2D and 3D Echocardiography-Derived Indices of Left Ventricular Function and Shape: Relationship With Mortality

Abstract

OBJECTIVES

This study hypothesized that left ventricular (LV) ejection fraction (EF) and global longitudinal strain (GLS) derived from 3-dimensional echocardiographic (3DE) images would better predict mortality than those obtained by 2-dimensional echocardiographic (2DE) measurements, and that 3DE-based LV shape analysis may have added prognostic value.

BACKGROUND

Previous studies have shown that both LVEF and GLS derived from 2DE images predict mortality. Recently, 3DE measurements of these parameters were found to be more accurate and reproducible because of independence of imaging plane and geometric assumptions. Also, 3DE analysis offers an opportunity to accurately quantify LV shape.

METHODS

We retrospectively studied 416 inpatients (60 18 years of age) referred for transthoracic echocardiography between 2006 and 2010, in whom good-quality 2DE and 3DE images were available. Mortality data through 2016 were collected. Both 2DE and 3DE images were analyzed to measure LVEF and GLS. Additionally, 3DE-derived LV endocardial surface information was analyzed to obtain global shape indices (sphericity and conicity) and regional curvature (anterior, septal, inferior, lateral walls). Cardiovascular (CV) mortality risks related to these indices were determined using Cox regression.

RESULTS

Of the 416 patients, 208 (50%) died, including 114 (27%) CV-related deaths over a mean follow-up period of 5 3 years. Cox regression revealed that age and body surface area, all 4 LV function indices (2D EF, 3D EF, 2D GLS, 3D GLS), and regional shape indices (septal and inferior wall curvatures) were independently associated with increased risk of CV mortality. GLS was the strongest prognosticator of CV mortality, superior to EF for both 2DE and 3DE analyses, and 2D EF was the weakest among the 4 functional indices. A 1% decrease in GLS magnitude was associated with an 11.3% increase in CV mortality risk.

CONCLUSIONS

GLS predicts mortality better than EF by both 3DE and 2DE analysis, whereas 3D EF is a better predictor than 2D EF. Also, LV shape indices provide additional risk assessment.

Left ventricular (LV) ejection fraction (EF) is the most commonly used echocardiographic parameter of LV function, known to be an independent predictor of mortality (1–4), and is routinely used to guide patient management. However, 2-dimensional (2D) echocardiographic assessment of LVEF, both qualitative and quantitative, is dependent on reader experience and imaging plane, and its accuracy varies with image quality (5). Newer techniques for quantitative evaluation of LV function include speckle tracking echocardiography (STE), which allows measurements of myocardial deformation parameters, such as global longitudinal strain (GLS). The strengths of GLS include better reproducibility and ability to detect subtle changes in myocardial function that precede changes in EF, as reported in a variety of disease states (6–8). Studies have shown that GLS can also predict mortality, potentially more accurately than EF (9–15). Most outcomes studies focusing on LV function were performed using 2D echocardiography (2DE).

Three-dimensional echocardiography (3DE) offers better reproducibility and higher accuracy than 2DE for the assessment of LV size and function (16) because it avoids apical foreshortening and is based on direct volumetric measurements without geometrical assumptions. Furthermore, because 3DE can track myocardial motion independently of the imaging plane, 3DE-derived GLS may also be more accurate and reproducible (17,18). Recently, LV shape has been gaining interest with the availability of 3DE analysis tools, and there is growing evidence that it may carry additional diagnostic and prognostic information (19–21). Accordingly, we hypothesized that: 1) 3DE parameters would be better predictors of cardiovascular (CV) mortality than 2DE; 2) 3D GLS would be a better predictor of CV mortality than 3D EF; and 3) 3DE-derived shape indices could also predict CV mortality. This study was designed to investigate the relationship between these indices and long-term survival.

