Hypertension and coronary heart disease are common in aortic stenosis (AS) and may impair prognosis for similar AS severity. Different changes in the electrocardiogram may be reflective of the separate impacts of AS, hypertension, and coronary heart disease, which could lead to enhanced risk stratification in AS. The aim of this study was therefore to examine if combining prognostically relevant electrocardiographic (ECG) findings improves prediction of cardiovascular mortality in asymptomatic AS. All patients with baseline electrocardiograms in the SEAS study were included. The primary end point was cardiovascular death. Backward elimination (p >0.01) identified heart rate, Q waves, and Cornell voltage-duration product as independently associated with cardiovascular death. Multivariate logistic and Cox regression models were used to evaluate if these 3 ECG variables improved prediction of cardiovascular death. In 1,473 patients followed for a mean of 4.3 years (6,362 patient-years of follow-up), 70 cardiovascular deaths (5%) occurred. In multivariate analysis, heart rate (hazard ratio [HR] 1.5 per 11.2 minute −1 [1 SD], 95% confidence interval [CI] 1.2 to 1.8), sum of Q-wave amplitude (HR 1.3 per 2.0 mm [1 SD], 95% CI 1.1 to 1.6), and Cornell voltage-duration product (HR 1.4 per 763 mm × ms [1 SD], 95% CI 1.2 to 1.7) remained independently associated with cardiovascular death. Combining the prognostic information contained in each of the 3 ECG variables improved integrated discrimination for prediction of cardiovascular death by 2.5%, net reclassification by 14.3%, and area under the curve by 0.06 (all p ≤0.04) beyond other important risk factors. ECG findings add incremental predictive information for cardiovascular mortality in asymptomatic patients with AS.
The Simvastatin and Ezetimibe in Aortic Stenosis (SEAS) study demonstrated that individual 12-lead electrocardiographic (ECG) findings, such as ECG left ventricular (LV) hypertrophy with ST-T repolarization abnormalities, are independently associated with poor prognosis in asymptomatic aortic stenosis (AS). On the population level, an “abnormal” ECG finding in asymptomatic patients with AS is likely to reflect a composite of AS and frequently coexisting hypertension and/or coronary heart disease, which may or may not be clinically recognized. We hypothesized that the separate impacts of AS, hypertension, and coronary heart disease would be conveyed through different changes on the electrocardiogram and that combining prognostically relevant ECG variables may therefore further improve risk stratification of cardiovascular (CV) mortality in asymptomatic AS. The aim of this study was therefore to examine if the combined information contained in separate ECG variables, relating independently to CV mortality, added predictive information of CV mortality beyond that obtained from the individual ECG findings during follow-up of asymptomatic patients with mild-to-moderate AS and preserved LV systolic function.
Methods
The SEAS study (NCT00092677) was a multicenter, randomized, double-blind, placebo-controlled study, investigating whether intensive lipid lowering with simvastatin and ezetimibe combination versus placebo could reduce the need for aortic valve replacement (AVR) and risk of CV morbidity and mortality in 1,873 patients, aged 45 to 85 years, with asymptomatic mild-to-moderate AS (defined as echocardiographic aortic valve thickening accompanied by Doppler-measured aortic peak flow velocity ≥2.5 and ≤4.0 m/sec and normal systolic LV function). The primary outcome including study design, organization, clinical measurements, exclusion criteria, and baseline characteristics and the main outcome have been published previously. This study uses post hoc analysis of SEAS data to test the usefulness of electrocardiography to predict CV mortality during follow-up of initially asymptomatic AS.
ECG study protocol, reading procedures, and reproducibility have been published. In brief, all electrocardiograms were read blinded to the randomization and all clinical data at a central ECG core laboratory. All coded ECG findings ( Supplementary Figure 1 ) on the baseline electrocardiogram (1 electrocardiogram per patient) were entered into a Cox regression model that used backward elimination (p >0.01) to identify the variables that were independently associated with subsequent CV mortality (heart rate at rest; Cornell voltage-duration product; and summed Q-wave amplitude in leads I, II, III, aVF, aVR, and aVL; Supplementary Table 1 ). Q waves were not coded in the presence of left bundle branch block (n = 43), which resulted in these patients being excluded from the present analyses. Missing data were due to 100% paced rhythm and lead switch and/or noise in an ECG lead needed for coding. Because of suspected colinearity, an alternative analysis was performed substituting ECG LV hypertrophy with ECG repolarization abnormalities.
