Association of Electrocardiographic and Imaging Surrogates of Left Ventricular Hypertrophy with Incident Atrial Fibrillation: The Multi-Ethnic Study of Atherosclerosis

Abstract

Objectives

To examine the association between LVH, defined by cardiovascular magnetic resonance (CMR) and electrocardiography (ECG), with incident AF.

Background

Previous studies of the association between atrial fibrillation (AF) and left ventricular hypertrophy (LVH) were based primarily on echocardiographic measures of LVH.

Methods

The Multi-Ethnic Study of Atherosclerosis (MESA) study enrolled 4942 participants free of clinically recognized cardiovascular disease. Incident AF was based on MESA ascertained hospital discharge ICD codes and Centers for Medicare and Medicaid Services (CMS) inpatient hospital claims. CMR-LVH was defined as left ventricular mass ≥ 95th percentile of the MESA population distribution. Eleven ECG-LVH criteria were assessed. The association of LVH with incident AF was evaluated using multivariable Cox proportional hazards models adjusted for CVD risk factors.

Results

During a median follow-up of 6.9 years, 214 incident AF events were documented. Participants with AF were more likely to be older, hypertensive, and overweight. The risk of AF was greater in participants with CMR-derived LVH [Hazard ratio (HR) 2.04, 95% CI 1.15-3.62]. AF was associated with ECG-derived LVH measure of Sokolow-Lyon voltage product after adjusting for CMR-LVH [HR=1.83 (1.06, 3.14), p= 0.02]. The associations with AF for CMR LVH and Sokolow-Lyon voltage product were attenuated when adjusted for CMR LA volumes.

Conclusion

In a multi-ethnic cohort of participants without clinically detected CVD, both CMR and ECG-derived LVH were associated with incident AF. ECG-LVH showed prognostic significance independent of CMR-LVH. The association was attenuated when adjusted for CMR LA volumes.

Introduction

Atrial fibrillation (AF) is the most common chronic dysrhythmia in the United States, affecting over 2 million people, and is associated with heart failure (HF), cardiovascular mortality, stroke and total mortality (3,4). Participants with AF are five times more likely to suffer from stroke and have a 1.5-1.9 fold increase in mortality (2,4,5). Due to the advancing age of the population, and improved survival from cardiovascular events and cardiac surgery, the burden of AF will likely increase. Importantly, up to 1 of 6 individuals over 40 will develop AF in the absence of HF or myocardial infarction (MI) (5). Known risk factors associated with the development of AF include advanced age, hypertension, diabetes, myocardial infarction, CHF, and valvular heart disease (2,3,5). Analysis in the Niigata Preventive Medicine Study showed electrocardiographic (ECG) left ventricular hypertrophy (defined by Minnesota code 3.1/3.3), ST-T abnormalities with left ventricular hypertrophy, and premature complexes are also associated with increased risk for AF (6).

A number of studies have evaluated the predictive ability of echocardiographic measurements as risk factors for the development of AF. Such predictive measures include left atrial enlargement, increased ventricular wall thickness and decreased left ventricular fractional shortening (2,7,8,9). Cardiovascular magnetic resonance (CMR) provides a more accurate assessment of myocardial size compared to echocardiography (10,11,13), but the association of CMR findings with incident risk of AF has not been explored. We also sought to define the association of baseline ECG defined left ventricular hypertrophy with future development of AF, and the extent to which these associations are mediated by CMR confirmed hypertrophy.

Methods

Study sample

The Multi-Ethnic Study of Atherosclerosis (MESA) is a prospective, longitudinal study initiated in July 2000, in six US centers, to evaluate the presence and progression of subclinical cardiovascular disease. The study objectives and design have been previously reported (17). The MESA study includes 6814 participants 45-84 years of age without clinically recognized cardiovascular disease (stroke, myocardial infarction or coronary heart disease) and with no history of AF at enrollment. A total of 4942 participants underwent ECG and CMR examinations at baseline during 2000-2002 and are included in the analysis. Incident AF events were based on MESA ascertained hospital discharge ICD-9 codes (427.31) and Centers for Medicare and Medicaid Services (CMS) inpatient hospital claims. AF events that occurred during a hospital stay with coronary artery bypass surgery or valve replacement surgery were not counted as incident events.

