Electrocardiographic Repolarization-Related Variables as Predictors of Coronary Heart Disease Death in the Women's Health Initiative Study
This article has a correction. Please see:
Background We evaluated 25 repolarization-related ECG variables for the risk of coronary heart disease (CHD) death in 52 994 postmenopausal women from the Women's Health Initiative study.
Methods and Results Hazard ratios from Cox regression were computed for subgroups of women with and without cardiovascular disease (CVD). During the average follow-up of 16.9 years, 941 CHD deaths occurred. Based on electrophysiological considerations, 2 sets of ECG variables with low correlations were considered as candidates for independent predictors of CHD death: Set 1, Ѳ(Tp|Tref), the spatial angle between T peak (Tp) and normal T reference (Tref) vectors; Ѳ(Tinit|Tterm), the angle between the initial and terminal T vectors; STJ depression in V6 and rate-adjusted QTp interval (QTpa); and Set 2, TaVR and TV1 amplitudes, heart rate, and QRS duration. Strong independent predictors with over 2-fold increased risk for CHD death in women with and without CVD were Ѳ(Tp|Tref) >42° from Set 1 and TaVR amplitude >−100 μV from Set 2. The risk for these CHD death predictors remained significant after multivariable adjustment for demographic/clinical factors. Other significant predictors for CHD death in fully adjusted risk models were Ѳ(Tinit|Tterm) >30°, TV1 >175 μV, and QRS duration >100 ms.
Conclusions Ѳ(Tp|Tref) angle and TaVR amplitude are associated with CHD mortality in postmenopausal women. The use of these measures to identify high-risk women for further diagnostic evaluation or more intense preventive intervention warrants further study.
Electrocardiographic depolarization- and repolarization-related abnormalities as predictors of coronary heart disease (CHD) mortality and morbidity have been a subject of many electrocardiographic investigations. From repolarization-related abnormalities, QT prolongation has been a common topic in studies on general populations and in clinical study groups, particularly with cardiovascular disease (CVD).1 Some newer reports from general populations have documented increased risk for CHD death for widened spatial angle between mean QRS and ST-T vectors (Ѳ(QRS|STT)).2–3 ST- and T-wave findings in women with CVD are generally considered as secondary abnormalities of little importance in clinical ECG interpretation, although some studies have associated them with CHD mortality risk,4–7 including the risk of sudden cardiac death (SCD). From depolarization-related ECG abnormalities, QRS duration increase even within its upper normal range has been found to be an independent predictor of CHD death, including SCD.4,8–9
A recently developed repolarization model introduced several novel repolarization-related variables from various repolarization time (RT) subintervals such as QT peak (QTp) interval, epicardial repolarization time (RTepi), left ventricular crossmural RT gradient (XMRTgrad) and, in addition to Ѳ(QRS|STT), several other spatial angles representing deviation of the repolarization sequence from normal direction during various RT subintervals.3,10–12 The primary objective of the present study was to evaluate the risk of CHD death for these novel ECG risk predictors in postmenopausal women from the Women's Health Initiative (WHI) study.
The WHI is a 40-center, national study of risk factors and the prevention of common causes of mortality, morbidity, and impaired quality of life in women. Postmenopausal women aged 50 to 79 years from various ethnic groups were recruited from 1994 to 1998. Details of the study design, protocol sampling procedures, and selection and exclusion criteria have been published previously.13 The present study group consisted of 68 133 women, a subgroup of the clinical trial component of WHI, which had digital ECGs and comprehensive documentation of outcome events available. Participants with missing or incomplete ECG data (n=966) were excluded; ECGs with inadequate quality or technical errors by visual inspection (n=614), bundle branch blocks (n=1739), electronic pacemakers or WPW pattern (n=109), and 47 ECGs with heart rate >100/min and 3 ECGs with incomplete data were also excluded. From the remaining group of 64 661 participants, 12 569 were found to have had a CVD event while 52 092 were CVD-free at baseline. The sequential steps in selection of the study group for risk analyses are shown in the block diagram in Figure 1.
