We performed a single center, prospective analysis of retrospectively acquired echocardiographic studies obtained from both in- and out-patients performed from October 2018 to December 2021 at our laboratory. Exclusion criteria were age less than 18 years, lack of 3DE acquisitions or incomplete 2DE data for LV volume quantitation, poor quality of the either 2DE or 3DE acquisitions (defined as the impossibility to visualize the endocardium of two or more adjacent LV myocardial segments without the use of contrast agents), echocardiographic studies performed for non-clinical indications (e.g., driving license, sports activity screening, etc.), and lack of follow-up data.

Clinical information at the time of the echocardiographic study included patients risk factors such as hypertension (either blood pressure >140/90 mm Hg or active antihypertensive treatment), hypercholesterolemia (LDL cholesterol >130 mg/dl or active statin treatment), diabetes (fasting glucose > 126 mg/dl), serum levels of creatinine, hemoglobin, atrial fibrillation, and history of ischemic heart disease (previous myocardial infarction or documented coronary artery disease). Data were obtained from the clinical records of our hospital (Table 1). This retrospective analysis of prospectively acquired data was approved by the Ethics Committee of the Istituto Auxologico Italiano, IRCCS (record #2021_05_18_13, approved on May 18, 2021). The need for patient written informed consent was waived due to the retrospective nature of the acquisitions.

All 2DE and 3DE acquisitions were obtained during the same echocardiographic study using a commercially available echocardiography system (Vivid E95, GE Healthcare, Horten, NO) equipped with both standard 2DE (M5S) and 3DE (4Vc) probes. All echocardiography studies were stored in a digital archive to be exported and analyzed offline using a commercially available software (Echopac BT204, GE Healthcare, Horten, NO).

From the apical four- and two-chamber views, the 2D LV volumes and EF were measured offline by a single experienced operator using the biplane method of disks’ summation (modified Simpson’s rule, Figure 1) (17). 3DE datasets of the LV were obtained from the apical approach using multi-beat full-volume acquisition during breath-holding and taking care to encompass the entire LV cavity in the dataset (16, 21). 3D LV volumes and EF (Figure 1) were measured offline by a single experienced operator using a dedicated software package for the LV analysis (4D AutoLVQ, GE Healthcare, Horten, NO). Measurement workflow started with the semi-automated detection of the LV endocardial borders. When needed, manual editing was used to optimize the endocardial contour identification (22). To trace the endocardial borders of both the 2DE and 3DE datasets, the end-diastolic frame was selected as the frame before the mitral valve closure, whereas the end-systolic frame was identified as the frame before mitral valve opening.

The primary clinical endpoint was the occurrence of death for any cause. The secondary endpoint was the composite of all-cause death and hospitalization for cardiac indication (either from heart failure, acute coronary syndromes, or arrhythmias). Information concerning both survival and hospitalization were obtained at regular intervals via: (i). review of electronic medical records of regular outpatient visits and hospital admission records; (ii). telephone interview with the patient, or if deceased, with family members; and (iii). contact with the patient’s physicians. Mortality status was verified independently through the Social Security Death Index and death certificates. For patients without events, the date of the last contact was used for survival analysis. Assignment of clinical events was performed by physicians unaware of the patients’ echocardiographic and clinical characteristics.

The normal distribution of continuous variables was tested with Kolmogorov–Smirnov test. Continuous variables were expressed as mean ± SD or as median (interquartile range). Categorical variables were expressed as absolute numbers (percentages). We compared the clinical and echocardiographic characteristics between patients who died and those who survived. Student’s t test for independent samples was used to compare differences between two groups of normally distributed continuous variables, while Mann-Whitney’s test was used to compare differences between two groups of non-normally distributed continuous variables. The chi-square test was used to assess differences between two groups of categorical variables. Pearson correlation coefficient was used to assess correlations between 2DE and 3DE parameters. Kaplan Meier curves were constructed to assess the prognostic stratification for the different 2DE and 3DE parameters. The log-rank test was used to assess the statistical significance between strata.

Since there was a high correlation between 2DE and 3DE LV volumes and EF, we built several pairs of models by separately adding the 2DE and 3DE parameters into the baseline clinical model. The first step used only the clinical variables available in our cohort (thus excluding the echocardiographic parameters), in a stepwise Cox proportional-hazard model, censoring data at first event. The clinical variables with significance level < 0.05 were included into the multivariate baseline clinical model (CM). Next, the 2DE parameters were added sequentially to the baseline CM, and then we built a third model by adding sequentially the 3DE parameters to the baseline CM, resulting in pairs of models for the end-diastolic volume (EDV), the end-systolic volume (ESV), and EF. The proportional hazards assumption for the Cox regression models was verified by visual assessment of Kaplan-Meier curves. The whole process was repeated for both the all-cause death and the composite endpoint. The independent and incremental value of each model compared to the previous one was assessed by comparing model χ² statistics. Model discrimination was further assessed using Harrel’s C-index. Comparison of the C statistics of the various multivariable models was performed using the method proposed by Newson RB (23).

