Refining Prediction of Atrial Fibrillation Risk in the General Population With Analysis of P-Wave Axis (from the Atherosclerosis Risk in Communities Study)

Abstract

Adverse atrial remodeling is associated with increased risk of atrial fibrillation (AF) and can be detected by a shift in P-wave axis. We aimed to determine whether an analysis of P-wave axis can be used to improve risk prediction of AF. We included 15,102 Atherosclerosis Risk in Communities Study participants who were free of AF at baseline. Abnormal P-wave axis (aPWA) was defined as any value outside 0 to 75 degrees on study visit 12-lead electrocardiograms. AF was determined using study visit electrocardiograms, death certificates, and hospital discharge records. Multivariable Cox regression was used to estimate hazard ratios and 95% confidence intervals (CIs) for the association of aPWA with AF. The Cohorts for Heart and Aging Research in Genomic Epidemiology-AF (CHARGE-AF) risk prediction model variables served as our benchmark. Improvement in 10-year AF prediction was assessed by C-statistic, category-based net reclassification improvement, and relative integrated discrimination improvement. During a mean follow-up of 20.2 years, there were 2,618 incident AF cases. aPWA was independently associated with a 2.34-fold (95% CI 2.12 to 2.58) increased risk of AF after adjusting for CHARGE-AF risk score variables. The use of aPWA improved the C-statistic from 0.719 (95% CI 0.702 to 0.736) to 0.722 (95% CI 0.705 to 0.739), which corresponded with a net reclassification improvement of 0.021 (95% CI 0.001, 0.040) and relative integrated discrimination improvement of 0.043 (95% CI 0.018, 0.069). In conclusion, aPWA is independently associated with AF in the general population. The use of this maker modestly improves AF prediction.

The P-wave axis is a standard automated measure reported on the 12-lead electrocardiogram (ECG) that reflects the net vector component of atrial depolarization in the frontal plane. Abnormal P-wave axis (aPWA) has been associated with atrial fibrillation (AF) development.^1,2 We aimed to determine if aPWA is associated with AF independent of the Cohorts for Heart and Aging Research in Genomic Epidemiology (CHARGE) consortium AF prediction score variables and if analysis of P-wave axis improves the prediction of AF risk in the general population using the CHARGE-AF score³ as a benchmark. We conducted our analysis using data from the Atherosclerosis Risk in Communities (ARIC) Study.

Methods

The ARIC study is a prospective community-based cohort study that was designed to identify and evaluate risk factors, etiology, and clinical manifestations of atherosclerotic coronary heart disease. Between 1987 and 1989, 15,792 men and women aged 45 to 64 years were enrolled and followed by 5 study visits (1987 to 2013) and annual telephone interviews. Further details regarding outcome ascertainment procedures, study design, and population statistics have been previously described.⁴ The present analysis is based on data through 2013. We evaluated all 15,792 participants at the baseline visit and excluded those with missing ECGs (n = 242), with missing P-wave axis data (n = 45), with missing covariates (n = 45), with prevalent AF (n = 37), and who were not white or black from all study sites and nonwhite from Minneapolis and the Washington County (due to small sample size; n = 103), resulting in a final cohort of 15,320 participants.

ECGs were obtained during each clinical exam. Echocardiograms were obtained in all field centers only during the fifth visit. Methods for the calculation of ECG⁵ and echocardiogram⁶ parameters in the ARIC study have been previously described. aPWA was defined as any value outside 0 to 75 degrees. Echocardiographic left atrial enlargement was defined by left atrial volume index >34.

A diagnosis of AF was confirmed after a review of each participant’s hospital discharge records, death certificates, and ECGs. Trained study personnel abstracted International Classification of Disease codes from hospital discharge data and death certificates to identify cases of AF. ECGs coded as AF were reviewed by a cardiologist. Further details regarding specific procedures for AF ascertainment in the ARIC study have been previously described.⁷

The covariates included in our analysis were CHARGE-AF simple model variables: age, race, height, weight, systolic and diastolic blood pressure, current smoking, use of anti-hypertensive medication, diabetes, and history of myocardial infarction or heart failure. Trained study personnel obtained demographic data and medical history from participants during the study visits and annual phone interviews. Age, race, and smoking status (current or noncurrent) were self-reported. Height was measured to the nearest centimeter. Standing weight was measured to the nearest kilogram. Blood pressure was measured using a random-zero sphygmomanometer after 5 minutes of rest. Antihypertensive medication use was self-reported. Prevalent myocardial infarction was determined by self-reported history of myocardial infarction or silent myocardial infarction on ECG.^8,9 Prevalent heart failure was defined as stage 3 “manifest heart failure” by the Gothenburg criteria or self-reported diagnosis of heart failure.¹⁰ Incident myocardial infarction and incident heart failure were identified from a review of hospital records, as previously described.¹¹ Diabetes was defined as a fasting (minimum of 8 hours) glucose ≥126 mg/dl, nonfasting glucose ≥200 mg/dl, self-reported use of oral hypoglycemic agents or insulin, or self-reported physician diagnosis of diabetes.¹²