METHODS

POPULATION AND STUDY DESIGN

We retrospectively studied 416 inpatients (60 ± 18 years of age; n = 213 men [51%]; body surface area [BSA]: 1.79 ± 0.28 m²), referred for a clinically indicated transthoracic echocardiography between 2006 and 2010, and who had 2DE and 3DE images of sufficient quality to allow both volume measurements and STE-based LV deformation analysis. Patients with atrial fibrillation or other arrhythmias during echocardiographic examinations were excluded. Clinical characteristics of our cohort are summarized in Table 1. Mortality data, including CV mortality, were collected from hospital records and the Social Security Death Index. 2DE and 3DE images were used to measure EF and GLS. In addition, 3D shape analysis was performed to obtain global and regional LV shape indices. The risks for CV mortality associated with these indices were determined using Cox regression and Kaplan-Meier analyses. The study was approved by the Institutional Review Board.

TABLE 1.

Clinical Characteristics of the Study Group

Dyslipidemia	50
Hypertension	69
Diabetes mellitus	26
Smoker	42
Paroxysmal atrial fibrillation	23
Status post ventricular tachycardia	11
Glomerular filtration rate <60 ml/min/1.73 m2	55
Coronary artery disease	44
s/p myocardial infarction	20
Ischemic cardiomyopathy	25
Nonischemic cardiomyopathy	21
Pulmonary hypertension	26
Valvular cardiomyopathy	9
Congenital cardiomyopathy	6
Implantable cardioverter-defibrillator	18
Pacemaker	19

Values are %.

ECHOCARDIOGRAPHIC IMAGING AND ANALYSIS

2DE and 3DE imaging was performed using commercial equipment (Philips iE33 imaging system with a fully sampled matrix array transducer, Philips Medical Systems, Andover, Massachusetts). 2DE imaging included apical 2-, 3-, and 4-chamber views, from which LVEF was measured using the biplane method of disks (Xcelera, Philips Medical Systems) and GLS using 2D STE in all 3 views (Philips QLab). 3DE imaging included multibeat full-volume datasets while maximizing frame rate, which was 18 ± 3 Hz. 3DE images were analyzed using commercial software (4D LV-Function, TomTec Imaging Systems, Unterschleissheim, Germany) to quantify LVEF by semi-automated detection of the endocardial boundaries with manual editing as necessary, and GLS by automated 3D STE analysis (Figure 1).

FIGURE 1. 3DE Volume and Deformation Analyses.

Example of 3DE dataset of the left ventricle (top). Endocardial boundaries initialized in 3 cross-sectional views extracted from the 3D dataset (middle) are used to create a dynamic 3D cast of the ventricle (bottom, center), from which both dynamic volume (bottom, left) and longitudinal strain (LS) (bottom, right) are calculated. 3DE = 3-dimensional echocardiographic; Ch = channel; LV = left ventricular.

LV shape analysis was performed using custom software that uses 3D endocardial surfaces exported from TomTec. Briefly, the endocardial surfaces expressed as a series of unstructured meshes of connected points were used as input for analysis of global and regional LV shape via an algorithm described in detail previously (19,20) and summarized in Figure 2. Global indices included sphericity and conicity, expressed as numbers between 0 and 1, reflecting the degree of similarity of the ventricle to a perfect sphere or cone, respectively (Figure 2, top). This was achieved by sampling along a helical pattern on the 3D LV surface and comparing the result with a signal obtained using the same procedure from an idealized, reference 3D shape, either a sphere or a cone.

FIGURE 2. Global and Regional Shape Indices.

Diagrammatic representation of the computation of sphericity and conicity shape indices of the left ventricle (top) and regional curvature of the 4 different ventricular walls (bottom). Abbreviations as in Figure 1.

