Variability in Ejection Fraction Measured By Echocardiography, Gated Single-Photon Emission Computed Tomography, and Cardiac Magnetic Resonance in Patients With Coronary Artery Disease and Left Ventricular Dysfunction

Key Points Question What is the variability in left ventricular ejection fraction (LVEF) as measured by different cardiac imaging modalities? Findings In this multicenter diagnostic study of 2032 patients with coronary artery disease and LVEF of 35% or less with imaging interpreted by core laboratories, correlation of LVEF between modalities ranged from r = 0.493 (for biplane echocardiography and cardiovascular magnetic resonance) to r = 0.660 (for cardiovascular magnetic resonance and gated single-photon emission computed tomography). There was no systematic overestimation or underestimation of LVEF for any modality. Meaning There is substantial variability in LVEF assessment between modalities, which should be considered in trial design and clinical management.

Meaning There is substantial variability in LVEF assessment between modalities, which should be considered in trial design and clinical management.

Introduction
Feasible, accurate, and reproducible assessment of left ventricular ejection fraction (LVEF) is an important objective of noninvasive cardiac imaging. Whether LVEF is preserved or reduced currently forms the basis for the classification of patients with heart failure. Additionally, LVEF is an important predictor of prognosis in patients with myocardial infarction, 1-3 heart failure, [4][5][6] and valve disease. 7 Moreover, current practice guidelines use LVEF thresholds for decision making in different clinical scenarios, such as the recommendation regarding device implantation or pharmacologic therapy in patients with heart failure 8,9 and the recommendation for valve replacement in patients with severe valvular heart disease. 10 Left ventricular ejection fraction is also a common enrollment criterion and/or end point for clinical trials. 11 Left ventricular ejection fraction can be determined by using multiple noninvasive imaging modalities, including echocardiography, cardiac magnetic resonance (CMR) imaging, and gated single-photon emission computed tomography (SPECT) imaging. All of these methods are routinely used for clinical decision making as well as research study enrollment. However, few data exist regarding the agreement between LVEF determined by these different methods. Prior studies have been limited by small numbers of participants, sometimes including only healthy volunteers, with imaging performed by a single center, and have compared only 2 imaging modalities. [12][13][14] As LVEF cut points are often the basis for clinical management decisions and trial eligibility, the implications of variability are substantial.
The Surgical Treatment for Ischemic Heart Failure (STICH) was an international multicenter trial aimed to compare coronary artery bypass grafting (CABG) and medical therapy for patients with heart failure, coronary artery disease (CAD), and left ventricular (LV) systolic dysfunction defined as LVEF of 35% or less. 15,16 In this trial, any of 3 diagnostic methods (echocardiography, gated SPECT imaging, or CMR) could be used by a local clinical site to measure LVEF in order to determine a patient's trial eligibility. All patients enrolled in the STICH trial were required to have a baseline determination of LVEF, and a subset of patients underwent this determination by multiple modalities, including echocardiography, gated SPECT imaging, and/or CMR. All LVEF data obtained by echocardiography, CMR, and SPECT were measured by respective core laboratories. Therefore, the STICH trial provides a unique opportunity to correlate core laboratory assessment of LVEF data between different modalities.
We conducted this study to determine the variability among imaging modalities and among different echocardiographic methods for assessing LVEF in patients with reduced LV systolic function, and to compare the association between these measurements and subsequent mortality in patients with ischemic cardiomyopathy. requirements before enrollment. When studies did not meet these requirements, additional studies were requested until requirements were met. Patients with baseline imaging data received by the core laboratories for 1 or more imaging modalities were considered for inclusion. Patients with LVEF measured 90 days or more from study randomization or with study quality deemed by the core laboratory as being unusable for measurement were excluded. Determination of LVEF was performed by a separate core laboratory for each modality, independent of clinical information, treatment assignment, and data from other modalities. 17 Each core laboratory provided oversight of quality control and assessed the quality of each study as excellent, good, fair, borderline, or unusable. Left ventricular end-systolic volume for each modality was indexed for body surface area.

