- Email: [email protected]
Correspondence of left ventricular ejection fraction determinations from two-dimensional echocardiography, radionuclide angiography and contrast cineangiography
JACC Vol. 25, No. 4 March 15, 1995:937-42
937
METHODS
Correspondence of Left Ventricular Ejection Fraction Determinations From Two-Dimensional Echocardiography, Radionuclide Angiography and Contrast Cineangiography M A N I S H M. NAIK, MD, G E O R G E A. D I A M O N D , MD, FACC, T I N A PAl, MD, A R I E L S O F F E R , MD, R O B E R T J. SIEGEL, MD, F A C C Los Angeles, California
Objectives. This study assessed the agreement of left ventricular ejection fraction determinations from two-dimensional echocardiography, radionuclide angiography and contrast cineangiography. Background. Previously published reports suggest that twodimensional echocardiography, radionuclide angiography and contrast cineangiography are equally acceptable methods of assessing left ventrieular ejection fraction on the basis of high coefficients of correlation. However, correlation of methods does not necessarily imply agreement. Methods. In a prospective analysis, 25 consecutive subjects all had two-dimensional echocardiography and radionuclide angiography performed within 10 days of each other in the cardiology department of a metropolitan community hospital. A retrospective computer search (Medline) revealed seven studies, using the coefficient of correlation (r), comparing two-dimensional echocardiographic left ventricular ejection fraction (n = 268) with radionuclide angiographic (n = 174) or contrast cineangiographic (n = 119) left ventricular ejection fractions. Results. The eight individual studies (n = 293) comparing
two-dimensional echocardiography with either radionuclide angiography or contrast cineangiography exhibited coefficients of correlation ranging from 0.78 to 0.93. Agreement analysis using the method of Bland and Altman was performed by averaging the results obtained from the two techniques and determining how disparate any single ejection fraction was (with 95% confidence limits) from the mean value. Agreement ranged from 23% to 42% around the mean ejection fraction. The average lack of agreement between the two methods for all studies involved was 17%, with an average r value of 0.86. Conclusions. Left ventricular ejection fraction determinations by means of two-dimensional echocardiography, radionuclide angiography and contrast cineangiography exhibit high correlation and only moderate agreement. High correlation does not always imply high agreement. These results suggest that, when validated by agreement analysis, multiple studies may not be necessary in appropriate clinical situations, potentially reducing costs.
Left ventricular ejection fraction is an important clinical variable with respect to diagnosis, prognosis and treatment in various clinical situations. Currently, there are three commonly used methods for determining left ventricular ejection fraction: 1) two-dimensional echocardiography, 2) radionuclide angiography, and 3) contrast cineangiography. Because the clinical situation may dictate using one method versus another, it is important for the clinician to know whether the results of ejection fraction estimates are comparable among the three methods and if they can be used interchangeably. If the answer is yes, this could reduce the need for multiple tests and thereby reduce the cost of health care to the patient. Previously published reports (1-3) suggest that twodimensional echocardiography, radionuclide angiography and
contrast cineangiography are equally acceptable methods of assessing left ventricular ejection fraction. This conclusion is based on high coefficients of correlation. However, correlation of two methods does not necessarily imply agreement (4,5). For example, if two methods of calculating ejection fraction were consistently 20% apart, they would exhibit high correlation but poor agreement. Therefore, although previous reports demonstrate high correlation among two-dimensional echocardiography, radionuclide angiography and contrast cineangiography, to our knowledge no previous report specifically analyzes the agreement among the values obtained by each of these methods. In seeking to analyze the agreement for the three methods of determining ejection fraction, a twofold approach was taken. First, a prospective study was designed to determine agreement of left ventricular ejection fractions determined by two-dimensional echocardiography and radionuclide angiography at our institution. Second, an analysis was performed of the available published data comparing left ventricular ejection fractions by two-dimensional echocardiography with radionuclide angiography or contrast cineangiography. The published
From the Division of Cardiolo~,, Department of Medicine, Cedars-Sinai Medical Center, Los Angeles, California. Manuscript received May 16, 1994; revised manuscript received October 26, 1994, accepted November 9, 1994. Address for correspondence: Dr. Robert J. Siegel, Division of Cardiology, Cedars-Sinai Medical Center, 8700 Beverly Boulevard, Room 5314, Los Angeles, California 90048-0750. ©1995 by the American College of Cardiology
(J Am Coil Cardiol 1995;25:937-42)
0735-1097/95/$9.50 0735-1097(94)00506-L
938
NAIK ET AL. C O R R E L A T I O N OF EJECTION FRACTION DETERMINATIONS
data were reanalyzed to evaluate agreement among the different methods.
