Author + information
- Received June 2, 2000
- Revision received November 8, 2000
- Accepted December 13, 2000
- Published online March 15, 2001.
- Gerard P Aurigemma, MD, FACC∗,* (, )
- John S Gottdiener, MD, FACC†,
- Lynn Shemanski, PhD‡,
- Julius Gardin, MD, FACC§ and
- Dalane Kitzman, MD, FACC∥
- ↵*Reprint requests and correspondence:
Dr. Gerard P. Aurigemma, Division of Cardiology, Department of Medicine, University of Massachusetts Medical School, 55 Lake Avenue North, Worcester, Massachusetts 01655
We sought to assess the ability of echocardiographic indices of systolic and diastolic function to predict incident congestive heart failure (CHF).
Noninvasive indices of subclinical systolic and/or diastolic dysfunction that can be used to identify patients in a transition phase between normal cardiac function and clinical CHF would be valuable. Though midwall shortening and Doppler mitral inflow patterns are seemingly well suited to predict subsequent CHF, the predictive value of these indices has not been investigated.
We studied 2,671 participants in the Cardiovascular Health Study who were free of coronary heart disease, CHF or atrial fibrillation. Clinical and quantitative echocardiographic data were obtained in all participants.
At a mean follow-up of 5.2 years (range 0 to 6 years), 170 participants (6.4% of the cohort) developed CHF. Although 96% of these participants had normal or borderline ejection fraction (EF) at baseline, only 57% had normal or borderline EF at the time of hospitalization. In multivariate modeling, fractional shortening at the endocardium (relative risk [RR] 1.85 per 10-unit decrease, confidence interval [CI] 1.27 to 2.39), fractional shortening at the midwall (RR 1.29 per five-unit decrease, 95% CI 1.11–1.51) and peak Doppler peak E (RR 1.15 for each 0.1 M/s increment; CI 1.02 to 1.21) independently predicted incident CHF. Both high and low Doppler E/A ratios were predictive of incident CHF.
Roughly half the occurrences of CHF in this population are associated with normal or borderline EF. Echocardiographic findings suggestive of subclinical contractile dysfunction and diastolic filling abnormalities are both predictive of subsequent CHF. The standard (FSendo) and refined (FSmw) parameters of systolic function performed similarly in this regard, though subjects with left ventricular hypertrophy and depressed FSmw are at particularly high risk for subsequent CHF.
The prevalence of congestive heart failure (CHF) increases along with the proportion of the elderly in the study population and is therefore increasing in prevalence in the U.S. (1). An important goal for clinicians should be to identify high-risk individuals, to better plan strategies to reduce the prevalence of CHF (2,3). In this regard, noninvasive indices of subclinical systolic and/or diastolic dysfunction that can be used to identify patients in a transition phase between normal cardiac function and clinical CHF would be valuable (4,5). Midwall shortening and Doppler mitral inflow patterns would both seem to be well suited for identifying patients with a normal ejection fraction who are at increased risk for subsequent CHF. de Simone et al. (6)have recently shown that in a hypertensive referral population, depressed midwall shortening better predicted cardiac mortality than standard systolic function parameters (e.g., ejection fraction [EF]). Doppler echocardiographic filling velocities (e.g., peak E and peak A velocity and their ratio) are commonly used to evaluate left ventricular (LV) diastolic dysfunction, and it is also thought that filling abnormalities may identify patients in a transition state between apparently normal cardiac function and CHF. However, whether Doppler filling abnormalities predict subsequent CHF is unknown.
The Cardiovascular Health Study (CHS), a prospective, population-based study of 5,201 men and women aged ≥65 years, provided an opportunity to assess the value of clinical, demographic and echocardiographic variables for the prediction of incident CHF. Accordingly, the purpose of this study was to investigate the value of echocardiographically determined systolic and diastolic function variables for the prediction of incident CHF.
The CHS baseline cohort was composed of 5,201 participants: 2,246 men and 2,955 women. Of this cohort, 4,850 participants were white, 307 were black and 44 were classified as “other nonwhite.”
