Prognosis of Adults With Borderline Left Ventricular Ejection FractionFree Access

Participants

The details of the selection criteria and examination of Framingham Heart Study (FHS) original and offspring cohorts have been described (10,11). We included original cohort participants who attended examinations 16 (1979 to 1981) or 20 (1988 to 1989) and offspring cohort participants who attended examinations 4 (1987 to 1990), 6 (1995 to 1998), or 8 (2005 to 2008) (Online Figure 1). Of 14,187 eligible person-observations, we excluded observations with a history of HF (n = 270), inadequate echocardiographic data (n = 3,592), and a lack of follow-up data (n = 104). Persons with missing measures were more likely to be obese and to have greater CVD risk factors (12). After exclusions, we included 10,221 person-observations representing 5,334 unique persons. The number of observations included at each examination is presented in Online Table 1.

Diabetes was defined as a fasting glucose concentration ≥126 mg/dl or the use of hypoglycemic medications. Systolic and diastolic blood pressures were measured as the average of 2 measurements made on seated participants by using a mercury column sphygmomanometer, an appropriately sized cuff, and a standardized protocol. Use of antihypertensive medications and antidiabetes medications was self-reported, and all medications were verified by the FHS clinic physician. Between January 1995 and September 1998, plasma brain natriuretic peptide levels were collected in offspring cohort participants (n = 2,552) at examination 6, in the morning after an overnight fast. Samples were stored at −70°C and were analyzed using sensitive noncompetitive immunoradiometric assays (Shionogi, Japan) in June 1999.

Echocardiography and calculation of LVEF

The following ultrasound machines were used for echocardiography: for original cohort examination cycles 16 and 20 and for offspring examination cycles 4 and 5, Hewlett Packard model 77020AC (Hewlett Packard, Palo Alto, California); and for offspring examinations 6 and 8, Hewlett Packard Sonos 1000 and Sonos 5500, respectively.

Measurements of M-mode LV end-diastolic dimension (LVEDD) and end-systolic dimension were performed by experienced sonographers using the leading edge technique according to American Society of Echocardiography guidelines (13). LVEF was calculated with these measures using the Z-volume formula by de Simone et al. (14):

$LVEF(\%)=\frac{\left[\left(4.5·LVED{D}^2\right)-\left(3.72·LVES{D}^2\right)\right]}{4.5·LVED{D}^2}·100$

The basis of this method is human (14) and experimental (15) evidence that the epicardial long axis–to–short axis ratio is constant through the cardiac cycle and has been widely applied in clinical studies (16–18). We selected this formula to include the longer follow-up of earlier cohorts that did not have routine 2-dimensional quantitation of chamber volume.

Additionally, in a subset of participants, both the de Simone method and the biplane Simpson method using 2-dimensional echocardiography (available in n = 2,315 of offspring cohort at examination 8) were used to quantitate LVEF by the summation of disks method in 4-chamber and 2-chamber views (3).

Follow-up

Participants’ medical records were reviewed and adjudicated for cardiovascular disease (CVD) and death. CVD included history of coronary artery disease, stable and unstable angina, myocardial infarction (MI), cerebrovascular accident (atherothrombotic brain infarct, transient ischemic attack, intracranial or subarachnoid hemorrhage, cerebral embolism), and peripheral arterial disease (intermittent claudication).

The diagnosis of HF was made using the FHS criteria (19), with sensitivity and specificity comparable to other HF criteria (20). The date of onset of HF was noted as the first episode of HF symptoms, physician visit, or hospitalization. HF with a reduced LVEF (HFREF) and HF with a preserved LVEF (HFPEF) were defined as HF symptoms with an LVEF <50% and ≥50%, respectively (21).

Our primary outcome was a composite of new-onset HF and death because death may be the first adverse event in persons with asymptomatic LV systolic dysfunction (6).

Statistical analysis

We pooled participants of FHS original cohort examinations 16 and 20 and of offspring cohort examination cycles 4, 6, and 8, and we retained participants free of prevalent HF. Participants were grouped by LVEF <50%, 50% to 55%, and >55%, calculated on the basis of their examination echocardiograms. We evaluated clinical correlates of borderline LVEF relative to LVEF >55% (excluding participants with LVEF <50%), by using multivariable logistic regression with the following covariates: age, sex, baseline CVD, diabetes, systolic blood pressure, diastolic blood pressure, and treatment of hypertension.