Methods Prospective analysis. Twenty-five consecutive patients meeting predefined inclusion criteria (15 men, 10 women; mean age 68.6 years, range 33 to 90) were entered into the study. Nineteen patients had a previous myocardial infarction with inferior posterior location in 12, anterior location in 7; 2 had dilated cardiomyopathy; 4 had no ventricular wall motion abnormalities. All patients fulfilled the following criteria for inclusion in this study: 1) both radionuclide angiographic and two-dimensional echocardiographic imaging were performed within 10 days of one another; and 2) the two-dimensional echocardiogram displayed sufficient endocardial definition to allow tracing of the outline of the ventricular cavity. The criterion for sufficient endocardial definition was visualization of 70% to 80% of the endocardium. However, this definition of endocardial visualization was somewhat subjective because the assessment was made by observing the endocardium in motion to fill in the gaps created by loss of resolution of still images. Three beats were computed and averaged. The interbeat variability was <5%. Four patients were excluded from the study because of technically inadequate echocardiograms. All patients in our study underwent radionuclide angiography within a mean (_SD) of 2 _+ 2.5 days (range 0 to 9) of two-dimensional echocardiography. The studies were interpreted independently of each other, in blinded manner. Two-dimensional echocardiograms were obtained using a Hewlett-Packard echocardiographic machine. Apical two- and four-chamber views were utilized. The MicroSonics Digital Echo Analyzer was used to analyze the echocardiograms. The fundamental components of the MicroSonics Analyzer are an image digitizer and a microcomputer (IBM AT) programmed to process and quantify two-dimensional echocardiograms. The digitizer converts analog video images to digital format. The MicroSonics Analyzer allowed the selection of a single, representative cardiac cycle (or a portion thereof), free of motion artifact and respiratory interference, for capture and storage on a floppy disk and subsequent evaluation. The captured cycle could then be played back in a continuous loop. The portion of the cardiac cycle of interest was that from end-diastole to end-systole; this allowed calculation of the maximal and minimal left ventricular volumes and left ventricular ejection fraction. The images were captured at a sequence of 8 frames/s, which could be viewed in motion or frame by frame. To calculate left ventricular volumes, the frame displaying the appropriately sized left ventricular volume was chosen, and the penlight was used to trace the endocardial outline and obtain other measurements, such as the long-axis length. In some cases, the endocardium was not clearly visible at all points; in these cases, after viewing of all frames, the best approximation was used. The MicroSonics Analyzer calculated the left ventricular volumes using a Simpson rule biplane formula, which divides the left ventricle into eight cylindric
JACC Vol. 25, No. 4 March 15, 1995:937-42
slices, and obtained the left ventricular ejection fraction from these volumes. For the radionuclide angiographic procedure, each patient received an injection of 25 mCi of autologous red cells labeled in vitro with technetium-99m. A gamma camera equipped with an all-purpose collimator was positioned in the left anterior oblique view with the exact angulation (40° to 50°) determined as that which best separated the left and right ventricles. Multiple-gated equilibrium blood pool scintigraphy was performed by acquisition of 20 frames of equal duration distributed uniformly over the entire cardiac cycle for a total of 2 rain/acquisition, resulting in -100,000 counts/frame. Count changes within a left ventricular region of interest were used to identify end-diastolic and end-systolic frames. Ejection fraction was then calculated by dividing stroke counts (end-diastolic minus end-systolic) by the background-corrected end-diastolic counts. Our prospective study did not include contrast cineangiography, but the studies reviewed retrospectively all used a similar method for determination of left ventricular ejection fraction. The studies used either a single-plane right anterior oblique view or biplane right and left anterior oblique views to examine a single sinus beat, excluding any premature beats. Left ventricular ejection fractions were calculated using the method of Sandler and Dodge (1,3,6-8). Retrospective analysis. A computer-assisted Medline search of published medical reports was conducted to find studies that compared two-dimensional echocardiographic left ventricular ejection fraction with radionuclide angiographic or contrast cineangiographic left ventricular ejection fraction. For inclusion in the analysis, the study must have a direct comparison of the left ventricular volumes from two-dimensional echocardiography with those from radionuclide angiography or contrast cineangiography. All studies that qualified for the analysis used the coefficient of correlation for comparison of left ventricular ejection fractions between the different methods. Three studies were found that compared two-dimensional echocardiography and radionuclide angiography (2,3,6); four compared two-dimensional echocardiography and contrast cineangiography (1,3,7,8). Left ventricular ejection fraction and linear regression analysis data were obtained from these reports. When left ventricular ejection fraction values were not explicitly provided, they were obtained from graphs included in the reports (3,6). A regression analysis was performed for each set of data to obtain a value for r, the coefficient of correlation. The "limits of agreement" described by Bland and Altman (4) were calculated with 95% confidence limits to assess the agreement between the left ventricular ejection fractions from twodimensional echocardiography, contrast cineangiography and radionuclide angiography. The method of Bland and Altman summarizes the lack of agreement by calculating the bias estimated by the mean difference (d) and the standard deviation of the difference (s). If the differences are normally distributed (Gaussian), 95% of the differences would be expected to lie between d - 2s and d + 2s (limits of agreement).
JACC Vol. 25, No. 4 March 15, 1995:937-42
NAIK ET AL. CORRELATION OF EJECTION FRACTION DETERMINATIONS
939
1.0 z
o.9. I-. 0 < n.1.1.
"~ z rr
0.8
I
w
2:
D
o. 9. 0.6 I-
oo
t'kl
o °
"-"
O
LL
LLI
w >
W a
• -
0
8
• O
•
•
4 0
J
• ° •
0.4
•
z__ - 4 w 0 z
0 Z
16 12
0.2
•
-8
•
•
•
•
w -12 £g
m
< rr
m -16 0 u--I u_ i
0.0 0.0
0.4
0.2
ECHO
EJECTION
i
L
0.6
0.8
I
I
I
20
I
I
40
M E A N LVEF B Y 2DE & R N A (%)
1.0
Figure 2. Difference in left ventricular ejection fraction (LVEF) from
FRACTION
Figure 1. Radionuclide angiographic versus two-dimensional echocardiographic (ECHO) left ventricular ejection fraction (r = 0.93).
Furthermore, because the range of values in our comparison is not 0 to infinity but rather 0 to 1.00 (range of ejection fractions), the limits of agreement narrow as one reaches either extreme of the range of ejection fractions. Intraobserver and interobserver variability were derived by calculating the mean difference between two observations and then the overall average mean difference, standard deviations and 95% confidence limits.
Results Comparison of two-dimensional echocardiography and ra. dionuclide angiography. Linear regression analysis of the data
in the prospective study of two-dimensional echocardiography and radionuclide angiography, as shown in Figure 1, revealed that r = 0.93, indicating good correlation. In addition, calculation of the limits of agreement with a 95% confidence interval yielded a lower limit of -11.5% and an upper limit of +11.7%, as shown in Figure 2. Thus, in our laboratory, a left
two-dimensional echocardiography (2DE) and radionuclide angiography (RNA) versus mean of the two values, with 95% confidence limits. Dotted lines = mean difference _+2 SD.
ventricular ejection fraction determined by two-dimensional echocardiography is consistently expected to be within 24% of the left ventricular ejection fraction obtained by radionuclide angiography in 95% of cases. Table 1 identifies the limits of agreement for the eight studies comparing two-dimensional echocardiography with radionuclide angiography and contrast cineangiography in 268 patients. As Table 1 shows, data from our laboratory yielded the highest r value (0.93) and the narrowest limits of agreement (+11.7 and -11.5). However, Table 1 also shows that high r values do not necessarily correlate with narrow limits of agreement. For example, the data from Carr et al. (7) also show an r value of 0.93 but broader limits of agreement of +15.3 and -18.3. In addition, the data from Starling et al. (6) yielded the worst upper limit of agreement (+26.2) but an intermediate r value of 0.81. In contrast, although the data from Folland et al. (2) had a narrower range of agreement, the r value was lower than that of Starling et al. (6). Table 1 clearly
Table 1. Limits of Agreement and Coefficient of Correlation Values for All Studies r
Study (ref. no.)