Ages ranged from 65 to 100 years in both men and women (mean ± SD, 73.3 ± 5.8 years in men and 72.4 ± 5.4 years in women). The overall design and recruitment strategy of this study, initiated and supported by the National Heart, Lung and Blood Institute, have been published (7,8). The principal objectives of the study were to investigate the associations between conventional and hypothesized risk factors, as well as ultrasonographic markers of subclinical disease, and outcome from coronary heart disease (CHD) and stroke.
The study population (n = 2,671) comprised all participants in CHS with M-mode echocardiographic data, exclusive of those with prevalent myocardial infarction (MI), angina, CHF or atrial fibrillation at baseline examination. The mean follow-up period for this cohort was 5.2 years.
M-mode measurements were made according to American Society of Echocardiography standards (9); LV mass and relative wall thickness (RWT), an index of LV geometry, were computed (10). Gender-specific partition values for LV mass index were derived using a linear model for the ratio of LV mass to height, which also accounted for the weight of the participant. The equations were derived from a subset (n = 1,281) of the CHS cohort who were free of cardiovascular disease. Left ventricular hypertrophy was defined as a ratio of observed to expected LV mass/height ratio exceeding 1.45 (11). Global LV function was qualitatively assessed from two-dimensional echo image and readers subjectively scored LVEF as normal, abnormal or borderline.
M-mode echocardiographic shortening indices alone and in relation to end-systolic stress were used to assess systolic function (12). Fractional shortening obtained at the endocardium (FSendo, %) and at the midwall (FSmw, %) were calculated (12). Both shortening parameters were related to circumferential systolic wall stress, an estimate of afterload (13), using mean blood pressure to approximate systolic pressure (14). Doppler peak E and peak A velocities and their ratio were measured as previously described (15).
EFs at the time of incident CHF
The reports for echocardiographic examinations performed to evaluate LV function during hospitalization for CHF were available in 145 of 170 (85%) of cases of CHF and were reviewed by the authors; EFs so derived were classified as normal if ≥55%, borderline if >45% but <55% and abnormal if <45%.
Carotid ultrasonography was performed with commercially available equipment. Measurement of the maximum intimal-medial thickness of the internal and common carotid arteries was made using previously described methods (16).
Clinical outcomes were assessed at semiannual contacts during which participants were asked about physician-diagnosed CHF in the past six months. If CHF was reported, medical records were reviewed by the CHS Events Committee for final adjudication (17,18). Congestive heart failure was defined as a constellation of symptoms (such as shortness of breath, fatigue, orthopnea and paroxysmal nocturnal dyspnea) and physical signs (such as edema, rales, tachycardia, a gallop rhythm and a displaced LV apical impulse). Cardiovascular Health Study criteria for CHF required that the participant have a diagnosis of CHF from a physician and be under medical treatment for CHF. In addition, any of the following criteria were considered sufficient but not necessary to validate a CHF diagnosis: the presence of cardiomegaly and pulmonary edema on chest X-ray, evidence of a dilated ventricle, and global or segmental wall-motion abnormalities with decreased systolic function either by echocardiography or contrast ventriculography.
For this analysis, we defined CHF as being due to MI if one or more of the following criteria were met: MI occurring within the two weeks before incident CHF, MI on the same date as presentation for CHF, coronary artery disease listed as a cause of CHF or CHF on date of death.
We determined the concordance between qualitative echocardiographic LVEF estimated by the core laboratory at the time of the baseline examination with that obtained by the treating physicians at the time of hospitalization.
Cox proportional hazard models were used to test the association between known risk factors and echocardiographic variables with incident CHF after adjustment for age and gender. Cox models were also used to model incident CHF in multivariate analyses, including both clinically recognized risk factors for CHF and echocardiographic variables. Clinical variables associated with incident CHF in age and gender-adjusted analyses that were forced into the multivariate Cox model included systolic and diastolic blood pressure, heart rate, height, weight, race, history of diabetes, history of hypertension, history of cigarette smoking, alcohol use and serum cholesterol. Left ventricular mass and end-systolic stress were also forced into the Cox models. Because FSendo and FSmw were highly correlated, these variables were analyzed in separate models. Wall thickness and LV cavity dimensions were excluded from the multivariate models, which included LV mass and shortening because of collinearity. Doppler variables of diastolic function were also allowed to enter the multivariate Cox models to assess their prognostic value.