We followed participants for incident HF or death during a follow-up period of up to 12 years. Ten-year age- and sex-adjusted incidence rates of HF by LVEF category were estimated using the data-augmentation method (22). Cumulative incidence curves describing the occurrence of these outcomes by ejection fraction category were presented. The proportions of participants in whom HFPEF and HFREF developed at follow-up were determined for each LVEF group. The hazard ratio (HR) for these outcomes was compared among the categories of LVEF, with LVEF >55% serving as the referent group. We estimated age- and sex-adjusted and multivariable-adjusted proportional hazards models for each outcome, after confirming that the assumption of proportionality of hazards was satisfied for each outcome. We also examined continuous LVEF as a risk factor for these outcomes. Primary models were stratified by cohort type and presence versus absence of prevalent MI. To address possible confounding from time (as a result of changes in echocardiography quality, CVD risk factor prevalence, and CVD treatment during the study), we examined similar models that stratified by examination and prevalent MI.

We also conducted a secondary analysis examining the association of LVEF calculated by the biplane Simpson method, available in offspring participants at examination 8, with our outcomes of interest. The Pearson correlation coefficient, Bland-Altman method (23), and weighted kappa statistic (24) were used to assess the agreement between LVEF calculated by the de Simone and Simpson biplane methods. Final models were repeated using the robust Lin-Wei covariance estimator to account for clustering of multiple periods of observations within persons. All multivariable models were adjusted for age, sex, body mass index, baseline CVD, systolic blood pressure, use of antihypertensive treatment, current smoking, and prevalent diabetes; a separate analysis additionally adjusted for LV cavity size. Covariates were selected on the basis of review of published data and clinical judgment of their probable associations with LVEF and the outcomes of HF and death, as well as on availability of these covariates in FHS examinations. Furthermore, we conducted several sensitivity analyses to evaluate for consistency with our primary results: using propensity score matching for CVD risk factor variables included in multivariable analysis, excluding patients with prevalent MI, and accounting for clustered observations among participants. Restricted penalized cubic splines were fitted to assess the linearity of the relationships between continuous LVEF and the outcomes. Statistical significance was considered at 2-tailed p ≤ 0.05. However, in light of multiple significance tests of association, one should interpret modest p values (e.g., 0.005 < p <0.05) as denoting modest associations. All analyses were performed using SAS version 9.3 (SAS Institute, Cary, North Carolina).

Prevalence and correlates of borderline LVEF in the study sample

Online Figure 2 displays the distribution of LVEF in our study sample. The prevalence of LVEF categories was as follows: <50%: 1.6%; 5% to 55%: 3.5%; and >55%: 94.9%. The characteristics of participants with an LVEF of 50% to 55% in comparison with groups with an LVEF <50% and an LVEF >55% are presented in Table 1. Participants in whom LVEF assessment was unavailable had a greater burden of CVD risk factors but a prevalence of MI similar to that of participants with available echocardiograms (Online Table 2). The proportions of mean prevalence of CVD, MI, and diabetes, and brain natriuretic peptide level were intermediate in participants with an LVEF of 50% to 55% compared with participants with an LVEF <50% and an LVEF >55% (Table 1). Male sex, prevalent CVD, and higher blood pressure were associated with greater odds of having a borderline, compared with a normal, LVEF, whereas higher mean age and use of antihypertensive medications were inversely associated with a borderline LVEF (Table 2).

Incidence of composite outcome (HF/all-cause mortality) and heart failure by LVEF category

The composite primary outcome (HF or death) occurred in 1,255 (12%) participants (cumulative incidence per LVEF group shown in Table 3 and Figure 1). Participants with an LVEF of 50% to 55% had an age- and sex-adjusted composite event rate of 0.24 per 10 years of follow-up. These event rates were intermediate between corresponding events in those participants with LVEF <50% and LVEF >55%. The adjusted hazard ratios for the primary outcome for the groups with an LVEF<50% and an LVEF of 50% to 55%, as compared with the referent group with an LVEF >55%, are shown in Table 4. Participants with a borderline LVEF had a greater risk of the composite outcome in all models compared with participants with an LVEF >55%. In analyses stratified by examination and prevalent MI, we observed similar, slightly higher hazard ratios for the composite outcome in all LVEF groups compared with the referent (Online Table 3). Participants with an unavailable LVEF had a greater risk of HF/death than did participants with an available LVEF (age- and sex-adjusted HR: 1.45; 95% confidence interval [CI]: 1.33 to 1.57; p < 0.0001; multivariable HR: 1.30; 95% CI: 1.19 to 1.42; p < 0.0001).