Person-years at risk were calculated from the date of baseline visit until the date of incident AF, death, loss to follow-up, or end of follow-up, whichever occurred first. For hazards analysis, we did not include P-wave axis measures from the fifth ARIC exam due to lack of AF follow-up data after 2013. The period between the baseline visit and aPWA diagnosis was considered as normal P-wave axis follow-up time. Initially, we explored the relationship between aPWA and incident AF using a restricted cubic spline (Figure 1). Cox proportional hazard models were used to estimate the hazard ratios (HRs) and 95% confidence intervals (CIs) of aPWA for incident AF. We constructed 2 models. Model 1 was adjusted for age and race. Model 2 was additionally adjusted for remaining CHARGE-AF variables: height, weight, systolic and diastolic blood pressure, current smoking, use of antihypertensive medication, diabetes, and history of myocardial infarction or heart failure. All covariates in models 1 and 2, as well as the aPWA exposure variable, were updated until the development of aPWA, administrative censoring, or incident AF, whichever occurred first. We also conducted a race- and gender-stratified analysis.

Figure 1.

P-wave axis and AF. Association between baseline P-wave axis and AF modeled using a restricted cubic spine adjusted for baseline age, race, height weight, systolic and diastolic blood pressure, antihypertensive medication use, smoking, diabetes, previous myocardial infarction, and previous heart failure. Data presented as hazard ratios (solid line) and 95% CIs (shaded area).

To evaluate whether aPWA improved AF risk prediction, we used 2 AF risk prediction models. The refitted CHARGE-AF score³ served as our benchmark (model 1) for a 10-year AF risk prediction. The terms included in this model were age, race, height, weight, systolic and diastolic blood pressure, smoking status, use of antihypertensive medication, diabetes, and history of myocardial infarction or heart failure. After adding aPWA to model 1 (model 2), we evaluated model performance by calculating the C-statistic, net reclassification improvement, and relative integrated discrimination improvement. Reclassification categories were defined as <2.5%, 2.5% to 5%, and >5% 10-year risk. We used the Hosmer–Lemeshow chi-square statistic to evaluate model calibration. For this analysis, we used visit 4 as the baseline because the majority of new AF events occurred in the 10-year period after visit 4.

Finally, to determine whether aPWA was associated with the structural remodeling of the atrium, we estimated the odds ratio of left atrial enlargement associated with aPWA using a cross-sectional analysis of all echocardiograms and ECGs obtained during the fifth study visit (2011 to 2013).

The proportional hazards assumption was assessed with scaled Schoenfeld residuals for both graphical and numerical tests, time interaction terms, and inspection of log negative log survival curves. Modeling assumptions were not violated in any model. Statistical analysis of ARIC data was performed using SAS version 9.3 (SAS Institute Inc., Cary, North Carolina) and STATA 13.0 (StataCorp LP, College Station, Texas). All p values reported are 2-sided, and the statistical significance threshold was chosen as 0.05.

Results

We identified 2,618 cases of incident AF over a mean follow-up time of 20.3 years. There were 1,236 cases of prevalent aPWA at the baseline visit and 2,590 cases of incident aPWA during follow-up. The average (SD) time from aPWA diagnosis to AF was 12.8 (6.8) years. The baseline characteristics of participants are listed in Table 1. Participants with aPWA were less likely to be diabetic and on antihypertensive medication. They were more likely to be current smokers and had a similar prevalence of coronary heart disease and heart failure to those without aPWA at baseline.

Table 1.

Baseline participant characteristics by baseline P-wave axis status Atherosclerosis Risk in Communities study, 1987–1989

Characteristic*	Normal PWA (n = 14,084)	aPWA (n = 1236)	p-value
Age (years)	54.1 ± 5.7	55.0 ± 5.8	<0.0001
Female	7767 (55%)	689 (56%)	0.69
Black	3748 (27%)	307 (25%)	0.17
Current smoker	3558 (25%)	443 (36%)	<0.0001
Weight (kg)	79.4 ± 16.7	71.8 ± 16.9	<0.0001
Height (cm)	168.4 ± 9.3	169.2 ± 9.5	0.006
Body Mass Index (kg/m2)	27.9 ± 5.3	25.0 ± 5.3	<0.0001
Diabetes Mellitus	1694 (12%)	110 (9%)	0.001
Hypertension medications	4370 (31%)	295 (24%)	<0.0001
Systolic blood pressure (mmHg)	121.3 ± 18.8	120.5 ± 20.0	0.10
Diastolic blood pressure (mmHg)	73.8 ± 11.2	72.2 ± 12.0	<0.0001
Heart Failure	650 (5%)	57 (5%)	0.99
Myocardial infarction	561 (4%)	60 (5%)	0.14

^*Data are presented as no. (%) or mean ± standard deviation.