Echocardiographic study protocol, reading procedures, and reproducibility have been published. In short, transthoracic echocardiograms were recorded at the local study centers, after which videotapes were read blinded to the randomization and study visit at the SEAS echocardiography core laboratory. Aortic valve area was calculated applying the continuity equation and averaged over 10 consecutive beats in patients with atrial fibrillation. LV dimensions and wall thicknesses were measured on 2-dimensional images following the American Society of Echocardiography guidelines, using an anatomically validated formula. Left atrial volume was measured in LV end-systole and end-diastole by the modified Simpson monoplane method in the apical 4-chamber view and indexed by body surface area. Mitral regurgitation was assessed by color Doppler using previously described 4-point grading scale (grade 0 to 3); grade ≥2 corresponding to a moderate-to-severe mitral regurgitation.
An end point committee blinded to randomized treatment classified end points according to a prespecified end point manual. The primary end point in this substudy was CV death (defined as death from complications of myocardial infarction, progressive heart failure, cerebrovascular disease, complications of cardiac surgery or intervention, or other forms of cardiac or CV diseases, including sudden cardiac death [defined as either witnessed instantaneous unexpected death occurring without any preceding symptoms, nonwitnessed unexpected death, if other causes of death can be excluded with reasonable certainty, or cardiac death occurring <24 hours after the onset of cardiac symptoms]).
Data were analyzed using the Statistical Analytical Software, version 9.2 (SAS, Cary, North Carolina). Continuous data are expressed as mean ± SD and categorical variables as proportions. Of all the coded ECG variables ( Supplementary Figure 1 ), Cox backward elimination (p >0.01 resulted in elimination) identified heart rate; summed Q-wave amplitude in leads I, II, III, aVL, aVR, and aVF; and Cornell voltage-duration product as ECG variables independently associated with CV mortality ( Supplementary Table 1 ). Fit of the model with the 3 ECG variables under study to predict CV mortality was evaluated by Hosmer and Lemeshow goodness-of-fit test (p = 0.40; Supplementary Table 2 ). Generalized linear models using backward elimination (p >0.01 resulted in elimination) of all baseline parameters, displaying significant univariate association with 1 of the 3 ECG variables (p >0.01 resulted in elimination), were used to identify parameters independently associated with each of the investigated ECG variables. The relations to each ECG variable are given as regression slopes (β) for continuous variables and estimated marginal means for discrete variables. Cox regression models were used to examine the event rate ratios for each ECG variable, presented as hazard ratios (HR) with 95% confidence intervals (CI). To allow for relative comparisons, risk was investigated as increase per 1 SD for each ECG variable. A multivariate Cox model was developed by backward elimination (p >0.01 resulted in elimination) entering the 3 ECG variables under study and all other variables differing in patients experiencing versus not experiencing CV death (age, systolic blood pressure, estimated glomerular filtration rate, left atrial volumes, aortic valve area, LV mass, LV stroke volume, LV ejection fraction, and diuretics; Table 1 ). The added discriminatory value for prediction of CV death obtained by adding the ECG variables to age and ejection fraction (non-ECG variables that remained in Cox-based backwards selection of CV death predictors) was evaluated as described by Pencina et al. Because there are no established pretest probabilities for CV death in asymptomatic mild-to-moderate AS, we used annual risks of 1% to 3% for CV death to estimate improvements in the net reclassification index; the groups correspond to that proposed by the American College of Cardiology/American Heart Association guidelines for the management of chronic stable angina. Differences in area under the receiver operating characteristic curve between the models with and without ECG data were compared by the method of DeLong. To assess if the predictive value of ECG findings of CV mortality was different in operated versus nonoperated patients, a test of interaction was performed with respect to time-varying AVR status. To avoid overfitting, a p >0.01 resulted in elimination from the multivariate models. For all other hypothesis testing, a 2-tailed p <0.05 was required for statistical significance.