Protocols for human studies were reviewed and approved by Institutional Review Boards of each participating center, and informed consent was obtained from each participant.
Standard 12-lead ECGs were recorded in all women in the supine position using MAC PC electrocardiographs (GE Marquette, Inc, Milwaukee, WI). ECG technicians in all participating centers were trained to use carefully standardized procedures for ECG acquisition including locating the chest electrodes in precise positions using a special chest electrode locator.14 All electrocardiograms received at a Central ECG Laboratory (EPICARE Center, University of Alberta; Edmonton, Alberta, Canada and later at Wake Forest University, Winston-Salem, NC) were inspected visually to detect technical errors, missing leads, and inadequate quality, and such records were rejected from ECG data files. The ECGs were processed by 2001 version of the Marquette 12SL program (GE Marquette, Inc, Milwaukee, WI).
Repolarization Parameters from the Repolarization Model
The orthogonal Frank XYZ leads were obtained from the 8 independent components (leads I, II, V1 to V6) using a transformation matrix from the 116-lead body surface map library of Horáĉek containing recordings for 892 adults aged 16 to 85 years.15 Repolarization measurements were made utilizing temporal reference points derived from the spatial T-vector magnitude curve derived from the XYZ leads (the “global” T wave), including QT end (QTe), QTpeak (QTp), and QTonset (QTo) intervals. QTe, QTp, and QTo intervals were rate adjusted (QTea, QTpa and QToa, respectively) as linear functions of the RR interval with the following formulas derived in the CVD-free group: QTea=QTe+184×(1−RR), QTpa=QTp+135×(1−RR) and QToa=QTo+113×(1−RR). Heart rate, QRS duration, QRS nondipolar voltage from singular value decomposition and a set of 22 repolarization-related ECG variables from our repolarization model were chosen for evaluation because of their functional role in generation of normal and abnormal repolarization waveforms or because of their previously shown value as risk predictors.2–3,10–12 QRS duration was included as the second depolarization-related parameter in addition to QRS nondipolar voltage from singular value decomposition because even moderate QRS prolongation is known to induce secondary repolarization abnormalities, which may be associated with adverse cardiac events over and above those induced by QRS prolongation alone.
The conceptual model used to derive RT subintervals and other model parameters for the present study has been described in detail in previous publications.2–3,10–12 A simplified summary description of the main model variables in nonstatistical terms is contained in Table 6. In more basic terms, RT of the subepicardial myocyte layers (RTepi), 1 of the main repolarization model parameters, is considered to represent RT of left ventricular (LV) myocytes at the time of global T-wave peak (Tp) when the majority of LV lateral wall myocytes are at some point of phase 3 of their action potential. RTepi is computed as a function of QTpa whereby RTepi=QTpa−(1−CosѲ(Tp|Tref)×(TpTxd))/2, where Ѳ(Tp|Tref) is the spatial angle between the Tp vector and Tref is the reference normal Tp vector with xyz components (0.75, 0.57, −0.33). TpTxd, in turn, is the interval from Tp to Txd, where Txd is the inflexion point (the steepest negative slope) at global T wave downstroke. Thus, RTepi is obtained from QTpa by modifying it by the degree of deviation of direction of the initial repolarization from the direction of normal repolarization. RT at time point Txd (RTxd) is obtained with an algorithm similar to that for RTepi, whereby RTxd=QTpa−(1+CosѲ(Tp|Tref)×(TpTxd))/2. In addition to Ѳ(Tp|Tref) noted above, a number of other spatial angles between various QRS and T vectors and other interval and amplitude variables were used in various phases of the study.
Ascertainment of Outcome Events
After baseline, deaths and hospitalization events were ascertained in each clinical center by annual follow-up calls, review of vital records, and community surveillance of hospitalized and fatal events. Detailed definitions for criteria for CHD death classification have been published previously.16 Briefly, CHD deaths included death with no known non-CHD death and either a history of chest pain within 72 hours before death or a history of chronic ischemic heart disease in the absence of valvular heart disease. The average follow-up was 15.9 years (up to 17 years).