Cardiovascular Magnetic Resonance

The MESA CMR protocol, image analysis, and inter- and intra reader reproducibility have been previously reported (14). Briefly, base to apex short-axis fast gradient echo images (slice thickness 6 mm, slice gap 4 mm, field of view 360-400 mm, matrix 256×160, flip angle 20, echo time 3-5 msec, repetition time 8-10 msec) were acquired using 1.5 Tesla CMR scanners (14). Left ventricular mass (LVM) was measured as the sum of the myocardial area (the difference between endocardial and epicardial contours) times slice thickness plus image gap in the end-diastolic phase multiplied by the specific gravity of the myocardium (1.05 g/mL) (15).The reproducibility of this protocol was assessed on 79 participants with a technical measurement error of 6% and an intra-class correlation coefficient of 0.98. The threshold for CMR LVH was set at greater than the 95^th percentile of the MESA population.

The original MESA CMR protocol did not measure left atrium (LA) size. Using the software cmr42 (Circle Cardiovascular Imaging, Cardiac MRI Software Version 4.1; Alberta, Canada), the baseline LA volume of all participants with atrial fibrillation and interpretable CMR images along with a 1:1 matched (age, sex and race) population were measured. Measurements were obtained at the end of atrial diastole (just prior to the opening of the mitral valve) in the long axis two and four chamber cine views. The software then calculated a final biplane measurement, which was used in the analysis.

Electrocardiography

LVH by ECG was assessed using eleven different criteria (Table 1). Left ventricular mass was estimated from the ECG based on the model by Rautaharju et al, which adjusts for weight, race, and sex based upon ECG and echocardiographic LVH associations in the multicenter Cardiovascular Health Study Cohort (16).

Table 1. ECG Criteria.

	Criteria
Sokolow-Lyon voltage18	SV1 + RV5/V6 ≥3.5 mV and/or RaVL ≥1.1 mV
Sex-specific Cornell voltage19	SV3 + RaVL >2.8 mV [for men] and >2.0 mV [for women]
Romhilt-Estes point score20	Diagnostic ≥5 points and probable ≥4 points
	Criteria	Points
	Voltage Criteria
	- R or S in limb leads ≥ 20mm	3
	- S in V1 or V2 ≥ 30mm	3
	- R in V5 or V6 ≥ 30mm	3
	ST-T Abnormality -ST-T vector opposite to QRS w/o digitalis	3
	-ST-T vector opposite to QRS w/digitalis	1
	Negative terminal P mode in V1 1 mm in depth and 0.04 sec in duration	3
	Left axis deviation	2
	QRS duration ≥0.09 sec	1
	Delayed intrinsicoid deflection in V5 or V6 (>0.05 sec)	1
Perugia score21	Positivity of at least 1 of the following 3 criteria: SV3 + RaVL >2.4 mV [men] or >2.0 mV [women], left ventricular strain, or Romhilt-Estes score of ≥5
Perugia 2 Score38	Positivity of at least 1 of the following 2 criteria: SV3 + RaVL >2.4 mV [men] or >2.0 mV [women], or left ventricular strain
Minnesota code 3.122	RV5/V6 >2.6 mV or RI/II/III/aVF >2 mV or RaVL >1.2 mV
Lewis index23	([RI + SIII] − [RIII + SI] >1.7 mV
Framingham-adjusted Cornell voltage24	men: [RaVL + SV3 + 0.0174*{age − 49} + 0.191*{body mass index (BMI) - 26.5}] ≥2.8 mV; women: [RaVL + SV3 + 0.0387*{age − 50} + 0.212* {BMI − 24.9}] ≥2.0 mV
Cornell voltage product25	([RaVL + SV3] *QRS duration ≥243,600 μVms
Sokolow- Lyon voltage product25	[SV1 + RV5/RV6]*QRS duration ≥371,000 μVms
Gubner and Ungerleider voltage26	RI + SIII ≥2.2 mV