Calculation of LVEF
Left ventricular ejection fraction was determined from LV end-diastolic volume and end-systolic volume using the following standard formula:

Imaging
Echocardiography was performed at most sites for patient enrollment. Determination of LVEF was attempted according to the recommendations of the American Society of Echocardiography 18 using the Simpson method. 17 Measurements were averaged over 3 cardiac cycles for patients in sinus rhythm, and 3 to 5 cardiac cycles for those in atrial fibrillation. If 2 apical views were not available for LV volume measurement, only 1 apical view (single-plane Simpson measurement) was used for determination of LVEF. For Simpson measurements, the LV endocardial border was traced contiguously from one side of the mitral annulus to the other side excluding the papillary muscles and trabeculations. Left ventricular ejection fraction was also estimated visually in most patients and when the definition of the LV endocardial border was not satisfactory from any of the apical views, visual estimate was the only echocardiographic determination of LVEF. 19 Gated myocardial perfusion SPECT imaging, predominantly using sestamibi, was performed at clinical sites using a standard protocol. The gated raw projections were reconstructed by the radionuclide core laboratory using automatic software (AutoSPECT). When appropriate, an algorithm was applied to correct for motion. Resting studies accounted for 82% of measurements of LVEF; the remainder was obtained from poststress studies. Gated short-axis images were reviewed by a core laboratory technologist to optimize the accuracy of automatically determined LV contours, which were measured in end-systole and end-diastole. Manual adjustment addressed incorrect valve plane placement or contour deviations because of extracardiac radioactivity. Gated SPECT images were analyzed for LVEF by the radionuclide core laboratory using quantitative gated SPECT (QGS) software. 20 Cardiac magnetic resonance imaging was performed by clinical sites that had the required CMR platform and software. 17,21 A minimum of 2 data sets of short-and long-axis views were required to allow the core laboratory to select images of the highest quality. All gated data were displayed and reviewed by CMR core laboratory expert technologists to verify that LV boundaries were accurately demarcated. Short-axis images allowed determination of LVEF, using software developed by the CMR core laboratory at the University of Southern California (USC Cardio) and based on the Simpson method. Papillary muscles and trabeculations were considered to be part of the LV cavity. All shortaxis data were reviewed by a technologist and adjusted manually, if necessary, to optimize accuracy of LV contour borders.

Statistical Analysis
In this secondary study of the STICH trial, agreement of the core laboratory determinations of LVEF between modalities were assessed using 5 indices of variability: mean signed difference, mean absolute difference, Pearson correlation coefficient, Bland-Altman plots in which the mean of 2 with SPECT performed after stress could be influenced by ischemia or stunning, Bland-Altman plots were repeated after exclusion of the 18% with SPECT LVEF assessed following stress. To assess the impact of nonsimultaneous imaging, the number of days between performances of various modalities was considered. Finally, the prognostic effect of the different measures of LVEF for association with all-cause mortality was assessed using Cox regression models. In each case, the relationship of LVEF with mortality was modeled using restricted cubic spline functions. 24 The relative risk of every 5% LVEF increment, expressed as hazard ratio and 95% confidence interval, was also calculated using the Cox model. The longest available follow-up information was used for each patient. 16,25 This was a 2-sided test with a P value less than .05 required for significance.

Clinical Characteristics
The

Imaging
The median time interval between echocardiography and SPECT was 3.0 days (interquartile range, 1.0-9.0 days); between echocardiography and CMR was 2.0 days (interquartile range, 1.0-6.0 days); and between SPECT and CMR was 1.0 days (interquartile range, 1.0-5.0 days). All patients qualified for participation in the study based on LVEF of 35% or less as assessed by the recruiting site. Thus, in some patients, an alternative modality of determination of LVEF yielded a result of 35% or greater.
The mean (SD) for the LVEF by the core laboratories for each modality and for the various echocardiographic methods and the numbers and percentages with LVEF of 35% or less for each modality are shown in Table 2.  Bland-Altman comparisons between modalities for determining LVEF are shown in the Figure. Limits of agreement between modalities were broad, ranging from 28.27% to 35.31%. Bland-Altman comparison with SPECT repeated after exclusion of LVEF assessed following stress with SPECT were similar (eFigure in the Supplement). The mean signed difference between LVEF measured by biplane and single-plane echocardiography was closest to 0, indicating no substantial overestimation or underestimation of LVEF by either echocardiographic method. The mean signed differences for LVEF between modalities were larger with wider 95% confidence intervals.
Correlations between LVEF as determined by quantitative vs visual echocardiographic methods Left ventricular ejection fraction measurements were within 5% in 54.0% for biplane echocardiography and SPECT, 48.5% for SPECT and CMR, and 43.1% for biplane echocardiography and CMR. Using CMR as the standard and comparing it with SPECT and echocardiography as to whether there was intermodality agreement for LVEF greater than 35% is shown in Table 3. The results of the 4 variability indices comparing LVEF and end-systolic volume index for the 3 modalities are summarized in Table 4.