PWA = P-wave axis; aPWA = abnormal P-wave axis.

Figure 1 displays the association between baseline P-wave axis and AF modeled as a restricted cubic spine overlaying the distribution of baseline P-wave axis in our cohort. At visit 1, participants had a mean ± SD P-wave axis of 52.4° ± 20.9°. The risk of AF increased as P-wave axis became <14° and >55°.

The results of our Cox proportional hazards models are listed in Table 2. Participants with aPWA had a higher incidence rate of AF. The risk of AF in participants with aPWA was increased approximately twofold after adjustment for demographic variables in model 1 and cardiovascular risk factors in model 2. Table 3 shows gender- and race-stratified analyses. We did not observe any statistically significant gender- or race-based interactions.

Table 2.

Hazard ratios of abnormal P-wave axis for incident atrial fibrillation. Atherosclerosis Risk in Communities study, 1987–2013

	Normal PWA (n = 12,730)	aPWA (n = 2590)	p-value
Incident AF Cases	2124	494
Person-years	259,515	44,162
Incidence Rate (95% CI)*	8.18 (7.84–8.54)	10.17 (9.25–11.15)
HR (95% CI), Model 1†,§	1 (REF)	1.99 (1.80–2.19)	<0.0001
HR (95% CI), Model 2‡,§	1 (REF)	2.34 (2.12–2.58)	<0.0001

^*Per 1000 person-years.

^†Model 1: Cox proportional hazards model adjusted for race and age.

^‡Model 2: Model 1 + additional adjustment for height, weight, systolic and diastolic blood pressure, current smoking, use of anti-hypertensive medication, diabetes, and history of myocardial infarction or heart failure.

^§Covariates were updated until the development of aPWA, censoring, or atrial fibrillation development, whichever occurred first. AF = Atrial Fibrillation; aPWA = abnormal P Wave axis; PWA = P-Wave axis; HR = Hazard Ratio; CI = Confidence Interval; REF = Reference.

Table 3.

Risk of atrial fibrillation by P-wave axis stratified by race and sex, Atherosclerosis Risk in Communities study, 1987–2012

	Normal PWA	aPWA	p-value
Male, HR (95% CI)*,†	1 (REF)	2.13 (1.86–2.45)	<0.0001
Female, HR (95% CI)*,†	1 (REF)	2.58 (2.24–2.97)	<0.0001
Sex Interaction			0.07
Black, HR (95% CI)*,†	1 (REF)	1.89 (1.48–2.40)	<0.0001
White, HR (95% CI)*,†	1 (REF)	2.46 (2.21–2.74)	<0.0001
Race Interaction			0.10

^*Cox proportional hazards model adjusted for age, height, weight, systolic and diastolic blood pressure, current smoking, use of anti-hypertensive medication, diabetes, and history of myocardial infarction or heart failure.

^†Covariates were updated until the development of aPWA, censoring, or atrial fibrillation development, whichever occurred first.

aPWA = abnormal P Wave axis; PWA = P-Wave axis; HR = Hazard Ratio; CI = Confidence Interval; REF = Reference.

At visit 4, there were 10,956 participants who met our inclusion and exclusion criteria. There were 975 cases of prevalent aPWA. We identified 810 cases of AF over a 10-year follow-up period. The addition of aPWA to the refitted CHARGE-AF risk score (Table 4) resulted in a modest improvement in model discrimination (measured by C-statistic and relative integrated discrimination improvement), calibration (measured by the Hosmer–Lemeshow chi-square statistic), and risk reclassification (measured by net reclassification improvement). In patients who developed AF, 28 participants (3.5%) experienced favorable reclassification, whereas 29 (3.6%) experienced unfavorable reclassification. In patients who did not develop AF, 506 (5.0%) experienced favorable reclassification, whereas 290 (2.9%) experienced unfavorable reclassification (Table 5). During the fifth clinical exam, 5,574 participants underwent echocardiograms and ECGs. Of these, 669 had aPWA. The odds of left atrial enlargement were 1.44 (95% CI 1.16 to 1.80) higher in those with aPWA.

Table 4.