Statistical analyses

Continuous data are presented as a mean ± SD. Categorical data are presented as frequency. The baseline characteristics and CMR and ECG derived variables were compared among participants with and without incident AF using the Chi Squared test and Student's t-test where appropriate. Univariable and multivariable Cox proportional hazards models were used to determine association with AF. Results are presented as hazard ratios (HR) with 95% confidence intervals. The multivariable models adjusted for cardiovascular risk factors (age, sex, race, BMI, cigarette smoking status, systolic blood pressure, diabetes, total cholesterol, high-density lipoprotein cholesterol, and use of digitalis, anti-arrhythmic, anti-hypertensive and lipid medications,) to examine the association of LVH as defined by CMR and ECG with incident AF. Each of the eleven ECG criteria for LVH was independently assessed for their association with AF. When appropriate the CMR-LVH group was compared to the CMR group with LVM <=50^th percentile.

We also tested for two-way interactions of LVH (by CMR and ECG) with sex and ethnicity. Finally, since the original MESA MRI measurement protocols did not measure LA volume, we performed a nested case control study to assess the potential mediating effect of LA volume for the relationship of CMR-LVH with incident AF. We measured LA volume in all incident cases of AF and in age, gender and ethnicity matched cases and controls. LA volume assessment was done blinded to case-control status. We then used Cox proportional hazards models with shared frailty (by matching variable) with time to incident AF as outcome and CMR-LVH (defined as 95^th percentile of the MESA cohort), systolic blood pressure, and use of antihypertensive medications as independent variables, followed by the addition of LA volume to examine the role of LA volume as a mediator of the association between CMR-LVH and incident AF. Statistical analyses were performed using STATA statistical software (Version 9.0, College Station, TX). A p-value <0.05 was considered significantly significant.

Results

The total number of MESA participants with CMR-LVM and ECG measures was 4942. There were 214 incident AF events documented during a median follow-up of 2533 days (6.9 years). Participants with AF were more likely to be older, Caucasian, male, taller, overweight, have underlying systolic hypertension, have a history of smoking and have slightly lower total cholesterol (Table 2). There were no differences in the prevalence of diabetes or HDL levels among participants with or without AF.

Table 2. Baseline characteristics of the study population (n=4942).

Variable	No Afib (n=4728)	Afib (n=214)	P-value
Age, years	61.0 ± 10.0	70.0 ± 7.7	<0.001
Men, n (%)	2223(47.02)	130(60.8)	<0.001
Ethnicity, n (%)			<0.001
Caucasians	1808 (38.2)	118 (54.1)
Chinese	633 (13.4)	15 (7.0)
African American	1229 (26.0)	42 (19.6)
Hispanics	1058 (22.4)	39 (18.2)
Height, cm	166.3 ± 9.9	168.6 ± 10.4	0.0008
Weight, kgs	76.9 ± 16.2	80.1 ± 16.5	0.005
Cigarette smoking status, n (%)			0.041
Never	2441 (51.8)	96 (45.3)
Former	1668 (35.4)	93 (43.9)
Current	607 (12.9)	23 (10.9)
Systolic blood pressure, mm Hg	124.9 ± 21.1	134.8 ± 23.4	<0.001
Diabetes, n (%)	538 (11.4)	27 (12.6)	0.59
Total cholesterol, mg/dl	194.6 ± 35.5	188.8 ± 32.5	0.02
HDL cholesterol, mg/dl	51.2 ± 15.0	50.0 ± 14.2	0.22
Hypertension medication, n (%)	1612 (34.1)	122 (57.3)	<0.001
Lipid-lowering medication, n (%)	742 (15.7)	40 (18.8)	0.22
Any antiarrhythmic drug, n (%)	20 (0.4)	4 (1.8)	0.003
Digitalis, n (%)	6 (0.13)	4 (1.9)	<0.001