No. of Pts
d
SD
ULA
A
LLA
Value
- 11.5 17.4 -12.5 -15.4
0.93 0.75 0.93 0.81
-18.3 -23.2 -10.7 -15.3
0.93 0.78 0.87 0.89
Two-Dimensional Echocardiography Versus Radionuclide Angiography Cedars-Sinai Folland et al. (2) Quinones et al. (3) Starling et al. (6)
25 35 55 59
0.1 1.4 1.6 5.4
5.6 9.2 7.0 10.3
11.7 20.2 15.7 26.2
11.6 18.8 14.1 20.8
Two-Dimensional Echocardiography Versus Cineangiography Carr et al. (7) Folland et al. (2) Schiller et al. (1) Stamm et al. (8)
22 35 34 28
I
60
-1.5 -3.8 5.8 1.1
8.1 9.5 8.1 8.0
15.3 15.6 22.3 17.5
16.8 19.4 16.5 16.4
Averaged lack of agreement (A) = (Upper limit of agreement [UEA] - Lower limit of agreement [LLA])/2 = 16.8. Averaged r = 0.86. d = average deviation; Pts = patients.
940
NAIK ET AL. C O R R E L A T I O N OF EJECTION FRACTION DETERMINATIONS
JACC Vol. 25, No. 4 March 15, 1995:937-42
100
80
60
4o
20
17.5
15.a ~
lS.S
I ~ 0
:~a~:.
-20 - ,1.5
-17.4 -1a.5 -15.4
-le.3~~ ~_1o.7 N -15.3
Figure 3. Limits of agreement for all studies comparing two-dimensional echocardiographicleft ventricular ejection fractionswith radionuclideangiographic (open bars) tions.°r contrast cineangiographic(stippled bars) ejectionfrac-
-23.2 CSMC -40
Quinones Fo=and
Can" Starling
Schiller Folland
Sl~r,m
-60
-80
-100
shows that little concordance exists between r values and agreement for the studies reviewed. Figure 2 (Cedars-Sinai data) shows the difference in left ventricular ejection fraction between two-dimensional echocardiography and radionuclide angiography plotted against the average left ventricular ejection fraction by the two methods. The absence of a pattern, or trend, in the placement of the points in Figure 2 indicates that the magnitude of difference of the left ventricular ejection fractions between the two methods is not related to the absolute magnitude of the left ventricular ejection fractions being measured. This is supported by linear regression analysis of the data that yields a very. low r value of 0.03. The same graphic and regression analyses performed on the other sets of data yielded similar results. Agreement analysis of left ventricular ejection fraction. Figure 3 shows a graphic representation of each study's limits of agreement and how they compare with each other. Limits of agreement ranged from a lower limit of -23.2 to an upper limit of +26.2. For the entire series of studies, the variance between methods was <26% units (95% confidence interval), pointing out a relatively narrow range. Agreement ranged from 23% to 42% around the mean ejection fraction as shown. Figure 3 also graphically exhibits the lack of concordance between r values and the limits of agreement. Figure 4 (left) is a series of graphs combining data from our institution and all of the studies reviewed. It shows that the graphic display of data from all studies is similar, consistent with the high coefficients of correlation (r) found in each
individual study. The r values for the cumulative data were 0.85 (n = 119) for contrast cineangiographic versus echocardiographic ejection fraction, 0.87 (n = 178) for contrast cineangiographic versus radionuclide angiographic ejection fraction and 0.86 (n = 174) for radionuclide angiographic versus echocardiographic ejection fraction. An ejection fraction of 40% by echocardiography could range anywhere from 20% to 60% (from abnormal to normal) by radionuclide angiography, despite a correlation coefficient of 0.86. Figure 4 (right) demonstrates cumulative data from the studies analyzed using the agreement technique. Although the coefficient of correlation describes the relation between two sets of values, Figure 4 (right) graphically shows that it does not provide any information on how similar the values are to one another.
Discussion Clinical implications of agreement analysis. Many published studies have based the utility of different methods on estimates of comparative accuracy using correlation coefficients. Meta-analysis of the eight studies reviewed shows that the utility of correlation coefficients has limitations and that agreement analysis would be a more appropriate method for comparison. This is clear if the results of our data are applied to a clinical example. If one were to complete a preoperative evaluation for coronary artery bypass grafting or cardiac valve replacement, a test would be performed to estimate the ejection fraction to risk stratify the patient. If one obtained an
J A C C Vol. 25, No. 4 March 15, 1995:937 42
1.0
(9
0.50
08
•
O T
/ .-,..,,, .-.:-;/ /.. ,i.,.".. ~"
i
•
0.6
•
•
alk ,,
0.25
•
'Be 0,00
W
/."
"zU
• "u,//
0.2
Z
0.0 00
0.2
0 4
06
0.8
°
-I
0°
•
e~
~tlP='%~,,-
"
•
"
-0.25
-0.50
1.0
0.0
ECHO EJECTION F R A C T I O N
0.2
0.4
0.6
0.8
1.0
AVERAGE EJECTION FRACTION
1.0
F i g u r e 4. Combined regression (left) agreement analysis (right)
00
I..K '."~.t-p.
.
<
0.4
models for all studies. Top, Echocardiography (x) versus angiography (y): y = 0.075 + 0.858x, r = 0.848, SEE 0.090, p < 0.0001. Middle, Radionuclide angiography (x) versus angiography (y): v = 0.123 + 0.836x, r = 0.873, SEE 0.097, p < 0.0001. Bottom, Echocardiography (x) versus radionuclide angiography (y): y = 0.045 + 0.848x, r = 0.861, SEE 0.084, p < 0.0001.
941
N A I K E T AL. CORRELATION OF EJECTION FRACTION DETERMINATIONS
0.50
0.8 ~ Z
o_
0.25 •
°
°
0.6 =
u
:" "
,~ o
-.,..-. -.:.. r ~ . ~ + . : . . "
0.00
C)
0.4
Ilg ~ -0.25
02
7 0,0 0.0
02
0.4
06
08
-o.50
10
0.0
0.2
0.4
0.5
0.8
1.0
RADIONUCLIDE EJECTION F R A C T I O N AVERAGE EJECTION FRACTION 0.50
1+0
0.8 0
0.25
•"- . . ~, - :o
06
g
•
0.00 0.4
/ .+,-.<:"U""
•
°eeoc|
e
~ , -•~ .