C-statistic, net reclassification improvement (NRI), and relative integrated discrimination improvement (IDI) of predictive models for incident atrial fibrillation, the Atherosclerosis Risk in Communities study

	C-statistic (95% CI)	Calibration x2 (P-value)§	Category-based NRI* (95% CI)	Relative IDI (95% CI)
Model 1†	0.719 (0.702–0.736)	16.9 (0.05)
Model 2‡	0.722 (0.705–0.739)	15.6 (0.08)	0.021 (0.001, 0.040)	0.043 (0.018, 0.069)

^*For categorical NRI, we used the following categories for 10-year atrial fibrillation risk: < 2.5%, 2.5%−5% and ≥5%.

^†CHARGE-AF variables: age, race, height, weight, systolic and diastolic blood pressure, current smoking, use of anti-hypertensive medication, diabetes, and history of myocardial infarction or heart failure.

^‡Model 1 + abnormal P-wave Axis.

^§Hosmer-Lemeshow X² statistic.

Table 5.

Reclassification of atrial fibrillation risk based on P-wave axis^*

Model 1†,*	Model 2‡,*
	<2.5%	2.5–5%	>5%	Total
Participants who developed atrial fibrillation within 10 years
<2.5%	247	17¶	0¶	264
2.5–5%	15§	124	11¶	150
>5%	0§	14§	382	396
Total	262	155	382	810
Participants who did not develop atrial fibrillation within 10 years
<2.5%	1522	113¶	0¶	1635
2.5–5%	201§	2241	177¶	2619
>5%	0§	305§	5587	5892
Total	1723	2659	5764	10,146

^*Categorical net reclassification improvement table for 10-year atrial fibrillation risk based on the following risk categories: < 2.5%, 2.5%−5% and ≥5%.

^†CHARGE-AF variables: age, race, height, weight, systolic and diastolic blood pressure, current smoking, use of anti-hypertensive medication, diabetes, and history of myocardial infarction or heart failure.

^‡Model 1 + abnormal P-wave Axis.

^§Incorrect reclassification.

^¶Correct reclassification.

Discussion

In this large community-based cohort of middle-aged, white and black individuals, we found that aPWA was independently associated with a doubling in the risk of AF and that the addition of this marker to the CHARGE-AF score enhanced the prediction of AF.

In the United States Third National Health and Nutrition Examination Survey, aPWA was associated with cardiovascular and all-cause mortality.¹³ Extremes of the P-wave axis on preoperative ECGs have been associated with postoperative atrial fibrillation in cardiac surgery patients.¹ In the Cardiovascular Health Study, aPWA was independently associated with a 1.17-fold (95% CI 1.03 to 1.33) increase in AF risk.² Our analysis confirms this association in an independent cohort and suggests that the risk of aPWA-associated AF in the general population may be higher than previously estimated in the Cardiovascular Health Study. Although not statistically significant, the risk of aPWA-related AF seemed higher in women and whites. Further research is needed to explore the possibility of gender- and race-based interactions.

Three major clinical models have been published for purposes of AF risk prediction.^3,14,15 The CHARGE-AF consortium, the largest and most recent study, used easily accessible clinical variables to create a benchmark for AF risk assessment in the primary care setting.³ Although abnormal P-wave indices—P-wave duration, P-wave area, P-wave terminal force, PR interval—have been associated with increased AF risk, the incorporation of these markers into the CHARGE-AF score did not improve AF risk prediction.¹⁶ The addition of aPWA to the CHARGE-AF risk score variables augmented model performance in our study. This was primarily driven by a correct reclassification of low-risk individuals who did not develop AF. Although the magnitude of improvement was modest and therefore unlikely to have a dramatic clinical impact, our findings are hypothesis generating.

Proarrhythmic atrial remodeling¹⁷ can be detected through an analysis of P-wave morphology.^5,18,19 Shifts in the P-wave axis have been linked with atrial enlargement, an established AF risk factor,²⁰ on echocardiography and cardiac magnetic resonance imaging.²¹ Our analysis supports this association, and our findings collectively demonstrate that ECG markers that may reflect underlying arrhythmogenic atriopathy can improve AF risk prediction tools built with traditional risk factors for cardiovascular disease.

The principal strengths of our study are the large size of our cohort, the long follow-up duration, and the rigorous measurement of covariates. There are some limitations to consider. First, the capture of aPWA may have been suboptimal because P-wave indices were only measured during ARIC study visits. Second, AF was primarily identified from a review of hospital discharge data. We were unable to capture subclinical AF or AF that was treated solely in an ambulatory setting and not reported to ARIC staff. However, AF incidence in ARIC is consistent with other population-based studies, and utilizing hospital discharge records for the purposes of AF detection has been previously validated in ARIC.^7,22–24 Finally, we could not account for residual confounding by subclinical or unmeasured variables. Thus, the relationship between aPWA and AF may also be, in part, explained by shared cardiovascular risk factors.