Regional shape indices included curvature of 4 LV walls (anterior, septal, inferior, lateral) reflecting 3D curvedness of the corresponding part of the endocardial surface, averaged over the cardiac cycle (Figure 2, bottom). This approach has been previously used in the context of 3D analysis of cardiac magnetic resonance images (22). To achieve this, a quadratic polynomial function was fit to the local neighborhood of each point belonging to the LV endocardial surface. This allowed computing, for each point, the values of curvature k1 and k2, each corresponding to the inverse of the radius of the 2 circles in orthogonal planes best fitting the local surface. Then, the mean normalized curvature was obtained by averaging these 2 curvature values and dividing by calculated curvature of a sphere having the same volume as the ventricle. Of note, 0 curvature indicates a flat surface, whereas the more positive or negative values signify more convexity or concavity of the surface, respectively, from the perspective of a reference point outside the ventricle.

STATISTICAL ANALYSIS

Continuous variables are presented as mean ± SD, and categorical variables as numbers and percentages. Baseline characteristics of the groups were compared using analysis of variance for continuous variables. When data were not normally distributed, groups were compared with the Kruskal-Wallis test. Univariate comparisons were performed using tests for paired data, including the paired t test for normally distributed continuous variables, the Wilcoxon matched pairs signed rank test for non-normally distributed continuous variables, and the chi-square test for categorical variables.

When collinearity between a pair of variables was detected by both Pearson and Spearman (to avoid effects of outliers) correlation analysis, separate models were created for multiple regression for each variable to determine the strength of the association and risk of mortality. Cox proportional hazards models were used to calculate hazard ratios (HRs) for CV mortality risk, whereas the non-CV deaths were censored. Covariates that could influence the survival risk were included if found significant at p < 0.05 or considered clinically relevant based on previous publications. The results of these separate analyses were compared using global measures of model fit, including the –2 logarithmic likelihood (–2LL) test and receiver-operating characteristic analysis area under the curve ± SD and 95% confidence interval (CI) to determine the diagnostic accuracy of 2D and 3D indices in predicting CV death. In addition, this analysis was repeated for a subgroup of 322 patients, after excluding non-CV deaths. Finally, the incremental contribution of 3D GLS and EF, compared with their respective 2D parameters, in predicting the risk of 5-year CV death was evaluated using categorical net reclassification improvement approach, using the survival data of the subjects with at least 5 years of follow-up.

Survival analysis over time included Kaplan-Meier survival curves for 5 years because this was the mean follow-up time of the study. These curves were constructed for LV function parameters (EF and GLS measured by both 2DE and 3DE in the entire cohort) and also for shape indices (in the subgroup that excluded the non-CV deaths) that were found significant in the multiple regression analysis. Comparisons of cumulative events across strata were performed using the log-rank test. Thresholds for EF, GLS, and shape indices were determined by dividing the study group into tertiles for each index and testing the separation between them.

All analyses were performed using SPSS software version 22 (IBM, Armonk, New York). Statistical significance was inferred at p < 0.05.

RESULTS

During the mean follow-up of 5 ± 3 years, 208 of 416 (50%) patients died. Ninety-four patients (23%) died of non-CV causes. Of the remaining 322 patients, 114 died of CV-related causes (35%). Table 2 shows the results of all 2DE- and 3DE-derived LV size, function, and shape parameters for the entire study group and for the survivors, and all-cause deaths and CV-related deaths. Both 2DE and 3DE measurements showed that LV volumes were lower, whereas LVEF and GLS magnitudes were higher in the survivors group compared with nonsurvivors, with statistical significance in most comparisons. Global shape indices indicated lower spherical and more conical LV shape in survivors compared with nonsurvivors, both reflecting a more physiological prolate ellipsoid shape. Regional shape analysis indicated that survivors had a significantly lower inferior wall curvature and higher septal wall curvature. All of these differences, except the regional curvatures, were even more pronounced when survivors were compared with the CV mortality group.

TABLE 2.

