Accuracy of echocardiographic estimates of pulmonary artery pressures in pulmonary hypertension: insights from the KARUM hemodynamic database

Accurate assessment of pulmonary artery (PA) pressures is integral to diagnosis, follow-up and therapy selection in pulmonary hypertension (PH). Despite wide utilization, the accuracy of echocardiography to estimate PA pressures has been debated. We aimed to evaluate echocardiographic accuracy to estimate right heart catheterization (RHC) based PA pressures in a large, dual-centre hemodynamic database. Consecutive PH referrals that underwent comprehensive echocardiography within 3 h of clinically indicated right heart catheterization were enrolled. Subjects with absent or severe, free-flowing tricuspid regurgitation (TR) were excluded. Accuracy was defined as mean bias between echocardiographic and invasive measurements on Bland–Altman analysis for the cohort and estimate difference within ± 10 mmHg of invasive measurements for individual diagnosis. In 419 subjects, echocardiographic PA systolic and mean pressures demonstrated minimal bias with invasive measurements (+ 2.4 and + 1.9 mmHg respectively) but displayed wide limits of agreement (− 20 to + 25 and − 14 to + 18 mmHg respectively) and frequently misclassified subjects. Recommendation-based right atrial pressure (RAP) demonstrated poor precision and was falsely elevated in 32% of individual cases. Applying a fixed, median RAP to echocardiographic estimates resulted in relatively lower bias between modalities when assessing PA systolic (+ 1.4 mmHg; 95% limits of agreement + 25 to − 22 mmHg) and PA mean pressures (+ 1.4 mmHg; 95% limits of agreement + 19 to − 16 mmHg). Echocardiography accurately represents invasive PA pressures for population studies but may be misleading for individual diagnosis owing to modest precision and frequent misclassification. Recommendation-based estimates of RAPmean may not necessarily contribute to greater accuracy of PA pressure estimates. Supplementary Information The online version contains supplementary material available at 10.1007/s10554-021-02315-y.


Introduction
Accurate hemodynamic evaluation of pulmonary hypertension (PH) is essential for early disease identification, selection for potential therapy and during follow-up. Although PH diagnosis is established using right heart catheterization (RHC), transthoracic echocardiography is recommended for screening patients [1] and routinely utilized to quantify pulmonary artery (PA) pressures in addition to offering prognostic insight [2].
The accuracy and precision of echocardiography to assess PA pressures has been debated. Multiple earlier studies suggest that echocardiographic estimates of PA pressures are frequently innacurate [3][4][5][6], while more recent publications suggest good diagnostic accuracy [7][8][9]. These paradoxical observations may be attributed to diverse methodological 1 3 approaches to assess accuracy in the aforementioned studies [9], and compounded to some degree by varying recommendations to quantify PH using echocardiography [1,10]. More recently, D'Alto and colleagues demonstrated high echocardiographic accuracy to estimate both PA mean (PAP mean ) and systolic pressures (PAP systolic ) employing Bland-Altman analysis, suggesting appropriate utility in population studies [9]. However, modest precision represented by wide limits of agreement in that study advocates greater caution when employing echocardiography to estimate PH severity on an individual basis.
Given the practicality, low cost and low risk of echocardiography, this study was undertaken to investigate its accuracy to estimate PA pressures in a large, prospective, dual-centre database of PH referrals. Further, we wished to study the contribution of mean right atrial pressure (RAP mean ) estimates to PA pressure estimation by comparing the accuracy of the recommended approach that takes into consideration patient-specific mean right atrial mean pressure (RAP mean ) estimates [10], and a simplified model that applies a fixed, median RAP to estimate PAP systolic and PAP mean in all subjects.