Participants with incident AF had significantly higher prevalence of LVH at baseline by six of eleven ECG criteria (Sokolow-Lyon, Sokolow-Lyon Voltage product, Cornell Voltage product, Perugia score, Perugia 2 score, Romhilt-Estes score) and CMR-LVH (table 3). The risk of incident AF was higher in participants with ECG-LVM greater than 95^th percentile compared to those with LVM <50^th percentile (2.7 (1.7, 4.1), p=<0.001), but this association was attenuated and lost its significance after adjustment for traditional cardiovascular risk factors (Table 4). Eleven ECG-LVH criteria were analyzed for their association with incident AF in both the unadjusted and adjusted models (Table 5). Six of the eleven models had a significant association in the unadjusted models; Sokolow-Lyon voltage, Sokolow-Lyon voltage product, Cornell voltage product, Romhilt-Estes score Perugia score and Perugia 2 score. After adjustment for cardiovascular risk factors only three of eleven ECG-LVH criteria had significant associations with AF; Sokolow-Lyon voltage, Sokolow-Lyon voltage product and Perugia score (HR 1.57 (1.06, 2.32), p= 0.02; HR= 2.24 (1.33, 3.76), p= 0.002; HR 1.71 (1.09, 2.81), p=0.03, respectively) (table 5). Further analysis adjusting for CMR-LVH showed Sokolow-Lyon voltage product retained significant associations with AF (HR=1.83 (1.06, 3.14), p= 0.02). Sokolow-Lyon voltage and Perugia score did not retain significance after adjusting for CMR-LVH. There were no multiplicative interactions with gender (P=0.504) or race (P=0.533).

Table 3. Baseline CMR and ECG-derived variables of the study population (n=4942).

Variable, n (%)	No Afib (n=4728)	Afib (n=214)	P-value
CMR-LVH*	217 (4.6)	28 (13.1)	<0.001
Sokolow-Lyon voltage	398 (8.5)	33 (15.7)	<0.001
Cornell voltage	170 (3.6)	11 (5.2)	0.22
Framingham-adjusted Cornell voltage	171 (3.6)	11 (5.2)	0.23
Minnesota code 3.1	247 (5.3)	18 (8.6)	0.04
Lewis index	566 (12.1)	26 (12.4)	0.88
Gubner and Ungerleider	278 (5.9)	18 (8.6)	0.115
Sokolow-Lyon voltage product	155 (3.3)	17 (8.1)	<0.001
Cornell voltage product	276 (5.9)	23 (11.0)	0.003
Romhilt-Estes score ≥ 4	58 (1.3)	6 (3.1)	0.03
Perugia score	229 (5.3)	19 (10.0)	0.004
Perugia 2 score	215 (4.8)	16 (8.4)	0.026

Table 4. Hazard ratios (HR) and 95% Confidence Intervals (CI) for Incident AF by LV mass.

Measure	HR (95% CI)
	Model 1* Unadjusted	P-value	Model 2† Multivariable-adjusted*	P-value
CMR-derived
LV mass†‡	1.50 (1.34, 1.687)	<0.001	1.45 (1.23, 1.70)	<0.001
LV mass (intervals)
≤50th percentile	1 (ref.)		1 (ref.)
50th - 90th percentile	1.33 (0.84, 2.10)	0.21	1.23 (0.76, 2.00)	0.39
90th - 95th percentile	1.55 (0.99, 2.43)	0.05	1.29 (0.76, 2.20)	0.33
>95th percentile	2.77 (1.84, 4.16)	<0.001	2.04 (1.15, 3.62)	0.015
ECG-derived‡
LV mass†	1.33 (1.17, 1.52)	<0.001	1.01 (0.75, 1.36)	0.11
LV mass (intervals)
≤50th percentile	1 (ref.)		1 (ref.)
50th - 90th percentile	1.87 (1.18, 2.95)	0.007	1.39 (0.84, 2.31)	0.19
90th - 95th percentile	1.93 (1.22, 3.05)	0.005	0.99 (0.54, 1.83)	0.99
>95th percentile	2.67 (1.73, 4.12)	<0.001	1.31 (0.61, 2.81)	0.47

^*Adjusted for cardiovascular risk factors (age, sex, ethnicity, weight, height, systolic BP, diabetes, total and HDL cholesterol, smoking, and hypertension/lipid/arrhythmia/digitalis medication

^†Standardized (centered at 0 and scaled to SD units)

^‡ECG-LV mass derived from Rautaharju models

Table 5. Hazard ratios (HR) and 95% Confidence Intervals (CI) for Incident AF by ECG-LVH criteria.