O_ r~
e° e
.
o o°
•
~ eo
-0.25
0,2
0,0 0.0
--o , n - ~ ; ~-o- =i/~,-o - -
0.2
0 4
0 6
OB
1.0
-0.50 0.0
0.2
0,4
0.6
O.B
ECHO EJECTION F R A C T I O N AVERAGE EJECTION FRACTION
ejection fraction of 40% by echocardiography, Figure 4 (left) makes it evident that by radionuclide angiography this ejection fraction may be anywhere from 20% to 60%, the corresponding values having very different clinical implications for the patient. In this situation, one might elect to obtain an ejection fraction by a second method to have more confidence in the estimate. However, if one were to obtain an ejection fraction of 70% or 15% initially, the implications of an error of _+20% may not be significant, and one could omit obtaining ejection fraction by means of a second method. These decisions would depend on two factors: 1) how precise a measurement of ejection fraction is required for the given clinical situation, and 2) the limits of agreement among different methods in the laboratory being utilized.
Observer variability. Part of the difference in ejection fraction determined by different methods may be accounted for by intraobserver and interobserver variability. In our echocardiography laboratory, the mean difference between two observations by the same observer was 4.4%, with a standard deviation of 2.5% and 95% confidence intervals of 1.1% to 13.2%. Between two observers, the mean difference was 6.1%, with a standard deviation of 3.0% and 95% confidence intervals of 1.9% to 15.9%. Similarly, radionuclide angiography in our laboratory had an intraobserver variability of 2.5% and an interobserver variability of 6.8%. Standard deviations and confidence limits were not calculated for radionuclide angiography. For the data in the studies retrospectively reviewed, the intraobserver variability ranged from 1.4% to 5.1% for echo-
942
NAIK ET AL. CORRELATION OF EJECTION FRACT1ON DETERMINATIONS
cardiography and 2.0% to 5.1% for radionuclide angiography. Interobserver variability ranged from 2.3% to 8.1% for echocardiography and 2.0% to 8.1% for radionuclide angiography. Only Stamm et al. (8) reported intraobserver and interobserver variability data for cineangiography that were 4.3% and 6.7%, respectively. Taking observer variability into account, the data from our laboratory, exhibiting the narrowest limits of agreement (=12%), suggest that two-dimensional echocardiography is a reliable method for estimating left ventricular ejection fraction. However, in several other studies, the limit of agreement was >20%, and, as stated earlier, this may not be clinically acceptable. For situations in which serial follow-up of left ventricular function and ejection fraction may have clinical importance (such as evaluation of adriamycin toxicity, timing of aortic or mitral valve replacement, follow-up of cardiac transplantation candidates, post-myocardial-infarction patients and those in whom left ventricular contractility is being evaluated for response to drug therapy), the reliability of the estimate of left ventricular ejection fraction is crucial to clinical decision making. If one were simply to look at the high correlation coetficients of the different methods as a judge of their reliability, the data could be misleading. In these situations one has to consider two components of possible error: the variation between methods and the variation between serial determinations. For example, the error between radionuclide angiography and echocardiography in our laboratory is 12%. The data from Kuecherer et al. (9) showed that the variation of serial determinations of echocardiographic ejection fractions was 6.6%. The combined error (by addition in quadrature) would be 13.7%. The importance of agreement analysis in the clinical situations described is evident. Two-dimensional echocardiographic instruments have improved since the late 1970s, when the original studies comparing echocardiographic with radionuclide and cineangiographic left ventricular ejection fractions were performed. In addition, analytic algorithms for modeling the left ventricle have been refined. This may account for at least part of the difference between the results in our laboratory and those retrospectively reviewed. Conclusions. The coetficient of correlation has important limitations as an index of comparative accuracy of measures, such as left ventricular ejection fractions by two-dimensional echocardiography, radionuclide angiography and contrast
JACC Vol. 25, No. 4 March 15, 1995:937-42
cineangiography. However, analysis of agreement provides a more appropriate numeric estimate of accuracy. On the basis of such analyses, left ventricular ejection fraction determinations by two-dimensional echocardiography, radionuclide angiography and contrast cineangiography exhibit high correlation but only moderate agreement. The magnitude of agreement may be adequate for gross determinations of cardiac function, distinguishing among severely reduced, moderately reduced and normal left ventricular ejection fractions, but it may not be adequate for more precise determinations of ejection fraction (8). Ideally, each institution should validate its own data by analysis of agreement rather than correlation. These results suggest that, when validated by agreement analysis, multiple studies may not be necessary in appropriate clinical situations, potentially reducing costs. We gratefully acknowledge the technical assistance of Dr. Toshihiko Nishioka and Dr. Huai Luo.
References 1. Schiller NB, Acquatella H, Ports TA, et al. Left ventricular volume from paired biplane two-dimensional echocardiography. Circulation 1979;60:54755. 2. Folland ED, Parisi AF, Moynihan PF, Jones DR, Feldman CL, Tow DE. Assessment of left ventricular ejection fraction and volumes by real time, two-dimensional echocardiography. Circulation 1979;60:760-6. 3. Quinones MA, Waggoner AD, Reduto LA, et al. A new simplified and accurate method for determining ejection fraction with two-dimensional echocardiography. Circulation 1981;64:744-53. 4. Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1986;1:30%10. 5. Bookbinder MMJ, Panosian KJMJ. Using the coelficient of correlation in method comparison studies. Clin Chem 1987;33:1170-6. 6. Starling MR, Crawford MH, Sorenson SG, Levi B, Richards KL, O'Rourke RA. Comparative accuracy of apical biplane cross-sectional echocardiography and gated equilibrium radionuclide angiography for estimating left ventricular size and performance. Circulation 1981;63:1075-84. 7. Carr KW, Engler RE, Forsyth JR, Johnson AD, Gosink B. Measurement of left ventricular ejection fraction by mechanical cross-sectional echocardiography. Circulation 1979;59:1196-206. 8. Stamm RB, Carabello BA, Mayers DL, Martin RP. Two-dimensional echocardiographic measurement of left ventricular ejection fraction: prospective analysis of what constitutes an adequate determination. Am Heart J 1983; 104:136-44. 9. Kuecherer HF, Kee LL, Modin G, Cbeitlin MD, Schiller NB. Echocardiography in serial evaluation of left ventricular systolic arid diastolic function: importance of image acquisition, quantitation, and physiologic variability in clinical and investigational applications. J Am Soc Echocardiogr. 1991;4: 203-14.