Frequency distributions of ECG variables from the repolarization model were first inspected to rule out anomalies and outliers. QTe distributions were skewed, but otherwise, no anomalies that would notably interfere with analyses were observed. ECG predictors were first evaluated individually in unadjusted single ECG variable models and subsequently in multivariable-adjusted models adjusted for age, race, smoking status, diabetes, hypertension, family history of CHD and stroke, body mass index, hypercholesterolemia, and study component/arm groups (hormone therapy/dietary modification/calcium, and vitamin D).
In the search for independent predictors for CHD death, 1 primary concern was the collinearity of variables with high correlations. A previous investigation in participants free from cardiovascular disease from the Atherosclerosis Research in Communities Study found Ѳ(Tp|Tref) and aVR amplitude as independent predictors of CHD death and SCD.3 These 2 correlated variables (r=0.56) were used as primary explanatory variables, and a search was performed separately for each to identify other variables with low correlations (r<0.4). This procedure produced 2 sets of predictors as candidates for independent predictors for CHD death: Set 1, Ѳ(Tp|Tref) and spatial angle between the initial and terminal T vectors Ѳ(Tinit|Tterm), respectively), STJ depression in V6, rate-adjusted QTp interval (QTpa); and Set 2, TaVR and TV1 amplitudes, heart rate, and QRS duration. An association was considered significant when the P-value (2-sided) was <0.05.
Consistent with our previous risk data in studies with different end points,11–12 it was observed that CHD risk in general started to increase after the 80th percentile of the ECG variable distribution. Therefore, hazard ratios were constructed to evaluate the risk for CHD death with quintile 5 as the test group with quintiles 1 to 4 as the reference group. However, for ST J point and T-wave amplitudes in aVL and V6, the risk of CVD death was observed to increase at values below the 20th percentile of the distribution and quintile 1 was used as the test group for these variables, with quintiles 2 to 5 as the reference group. Risk for CHD death was first evaluated using group-specific cut points for the test group at 80th or 20th percentile. The quintiles were chosen for evaluation with the expectation that significant predictors for CHD death in test quintiles would be strong predictors with higher cut points. Finally, cut points were set at values representing upper or lower fifth percentiles of the CVD-free group, and these dichotomized cut points were used also for the CVD group.
All analyses were performed with SAS version 9.2 (SAS Institute, Cary, NC).
Characteristics of the Study Population
The subgroup of women considered CVD-free after exclusion of women with clinical or ECG evidence of any CVD was relatively healthy (Table 1). Still, 31% were hypertensive or on antihypertensive medication. All women with ECG-LVH (RaVL+SV3 >2200 μV with ST depression including the so-called LV strain pattern (ECG-LVH with down sloping ST and negative T wave) had been transferred to the CVD group. Five percent of the CVD-free women had diabetes, and 0.1% had atrial fibrillation by self-report. As expected, most differences between CVD and CVD-free groups were statistically significant. Nearly one half of the women with CVD were hypertensive, 11% had diabetes, 20% had atrial fibrillation by self-report and 1.7% had atrial fibrillation in the baseline ECG. Approximately one half of women in both groups had never smoked and about 8% were current smokers.
Single ECG Variables as Predictors of CHD Death
More than one half of the 25 ECG variables evaluated were significant predictors of CHD death in unadjusted single ECG variable risk models (not shown) and remained significant predictors in multivariable adjusted (Table 2). Four ECG variables in CVD-free women had an over 1.50-fold increased risk for CHD death. The strongest predictor in CVD-free women was ToV/TpV (hazard ratios 1.93 [1.42 to 2.63], P<0.001), the ratio of the spatial magnitudes of T vectors at T wave onset and T wave peak. Increased value of this variable reflects reduced convexity of ST magnitude curve which in turn is thought to reflect triangularization of phase 3 of LV lateral wall action potentials. Many ECG variables in women with CVD were strong predictors of CHD death, including ToV/TpV and STJ-point and T wave amplitudes in several ECG leads. Twelve of the ECG variables had an over 1.5-fold increased risk for CHD death, and for 2 of them, Ѳ(Tp|Tref) and STJV6, there was an over 2-fold increased risk.