ECG measure	HR (95% CI)
	Model 1*	P-value	Model 2†	P-value	Model 3‡	P-Value
Voltage-only criteria
Sokolow-Lyon voltage	1.97 (1.36, 2.80)	<0.001	1.57 (1.06, 2.32)	0.02	1.37 (0.92, 2.07)	0.12
Cornell voltage	1.49 (0.81, 2.73)	0.19	1.36 (0.72, 2.58)	0.33	-	-
Framingham adjusted Cornell voltage	1.48 (0.80, 2.72)	0.20	1.36 (0.76, 2.58)	0.33	-	-
Minnesota code 3.1	1.65 (1.01, 2.69)	0.04	1.26 (0.76, 2.08)	0.35	-	-
Lewis index	1.02 (0.68, 1.54)	0.89	0.72 (0.47, 1.11)	0.14	-	-
Gubner and Ungerleider	1.45 (0.89, 2.36)	0.12	1.02 (0.62, 1.68)	0.91	-	-
Voltage-duration product
Sokolow-Lyon voltage product	2.56 (1.56, 4.21)	<0.001	2.24 (1.33, 3.76)	0.002	1.83 (1.06, 3.14)	0.02
Cornell voltage product	1.97(1.28, 3.04)	0.002	1.69 (0.94, 2.31)	0.09	-	-
Composite criteria
Romhilt-Estes score ≥ 4	2.45 (1.08, 5.53)	0.03	1.48 (0.64, 3.39)	0.94	-	-
Perugia score	2.07 (1.28, 3.32)	0.003	1.71 (1.09, 2.81)	0.03	1.35(0.7 9, 2.28)	0.26
Perugia 2 score	1.83 (1.10, 3.06)	0.020	1.38 (0.80, 2.38)	0.25	-	-

^*Unadjusted

^†Adjusted for cardiovascular risk factors (age, sex, ethnicity, weight, height, systolic BP, diabetes, total and HDL cholesterol, smoking, and hypertension/lipid/arrhythmia/digitalis medication

^‡Adjusted for cardiovascular risk factors (age, sex, ethnicity, weight, height, systolic BP, diabetes, total and HDL cholesterol, smoking, hypertension/lipid/arrhythmia/digitalis medication and CMR-LVH)

The risk of incident AF increased with CMR LVM in both the unadjusted and adjusted models (Table 4). In the unadjusted and adjusted models, participants with LVM ≥ 95^th percentile were more likely to have incident AF (HR = 2.77 (1.84, 4.16), p = <0.001; and HR = 2.04 (1.15, 3.62), p= 0.01, respectively) compared to those with LVM <50^th percentile.

A total of 206 of 214 participants with AF had an interpretable CMR LA volume at baseline. The average baseline LA volume for participants with AF was 65.63 mL and the average LA volume for the matched control group was 56.77 mL. When measurements of LA volume were incorporated into the conditional (shared frailty) Cox proportional hazards model the association of CMR-LVH (adjusted for hypertension medications and systolic blood pressure) with incident AF was attenuated but statistical significance was preserved (HR 2.17, 95% CI (1.42-3.31), p<0.001 to HR 1.67, 95% CI (1.07-2.60), p=0.024).

Discussion

The main finding of this study is that CMR-defined LVH in the MESA population is associated with development of incident AF. Participants with a LVM =>95^th percentile were two times more likely to develop AF in this population. In addition, we found that LVH defined by certain ECG criteria can also be predictive of AF, and the Sokolow-Lyon voltage product ECG criteria retained association with incident AF independent of CMR-LVH.