937
METHODS
Correspondence of Left Ventricular Ejection Fraction Determinations From Two-Dimensional Echocardiography, Radionuclide Angiography and Contrast Cineangiography M A N I S H M. NAIK, MD, G E O R G E A. D I A M O N D , MD, FACC, T I N A PAl, MD, A R I E L S O F F E R , MD, R O B E R T J. SIEGEL, MD, F A C C Los Angeles, California
Objectives. This study assessed the agreement of left ventricular ejection fraction determinations from two-dimensional echocardiography, radionuclide angiography and contrast cineangiography. Background. Previously published reports suggest that twodimensional echocardiography, radionuclide angiography and contrast cineangiography are equally acceptable methods of assessing left ventrieular ejection fraction on the basis of high coefficients of correlation. However, correlation of methods does not necessarily imply agreement. Methods. In a prospective analysis, 25 consecutive subjects all had two-dimensional echocardiography and radionuclide angiography performed within 10 days of each other in the cardiology department of a metropolitan community hospital. A retrospective computer search (Medline) revealed seven studies, using the coefficient of correlation (r), comparing two-dimensional echocardiographic left ventricular ejection fraction (n = 268) with radionuclide angiographic (n = 174) or contrast cineangiographic (n = 119) left ventricular ejection fractions. Results. The eight individual studies (n = 293) comparing
two-dimensional echocardiography with either radionuclide angiography or contrast cineangiography exhibited coefficients of correlation ranging from 0.78 to 0.93. Agreement analysis using the method of Bland and Altman was performed by averaging the results obtained from the two techniques and determining how disparate any single ejection fraction was (with 95% confidence limits) from the mean value. Agreement ranged from 23% to 42% around the mean ejection fraction. The average lack of agreement between the two methods for all studies involved was 17%, with an average r value of 0.86. Conclusions. Left ventricular ejection fraction determinations by means of two-dimensional echocardiography, radionuclide angiography and contrast cineangiography exhibit high correlation and only moderate agreement. High correlation does not always imply high agreement. These results suggest that, when validated by agreement analysis, multiple studies may not be necessary in appropriate clinical situations, potentially reducing costs.
Left ventricular ejection fraction is an important clinical variable with respect to diagnosis, prognosis and treatment in various clinical situations. Currently, there are three commonly used methods for determining left ventricular ejection fraction: 1) two-dimensional echocardiography, 2) radionuclide angiography, and 3) contrast cineangiography. Because the clinical situation may dictate using one method versus another, it is important for the clinician to know whether the results of ejection fraction estimates are comparable among the three methods and if they can be used interchangeably. If the answer is yes, this could reduce the need for multiple tests and thereby reduce the cost of health care to the patient. Previously published reports (1-3) suggest that twodimensional echocardiography, radionuclide angiography and
contrast cineangiography are equally acceptable methods of assessing left ventricular ejection fraction. This conclusion is based on high coefficients of correlation. However, correlation of two methods does not necessarily imply agreement (4,5). For example, if two methods of calculating ejection fraction were consistently 20% apart, they would exhibit high correlation but poor agreement. Therefore, although previous reports demonstrate high correlation among two-dimensional echocardiography, radionuclide angiography and contrast cineangiography, to our knowledge no previous report specifically analyzes the agreement among the values obtained by each of these methods. In seeking to analyze the agreement for the three methods of determining ejection fraction, a twofold approach was taken. First, a prospective study was designed to determine agreement of left ventricular ejection fractions determined by two-dimensional echocardiography and radionuclide angiography at our institution. Second, an analysis was performed of the available published data comparing left ventricular ejection fractions by two-dimensional echocardiography with radionuclide angiography or contrast cineangiography. The published
From the Division of Cardiolo~,, Department of Medicine, Cedars-Sinai Medical Center, Los Angeles, California. Manuscript received May 16, 1994; revised manuscript received October 26, 1994, accepted November 9, 1994. Address for correspondence: Dr. Robert J. Siegel, Division of Cardiology, Cedars-Sinai Medical Center, 8700 Beverly Boulevard, Room 5314, Los Angeles, California 90048-0750. ©1995 by the American College of Cardiology
(J Am Coil Cardiol 1995;25:937-42)
0735-1097/95/$9.50 0735-1097(94)00506-L
938
NAIK ET AL. C O R R E L A T I O N OF EJECTION FRACTION DETERMINATIONS
data were reanalyzed to evaluate agreement among the different methods.