Echocardiographic evaluation
All patients underwent comprehensive echocardiography within 3 h of catheterization at both centers employing a Vivid E9 ultrasound system (GE Ultrasound, Horten, Norway) in keeping with current recommendations [10]. Pharmacological status was unaltered between echocardiography and RHC examinations. All studies were performed by credentialed echocardiographers with > 15 years' experience at each center (PL/AV). 2D gray-scale images were acquired at 50-80 frames/sec and Doppler tracings were recorded using a sweep speed of 100 mm/sec. Three consecutive heart cycles were acquired in sinus rhythm and 5 in the setting of atrial fibrillation (AF). TRV max was measured with Continuous wave Doppler, considering the most optimal signal obtained from multiple echocardiographic windows. RAP mean was estimated by evaluating inferior vena cava (IVC) size and collapsibility with patients in a supine position, taking care to maximize IVC diameter both during relaxed respiration and with rapid inspiration. All images were subsequently exported and analyzed offline (EchoPAC PC, version 11.0.0.0 GE Ultrasound, Waukesha, Wisconsin) by experienced investigators (PL/AV) blinded to catheterization data.
Subjects with absent or poor TR signal quality, in addition to those with a flail tricuspid valve, endocarditis or a coaptation defect resulting in massive, free-flowing jet were subsequently excluded from the analysis. TR severity was assessed semi-quantitatively and graded as mild (grade 1), moderate (grade 2) and moderately-severe to severe (grade 3). In keeping with American Society of Echocardiography/European Association of Cardiovascular Imaging (ASE/EACVI) recommendations, RAP mean was estimated as 3 mmHg if the IVC diameter was < 2.1 cm and collapsed > 50% during rapid inspiration and 15 mmHg if the IVC diameter was ≥ 2.1 cm and collapsed < 50%. In scenarios where IVC size and dynamics did not fit this paradigm (IVC diameter < 2.1 cm with < 50% collapse and IVC diameter ≥ 2.1 cm with > 50% collapse), RAP mean was estimated as 8 mmHg [10]. In the simplified model, median RAP was uniformly applied to corresponding TRV max gradients to estimate PAP systolic . PAP mean was calculated as 0.6 × PAP systolic + 2 [11] using both recommended [10] and fixed, median RAP estimates.

Invasive evaluation
RHC was performed by experienced operators blinded to echocardiography examinations at each center using a 6F Swan Ganz catheter employing jugular or femoral vein access. After suitable calibration with the zero-level set at the mid-thoracic line, pressure measurements were taken from the right atrium (RA), right ventricle (RV) and PA during end-expiration. Five to ten cardiac cycles were acquired and all pressure tracings were stored and analyzed offline using a standard hemodynamic software package (WITT Series III, Witt Biomedical Corp., Melbourne, FL).

Statistical analysis
Normality was tested using the Shapiro-Wilk test and visually reaffirmed using QQ plots. Continuous variables were expressed as mean ± SD for parametric variables or median (interquartile range) for non-parametric variables and categorical variables were expressed as numbers and percentage. Correlations between echocardiographic and corresponding invasive measurements was performed using the Pearson's 2-tailed test (correlation between 2 continuous variables). Receiver operating characteristics (ROC) curve was employed to illustrate diagnostic potential of each chosen variable. An invasive PA mean pressure (PAP mean ) ≥ 25 mmHg was chosen to represent PH and RAP mean ≥ 7 mmHg, to represent an elevated RAP mean [12]. Sensitivity, specificity, negative predictive value (NPV) and positive predictive value (PPV) were measured. Accuracy was assessed both for individual diagnosis and at the cohort level. Accuracy for individual diagnosis was predefined as an estimate difference within ± 10 mmHg of invasive measurements. Accuracy at the cohort level was defined as the mean bias between echocardiographic and invasive measurements on Bland-Altman analysis. IBM SPSS statistics version 23.0 was employed for analysis.

Study population
Of 480 subjects enrolled across the two sites, 47 (10%) patients with no TR and 14 (3%) with a coaptation defect resulting in severe, free-flowing TR were excluded, yielding 419 patients [Karolinska: n = 296 (70%); Umeå: n = 123 (30%)] for analysis. Clinical characteristics, invasive and echocardiographic data are provided in Table 1. Fifty-two percent of the subjects were female. Twenty percent (n = 86) presented with AF and 7% (n = 31) were on pacemaker therapy. A wide range of invasive pressures were observed for RAP mean (1-29 mmHg), PAP mean (7-99 mmHg) and PAP systolic (12-136 mmHg). One hundred and seventy-nine patients (44%) presented with reduced RV systolic function as suggested by TAPSE < 16 mm [10]. Echocardiographic images of the IVC were either not available or did not permit optimal evaluation in a small fraction (n = 32; 7.6%). Two hundred and forty patients (57%) presented with mild TR, 122 (29%) with moderate TR, and 57 (14%) with severe TR. An illustration of echocardiographic evaluation of PA pressures is provided in Fig. 1.