Time-dependent Receiver Operating Characteristics (ROC) curve analyses for censored event times were used to compare the prognostic value of the LVEF severity grading threshold values obtained with 2DE and 3DE and Areas Under Curve (AUC) were derived. The De Long test was used to compare the AUCs of tested threshold values (24).

Data were analyzed using SPSS v.24.0 (SPSS Inc., Chicago, IL, USA) and MedCalc 20, (MedCalc Software Ltd., Ostend, Belgium). Statistical significance was defined as p < 0.5.

The final study population included 725 patients with various cardiac conditions (Table 1) who underwent echocardiography for various clinical reasons (Figure 2).

During the follow-up period of (median 3.39 years, IQR = 2.6 years), 111 (15.3%) deaths (83 of them, 75% were cardiac deaths), and 248 (34.2%) cardiac hospitalizations occurred. The latter were distributed as follows: 151 (20.8%) for heart failure, 51 (7.0%) for acute coronary syndrome, and 46 (6.3%) for arrhythmias. Finally, 304 (41.9%) patients reached the composite endpoint of all-cause death or cardiac hospitalization.

Patients who died were older and had higher heart rate than survivors (Table 1). In addition, patients who died had a higher prevalence of diabetes, chronic kidney disease, anemia, permanent atrial fibrillation, and history of ischemic heart disease than the survivors (Table 1). Conversely, sex and body mass index distribution, as well as the prevalence of smoking, hypertension, and dyslipidemia were similar between the two groups.

As expected, there was a close correlation between the 2DE and 3DE LV EDVs (r = 0.964; p < 0.001), ESVs (r = 0.972; p < 0.001), and EFs (r = 0.925; p < 0.001). However, 3DE LV volumes were larger than the 2DE LV volumes (bias = + 21 ml, LOI ± 18 ml for the EDV, and bias = + 19 ml, LOI ± 10 ml for the ESV, respectively), whereas the 3D LVEF was lower than the 2D LVEF (bias = −2%, LOI ± 10).

2DE and 3DE LV volumes were categorized into normal or dilated according to technique-specific threshold values. The incidence of LV dilatation was significantly higher when LV EDV and ESV were calculated by 2DE than when they were measured with 3DE (45% vs. 35%, X² = 12.93; p = 0.0003 and 56 vs. 51%, X² = 4.237; p = 0.034, respectively).

The patients who died had larger LV EDV and ESV, as well as lower EF, by both 2DE and 3DE than the survivors (Table 1). Although 2DE LV volumes were associated to worse all-cause death survival and event-free survival, 3DE LV volumes were stronger predictors of both. 3DE EDV ≤ 85 ml/m² in men and ≤ 78 ml/m² in women (X² = 15.661; p < 0.001) and 3DE ESV ≤ 34 ml/m² in men and ≤ 28 ml/m² in women (X² = 20.173; p < 0.001) were more strongly associated (all p = 0.001) with all-cause death than either 2DE EDV ≤ 74 ml/m² in men and ≤ 61 ml/m² in women (X² = 9.595; p = 0.002) or 2DE ESV ≤ 31 ml/m² in men and ≤ 24 ml/m² in women (X² = 12.544; p < 0.001) (Figure 3). Similarly, 3DE EDV ≤ 85 ml/m² in men and ≤ 78 ml/m² in women (X² = 25.998; p < 0.001) and 3DE ESV ≤ 34 ml/m² in men and ≤ 28 ml/m² in women (X² = 19.491; p < 0.001) were more strongly associated (all p = 0.001) with the composite endpoint of all-cause death or cardiac hospitalizations than either 2DE EDV ≤ 74 ml/m² in men and ≤ 61 ml/m² in women (X² = 16.794; p = 0.002) or 2DE ESV ≤ 31 ml/m² in men and ≤ 24 ml/m² in women (X² = 12.092; p < 0.001) (Figure 4).

We divided our study population into four groups, according to the ranges of the LVEF recommended by current guidelines: normal LVEF (> 52% for men, > 54% for women), mildly reduced LVEF (51-41% for men, 53%-41% for women), moderately reduced LVEF (40-30% for men and women), and severely reduced LVEF (< 30% for men and women). The grading of LV EF was more severe when 3DE was used to measure LV volumes (X² = 13.22; p = 0.0042, Figure 3) than when LV volumes were calculated using 2DE.