In order to assess the interaction of LV mass index and midwall fractional shortening, these two parameters were categorized into quartiles after adjustment for end-systolic stress. The significance of the interactions was analyzed by comparing the log likelihood of the Cox model with and without the interaction terms forced in.
Statistical tests were considered significant at p < 0.05; 95% confidence intervals (CIs) were also computed. All analyses were performed using SAS (SAS, Version 6.12, Cary, North Carolina).
During the 5.2 years of follow-up, a total of 170 (6.4%) of the original cohort developed either fatal (n = 6) or nonfatal (n = 164) CHF. Of these CHF events, 18.2% were related to MI. Table 1shows the breakdown of EFs for both baseline and at the time of hospitalization for CHF. At the baseline study, 96% of patients destined to develop CHF had normal or mildly depressed EF. At the time of hospitalization for CHF, 57% of participants were judged by the physician interpreting the echocardiogram to have normal or borderline EF.
Characteristics of subjects with incident CHF during follow-up
Table 2shows baseline clinical and demographic features of subjects with and without incident CHF during follow-up. Participants with incident CHF were older, more likely to be male, and had higher blood pressure and measures of body mass when compared with those participants free of CHF. Participants with incident CHF also had lower ankle-arm index and greater common carotid artery thickness at baseline. Participants with incident CHF had a greater prevalence of diabetes and hypertension and had higher values for weight and body mass index.
Table 3shows baseline echocardiographic data for the participants developing CHF. Participants with incident CHF had greater LV mass as well as greater values for LV dimensions, wall thickness and RWT. Participants who developed CHF at follow-up had lower values for FSendo and FSmw and higher values for end-systolic stress. Of the Doppler indices of diastolic function, peak A and the E/A ratio were significantly different between participants with CHF versus those without CHF.
Functional predictors of incident CHF
Variables of LV systolic function and their relationship to incident CHF were examined. Systolic function, as measured by either fractional shortening index, was associated with increased risk of CHF, which persisted after adjustment for age, race, gender, blood pressure, indices of body size, alcohol use, smoking status, diabetes, hypertension status, LV mass and end-systolic stress. FSendo was a stronger predictor of incident CHF than FSmw, with a relative risk (RR) of 2.51 (95% CI 1.6 to 3.9, p = 0.006) for every 10-unit decrease. By contrast, the relative risk for CHF associated with a five-unit decrease in FSmw was 1.81 (95% CI 1.1 to 3.0, p = 0.06).
To investigate the relationship between Doppler filling patterns and incident CHF, we classified E/A ratio as low (<0.7), intermediate (0.7 to 1.5) or high (>1.5). The corresponding CHF event rates are shown in Figure 1.
Multivariate models of incident CHF
Cox proportional hazard analyses showed that FSendo and peak E velocity remained predictive of incident CHF even after adjustments. When FSendo was excluded from the model, FSmw and peak E velocity were independently predictive of incident CHF. Doppler peak E velocity was associated with an RR of 1.11 per 0.1 M/s increment for both models (including either FSendo or FSmw). The increments of 10% for FSendo and 5% for FSmw were chosen for this analysis; repeating the analysis using one standard deviation change for FSendo (7.7%) and FSmw (5.2%) did not affect the results. A separate multivariate model (not including early and late diastolic velocities) tested the predictive value of the E/A ratio. After adjustment for covariates, incident CHF was highest at the extremes, with a RR for ratios <0.7 of 1.88 (1.33 to 2.68) and at high ratios (>1.5) with an RR of 3.50 (CI 1.80 to 6.80), compared with participants who had intermediate values for E/A ratio. Thus, both low and high E/A ratios entered the model, indicating that participants at either extreme of the velocity profile spectrum were at increased risk of incident CHF.
Interaction of midwall fractional shortening with LV mass index
The incident rates for CHF were greatest in those participants with the highest values for LV mass index and the lowest values for FSmw (Fig. 2). In analyses that adjusted for covariates reflecting clinical and subclinical disease, participants in the highest quartile of LV mass/height at baseline had an RR of 1.40 associated with each five-unit decrement in midwall shortening. In contrast, participants in the lowest quartile of LV mass index had an RR of 0.87 for each five-unit decrement in FSmw (p < 0.01 for interaction). No such interaction was observed when the analysis was repeated using quartiles of FSendo.