| Characteristic | All Patients (n = 1,473) | No CV Death (n = 1,403) | CV Death (n = 70) | p-Value |
|---|---|---|---|---|
| Clinical and biochemistry | ||||
| Age (years) | 67.3 ± 9.7 | 67.0 ± 9.6 | 73.4 ± 8.7 | <0.001 |
| Women | 573 (39%) | 546 (39%) | 27 (39%) | 0.95 |
| Body mass index (kg/m 2 ) | 27.0 ± 4.4 | 27.0 ± 4.4 | 26.7 ± 4.5 | 0.58 |
| Systolic BP (mm Hg) | 145.1 ± 20.1 | 144.7 ± 20.0 | 152.8 ± 20.8 | <0.001 |
| Diastolic BP (mm Hg) | 82.1 ± 10.4 | 82.0 ± 10.3 | 84.2 ± 12.1 | 0.13 |
| LDL cholesterol (mg/dl) | 137.3 ± 34.1 | 137.5 ± 34.0 | 133.4 ± 35.6 | 0.33 |
| LDL cholesterol (mmol/l) | 3.6 ± 0.9 | 3.56 ± 0.88 | 3.45 ± 0.92 | — |
| HDL cholesterol (mg/dl) | 58.4 ± 16.8 | 58.4 ± 16.8 | 58.8 ± 16.8 | 0.80 |
| HDL cholesterol (mmol/l) | 1.5 ± 0.4 | 1.51 ± 0.43 | 1.52 ± 0.45 | — |
| eGFR (ml/min/1.73 m 2 ) | 68.2 ± 12.1 | 68.4 ± 12.0 | 64.2 ± 14.8 | 0.03 |
| 12-Lead ECG | ||||
| Baseline atrial fibrillation | 50 (3%) | 45 (3%) | 5 (7%) | 0.09 |
| T-wave inversion V 4–6 | 336 (24%) | 312 (23%) | 24 (36%) | 0.01 |
| Sum T-wave inversion (mm) † | 0 (−3.5 to 0) | 0 (−3.5 to 0) | −1 (−5.5 to 0) | 0.01 |
| ST-depression V 4–6 | 241 (17%) | 224 (16%) | 17 (26%) | 0.04 |
| Sum ST-depression (mm) † | 0 (0–0) | 0 (0–0) | 0 (−1.5 to 0) | 0.04 |
| Cornell product (mm × ms) † | 1600 (1160–2070) | 1575 (1160–2040) | 1710 (1320–2320) | 0.01 |
| S V1 + R V5–6 (mm) | 26.6 ± 9.5 | 26.5 ± 9.4 | 27.5 ± 10.4 | 0.40 |
| Q-wave precordial leads | 351 (24%) | 330 (24%) | 21 (30%) | 0.23 |
| Sum Q-wave precordial (mm) † | 0 (0–0) | 0 (0–0) | 0 (0–2) | 0.18 |
| Q-wave limb leads | 438 (30%) | 413 (29%) | 25 (36%) | 0.26 |
| Sum Q-wave limb (mm) † | 0 (0–2) | 0 (0–2) | 0 (0–3) | 0.09 |
| Heart rate (beats/min) | 65.1 ± 11.2 | 64.9 ± 11.0 | 70.3 ± 12.7 | <0.001 |
| Echocardiography | ||||
| LVIDD (cm) | 5.0 ± 0.6 | 5.03 ± 0.62 | 5.12 ± 0.79 | 0.40 |
| LVIDS (cm) | 3.20 ± 0.57 | 3.19 ± 0.56 | 3.35 ± 0.68 | 0.06 |
| LA S volume/BSA (ml/m 2 ) † | 34.0 (26.2–42.8) | 33.9 (26.0–42.4) | 34.8 (29.5–54.7) | 0.04 |
| LA D volume/BSA (ml/m 2 ) † | 17.0 (11.6–23.8) | 16.8 (11.6–23.7) | 19.6 (13.1–31.3) | 0.02 |
| Peak aortic jet velocity (m/s) | 3.1 ± 0.5 | 3.08 ± 0.54 | 3.19 ± 0.55 | 0.12 |
| AVA/BSA (cm 2 /m 2 ) | 0.6 ± 0.2 | 0.61 ± 0.19 | 0.56 ± 0.19 | 0.04 |
| Mean aortic gradient (mm Hg) | 22.8 ± 8.8 | 22.7 ± 8.8 | 24.9 ± 9.2 | 0.05 |
| Mitral regurgitation (grade ≥2) | 146 (11%) | 135 (10%) | 11 (17%) | 0.08 |
| LV stroke volume (ml) | 75.8 ± 17.4 | 76.1 ± 17.5 | 70.9 ± 15.0 | 0.02 |
| LV mass index (g/m 2 ) | 99.5 ± 30.0 | 98.9 ± 29.3 | 111.1 ± 40.2 | 0.02 |
| LV ejection fraction (%) | 65.8 ± 8.3 | 66.0 ± 8.2 | 63.2 ± 8.9 | 0.01 |
| Medicine | ||||
| Simvastatin/ezetimibe | 733 (50%) | 701 (50%) | 32 (46%) | 0.49 |
| RAS inhibitor | 600 (41%) | 570 (41%) | 30 (43%) | 0.71 |
| Ca 2+ -blocker | 418 (28%) | 394 (28%) | 24 (34%) | 0.26 |
| β-blocker | 738 (50%) | 704 (50%) | 34 (49%) | 0.79 |
| Diuretic | 674 (46%) | 630 (45%) | 44 (63%) | 0.003 |
| Aspirin | 386 (26%) | 365 (26%) | 22 (31%) | 0.32 |
† Values reflect medians with twenty-fifth to seventy-fifth percentiles and the corresponding p-values results from the Wilcoxon test.