Summary of Results of All 2DE- and 3DE-Derived LV Size, Function, and Shape Parameters for Entire Study Group and Survivors and All-Cause and CV-Related Deaths

	Total (N = 416)	Survivors (n = 208)	All-Cause Mortality (n = 208)	CV Mortality (n = 114)	p Value
					Survivors vs. All-Cause Mortality	Survivors vs. CV Mortality
Age, yrs	60 ± 18	54 ± 17	65 ± 17	67 ± 17	<0.01	<0.01
2DE
Biplane EDV, ml	166 ± 85	160 ± 76	172 ± 89	184 ±104	0.12	0.03
Biplane ESV, ml	88 ± 77	79 ± 66	96 ± 82	111 ± 97	0.01	<0.01
Biplane EF, %	52 ± 16	54 ± 14	50 ± 17	46 ± 18	<0.01	<0.01
3DE
EDV, ml	187 ± 89	182 ± 81	191 ± 92	206 ± 107	0.22	0.03
ESV, ml	99 ± 84	88 ± 75	109 ± 87	128 ± 103	0.01	<0.01
EF, %	52 ± 17	55 ± 15	48 ± 17	44 ± 18	<0.01	<0.01
2DE
GLS, %	−1.56 ± 4.9	−16.8 ± 4.7	−14.4 ± 4.6	−13.4 ± 4.5	<0.01	<0.01
3DE
GLS, %	−17.2 ± 6.6	−1.90 ± 6.1	−15.6 ± 6.4	−14.0 ± 6.6	<0.01	<0.01
Shape
Sphericity	0.64 ± 0.06	0.63 ± 0.05	0.65 ± 0.06	0.65 ± 0.06	0.01	0.01
Conicity	0.79 ± 0.03	0.79 ± 0.03	0.78 ± 0.03	0.78 ± 0.03	<0.01	<0.01
Anterior curvature	1.02 ± 0.13	1.02 ± 0.12	1.03 ± 0.14	1.02 ± 0.13	0.32	0.66
Septal curvature	0.98 ± 0.13	1.00 ± 0.11	0.96 ± 0.15	0.98 ± 0.14	0.01	0.15
Inferior curvature	1.00 ± 0.14	0.98 ± 0.14	1.02 ± 0.13	1.02 ± 0.14	<0.01	0.03
Lateral curvature	0.98 ± 0.05	0.98 ± 0.05	0.98 ± 0.05	0.98 ± 0.05	0.65	0.60

Values are mean ± SD. Bold indicates p < 0.05.

2DE = 2-dimensional echocardiographic; 3DE = 3-dimensional echocardiographic; CV = cardiovascular; EDV = end-diastolic volume; EF = ejection fraction; ESV = end-systolic volume; GLS = global longitudinal strain; LV = left ventricular.

As expected, strong correlations between 2D EF, 3D EF, 2D GLS, and 3D GLS and multicollinearity were detected, necessitating 4 separate regression models for the Cox analysis. Each regression model included age, BSA, and all global and regional indices of LV shape. Table 3 shows the results of the multiple regression for CV mortality; Table 4 summarizes the results of a subgroup analysis (n = 322), in which non-CV deaths were excluded. Both tables include the results of both unadjusted (left) and adjusted (right) analyses.

TABLE 3.

Results of Cox Regression Analyses of LV Function and Shape Indices From 2DE and 3DE Images: Risk Assessment of Long-Term CV Mortality in Patients Referred for Echocardiographic Examination