Methods Prospective analysis. Twenty-five consecutive patients meeting predefined inclusion criteria (15 men, 10 women; mean age 68.6 years, range 33 to 90) were entered into the study. Nineteen patients had a previous myocardial infarction with inferior posterior location in 12, anterior location in 7; 2 had dilated cardiomyopathy; 4 had no ventricular wall motion abnormalities. All patients fulfilled the following criteria for inclusion in this study: 1) both radionuclide angiographic and two-dimensional echocardiographic imaging were performed within 10 days of one another; and 2) the two-dimensional echocardiogram displayed sufficient endocardial definition to allow tracing of the outline of the ventricular cavity. The criterion for sufficient endocardial definition was visualization of 70% to 80% of the endocardium. However, this definition of endocardial visualization was somewhat subjective because the assessment was made by observing the endocardium in motion to fill in the gaps created by loss of resolution of still images. Three beats were computed and averaged. The interbeat variability was <5%. Four patients were excluded from the study because of technically inadequate echocardiograms. All patients in our study underwent radionuclide angiography within a mean (_SD) of 2 _+ 2.5 days (range 0 to 9) of two-dimensional echocardiography. The studies were interpreted independently of each other, in blinded manner. Two-dimensional echocardiograms were obtained using a Hewlett-Packard echocardiographic machine. Apical two- and four-chamber views were utilized. The MicroSonics Digital Echo Analyzer was used to analyze the echocardiograms. The fundamental components of the MicroSonics Analyzer are an image digitizer and a microcomputer (IBM AT) programmed to process and quantify two-dimensional echocardiograms. The digitizer converts analog video images to digital format. The MicroSonics Analyzer allowed the selection of a single, representative cardiac cycle (or a portion thereof), free of motion artifact and respiratory interference, for capture and storage on a floppy disk and subsequent evaluation. The captured cycle could then be played back in a continuous loop. The portion of the cardiac cycle of interest was that from end-diastole to end-systole; this allowed calculation of the maximal and minimal left ventricular volumes and left ventricular ejection fraction. The images were captured at a sequence of 8 frames/s, which could be viewed in motion or frame by frame. To calculate left ventricular volumes, the frame displaying the appropriately sized left ventricular volume was chosen, and the penlight was used to trace the endocardial outline and obtain other measurements, such as the long-axis length. In some cases, the endocardium was not clearly visible at all points; in these cases, after viewing of all frames, the best approximation was used. The MicroSonics Analyzer calculated the left ventricular volumes using a Simpson rule biplane formula, which divides the left ventricle into eight cylindric
JACC Vol. 25, No. 4 March 15, 1995:937-42
slices, and obtained the left ventricular ejection fraction from these volumes. For the radionuclide angiographic procedure, each patient received an injection of 25 mCi of autologous red cells labeled in vitro with technetium-99m. A gamma camera equipped with an all-purpose collimator was positioned in the left anterior oblique view with the exact angulation (40° to 50°) determined as that which best separated the left and right ventricles. Multiple-gated equilibrium blood pool scintigraphy was performed by acquisition of 20 frames of equal duration distributed uniformly over the entire cardiac cycle for a total of 2 rain/acquisition, resulting in -100,000 counts/frame. Count changes within a left ventricular region of interest were used to identify end-diastolic and end-systolic frames. Ejection fraction was then calculated by dividing stroke counts (end-diastolic minus end-systolic) by the background-corrected end-diastolic counts. Our prospective study did not include contrast cineangiography, but the studies reviewed retrospectively all used a similar method for determination of left ventricular ejection fraction. The studies used either a single-plane right anterior oblique view or biplane right and left anterior oblique views to examine a single sinus beat, excluding any premature beats. Left ventricular ejection fractions were calculated using the method of Sandler and Dodge (1,3,6-8). Retrospective analysis. A computer-assisted Medline search of published medical reports was conducted to find studies that compared two-dimensional echocardiographic left ventricular ejection fraction with radionuclide angiographic or contrast cineangiographic left ventricular ejection fraction. For inclusion in the analysis, the study must have a direct comparison of the left ventricular volumes from two-dimensional echocardiography with those from radionuclide angiography or contrast cineangiography. All studies that qualified for the analysis used the coefficient of correlation for comparison of left ventricular ejection fractions between the different methods. Three studies were found that compared two-dimensional echocardiography and radionuclide angiography (2,3,6); four compared two-dimensional echocardiography and contrast cineangiography (1,3,7,8). Left ventricular ejection fraction and linear regression analysis data were obtained from these reports. When left ventricular ejection fraction values were not explicitly provided, they were obtained from graphs included in the reports (3,6). A regression analysis was performed for each set of data to obtain a value for r, the coefficient of correlation. The "limits of agreement" described by Bland and Altman (4) were calculated with 95% confidence limits to assess the agreement between the left ventricular ejection fractions from twodimensional echocardiography, contrast cineangiography and radionuclide angiography. The method of Bland and Altman summarizes the lack of agreement by calculating the bias estimated by the mean difference (d) and the standard deviation of the difference (s). If the differences are normally distributed (Gaussian), 95% of the differences would be expected to lie between d - 2s and d + 2s (limits of agreement).
JACC Vol. 25, No. 4 March 15, 1995:937-42
NAIK ET AL. CORRELATION OF EJECTION FRACTION DETERMINATIONS
939
1.0 z
o.9. I-. 0 < n.1.1.
"~ z rr
0.8
I
w
2:
D
o. 9. 0.6 I-
oo
t'kl
o °
"-"
O
LL
LLI
w >
W a
• -
0
8
• O
•
•
4 0
J
• ° •
0.4
•
z__ - 4 w 0 z
0 Z
16 12
0.2
•
-8
•
•
•
•
w -12 £g
m
< rr
m -16 0 u--I u_ i
0.0 0.0
0.4
0.2
ECHO
EJECTION
i
L
0.6
0.8
I
I
I
20
I
I
40
M E A N LVEF B Y 2DE & R N A (%)
1.0
Figure 2. Difference in left ventricular ejection fraction (LVEF) from
FRACTION
Figure 1. Radionuclide angiographic versus two-dimensional echocardiographic (ECHO) left ventricular ejection fraction (r = 0.93).
Furthermore, because the range of values in our comparison is not 0 to infinity but rather 0 to 1.00 (range of ejection fractions), the limits of agreement narrow as one reaches either extreme of the range of ejection fractions. Intraobserver and interobserver variability were derived by calculating the mean difference between two observations and then the overall average mean difference, standard deviations and 95% confidence limits.
Results Comparison of two-dimensional echocardiography and ra. dionuclide angiography. Linear regression analysis of the data
in the prospective study of two-dimensional echocardiography and radionuclide angiography, as shown in Figure 1, revealed that r = 0.93, indicating good correlation. In addition, calculation of the limits of agreement with a 95% confidence interval yielded a lower limit of -11.5% and an upper limit of +11.7%, as shown in Figure 2. Thus, in our laboratory, a left
two-dimensional echocardiography (2DE) and radionuclide angiography (RNA) versus mean of the two values, with 95% confidence limits. Dotted lines = mean difference _+2 SD.
ventricular ejection fraction determined by two-dimensional echocardiography is consistently expected to be within 24% of the left ventricular ejection fraction obtained by radionuclide angiography in 95% of cases. Table 1 identifies the limits of agreement for the eight studies comparing two-dimensional echocardiography with radionuclide angiography and contrast cineangiography in 268 patients. As Table 1 shows, data from our laboratory yielded the highest r value (0.93) and the narrowest limits of agreement (+11.7 and -11.5). However, Table 1 also shows that high r values do not necessarily correlate with narrow limits of agreement. For example, the data from Carr et al. (7) also show an r value of 0.93 but broader limits of agreement of +15.3 and -18.3. In addition, the data from Starling et al. (6) yielded the worst upper limit of agreement (+26.2) but an intermediate r value of 0.81. In contrast, although the data from Folland et al. (2) had a narrower range of agreement, the r value was lower than that of Starling et al. (6). Table 1 clearly
Table 1. Limits of Agreement and Coefficient of Correlation Values for All Studies r
Study (ref. no.)