Discussion
Left ventricular ejection fraction refers to the fraction of LV end-diastolic volume ejected during systole. It is the most widely used measure of assessment for LV systolic function and is familiar to   Abbreviations: CMR, cardiovascular magnetic resonance; EF, ejection fraction; LVEF, left ventricular ejection fraction; SPECT, single-photon emission computed tomography. patients and clinicians. This is the first study to compare echocardiographic, CMR, and SPECT methods for determination of LVEF in a large, international, multicenter cohort of patients included in a clinical trial with extensive follow-up. In this population with LV dysfunction and CAD, we found that there was substantial variation among modalities for determination of LVEF even though these measures were made by specialized core laboratories, each of which followed specific plans for image analysis and measurements. 19,26 Acquisition of data at 127 sites worldwide may have contributed to this variability. Moreover, variation was not predictable; there was no substantial overestimation or underestimation of LVEF by any modality relative to another.

JAMA Network Open | Cardiology
Surprisingly, few studies have assessed the differences in LVEF as measured by different imaging modalities. [12][13][14]27 Previous studies were predominantly single center, most with fewer than 100 participants, and most recent studies have focused on 3-dimensional or contrast echocardiography. These newer techniques were associated with improved reproducibility and agreement with CMR. 28,29 Correlations between various methods of determination of LVEF by a single modality, echocardiography, were similar and better (r = 0.898 for biplane and visual estimation; r = 0.874 for single plane and biplane; and r = 0.876 for single plane and visual) than were correlations between different modalities, which ranged from r = 0.493 (for biplane echocardiography and CMR) to r = 0.660 (for CMR and SPECT). The good correlation between visual estimation and measurement of echocardiographic images is of interest; however, it should be noted that visual estimates were made by echocardiographic core laboratory staff, who had advanced training and extensive experience. Results might be worse with less experienced reviewers. Although the 3 echocardiographic methods assessed were well correlated, LVEF by biplane method of disks should be used when feasible as it correlated best with LVEF by other modalities.
Bland-Altman analysis showed no substantial overestimation or underestimation of LVEF by different modalities. Biplane quantitation with echocardiography averaged only 2.5% higher than CMR and 2.2% higher than gated SPECT, and SPECT was 0.8% higher than CMR. Limits of agreement were broad. Variability between modalities for measures of LV end-systolic volume index were also broad.
Variation of LVEF within 5% between modalities might be considered clinically acceptable.
However, the percentage of observations that fell within a range of 5% ranged from 43% to 54% between different imaging modalities. Discordance between modalities as to whether LVEF was greater than 35% was present in about 20% to 25% of cases. Had core laboratory determination of LVEF been used for enrollment in the STICH trial, which required LVEF of 35% or less, many of these patients would have been ineligible.

Conclusions
Although LVEF is a widely reported measurement and is the cornerstone of many treatment decisions, there is substantial variability in its measurement using different modalities, even when assessed by core laboratories. In this international study in which LVEF imaging was performed at 127 clinical sites using up to 3 widely used imaging modalities and LVEF was independently measured by core laboratories according to standard protocols, the variability in LVEF measurement exceeded 5% in about half of the patients. Variability was less for different methods of determining LVEF when a single imaging modality (echocardiography, in this case) was used. Longitudinal assessments of a given patient may best be accomplished using a single imaging modality.
The diagnostic and prognostic importance of LVEF as well documented in numerous studies is not disputed. Left ventricular ejection fraction by each modality was associated with mortality.
However, variability in LVEF assessment by different imaging modalities should be considered in trial design and clinical management. Considering this variability, cut points in LVEF should not be the sole basis for decision making.