Cox Regression: CV Mortality (N = 416)
Unadjusted				Adjusted
	p Value	HR	95% CI	2D EF Model 1,140.9* 0.628 ± 0.32 (0.565–0.691)†			2D GLS Model 1,134.0* 0.676 ± 0.28 (0.620–0.731)†			3D EF Model 1,133.1* 0.658 ± 0.31 (0.597–0.719)†			3D GLS Model 1,124.2* 0.690 ± 0.29 (0.633–0.747)†
				p Value	HR	95% CI	p Value	HR	95% CI	p Value	HR	95% CI	p Value	HR	95% CI
Age	<0.01	1.036	1.02–1.05	<0.01	1.032	1.02–1.05	<0.01	1.028	1.02–1.04	<0.01	1.033	1.02–1.05	<0.01	1.032	1.02–1.05
BSA	0.031	0.474	0.24–0.93	0.031	0.431	0.20–0.93	0.021	0.406	0.19–0.87	0.063	0.508	0.25–1.04	0.033	0.462	0.23–0.94
Sphericity	0.113	11.6	0.56–242	0.063	0.000	0.00–1.76	0.201	0.001	0.00–47.4	0.069	0.000	0.00–2.19	0.105	0.000	0.00–6.46
Conicity	0.031	0.002	0.00–0.58	0.208	0.000	0.00–1.5K	0.332 0.000		0.00–36K	0.385	0.000	0.00–100K	0.284	0.000	0.00–9.3K
Anterior	0.661	1.415	0.30–6.67	0.875	1.168	0.17–8.10	0.916	0.899	0.12–6.51	0.933	1.086	0.16–7.47	0.964	0.956	0.14–6.63
Septum	0.151	0.351	0.08–1.47	0.262	0.328	0.05–2.30	0.427	0.465	0.07–3.07	0.234	0.312	0.05–2.13	0.270	0.338	0.05–2.32
Inferior	0.027	4.835	1.20–19.5	0.058	4.959	0.95–25.9	0.109	3.827	0.74–19.8	0.039	5.47	1.09–27.4	0.045	4.939	1.03–23.6
Lateral	0.646	2.240	0.07–70.3	0.123	27.70	0.41–1.9K	0.820	1.655	0.02–126	.240	12.48	0.19–838	0.383	6.47	0.10–431
2D EF	<0.01	0.975	0.96–0.99	<0.01	0.963	0.95–0.98
2D GLS	<0.01	1.123	1.08–1.17				<0.01	1.140	1.09–1.19
3D EF	<0.01	0.973	0.96–0.98							<0.01	0.961	0.95–0.97
3D GLS	<0.01	1.091	1.06–1.12										<0.01	1.113	1.08–1.15

^*−2LL AUC.

^†AUC ROC – SE (95% CI). Bold indicates p < 0.05.

−2LL = −2 logarithmic likelihood; AUC ROC = area under curve in receiver-operating characteristic analysis; BSA = body surface area; CI = confidence interval; HR = hazard ratio.

TABLE 4.

Results of Subgroup Analyses Excluding Noncardiovascular Deaths

Cox Regression: CV Mortality (N = 322)
Unadjusted				Adjusted
	p Value	HR	95% CI	2D EF Model 1,092.5* 0.635 ± 0.34 (0.569–0.701)†			2D GLS Model 1,081.4* 0.698 ± 0.30 (0.640–0.756)†			3D EF Model 1,082.1* 0.672 ± 0.32 (0.608–0.735)†			3D GLS Model 1,073.7* 0.711 ± 0.30 (0.652–0.769)†
				p Value	HR	95% CI	p Value	HR	95% CI	p Value	HR	95% CI	p Value	HR	95% CI
Age	<0.01	1.038	1.03–1.05	<0.01	1.034	1.02–1.05	<0.01	1.029	1.02–1.04	<0.01	1.035	1.02–1.05	<0.01	1.034	1.02–1.05
BSA	0.012	.428	0.22–0.83	0.009	0.363	0.17–0.78	0.003	0.318	0.15–0.68	0.013	0.408	0.20–0.83	0.008	0.384	0.19–0.78
Sphericity	0.069	18.0	0.80–403	0.075	0.000	0.00–2.67	0.258	0.002	0.00–95.3	0.059	0.000	0.00–1.49	0.095	0.000	0.00–4.68
Conicity	0.017	.001	0.00–0.29	0.198	0.000	0.00–1.0K	0.414	0.000	0.00–14K	0.275	0.000	0.00–6.4K	0.211	0.000	0.00–1.1K
Anterior	0.429	1.913	0.38–9.56	0.809	1.268	0.18–8.71	0.991	1.011	0.15–6.95	0.923	1.098	0.17–7.29	0.812	0.791	0.11–5.46
Septum	0.039	.215	0.05–0.93	0.061	0.160	0.02–1.09	0.077	0.187	0.03–1.20	0.026	0.117	0.02–0.77	0.046	0.145	0.02–0.97
Inferior	0.005	7.099	1.80–27.9	0.017	7.523	1.43–39.5	0.039	5.723	1.10–29.9	0.008	9.40	1.82–48.6	0.012	7.631	1.57–37.0
Lateral	0.912	1.217	0.04–39.9	0.163	27.09	0.26–2.8K	0.566	3.857	0.04–389	0.107	44.57	0.44–4.5K	0.193	21.30	0.21–2.1K
2D EF	<0.01	.976	0.96–0.99	<0.01	0.962	0.95–0.98
2D GLS	<0.01	1.124	1.08–1.17				<0.01	1.154	1.10–1.21
3D EF	<0.01	.974	0.96–0.98							<0.01	.960	0.95–0.97
3D GLS	<0.01	1.089	1.06–1.12										<0.01	1.116	1.08–1.15