Independent ECG Predictors of CHD Death
It was observed that many of the repolarization variables including several of T wave and STJ-point amplitudes were correlated with Ѳ(Tp|Tref) (r>0.4) (Table 3). A smaller subset of variables with lower correlation with Ѳ(Tp|Tref) (r<0.4) were chosen initially to search for independent predictors for CHD death. Two sets of predictors were identified as independent predictors of CHD death (Table 4). Set 1, spatial angles between T peak (Tp) and normal T reference (Tref) vectors and between the initial and terminal T vectors (Ѳ(Tp|Tref) and Ѳ(Tinit|Tterm), respectively), STJ depression in V6; and Set 2, TaVR and TV1 amplitudes and QRS duration. The strongest independent predictors in women with and without CVD were Ѳ(Tp|Tref) >42° in Set 1 and TaVR amplitude less negative than −100 μV in Set 2 with an over 2-fold increased risk for both, and also heart rate >84 had an over 2-fold increased risk among Set 2 variables in CVD-free women. For the other independent predictors of CHD death in Set 1, risk increase ranged from 30% for ST J-point amplitude in V6 to 87% for heart rate and in Set 2 from 56% for TV1 amplitude to 64% for QRS duration. These independent predictors of CHD death in multivariable Model 1 remained significant with additional multivariate adjustment for demographic and clinical factors in Model 2. It is noteworthy that Set 2 ECG variables TaVR and heart rate were as strong predictors for the risk of CHD death as the computationally more complex best Set 1 variable Ѳ(Tp|Tref).
Clinical Diseases and Related ECG Findings as Predictors of CHD Death
Hazard ratios are listed in Table 5 for selected clinical classification categories and related ECG findings of interest. Atrial fibrillation by self-report was the strongest predictor in the remaining classification categories in CVD-free women, with an over 4-fold increased risk for CHD death in multivariable-adjusted model. However, the prevalence of this condition was low in CVD-free women (0.1%, Table 1). Diabetes was a strong predictor of CHD death, with a 2.7-fold increased multivariable-adjusted risk in both groups of women. Hypertension in CVD-free women had a 1.59-fold multivariable-adjusted increased risk for CHD death and a 1.81-fold increased risk in women with CVD. Of interest is that ventricular ectopic complexes and left atrial enlargement were both significant predictors of CHD risk in both groups of women.
Key results of this investigation can be summarized as follows: (1) A majority of the ECG variables were significant predictors of CHD death in women when evaluated as single ECG variables and remained significant in multivariable-adjusted risk models; (2) 2 sets were considered as candidates for independent predictors of CHD death: Set 1, spatial angles between Tpeak (Tp) and normal T reference (Tref) vectors and between the initial and terminal T vectors (Ѳ(Tp|Tref) and Ѳ(Tinit|Tterm), respectively), STJ depression in V6 and rate-adjusted QTp interval (QTpa); and Set 2, TaVR and TV1 amplitudes, heart rate and QRS duration; (3) The strongest independent predictors in women with and without CVD with an over 2-fold increased risk were Ѳ(Tp|Tref) >42° in Set 1 and TaVR amplitude less negative than −100 μV in Set 2; (4) Among Set 2 variables also heart rate >84 had an over 2-fold increased risk in CVD-free women; (5) The risk for these strong CHD death predictors remained significant after multivariable adjustment for demographic/clinical factors; and (6), Set 2 variable TaVR was as strong predictor as the computationally more complex Set 1 best predictor Ѳ(Tp|Tref).