The principal finding of this study is that depressed systolic function and abnormalities in LV diastolic filling were predictive of CHF in elderly subjects without clinically evident atherosclerotic vascular disease or CHD at the time of their baseline study. The predictive value of these parameters persisted after adjustment for clinical covariates known to be associated with CHF, such as systolic blood pressure, male gender and age. Inclusion of measures of subclinical atherosclerosis such as carotid intimal-medial thickness and ankle arm index did not weaken the association of LV function with incident CHF. These findings were also independent of LV mass index, an established independent predictor of cardiac morbidity and mortality (19). However, LV mass index and systolic dysfunction appeared to amplify each other’s predictive power (Fig. 2).
Systolic function variables as predictors of incident CHF
Of the 170 participants who developed CHF during the observation period, 150 (96%) had preserved LVEF at baseline study and 57% had preserved EF at the time of hospitalization for CHF. Acute MI precipitated CHF in less than one-fifth of participants. Thus, most CHF in the elderly is associated with an EF that is in (or close to) the normal range (4,20). In a recent study whose results are strikingly similar to those presented herein, Vasan reported on 73 patients, mean age 73 years, participating in the Framingham study, who were hospitalized for CHF. On the basis of their first posthospitalization echo, 51% were judged to have a normal EF, that is, >50% (21). Our study extends these results in that the sample size is twice as large (170 CHF events) and the echocardiograms were obtained during hospitalization for CHF. Evidently subclinical deterioration in LV systolic function occurred in a substantial number of participants.
Predictive value of midwall shortening
Our study was designed to determine which echocardiographic variables of LV function predicted CHF. We hypothesized that depressed midwall shortening would identify individuals with a normal EF at risk for subsequent CHF. This hypothesis was stimulated by previous studies that showed that normal FSendo overestimates myocardial function (12,22)when RWT is increased and that there is significant incremental value of FSmw (compared with FSendo) in the prediction of cardiovascular mortality in hypertensives (6). Although our results confirm the predictive value of midwall shortening in a much larger population, this variable does not perform any better than conventional endocardial fractional shortening (FSendo). In seeking to reconcile our findings with those of de Simone et al. (6), we conclude that the predictive power of midwall shortening is greater in a population with a higher mean RWT (e.g., 0.52 ± 12 in the study of de Simone et al. (6)compared with 0.37 ± 0.10 in our study), because FSmw integrates LV RWT andFSendo (12). Thus, in a large cross-sectional population with RWT in the normal range, simple systolic function parameters can predict incident CHF, as has been shown by the Framingham investigators (23,24). The present study also suggests that LV mass and depressed contractile function act synergistically to predict CHF.
Diastolic heart failure, Doppler filling patterns, and outcome
The development of congestive symptoms attributable to abnormal diastolic properties of the LV is supported by the following evidence. First, as stated earlier, LVEF was normal in most participants at baseline and at the time of hospitalization for CHF. Second, the Doppler data suggest that abnormal LV filling was related to incident CHF. Doppler peak E velocity was positively associated with incident CHF in the multivariable regression and the E/A ratio, which is commonly used clinically to assess LV diastolic function, was related to incident CHF in a U-shaped manner (Fig. 1). This relationship between Doppler filling and incident CHF persisted even after adjustment for LV mass and other covariates. This is consistent with the notion that increases in E velocity under certain circumstances indicate decreased LV compliance, the so-called pseudonormal or restrictive filling pattern (25). Conversely, at the other extreme, a low E/A ratio likely indicates an abnormality in LV relaxation (25). Both patterns in the present study indicate an increased risk for incident CHF, independent of age, blood pressure and level of stress-corrected LV systolic function. Whereas a high E/A ratio (>2) has been shown to portend a poor outcome in patients with established CHF (25), to our knowledge our study is the first to establish that extremes of the Doppler filling ratio independently predict CHF (or any cardiovascular morbid event) in a large prospective study.