Cox regression of LV function and shape indices obtained from 2DE and 3DE images: assessment of risks of long-term CV mortality.

^*−2LL AUC.

^†AUC ROC ± SE (95% CI). Bold indicates p < 0.05. Abbreviations as in Tables 2 and 3.

The unadjusted Cox regression revealed that in addition to age and BSA, all 4 LV function indices (2D EF, 3D EF, 2D GLS, 3D GLS), 1 of the 2 global shape indices (namely conicity), and 1 of the regional shape indices (inferior wall curvature) were associated with CV mortality (Table 3). The adjusted regression indicated that the global shape indices were no longer significant, but the remaining indices were independently associated with CV mortality, although the inferior wall curvature was significant only in the 3D models. Of note, reduced EF and GLS were associated with increased risk of CV mortality (as reflected by HRs <1 for EF and >1 for GLS).

Similarly, in the subgroup analysis excluding the non-CV deaths, unadjusted Cox regression analysis showed that the same indices with the addition of septal curvature were associated with increased risks of CV mortality (Table 4). The adjusted regression analyses indicated that, similar to the entire study cohort, none of the global shape indices were significant, but the remaining indices were independently associated with CV mortality, with the inferior wall curvature being significant in all models and septal curvature only in the 3D models (as reflected by HRs >1 for the inferior wall and <1 for the septum).

Comparisons between the 4 models for both regression analyses revealed that 3D GLS was the strongest predictor of CV mortality, reflected by the lowest –2LL value (Tables 3 and 4). Although this was also reflected by a higher receiver-operating characteristic area under the curve value, the 95% CI showed overlap between these 4 multivariate analysis models of the respective LV function indices. Interestingly, 2D EF was the weakest among the 4 indices and GLS was superior to EF for both 2DE and 3DE analyses. Additionally, 3DE-derived GLS and EF were superior to their 2DE counterparts. Importantly, each 1% decrease in GLS magnitude corresponded to an 11.3% increase in CV mortality in the entire cohort (Table 3) (HR: 1.113; 95% CI: 1.08 to 1.15; p < 0.001). Categorical net reclassification improvement analysis showed an overall similarity between 3D GLS and 2D GLS (+0.3%), whereas improvement of +6.7% was noted in the accuracy of classification between 3D EF and 2D EF.

