5.0 | 0.78 | 27 ± 4.8 | 28 ± 4.6 | 0.016 | 27 ± 4.2 | 26 ± 4 | <0.001 |
| Hypertension | 1,933 (42%) | 37 (32%) | 0.021 | 1,899 (42%) | 71 (51%) | 0.028 | 280 (44%) | 220 (37%) | 0.007 |
| Diabetes mellitus | 655 (14%) | 29 (25%) | 0.002 | 660 (15%) | 24 (17%) | 0.39 | 100 (16%) | 78 (13%) | 0.17 |
| Heart failure | 116 (2.5%) | 6 (5.1%) | 0.08 | 110 (2.4%) | 12 (8.6%) | <0.001 | 14 (2.2%) | 17 (2.8%) | 0.48 |
| Coronary disease | 757 (17%) | 25 (21%) | 0.17 | 747 (16%) | 35 (25%) | 0.006 | 117 (18%) | 122 (20%) | 0.40 |
| Myocardial infarction | 340 (7.4%) | 17 (14.5%) | 0.004 | 341 (7.5%) | 16 (12%) | 0.078 | 54 (8.5%) | 74 (12%) | 0.028 |
| Incident atrial fibrillation | 1,211 (26%) | 39 (33%) | 0.10 | 1,199 (26%) | 51 (37%) | 0.006 | 114 (18%) | 221 (37%) | <0.001 |
| Left anterior fascicular block | – | – | 113 (2.5%) | 4 (2.9%) | 0.77 | 5 (.79%) | 21 (3.5%) | 0.001 | |
| Prolonged QT C | 135 (3%) | 4 (3%) | 0.77 | – | – | 21 (3.8%) ‡ | 36 (7.1%) | 0.016 | |
| Median atrial premature complex count, beats/h § (IQR) | 2.5 (0.8-9.4) | 5.3 (3.0-15) | 0.018 | 2.4 (.75-8.4) | 3.9 (1.8-12) | 0.007 | – | – | |
∗ QT corrected by Framingham formula.
† For the comparison of the indicated characteristic in participants with and without the specified ECG marker.
‡ QT corrected by Hodge formula. Below or equal to the median of percent APCs (n = 554). Above the median of percent APCs (n = 505).
Stay updated, free articles. Join our Telegram channel
Full access? Get Clinical Tree