Kaplan Meier survival curves were derived for the different ranges of the 2DE LVEF and 3DE LVEF, respectively (Figures 3, 4).

Both 2DE and 3DE LVEF curves could significantly stratify the risk of death among the different ranges of the LVEF. However, the Kaplan Meier curves for the 3DE LVEF thresholds of LV dysfunction had a higher χ² at the log-rank discrimination analysis by comparison with the 2DE LVEF threshold values (χ² = 83.706, p < 0.001 vs. χ² = 59.752, p < 0.001). Moreover, the Receiver Operating Curves for both 2DE and 3DE LVEF threshold values showed a higher area under the curve (AUC) for the 3DE LVEF ranges than for the 2DE LVEF ranges in predicting death (0.76 ± 0.03 vs. 0.69 ± 0.04; p < 0.001).

Furthermore, only the 3DE LVEF was a significant predictor of time-to-event for each change in the LVEF category [2DE LVEF: HR 1.33 (0.88-2.01), p = 0.169; 3DE LVEF: HR 0.35 (0.24-0.52), p < 0.001]. Moreover, the HR increase for the three categories with reduced LVEF compared with the group with normal LVEF were higher when using the 3DE LVEF threshold values (mildly reduced LVEF: HR 2.40 (1.23-4.65), p = 0.01; moderately reduced LVEF: HR 7.68 (4.15-14.24), p < 0.001; severely reduced LVEF: HR 10.4 (5.67-19.09), p < 0.001; χ² = 98.26, p < 0.001) than when using the 2DE LVEF ones (mildly reduced LVEF: HR 1.46 (0.77-2.77), p < 0.25; moderately reduced LVEF: HR 3.68 (2.03-6.7), p < 0.001; severely reduced LVEF: HR 7.66 (4.43-13.25), p < 0.001; χ² = 77.10, p < 0.001).

The clinical parameters associated with all-cause death and the composite endpoint are listed in Table 2. Clinical parameters associated with all-cause death were age, dyslipidemia, permanent atrial fibrillation, ischemic heart disease, and serum levels of creatinine and hemoglobin (Table 2). Clinical parameters associated with the composite endpoint of all-cause death and cardiac hospitalization were permanent atrial fibrillation, ischemic heart disease, and serum levels of creatinine and hemoglobin (Table 2).

Clinical predictors identified at multivariable analysis as independently associated with outcome were placed in two separate stepwise regression models (CM_death and CM_composite) for all-cause death and the composite endpoint, respectively. For all-cause death, the Harrel’s C-index (HC) of the CM_death was 0.78 (95%CI 0.72-0.84). The addition of the EDV to CM_death increased the HC at the same value using both 2DE (HC = 0.79, 95%CI 0.73-0.85) and 3DE (HC = 0.79, 95%CI 0.73-0.85). Conversely, the addition of ESV (HC = 0.83, 95%CI 0.74-0.91) and EF (HC = 0.84, 95%CI 0.78-0.9) by 3DE was associated with greater HC than the same parameters obtained with 2DE (HC = 0.80, 95%CI 0.75-0.85, and HC = 0.81, 95%CI 0.76-0.86) (Figure 5).

For the composite endpoint, the HC of the CM_composite was 0.64 (95%IC 0.61-0.68). The addition of the EDV by 2DE increased the HC of the model (0.79, 95%CI 0.73-0.85), but the 3DE EDV was associated with a larger increase of the HC (0.81, 95%CI 0.75-0.88). Conversely the increase of the HC of the CM_composite obtained by adding the ESV and EF was similar for the 2DE (HC = 0.80, 95%CI 0.74-0.85) and HC = 0.84, 95%CI 0.77-0.92, respectively for 2DE ESV and EF) and 3DE (HC = 0.80, 95%CI 0.75-0.86) and HC = 0.84, 95%CI 0.78-0.90, respectively for 3DE ESV and EF) (Figure 6).

According to the reports of epidemiologic studies, cardiovascular diseases are the leading cause of death globally, and, particularly, in the most developed countries (25). In the European Union, cardiovascular diseases cause 35% of the deaths in women and men under the age of 75 years (26). In the next decade, the expected number of disability-adjusted-life-years that will be lost because of cardiovascular diseases will increase from 169 million in 2020 to 187 million in 2030¹.

Cardiovascular morbidity and mortality in the general population have been related to both non-modifiable (e.g., age, sex, genetics) (27) and potentially modifiable risk factors (28), and to the underlying cardiac condition². Accordingly, in our study patients, age, chronic kidney disease, anemia, atrial fibrillation, and ischemic heart disease were significantly associated with clinical outcomes. In addition, also the LV volumes and EF measured by both 2DE and 3DE have been associated with clinical outcomes (19, 29, 30).