No. of Pts
d
SD
ULA
A
LLA
Value
- 11.5 17.4 -12.5 -15.4
0.93 0.75 0.93 0.81
-18.3 -23.2 -10.7 -15.3
0.93 0.78 0.87 0.89
Two-Dimensional Echocardiography Versus Radionuclide Angiography Cedars-Sinai Folland et al. (2) Quinones et al. (3) Starling et al. (6)
25 35 55 59
0.1 1.4 1.6 5.4
5.6 9.2 7.0 10.3
11.7 20.2 15.7 26.2
11.6 18.8 14.1 20.8
Two-Dimensional Echocardiography Versus Cineangiography Carr et al. (7) Folland et al. (2) Schiller et al. (1) Stamm et al. (8)
22 35 34 28
I
60
-1.5 -3.8 5.8 1.1
8.1 9.5 8.1 8.0
15.3 15.6 22.3 17.5
16.8 19.4 16.5 16.4
Averaged lack of agreement (A) = (Upper limit of agreement [UEA] - Lower limit of agreement [LLA])/2 = 16.8. Averaged r = 0.86. d = average deviation; Pts = patients.
940
NAIK ET AL. C O R R E L A T I O N OF EJECTION FRACTION DETERMINATIONS
JACC Vol. 25, No. 4 March 15, 1995:937-42
100
80
60
4o
20
17.5
15.a ~
lS.S
I ~ 0
:~a~:.
-20 - ,1.5
-17.4 -1a.5 -15.4
-le.3~~ ~_1o.7 N -15.3
Figure 3. Limits of agreement for all studies comparing two-dimensional echocardiographicleft ventricular ejection fractionswith radionuclideangiographic (open bars) tions.°r contrast cineangiographic(stippled bars) ejectionfrac-
-23.2 CSMC -40
Quinones Fo=and
Can" Starling
Schiller Folland
Sl~r,m
-60
-80
-100
shows that little concordance exists between r values and agreement for the studies reviewed. Figure 2 (Cedars-Sinai data) shows the difference in left ventricular ejection fraction between two-dimensional echocardiography and radionuclide angiography plotted against the average left ventricular ejection fraction by the two methods. The absence of a pattern, or trend, in the placement of the points in Figure 2 indicates that the magnitude of difference of the left ventricular ejection fractions between the two methods is not related to the absolute magnitude of the left ventricular ejection fractions being measured. This is supported by linear regression analysis of the data that yields a very. low r value of 0.03. The same graphic and regression analyses performed on the other sets of data yielded similar results. Agreement analysis of left ventricular ejection fraction. Figure 3 shows a graphic representation of each study's limits of agreement and how they compare with each other. Limits of agreement ranged from a lower limit of -23.2 to an upper limit of +26.2. For the entire series of studies, the variance between methods was <26% units (95% confidence interval), pointing out a relatively narrow range. Agreement ranged from 23% to 42% around the mean ejection fraction as shown. Figure 3 also graphically exhibits the lack of concordance between r values and the limits of agreement. Figure 4 (left) is a series of graphs combining data from our institution and all of the studies reviewed. It shows that the graphic display of data from all studies is similar, consistent with the high coefficients of correlation (r) found in each
individual study. The r values for the cumulative data were 0.85 (n = 119) for contrast cineangiographic versus echocardiographic ejection fraction, 0.87 (n = 178) for contrast cineangiographic versus radionuclide angiographic ejection fraction and 0.86 (n = 174) for radionuclide angiographic versus echocardiographic ejection fraction. An ejection fraction of 40% by echocardiography could range anywhere from 20% to 60% (from abnormal to normal) by radionuclide angiography, despite a correlation coefficient of 0.86. Figure 4 (right) demonstrates cumulative data from the studies analyzed using the agreement technique. Although the coefficient of correlation describes the relation between two sets of values, Figure 4 (right) graphically shows that it does not provide any information on how similar the values are to one another.
Discussion Clinical implications of agreement analysis. Many published studies have based the utility of different methods on estimates of comparative accuracy using correlation coefficients. Meta-analysis of the eight studies reviewed shows that the utility of correlation coefficients has limitations and that agreement analysis would be a more appropriate method for comparison. This is clear if the results of our data are applied to a clinical example. If one were to complete a preoperative evaluation for coronary artery bypass grafting or cardiac valve replacement, a test would be performed to estimate the ejection fraction to risk stratify the patient. If one obtained an
J A C C Vol. 25, No. 4 March 15, 1995:937 42
1.0
(9
0.50
08
•
O T
/ .-,..,,, .-.:-;/ /.. ,i.,.".. ~"
i
•
0.6
•
•
alk ,,
0.25
•
'Be 0,00
W
/."
"zU
• "u,//
0.2
Z
0.0 00
0.2
0 4
06
0.8
°
-I
0°
•
e~
~tlP='%~,,-
"
•
"
-0.25
-0.50
1.0
0.0
ECHO EJECTION F R A C T I O N
0.2
0.4
0.6
0.8
1.0
AVERAGE EJECTION FRACTION
1.0
F i g u r e 4. Combined regression (left) agreement analysis (right)
00
I..K '."~.t-p.
.
<
0.4
models for all studies. Top, Echocardiography (x) versus angiography (y): y = 0.075 + 0.858x, r = 0.848, SEE 0.090, p < 0.0001. Middle, Radionuclide angiography (x) versus angiography (y): v = 0.123 + 0.836x, r = 0.873, SEE 0.097, p < 0.0001. Bottom, Echocardiography (x) versus radionuclide angiography (y): y = 0.045 + 0.848x, r = 0.861, SEE 0.084, p < 0.0001.
941
N A I K E T AL. CORRELATION OF EJECTION FRACTION DETERMINATIONS
0.50
0.8 ~ Z
o_
0.25 •
°
°
0.6 =
u
:" "
,~ o
-.,..-. -.:.. r ~ . ~ + . : . . "
0.00
C)
0.4
Ilg ~ -0.25
02
7 0,0 0.0
02
0.4
06
08
-o.50
10
0.0
0.2
0.4
0.5
0.8
1.0
RADIONUCLIDE EJECTION F R A C T I O N AVERAGE EJECTION FRACTION 0.50
1+0
0.8 0
0.25
•"- . . ~, - :o
06
g
•
0.00 0.4
/ .+,-.<:"U""
•
°eeoc|
e
~ , -•~ .