On follow-up, new-onset HF occurred in 355 (3.5%) participants. Table 3 and Figure 2 show the cumulative incidence of HF according to baseline LVEF category. The HF event rate in participants with a borderline LVEF was 0.13 per 10 years follow-up, which was intermediate between those with reduced LVEF and normal LVEF. Participants with an LVEF of 50% to 55% had a >2-fold increased risk for HF compared with those with an LVEF >55% in all models (Table 4). Similar results were seen in analyses stratified by examination and prevalent MI (Online Table 3). The addition of brain natriuretic peptide to multivariable models also yielded similar results (Online Table 4).

Because LVEDD is associated with CVD, we examined the relations of LVEDD in our models. LVEDD was associated with HF/mortality and HF in multivariable-adjusted models not including LVEF as a covariate (for HF/mortality, HR:1.33; 95% CI: 1.18 to 1.50; p < 0.0001; for HF, HR: 2.25; 95% CI: 1.81 to 2.79; p < 0.0001), but inclusion in models including LVEF group only modestly attenuated the association of LVEF with the outcomes (Online Table 5).

Among participants in whom HF developed, the prevalence of HFPEF, HFREF, and interim MI are presented in Table 5. Of the participants with a baseline LVEF of 50% to 55% in whom HF developed and in whom LVEF was available, HFPEF developed in one-third, whereas HFREF developed in nearly two-thirds, a prevalence intermediate between those with an LVEF <50% and >55%. Interim MI (between echocardiography and follow-up) occurred in 78 (22%) participants in whom HF developed: n = 7 (24%) among those with a baseline LVEF <50%; n = 2 (8%) among those with an LVEF of 50% to 55%; and n = 69 (23%) among those with an LVEF >55%.

Risk of outcomes according to continuous LVEF

In multivariable-adjusted analyses, every 5% decline in LVEF (modeled as a continuous variable) was associated with a 12% and 23% increase in the risk for the composite outcome and HF, respectively (Table 4). Splines revealed a nearly linear relationship of the risk of both the composite outcome and HF with decreasing LVEF (Figures 3 and 4). Results of tests for nonlinearity of these associations were not statistically significant (p = 0.28 for HF or death; p = 0.11 for HF).

Secondary analyses using biplane Simpson’s LVEF

LVEF calculated by the de Simone and biplane Simpson methods correlated well (Pearson correlation 0.82; p < 0.0001) with good agreement (94% of observations falling within ±5%) (Online Figure 3). The weighted kappa statistic for categorical LVEF between the de Simone and biplane methods was 0.64, indicating very good agreement between the 2 methods.

HF or mortality occurred in 225 of 2,315 (10%) of FHS offspring participants. A borderline LVEF was associated with a >2-fold risk of HF/mortality (HR: 2.15; 95% CI: 1.04 to 4.45; p = 0.04). Each 5% decrement in LVEF was associated with a 17% risk of the composite outcome (p = 0.0004). There were 78 HF events. In logistic regression analyses, a trend toward risk of HF was seen, but it was not statistically significant (HR: 2.61; 95% CI: 0.91 to 7.50; p = 0.075). However, every 5% lower LVEF was associated with a 29% greater risk of HF (95% CI: 1.13 to 1.47; p = 0.0001).

Examination for sex-interaction and sensitivity analyses

We did not observe effect modification by sex for either the composite outcome of HF/mortality or HF (p = 0.60 and p = 0.10, respectively). In sensitivity analyses using propensity-score matching, similar results were observed (Online Table 6). In additional sensitivity analyses excluding participants with prevalent MI and accounting for clustered observations within participants, results were similar (data not shown).

Characteristics of the borderline LVEF group

The prevalence of HFPEF and of HFREF at follow-up among participants with a baseline LVEF of 50% to 55% was intermediate between participants with an LVEF <50% and those with an LVEF >55%. The cause of HF in participants with borderline LVEF is unclear, but it cannot be explained as a consequence of MI, which occurred in a very small proportion of the participants in whom HF developed. In contrast, among those with LVEF <50%, development of symptomatic HF may be more attributable to ischemic events. Our findings are consistent with earlier reports suggesting that preclinical LV systolic dysfunction may progress to clinical HF along a continuum of incrementally lower LVEF, with possible neurohormonal derangements that may lead to adverse changes in the cellular and extracellular environment (6,26–29).