Non-invasive assessment of LV volumes and EF are critically important for clinical decision-making and represents the most frequent indication for an echocardiographic study. Eligibility to device implantation of patients with LV dysfunction, discontinuation of potentially cardiotoxic chemotherapy in cancer patients, indications to cardiac surgery or to treatment initiation in asymptomatic patients are among the most important clinical decisions that rely on an accurate measurement of LV EF. LV volume calculations by 2DE is highly operator dependent, uses only limited data contained in a few predetermined tomographic planes of the LV to assess global myocardial function, and relies on geometrical assumptions that may not be necessarily valid in every patient. The geometric assumptions about LV shape associated with the 2DE algorithms make the calculations of LV volumes and EF more inaccurate in patients in whom this information is more critical (i.e., patients in whom the LV geometry is distorted because of aneurysms, or in those with extensive wall motion abnormalities) (31, 32). With 3DE, LV volumes are actually measured (and not calculated anymore) without any assumption regarding LV shape (14). This technique has been extensively validated using the cardiac magnetic resonance as a reference modality (10, 15), and was demonstrated to be more timesaving, reproducible, repeatable and accurate than conventional 2DE for both LV volumes and EF measurements (11, 33–35).

Caselli et al. (36) showed that, in a limited cohort of 178 patients, 3DE LV volumes and EF had a significant association with the composite endpoint of death, myocardial infarction or stroke but not with the primary endpoint of cardiovascular death. Similarly, Mancuso et al. (37) showed that, in 89 patients with systolic heart failure, the LVEF measured by 3DE was an independent predictor for a composite endpoint of death, cardiac transplantation and hospitalization for HF, whereas LV volumes were not. However, in addition to the smaller number of patients and the specific clinical settings of this study, it should be noted that, in contrast with available data, the LV volumes measured by 3DE were smaller than those obtained with 2DE. In our study, the LV volumes measured by 3DE were significantly larger than those obtained by 2DE, in accordance to other studies that showed the higher accuracy of the 3DE for the measurement of the LV volumes in comparison to cardiac magnetic resonance (10). New 3DE technologies combined with an increased experience in 3DE of the cardiologists who perform echocardiography allow encompassing larger LVs into the 3DE dataset at a good volume rate, allowing good endocardial delineation and better measurements of the LV volumes. A previous study performed in unselected patients with a wide range of LV volumes showed that the 3DE LV volumes measured by different echocardiography systems were similar and had a better accuracy than 2DE when compared to cardiac magnetic resonance (38).

The present study adds to previous ones by showing that LV volumes and EF measured by 3DE provide incremental prognostic value over 2DE also in patients referred for a routine and clinically-indicated echocardiography study. Both LV volumes and EF calculated by 2DE were independently associated with either cardiac death or the composite of cardiac death and hospitalization for cardiac causes after adjusting for covariates. However, 3DE LV ESV and EF were able to significantly increase the power of the predictive model when added to a model including the clinical variables. These findings suggested that LV ESV and EF obtained from 3DE were superior to those derived from 2DE to predict cardiac death and the need of hospitalization for cardiac issues.

When using LVEF values obtained from 3DE to grade the extent of LV dysfunction severity into normal function, and mild, moderate and severe LV dysfunction, LVEF measured by 3DE showed higher discriminant power for survival than LVEF measured by 2DE. These results are consistent with the findings by Stanton et al. (19) who compared the occurrence of the composite endpoint between patients with “normal” and “abnormal” LVEF measured by 2DE and 3DE. LVEF measured by 3DE was also an independent predictor of major arrhythmic events and improved the ability to predict the arrhythmic risk in 172 patients with LVEF below 50% (39). In that study, when compared with 2DE LVEF, 3DE-measured LVEF changed the indication to implantable cardioverter-defibrillator implant in 20% of the patients. Finally, in our patients, only the 3DE measurement of the LVEF was an independent predictor of time to event for each change in the LVEF.

The 3D LVEDV threshold values used to identify patients with dilated LV provided better stratification of the risk of death compared to 2D LVEDV. Dilated LVESV according to the 3DE cut-offs proposed by Muraru et al. (21) in normal individuals showed to be a predictor for time to event, as well, while the 2D LVESV cut-offs offered by the current guidelines for LV dilation did not (17).

Our findings showed that, although LV volumes obtained by 2DE and 3DE are significantly correlated, 3DE might be better than 2DE in assessing the severity of LV dysfunction and the prognosis of the patients referred to the echocardiography laboratory for routine, clinically indicated study, and physicians should use parameters measured by 3DE to better guide patients’ management. Future research should focus on whether 3DE can improve the predictive value in larger and prospective cohorts of consecutive patients, and whether 3DE guided therapy can improve clinical outcomes.