O_ r~
e° e
.
o o°
•
~ eo
-0.25
0,2
0,0 0.0
--o , n - ~ ; ~-o- =i/~,-o - -
0.2
0 4
0 6
OB
1.0
-0.50 0.0
0.2
0,4
0.6
O.B
ECHO EJECTION F R A C T I O N AVERAGE EJECTION FRACTION
ejection fraction of 40% by echocardiography, Figure 4 (left) makes it evident that by radionuclide angiography this ejection fraction may be anywhere from 20% to 60%, the corresponding values having very different clinical implications for the patient. In this situation, one might elect to obtain an ejection fraction by a second method to have more confidence in the estimate. However, if one were to obtain an ejection fraction of 70% or 15% initially, the implications of an error of _+20% may not be significant, and one could omit obtaining ejection fraction by means of a second method. These decisions would depend on two factors: 1) how precise a measurement of ejection fraction is required for the given clinical situation, and 2) the limits of agreement among different methods in the laboratory being utilized.
Observer variability. Part of the difference in ejection fraction determined by different methods may be accounted for by intraobserver and interobserver variability. In our echocardiography laboratory, the mean difference between two observations by the same observer was 4.4%, with a standard deviation of 2.5% and 95% confidence intervals of 1.1% to 13.2%. Between two observers, the mean difference was 6.1%, with a standard deviation of 3.0% and 95% confidence intervals of 1.9% to 15.9%. Similarly, radionuclide angiography in our laboratory had an intraobserver variability of 2.5% and an interobserver variability of 6.8%. Standard deviations and confidence limits were not calculated for radionuclide angiography. For the data in the studies retrospectively reviewed, the intraobserver variability ranged from 1.4% to 5.1% for echo-
942
NAIK ET AL. CORRELATION OF EJECTION FRACT1ON DETERMINATIONS
cardiography and 2.0% to 5.1% for radionuclide angiography. Interobserver variability ranged from 2.3% to 8.1% for echocardiography and 2.0% to 8.1% for radionuclide angiography. Only Stamm et al. (8) reported intraobserver and interobserver variability data for cineangiography that were 4.3% and 6.7%, respectively. Taking observer variability into account, the data from our laboratory, exhibiting the narrowest limits of agreement (=12%), suggest that two-dimensional echocardiography is a reliable method for estimating left ventricular ejection fraction. However, in several other studies, the limit of agreement was >20%, and, as stated earlier, this may not be clinically acceptable. For situations in which serial follow-up of left ventricular function and ejection fraction may have clinical importance (such as evaluation of adriamycin toxicity, timing of aortic or mitral valve replacement, follow-up of cardiac transplantation candidates, post-myocardial-infarction patients and those in whom left ventricular contractility is being evaluated for response to drug therapy), the reliability of the estimate of left ventricular ejection fraction is crucial to clinical decision making. If one were simply to look at the high correlation coetficients of the different methods as a judge of their reliability, the data could be misleading. In these situations one has to consider two components of possible error: the variation between methods and the variation between serial determinations. For example, the error between radionuclide angiography and echocardiography in our laboratory is 12%. The data from Kuecherer et al. (9) showed that the variation of serial determinations of echocardiographic ejection fractions was 6.6%. The combined error (by addition in quadrature) would be 13.7%. The importance of agreement analysis in the clinical situations described is evident. Two-dimensional echocardiographic instruments have improved since the late 1970s, when the original studies comparing echocardiographic with radionuclide and cineangiographic left ventricular ejection fractions were performed. In addition, analytic algorithms for modeling the left ventricle have been refined. This may account for at least part of the difference between the results in our laboratory and those retrospectively reviewed. Conclusions. The coetficient of correlation has important limitations as an index of comparative accuracy of measures, such as left ventricular ejection fractions by two-dimensional echocardiography, radionuclide angiography and contrast
JACC Vol. 25, No. 4 March 15, 1995:937-42
cineangiography. However, analysis of agreement provides a more appropriate numeric estimate of accuracy. On the basis of such analyses, left ventricular ejection fraction determinations by two-dimensional echocardiography, radionuclide angiography and contrast cineangiography exhibit high correlation but only moderate agreement. The magnitude of agreement may be adequate for gross determinations of cardiac function, distinguishing among severely reduced, moderately reduced and normal left ventricular ejection fractions, but it may not be adequate for more precise determinations of ejection fraction (8). Ideally, each institution should validate its own data by analysis of agreement rather than correlation. These results suggest that, when validated by agreement analysis, multiple studies may not be necessary in appropriate clinical situations, potentially reducing costs. We gratefully acknowledge the technical assistance of Dr. Toshihiko Nishioka and Dr. Huai Luo.
References 1. Schiller NB, Acquatella H, Ports TA, et al. Left ventricular volume from paired biplane two-dimensional echocardiography. Circulation 1979;60:54755. 2. Folland ED, Parisi AF, Moynihan PF, Jones DR, Feldman CL, Tow DE. Assessment of left ventricular ejection fraction and volumes by real time, two-dimensional echocardiography. Circulation 1979;60:760-6. 3. Quinones MA, Waggoner AD, Reduto LA, et al. A new simplified and accurate method for determining ejection fraction with two-dimensional echocardiography. Circulation 1981;64:744-53. 4. Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1986;1:30%10. 5. Bookbinder MMJ, Panosian KJMJ. Using the coelficient of correlation in method comparison studies. Clin Chem 1987;33:1170-6. 6. Starling MR, Crawford MH, Sorenson SG, Levi B, Richards KL, O'Rourke RA. Comparative accuracy of apical biplane cross-sectional echocardiography and gated equilibrium radionuclide angiography for estimating left ventricular size and performance. Circulation 1981;63:1075-84. 7. Carr KW, Engler RE, Forsyth JR, Johnson AD, Gosink B. Measurement of left ventricular ejection fraction by mechanical cross-sectional echocardiography. Circulation 1979;59:1196-206. 8. Stamm RB, Carabello BA, Mayers DL, Martin RP. Two-dimensional echocardiographic measurement of left ventricular ejection fraction: prospective analysis of what constitutes an adequate determination. Am Heart J 1983; 104:136-44. 9. Kuecherer HF, Kee LL, Modin G, Cbeitlin MD, Schiller NB. Echocardiography in serial evaluation of left ventricular systolic arid diastolic function: importance of image acquisition, quantitation, and physiologic variability in clinical and investigational applications. J Am Soc Echocardiogr. 1991;4: 203-14.