CMR

The association of CMR defined LVH with AF risk is consistent with prior echocardiographic studies. Previous echocardiographic studies evaluating LVH with the development of AF, showed a HR of 1.28 (95% CI, 1.03-1.6) with each 4 mm incremental increase in septal or posterior left ventricular wall thickness (7). To the best of our knowledge, no previous studies have evaluated the association of CMR LVH with the development of AF. CMR has been shown to be superior to echocardiography in its multi-planar capabilities, soft tissue resolution, and accuracy of measuring left ventricular mass and volumes. Confirmation of the association of LVH with AF using this superior imaging modality reinforces the importance of LVH as a risk factor for the development of AF. Importantly, the association of CMR-LVH with incident AF was attenuated when adjusted for CMR LA volume, however borderline statistical significance was retained. This suggests that some, but not all, of the association of CMR-LVH with incident AF is mediated by LA enlargement.

ECG

There are a number of established ECG criteria for the diagnosis of LVH and recent publications have shown that ECG based criteria have a low sensitivity but high specificity for MRI-defined LVH (27,28). Among the eleven ECG criteria, the three that retained a significant association with AF in adjusted models were the Sokolow-Lyon voltage, Sokolow-Lyon voltage product and Perugia score. Importantly, however, the Sokolow-Lyon voltage product remained predictive of AF even after adjusting for CMR-LVH. This suggests that the Sokolow-Lyon voltage product may be a surrogate of other electrical or structural features associated with the development of AF beyond anatomical LVH. Further, previous studies analyzed ECG time-voltage product in the assessment of LVH and found QRS duration to be an independent predictor of LVH (25,29). This suggests that compared to the Sokolow-Lyon and Perugia score, the association with incident AF of the Sokolow-Lyon voltage product may be due to the inclusion of the QRS duration in its determination.

Limitations

Our study has limitations. First, there were a limited number of AF events, which may explain why some ECG-LVH definitions were not significantly associated with AF risk. Second, the ascertainment of AF was based on US inpatient data. This might have led to underestimation of AF cases not requiring hospitalization, or managed abroad. Additionally, in some cases of asymptomatic AF, time to incident AF may be overestimated. Third, this study does not differentiate between cases of paroxysmal, persistent or permanent AF. Fourth, LA volume was measured in a subset of the total population. Finally, the model we used for determination of ECG-LVM was developed using Caucasian and African-American populations and adjusted only for those ethnicities. In our analysis we combined Chinese and Hispanic patients into the same group with Caucasians. The assumption was made that ECG criterion for LVH is similar within those particular groups, which may not be appropriate and thus may have contributed to the lack of association between ECG-LVM and incident AF.

Clinical Implications

The findings of this study demonstrate that LVH by CMR is associated with future risk of AF in participants with no clinically evident underlying cardiovascular disease. The recent statement from the Working Group on Electrocardiographic LVH called for validation of ECG criteria as prognostic determinants. Here we have validated three ECG algorithms for LVH as prognostic determinants of incident AF in a multi-ethnic population (15,26,27). We have also demonstrated that one algorithm (the Sokolow-Lyon voltage product) has prognostic value beyond structural LVH as defined by CMR. Importantly, the Sokolow-Lyon and Perugia score are simple ECG measures of LVH that can be performed at the bedside without the need for digital acquisition of tracings. Given that most individuals in the general practice do not have CMR images to assess for LVH, the previously mentioned ECG criteria can more widely be used to identify individuals at higher risk for AF. There has been promising evidence in a number of secondary analysis in large clinical trials (LIFE, VALUE, CHARM, Val-HeFT), and meta-analyses suggesting the role of inhibiting the renin-angiotensin system (RAS) in reducing the incidence of AF (30-35). Inhibition of the RAS has also been shown to decrease LVM, particularly in individuals with hypertension (36). Further research is needed to analyze preventive strategies for the development of AF in participants with subclinical cardiovascular disease.

Common Abbreviations

AF

Atrial Fibrillation

ECG

Electrocardiogram

LVH

Left Ventricular Hypertrophy

CMR

Cardiovascular Magnetic Resonance

LVM

Left Ventricular Mass

LA

Left Atrium

MI

Myocardial Infarction

LIFE

Losartan Intervention For Endpoint Reduction in Hypertension

VALUE

Valsartan Antihypertensive Long-Term Use Evaluation

CHARM

Candesartan in heart failure--assessment of reduction in mortality and morbidity

Val-HeFT

Valsartan Heart Failure Trial