Accuracy of TRV max to identify presence of pH
TRV max demonstrated strong association with invasive PAP mean (r = 0.75, p < 0.001) and a cut-off of 2.8 m/sec demonstrated good ability to identify PH, defined as invasive PAP mean ≥ 25 mmHg (AUC = 0.87, CI 0.84-0.91, p < 0.001). Sensitivity analysis for different echocardiographic cut-offs to is presented in Table 2. At 2.8 m/sec, TRV max demonstrated 89% sensitivity and 62% specificity to identify PH, with a 38% false positive rate. Forty-five patients (15%) with a TRV max > 2.8 m/sec demonstrated normal PA pressures on RHC. At 3.4 m/sec, TRV max demonstrated 94% specificity and 62% sensitivity, and false positive rate fell to 5.9%. Even when balanced sensitivity and specificity was identified at a 3.0 m/sec cut-off (80% sensitivity, 80% specificity), a 20% false positive rate was observed. Supplementary sensitivity analysis was also performed considering PAP mean ≥ 20 mmHg which revealed similar results (Online Appendix Table 1). On Bland-Altman analysis, echocardiographic TR gradient demonstrated a mean bias of + 2.5 mmHg with invasive RV-RA gradient (95% limits of agreement + 23 to − 18 mmHg).

Accuracy of echocardiography to evaluate invasive PAP mean
Echocardiographic PAP mean incorporating recommendationbased RAP estimates [10] demonstrated strong association with invasive PAP mean (r = 0.81, p < 0.001) (Fig. 4a). Bias and limits of agreement between echocardiographic estimates of PAP mean and RHC are presented in Table 3.