Possible Mechanisms for the Association of ECG Predictors With the Risk of CHD Death
Three mechanisms possibly accounting for increased risk for CVD death are summarized in Table 6. The first mechanism is related to myocardial ischemia in chronic CHD. Myocardial ischemia most commonly located in left-anterior-descending coronary artery perfusion area shortens action potential duration and alters spatial direction of the repolarization sequence during the initial LV lateral wall repolarization. A previous report from the Cardiovascular Health Study demonstrated that anterior-right rotation of the Tp vector is associated with QRS|T angle widening in CVD free men and women.17 Anterior-right rotation of the Tp vector also accounts for the increased (less negative) amplitude in aVR and increased V1 amplitude (Figure 2). Thus, same pathophysiological mechanisms relating to altered regional repolarization times may account for increased risk of CHD death associated with Set 1 main predictor Ѳ(Tp|Tref) and Set 2 main predictor TaVR. The second mechanism in Table 6 is related to LV overload in hypertensive heart disease. In LVH, with increased epicardial excitation time (ETepi) and RTepi due to increased LV mass and possibly also with slowed myocardial conduction velocity leads into widening of Ѳ(Tp|Tref) and the repolarization sequence changes progressively from normal predominantly reverse to predominantly concordant with respect to depolarization sequence generating the so called LV strain pattern. A predominantly concordant repolarization sequence results in increase (less negative) aVR amplitude, again suggesting that widened Ѳ(Tp|Tref) angle and decreased TaVR amplitude are produced by the same pathophysiological mechanism. Increasing dyssynchrony of depolarization 11–12 may in turn, lead into dyssynchrony of ventricular relaxation with impairment of diastolic function.18 These ECG abnormalities are associated with increased risk of CHD death and heart failure. The third mechanism postulated is associated with derailed ionic channel dynamics due to possible adverse effects of cardioactive drugs and a multiplicity of other factors inducing regional QT prolongation (increased QTpa, RTepi) or diffuse global QT prolongation, known to be associated with increased risk of CHD death, including sudden cardiac death.
Ѳ(Tinit|Tterm) was the second spatial angle as a significant predictor of CHD death. Ѳ(Tinit|Tterm) reflects increased difference in the spatial direction of repolarization during initial and terminal repolarization as a manifestation of a widened, rounder T vector loop related to T wave complexity which has been suggested as an indicator of subclinical myocardial ischemia in asymptomatic adults.19
Relation of the Present Study With Previous Investigations
The risk of CHD death in CVD-free men and women 45 to 65 years old was evaluated in a report from the Atherosclerosis Research in Communities Study excluding men and women with a history or clinical manifestations of CHD or other CVD.3 ECG-based exclusions from the CVD-free group included QRS duration 120 ms or longer or major Q waves by Minnesota Code20 (MC 1.1). In women, independent predictors of the risk of CHD death were (QRSm|Tm) and (Tp|Tref), with a 2-fold increased risk for the former, and with a 1.7-fold increased risk for the latter variable. QTea was an independent predictor in men but not in women. A notable finding in the Atherosclerosis Research in Communities Study was that the risk levels for independent predictors for CHD death were stronger in women than in men. In the present investigation Ѳ(Tp|Tref) and Ѳ(Tinit|Tterm) were independent predictors of CHD death in addition to heart rate. In the selection of CHD-free women in the present study a more extensive set of ECG-based exclusions were made, including ECG evidence of an old MI, atrial fibrillation in baseline ECG, high-amplitude QRS (Cornell voltage) with even minor T-wave abnormalities (MC 5.1 to 5.3) so that the repolarization measures used can be considered as isolated independent predictors of CHD death.
A report from the Seven Countries Study in a male cohort with no manifest cardiac diseases at baseline evaluated the risk of CHD death for isolated inverted T waves with no other codable ECG abnormalities.5 The risk of CHD death for inverted T waves was over 3-fold in 5-year follow-up, decreasing with the length of follow-up but still significant at 40-year follow-up.