These findings were confirmed by Kaplan-Meier curves (Figure 3), which demonstrated that both 2D and 3D EF were not able to differentiate normal from mild-to-moderately reduced EF groups (left panels), whereas both 2D and 3D GLS were able to differentiate all these groups (right panels). Figure 4 shows the survival curves for the 2 regional curvature indices (inferior and septal walls) that were found to be significant in the subgroup Cox regression analysis. Survival curves were significantly different for the patients with the upper-tertile inferior wall curvatures (left panel), and for those with the lower-tertile septal curvatures (right panel), indicating that these patients were at higher risk of CV death. Of note, the division into tertiles resulted in slightly different cutoffs for both EF and GLS in 2D and 3D.

FIGURE 3. Results of Survival Analysis for EF and Strain.

Kaplan-Meier survival curves for 2DE- and 3DE-based LVEF and GLS, stratified by tertiles both for EF and GLS. 2DE = 2-dimensional echocardiography; EF = ejection fraction; GLS = global longitudinal strain; other abbreviation as in Figure 1

FIGURE 4. Results of Survival Analysis for Regional Shape Indices.

Kaplan-Meier survival curves for regional shape indices: inferior wall (left) and septal wall (right) curvature, stratified by tertiles for each index.

DISCUSSION

LV function has important implications for clinical management and clinical trials. The parameter most frequently used to assess LV function is EF, which has been shown to correlate with morbidity and mortality and thus is used as a guide for the management of the individual patient. Another parameter of LV function that emerged in the past decade is GLS. The recent guidelines emphasize recommendations for LV function quantification by measuring both EF and GLS (23); however, the relative prognostic value of these 2 indices has not been determined in the context of 3DE, which was the focus of this study.

This study was possible because of the availability of a unique 3DE database including hundreds of studies dating from 2006, which allowed us an up to decade-long follow-up to investigate the relationship between 3DE-derived indices of LV function and long-term mortality. The inclusion criteria for this study were the availability of good-quality 2DE and 3DE images performed in any inpatients referred for a clinically indicated echocardiography study. This reflects the feasibility of acquiring 3D LV full-volume datasets of adequate quality as early as 2006, which provided the basis for the new chamber quantification guidelines (23) that recommend this methodology whenever possible. We also followed these guidelines’ recommendation for 2D GLS measurements using 2-, 3-, and 4-chamber views, which underscores the true superiority of 3D over 2D GLS measurements.

One might question the high all-cause mortality rate in our patient cohort, which was approximately 50% over a 10-year period. We believe that this reflects the high acuity of inpatients referred for cardiac ultrasound examinations in a tertiary referral hospital and the long-term follow-up in our study. This may also reflect that our hospital is located in an under-served urban area, where patient noncompliance is known to be high because of socioeconomic factors, as well as the high prevalence of comorbidities (Table 1).

Our findings confirmed that both EF and GLS can be used to predict CV mortality and that the predictive power of GLS is better than that of EF measured by 2DE (1–4,9–15). In addition, to our knowledge, this study is the first to show several clinically important new findings: 1) GLS was a better predictor of CV mortality than EF not only by 2DE, but also by 3DE; 2) 3DE is a better technique than 2DE to predict CV mortality for EF, although only minimally better for GLS; and 3) although global sphericity and conicity indices did not independently predict CV mortality, regional shape indices, namely curvature of the septal and inferior walls, were independent predictors of mortality in addition to the previously mentioned LV function parameters.

Our finding that the predictive power of GLS is better than that of EF may be related to its superior accuracy and reproducibility, when measured by both 2DE (6–8) and 3DE (17,18). A possible reason for the improved reproducibility and accuracy compared with EF may be related to the methodology of the semiautomated speckle tracking algorithm. This is because 1 source of variability of the conventional technique for EF measurement is the need to visually identify the end-diastolic and end-systolic frames for analysis, which is not perfectly reproducible. Even the chamber quantification guidelines from the American Society of Echocardiography are ambiguous with regard to this frame selection (23). In contrast, GLS measurements are based on speckle tracking technology, which follows the myocardium throughout the cardiac cycle and thus includes every single frame in each analysis, avoiding this source of variability.