Discussion
In the large, dual-centre KARUM hemodynamic database, echocardiographic PA pressures were reasonably accurate, demonstrating strong association and minimal bias with corresponding pressures obtained by RHC. However, wide limits of agreement in addition to frequent misclassification suggests modest precision and precludes wider echocardiographic utilization to quantify PH severity in individual cases. An important observation was that recommended echocardiographic estimates of RAP mean were falsely elevated in more than 1 in 3 subjects and incorporation of these estimates to calculate PAP systolic and PAP mean resulted in relatively higher mean bias with RHC than when the median estimate was considered for all subjects.
Interest in the utility of echocardiography to estimate PA pressures emerged with early studies suggesting a significant correlation between TR-derived estimates and invasive pulmonary pressures [13][14][15]. Since then, estimation of TRV max has been routinely utilized to grade PH probability [1,16] and to derive PA systolic pressures using the Bernoulli equation. In keeping with earlier studies, we demonstrate that despite a significant correlation between invasive and echocardiographic measurements, frequent misclassifications may occur when individual cases are considered [3,4]. A number of reasons have previously been proposed to explain poor accuracy in the setting of specific cases, and these have been considered and explored in our work. First, Fig. 4 a Scatter plot demonstrating correlation between PAP mean estimated by echocardiography employing current ASE/EACVI recommendations and RHC. b Bland-Altman plot demonstrating agreement between PAP mean estimated by echocardiography employing current ASE/EACVI recommendations and RHC Fig. 5 a Scatter plot demonstrating correlation between PAP mean estimated by echocardiography employing RAP = 7 mmHg and RHC. b Bland-Altman plot demonstrating agreement between PAP mean estimated by echocardiography employing RAP = 7 mmHg and RHC 1 3 poor agreement with invasive pressures has been previously documented in subjects with absent [17] and severe, freeflowing TR secondary to a coaptation defect [18]. Both these groups were excluded from our analysis to circumvent any bias or negative influence on our results. Second, application of the Bernoulli equation to TR velocity to calculate pressure gradient is inherently error-prone, as even small errors in absolute measurement result in exponential differences in PAP systolic estimates. Certain international recommendations hence encourage the use of absolute velocities instead [1,16]. Our data shows that the recommended 2.8 m/sec cut-off for intermediate-probability PH misclassified one in three subjects and raising the cut-off to 3.4 m/sec resulted in a drop in sensitivity, thereby warranting re-evaluation of these recommended values. Finally, the integration of IVC-derived RAP estimates to corresponding TR gradients is recommended to calculate PAP systolic [10] . The reliability of this method to represent invasive RAP mean has been debated [19][20][21][22][23][24], with certain studies suggesting modest or no association [20,21,24]. While IVC-estimated RAP demonstrated good ability to identify elevated invasive RAP mean in our study, accuracy of recommended cut-offs to represent invasive pressures was poor and may explain the frequent misrepresentation of pulmonary systolic pressures. When a fixed median RAP of 7 mmHg was considered in the population, association was still strong and bias between echocardiographic and invasive PAP systolic and PAP mean readings was actually lower in our study, suggesting that these recommended estimates may offer no additional advantage to PA pressure assessment [25]. A recent study suggests that echocardiography frequently underestimates PA pressures owing to the inability to accurately assess elevated RA pressures during exercise [7]. Our findings suggest that the echocardiographic assessment of RAP mean is frequently inaccurate even during rest, and results in frequent overestimation of invasive pressures. Objective assessment of the IVC demonstrates inherent technical limitations related to excessive translation during rapid inspiration [26] and has been previously reported to be inaccurate in athletes [27] and patients on mechanical ventilation [24]. Recent studies suggest that advanced techniques such as speckle-tracking based right atrial reservoir strain [28] and 3D volumes [29] may provide a more accurate measure of RAP mean , but these findings require further validation in larger cohorts.
Our study also corroborates earlier findings that suggest modest echocardiographic precision to reflect invasive PA pressures when individual cases are concerned [3,4,9,30], but an appropriate method for population studies given its high accuracy at a cohort level [9]. Echocardiography remains a practical, inexpensive and safe screening tool to arouse suspicion of PH, offers additional etiological insight in addition to complementary information on ventricular structural and/or functional aberrations. Further, aggravations in TR severity assessed by Doppler have been associated with worsening prognosis irrespective of PA pressures and right heart failure [31]. From a clinical stand point, our study suggests that echocardiography is useful to raise suspicion of PH and accurately represents invasive PA pressures for population studies, but sole reliance to quantify PA pressure elevations for individual diagnosis may be frequently inaccurate. A diagnostic algorithm that combines hemodynamic information with structural indices of right-heart structure and function may vastly improve accuracy of non-invasive PH estimation for individual cases and needs to be explored.
The use of fluid-filled catheters instead of high-fidelity manometer-tipped catheters for pressure measurement might introduce additional error and may be considered a limitation in this study. Additionally, we did not adopt a core lab approach to evaluation of echocardiographic images in this study and inter-operator and inter-evaluator variability may be considered a limitation. However, a standard international acquisition and analysis protocol was followed by two experienced echocardiographers with over 15 years' experience. Finally, we did not employ agitated saline to boost weak TR signals in this study, but chose instead to exclude unanalyzable signal registrations from the analysis. Fewer cases may have been excluded with contrast use and this may be considered a limitation.

Conclusions
Echocardiography accurately represents invasive PA pressures and is, hence, appropriate for population studies. However, modest precision and frequent misclassification preclude its utility for evaluation of PH severity in individual cases. Recommendation-based estimates of RAP mean frequently overestimate corresponding invasive measurements and do not necessarily contribute to greater accuracy of pulmonary artery (PA) pressure estimates.
Author contributions AV and PL conceptualized the study, acquired and analysed data and drafted the work; NS and HT analysed data, BK, LHL and ET critically reviewed the draft and provided intellectual inputs. All authors reviewed and approved the final manuscript.