The Doppler mitral inflow analysis did not include either the E-wave deceleration time or the isovolumic relaxation time. Thus, the suggestion that patients with an extremely high E/A ratio have restrictive pathophysiology is an extrapolation. It is possible that some of these individuals had other conditions (e.g., significant mitral regurgitation, constrictive pericarditis) that might have independently led to a high E/A ratio. It is unlikely, however, in view of the well-known diminution in the E/A ratio with aging, that older individuals with such high ratios would have normal diastolic filling at the time of their baseline examination.
Although we have examined EF at the time of CHF, it is important to note that this determination was not made in the CHS core laboratory. Therefore, we were unable to calculate the M-mode indices of systolic function (FSendo, FSmw, systolic stress) or measure diastolic filling parameters, precluding a comparison with baseline values.
The sensitivity of all echocardiography parameters of LV function for prediction of incident CHF was relatively modest (Table 4).
Summary and clinical implications
Most CHF in the elderly is associated with normal EF, implying that the “congested state” is more closely related to abnormalities in diastolic function of the LV than to poor pump function. In view of the relatively poor positive predictive value (for CHF) of the parameters studied, the routine use of echocardiography for screening purposes does not appear warranted. However, when echocardiography is performed, the findings of LV hypertrophy, particularly when accompanied by reduced systolic function (assessed by either FSendo or FSmw) and/or extremes of the Doppler E/A ratio, appear to identify a higher risk cohort. In these individuals a more aggressive approach to blood pressure control (26)and more careful monitoring of volume status might be prudent.
☆ This study was supported by contracts N01-HC-85079-85086 and N01-HC-15103 from the National Heart, Lung, and Blood Institute, National Institutes of Health, Bethesda, Maryland.
Presented in part at the 69th Annual Scientific Sessions of the American Heart Association, New Orleans, Louisiana, November 1996.
- coronary heart disease
- congestive heart failure
- Cardiovascular Health Study
- confidence intervals
- ejection fraction
- fractional shortening measured at the endocardium
- fractional shortening measured at the midwall
- left ventricular
- myocardial infarction
- relative risk
- relative wall thickness
- Received June 2, 2000.
- Revision received November 8, 2000.
- Accepted December 13, 2000.
- American College of Cardiology
- St. John Sutton M.,
- Pfeffer M.A.,
- Plappert T.,
- et al.
- de Simone G.,
- Devereux R.B.,
- Koren M.J.,
- et al.
- ↵Tell G, Fried L, Hermanson B, Manolio T, Newman A, Borhani N, for the Cardiovascular Health Study Collaborative Research Group. Recruitment of adults 65 years and older as participants in The Cardiovascular Health Study. Ann Epidemiol 1993;3358–66.
- Sahn D.J.,
- DeMaria A.N.,
- Kisslo J.,
- Weyman A.
- Devereux R.,
- Alonso D.R.,
- Lutas E.M.,
- et al.
- Gardin J.M.,
- Siscovick D.,
- Anton-Culver H.,
- et al.
- Aurigemma G.P.,
- Silver K.H.,
- Priest M.A.,
- Gaasch W.H.
- Gaasch W.H.,
- Zile M.R.,
- Hoshino P.K.,
- Apstein C.S.,
- Blaustein A.S.
- Rozich J.D.,
- Carabello B.A.,
- Usher B.W.,
- Kratz J.M.,
- Bell A.E.,
- Zile M.R.
- Stoddard M.F.,
- Pearson A.C.,
- Kern M.J.,
- Ratcliff J.,
- Mrosek D.G.,
- Labovitz A.J.
- O’Leary D.H.,
- Polak J.F.,
- Wolfson S.K. Jr.,
- et al.
- Psaty B.,
- Kuller L.,
- Bild D.,
- et al.
- Ives D, Fitzpatrick A, Bild DE, et al. Surveillance and ascertainment of cardiovascular events. The Cardiovascular Health Study. Ann Epidemiol 1995;4:278–5.
- Vasan R.S.,
- Benjamin E.J.,
- Levy D.
- Vasan R.S.,
- Larson M.G.,
- Benjamin E.J.,
- Evans J.C.,
- Reiss C.K.,
- Levy D.
- de Simone G.,
- Devereux R.B.,
- Roman M.J.,
- et al.
- Nishimura R.A.,
- Tajik A.J.