The changes in regional LV curvature that were associated with CV mortality may be a result of septal shift toward the left ventricle in patients with right ventricular pressure overload that is known to affect survival (24); such patients were relatively common in our study cohort (26%). HR <1 for the septal wall indicated that the lower the curvature (the flatter the septum) was, the higher the likelihood of CV mortality would be. Conversely, HR >1 for the inferior wall indicated that the higher the curvature (the more convex the inferior wall) was, the higher the likelihood of CV mortality would be.

Interestingly, Kaplan-Meier curves (Figure 3) showed that both 2D and 3D EF were unable to differentiate normal from mild-to-moderately reduced EF groups. Importantly, both 2D and 3D GLS were able to differentiate among all 3 GLS tertiles, reflecting the added predictive value of this index over EF. This finding may reflect the improved sensitivity of GLS, which may manifest itself in better predictive power for even lower degrees of LV dysfunction. This is as opposed to EF, which was associated with higher mortality only in the severe LV dysfunction group.

One might question whether the small intergroup differences in global shape indices are meaningful. Although these differences were statistically significant because of extremely low intersubject variability reflected by the very small SDs (Table 2), whether these indices can be clinically significant remains to be determined. Of note, this study showed that these small differences were insufficient to give these indices predictive power for CV mortality.

Our choices of statistical methodology used in this study need explanation. The 4 LV function indices we aimed at comparing with respect to predictive power, namely 2D and 3D LVEF and GLS, are not independent of each other, but instead are strongly correlated between them, which was confirmed by collinearity tests. One way to overcome this limitation and determine which parameter is the best was to compare separate regression models that included 1 of these parameters at a time. The strength of these models was compared using advanced statistical measures specifically designed for such circumstances. The reason we performed a subgroup analysis excluding the non-CV deaths was to eliminate data “contamination” by irrelevant factors.

STUDY LIMITATIONS

One limitation is the retrospective nature of this study, which might have biased the patient selection. For example, we focused on inpatients with good-quality images, thus limiting the generalizability of our findings. Morbidly obese patients and those with suboptimal images because of other conditions, such as lung disease, may not be adequately represented in our cohort. Accordingly, our results cannot be extrapolated to consecutive patients or outpatients with a wide range of image quality. Also, we cannot estimate the feasibility of 3D analysis in this historical cohort of patients who were selected on the basis of image quality over a decade using imaging equipment available at the time. Finally, this study only analyzed patients in sinus rhythm during imaging; thus, these results also cannot be extrapolated to patients with atrial fibrillation or other types of arrhythmia.

CONCLUSIONS

In summary, our results are the first to demonstrate a superior predictive ability of 3D EF over 2D EF, and also the value of adding to established risk factors GLS, which was found to be a superior survival prognostic factor. This study adds weight to the use of 3DE functional indices in echocardiographic examinations for prognostic and not only diagnostic purposes.

PERSPECTIVES.

COMPETENCY IN MEDICAL KNOWLEDGE

We compared the value of LVEF and GLS measured using both 2DE and 3DE for predicting long-term mortality in a cohort of inpatients who underwent clinically indicated echocardiographic examinations. We found that 3D measurements of LVEF predicted 5-year mortality better than 2D LVEF, and that GLS was a better predictor than EF. We also found that 3D analysis of LV shape may provide additional risk assessment.

TRANSLATIONAL OUTLOOK

Our study focused on inpatients with good-quality images, thus limiting the generalizability of our findings to consecutive patients, outpatients with a wide range of image quality, or patients with atrial fibrillation or other types of arrhythmia during echocardiographic examinations.

ABBREVIATIONS AND ACRONYMS

2DE

2-dimensional echocardiography

3DE

3-dimensional echocardiography

CI

confidence interval

CV

cardiovascular

EF

ejection fraction

GLS

global longitudinal strain

HR

hazard ratio

−2LL

−2 logarithmic likelihood

LV

left ventricular

STE

speckle tracking echocardiography