Implications of small decrements in LVEF

Previous findings suggested that 3.0% to 7.3% of the population has asymptomatic LV systolic dysfunction (6,8), mostly mild LV systolic dysfunction (LVEF 45% to 54%). We observed a 3.5% prevalence of an LVEF of 50% to 55% that reflects a small but significant proportion of the general population. We noted a significantly increased risk for both HF and death with every 5% decline in LVEF. Whereas a 5% decrement in LVEF may seem to represent a small differential change, the implications at a population level may be large. For example, a 5% difference in LVEF represented a difference between LVEF quartiles in our sample. Thus, our findings may affect a significant proportion of the general population.

Comparison with previous studies and impact of our findings

Even a mildly decreased LVEF in asymptomatic persons is associated with HF and mortality (6–8,25,30). To our knowledge, the long-term prognosis of persons with a borderline LVEF is uncertain. Our finding of a risk for incident HF and death intermediate between LVEF <50% and >55% is consistent with earlier studies and extends the findings to a large population whose members currently may not receive attention because of their nearly normal LVEF. In addition, the rising risk of outcome events with progressively decreasing LVEF suggests a continuum of cardiovascular risk across the range of LVEF distribution. Clinically, a borderline LVEF may be referred to as “low-normal.” However, our results suggest that this term for an LVEF of 50% to 55% may be misleading, and it may not convey the increased adverse risks associated with borderline LV systolic function.

Our findings are supported by consistency between our primary results and several secondary and sensitivity analyses including use of the biplane Simpson LVEF measurement, exclusion of persons with MI at baseline, accounting for clustered observations within participants, additional adjustment for LV cavity size, and propensity score matching. Substantial changes in imaging techniques and in CVD risk factor prevalence and treatment have evolved over the 3 decades encompassed in this study. Nevertheless, our results comparing participants within the same examination cycle (stratification by examination status and MI) were similar, albeit stronger, than results of our primary analyses, thus strengthening our conclusions. In addition, we examined for and did not observe effect modification by sex in these associations.

The optimal therapy, if any, for patients with a borderline LVEF of 50% to 55% remains uncertain. Currently, the borderline group represents persons who, under current guidelines, do not merit medical surveillance or therapy for HF prevention compared with persons with an LVEF<40%. The absolute event rates in this group are low and would preclude a randomized controlled clinical trial targeting this group to define optimal management strategies, beyond aggressive control of prevalent CVD risk factors.

Study limitations

The FHS cohorts represent large, community-based samples in which echocardiographic measurements have been well validated and participants are under regular surveillance for the detection of events. In primary analyses, we used the method of de Simone for the estimation of LVEF that has been applied in numerous clinical studies (16–18); intercorrelation and intracorrelation measures of the echocardiographic measures were excellent (31). Although this method assesses change in LV diameter at the base, it can overestimate or underestimate LVEF if regional wall motion abnormalities exist in the apical or basal regions, respectively. However, these misclassification errors would likely have biased our findings toward the null hypothesis of no association of borderline LVEF with the risk of developing outcome events. Furthermore, we observed similar results using the biplane Simpson LVEF calculation, and our data suggest very good agreement between the de Simone and biplane Simpson methods of LVEF calculation.

LV diastolic dysfunction is a feature of HF, both HFPEF and HFREF. We were unable to evaluate the separate and distinct contribution of diastolic dysfunction to HF because serial diastolic function parameters were unavailable at the baseline examination we chose. However, it is often challenging to differentiate the extent to which clinical HF exacerbations result from diastolic, as compared with systolic, dysfunction in patients with HFREF. We assessed HF outcomes that included the spectrum of LVEF (both preserved and reduced LVEF subtypes). Future investigation may elucidate the extent to which diastolic dysfunction contributes to clinical HF and mortality in persons with a borderline LVEF.

Whereas our observations suggest greater risk for the borderline LVEF group in the population, the LVEF measures are less precise at the level of an individual patient. Future studies using modern imaging modalities with greater reproducibility may reduce measurement variability, increase the sample size with assessable LVEF, and validate our findings. However, our observed low absolute event rates necessitate a long follow-up period, and thus a similar study may not be feasible with newer echocardiographic methods at this time.

Finally, additional epidemiological considerations must be made. Although we included clinical risk factors for HF in our statistical models, we cannot exclude residual confounding from unmeasured variables. Given the observational nature of this study, causal inferences cannot be made. We highlight that the majority of our sample consisted of middle-aged to older adults of Northern European descent, thus limiting generalizability to persons of other races or ethnicities. It would be of interest to evaluate the prognosis of a borderline LVEF in different population samples, including those identified in the clinical setting and in ethnically and racially diverse groups.