Laukkanen et al evaluated the association of isolated T wave inversion and widened QTS|T angle with the risk of SCD in a male cohort from a general Finnish population with a 20-year follow-up.4 In a multivariable adjusted single ECG variable model, T wave inversion and widened QRS|T angle were both associated with an over 3-fold risk for SCD. QRS duration from 110 to 119 ms was also a significant predictor of SCD compared to men with QRS duration <110 ms. Anttila et al, in another report from a nationally representative sample of the general Finnish population of adult men and women, documented that a positive T wave in aVR was a strong predictor of CVD death in fully adjusted risk models.7 TaVR was also reported in an earlier study to be a predictor of CVD death in a large clinical male population.21
Ѳ(Tp|Tref) was a strong predictor of CHD death in women with CVD as well as in CVD-free women. From a practical clinical point of view, a potentially more important observation was that ECG variables such as TV1 and TaVR amplitudes (from the alternative Set 2 in Table 4) were practically as strong predictors of CHD death as the computationally more complex angular measures of deviant repolarization. This finding suggests that these simple variables may be potentially useful clinical tools for identification of high-risk women for preventive intervention on CHD death.
Limitations of the Study
Data were not available from echocardiographic evaluation of cardiac function to permit a more refined identification of silent CVD. T waves, particularly in women, are considered to be sensitive to variations in sympathetic tone as reflected by increased heart rate. However, the correlations between heart rate and the angular measures of deviant spatial direction of repolarization and also QTpa were low (r<0.4). Since the primary focus of our study was on a limited number of independent predictors of CHD death, no provision was made to adjust for multiple comparisons for mean differences between CVD and CVD-groups. No competing risk analysis was done to evaluate additional risk of CHD death for ECG variables beyond the risk information contained in diabetes, hypertension, and other clinical conditions. The primary aim of our study was not diagnostic discrimination but rather an exploratory analysis to establish associations as the first-line predictors of CHD death and to consider possible mechanisms accounting for the observed excess risk found for them.
The authors thank the WHI investigators and staff for their dedication, and the study participants for making the program possible. A full listing of WHI investigators can be found at: www.whi.org/researchers/Documents%20%20Write%20a%20Paper/WHI%20Investigator%20Short%20List.pdf.
Sources of Funding
The WHI Sequencing Project is funded by the National Heart, Lung, and Blood Institute (HL-102924) as well as the National Institutes of Health (NIH), U.S. Department of Health and Human Services through contracts HHSN268201100046C, HHSN268201100001C, HHSN268201100002C, HHSN268201100003C, HHSN268201100004C, and HHSN271201100004C.
- Rautaharju PM,
- Kooperberg C,
- Larson JC,
- LaCroix A
- Rautaharju PM,
- Zhang ZM,
- Warren J,
- Haisty WK,
- Prineas RJ,
- Kurcharska-Newton AM,
- Rosamond WD,
- Soliman EZ
- Rautaharju PM,
- Menotti A,
- Blackburn H,
- Parapid B,
- Kircanski B
- Kumar A,
- Prineas RJ,
- Arnold AM,
- Psaty BM,
- Furberg CD,
- Robbins J,
- Lloyd-Jones DM
- Kurl S,
- Mäkikallio TH,
- Rautaharju P,
- Kiviniemi V,
- Laukkanen JA
- Rautaharju PM,
- Gregg RE,
- Zhou SH,
- Startt-Selvester RH
- Rautaharju PM,
- Zhou SH,
- Gregg RE,
- Startt-Selvester RH
- Rautaharju PM,
- Clark-Nelson J,
- Kronmal RA,
- Zhang ZM,
- Robbins J,
- Gottdiener J,
- Furberg C,
- Manolio T,
- Fried L
- Zhu TG,
- Patel C,
- Martin S,
- Quan X,
- Wu Y,
- Burke JF,
- Chernick M,
- Kowey PR,
- Yan GX
- Blackburn H,
- Keys A,
- Simonson E,
- Rautaharju PM,
- Punsar S