Results
Baseline electrocardiograms were available in 1,563 patients (83.4%), among which the ECG abnormalities under study could be assessed in 94% (n = 1,473). Compared with patients without ECG data for this study (n = 400), subjects with ECG data did not differ in peak aortic jet velocity (p = 0.95) but were likely to be younger subjects (mean difference 1.2 years) with slightly higher ejection fraction (mean difference 2%, both p ≤0.04). The 1,473 patients with ECG data for this study included 900 men (61%) and 573 women (39%), followed for a mean of 4.3 years (6,362 patient-years of follow-up). Among all screened ECG variables ( Supplementary Figure 1 ) displaying univariate association with CV mortality, the ECG variables that were independently associated with CV mortality, at the 99% confidence level, were heart rate; sum of Q-wave amplitudes in leads I, II, III, aVL, aVR, and aVF; and the Cornell voltage-duration product ( Supplementary Table 1 ).
Baseline characteristics independently associated with each of the 3 ECG multivariate predictors of CV death, at the 99% confidence level, are listed in Table 2 . There was little or no colinearity in pairwise comparisons of the 3 ECG variables under study (all Pearson correlation coefficients ≤0.08). During the course of the study, 139 participants (9%) died, including 70 CV deaths (5%) and 26 categorized as sudden cardiac death. Unadjusted heart rate (HR 1.5 per 11.2 minute −1 [1 SD], 95% CI 1.2 to 1.8), summed Q-wave amplitude (HR 1.2 per 2.0 mm [1 SD], 95% CI 1.0 to 1.4), and Cornell voltage-duration product (HR 1.4 per 763 mm × ms [1 SD], 95% CI 1.2 to 1.7, all p ≤0.01) were associated with clear excess in the risk of CV death ( Supplementary Table 1 ). Adjusted by each other in addition to age and LV ejection fraction (non-ECG variables remaining in backward elimination with a p >0.01 resulting in removal from the model), heart rate (HR 1.5 per 11.2 minute −1 [1 SD], 95% CI 1.2 to 1.8), summed Q-wave amplitude (HR 1.3 per 2.0 mm [1 SD], 95% CI 1.1 to 1.6), and Cornell voltage-duration product (HR 1.4 per 763 mm × ms [1 SD], 95% CI 1.2 to 1.7, all p <0.001) remained significantly associated with CV mortality. The combined prognostic information contained in each of the 3 ECG variables improved prediction of CV mortality beyond the individual ECG findings ( Figure 1 ). Adding ECG strain to heart rate and summed Q-wave amplitude in lieu of the Cornell product provided a nearly identical amount of predictive information (overall model chi-square 35.8 vs 37.9, respectively). However, when adjusting by non-ECG variables that were also associated with CV mortality, ECG strain was, unlike the Cornell product, removed by backward elimination. In a model with CV death as the outcome, the addition of ECG data to other important baseline covariates significantly increased area under the curve ( Figure 2 ), improved integrated discrimination by 2.5%, and the overall net reclassification index by 14.3% (all p ≤0.04; Table 3 ). Of note, the net reclassification index improvement was largely due to reclassification of subjects with a priori intermediate risk of CV death ( Supplementary Table 3 ). There was no detectable risk dependency between the ECG findings and time-varying AVR on risk of CV mortality (all p ≥0.09 for interaction). The latter may be reflective of the fact that the ECG variables under study were not included in the clinical evaluation of the need for AVR. This hypothesis was supported in 2 separate explorative analyses. First, when restricting the multivariate analyses to the 52 CV deaths before AVR (74%), heart rate (HR 1.5 per 11.2 minute −1 [1 SD], 95% CI 1.2 to 1.9), summed Q-wave amplitude (HR 1.4 per 2.0 mm [1 SD], 95% CI 1.2 to 1.7), and Cornell voltage-duration product (HR 1.5 per 763 mm × ms [1 SD], 95% CI 1.2 to 1.8, all p ≤0.001) remained highly associated with CV death. Second, in univariate comparisons of the 18 CV deaths occurring after AVR, the heart rate was the only ECG variable displaying an association with risk of postoperative CV death (HR 1.6 per 11.2 minute −1 [1 SD], 95% CI 1.1 to 2.4, p = 0.01).
