Left atrial reservoir strain improves diagnostic accuracy of the 2016 ASE/EACVI diastolic algorithm in patients with preserved left ventricular ejection fraction: insights from the KARUM haemodynamic database

Abstract Aims This study aimed to investigate the incremental value offered by left atrial reservoir strain (LASr) to the 2016 American Society of Echocardiography/European Association of Cardiovascular Imaging (ASE/EACVI) diastolic algorithm to identify elevated left ventricular (LV) filling pressure in patients with preserved ejection fraction (EF). Methods and results Near-simultaneous echocardiography and right heart catheterization were performed in 210 patients with EF ≥50% in a large, dual-centre study. Elevated filling pressure was defined as invasive pulmonary capillary wedge pressure (PCWP) ≥15 mmHg. LASr was evaluated using speckle-tracking echocardiography. Diagnostic performance of the ASE/EACVI diastolic algorithm was validated against invasive reference and compared with modified algorithms incorporating LASr. Modest correlation was observed between E/e′, E/A ratio, and LA volume index with PCWP (r = 0.46, 0.46, and 0.36, respectively; P < 0.001 for all). Mitral e′ and TR peak velocity showed no association. The ASE/EACVI algorithm (89% feasibility, 71% sensitivity, 68% specificity) demonstrated reasonable ability (AUC = 0.69) and 68% accuracy to identify elevated LV filling pressure. LASr displayed strong ability to identify elevated PCWP (AUC = 0.76). Substituting TR peak velocity for LASr in the algorithm (69% sensitivity, 84% specificity) resulted in 91% feasibility, 81% accuracy, and stronger agreement with invasive measurements. Employing LASr as per expert consensus (71% sensitivity, 70% specificity) and adding LASr to conventional parameters (67% sensitivity, 84% specificity) also demonstrated greater feasibility (98% and 90%, respectively) and overall accuracy (70% and 80%, respectively) to estimate elevated PCWP. Conclusions LASr improves feasibility and overall accuracy of the ASE/EACVI algorithm to discern elevated filling pressures in patients with preserved EF.


Introduction
Chronic dyspnoea and fatigue are common symptoms in heart failure (HF) and pose a significant diagnostic challenge. These typical HF symptoms are often non-specific and lack accuracy to be independently employed for diagnosis, necessitating additional investigations. 1 Doppler echocardiography is integral to routine evaluation in the setting of suspected HF and provides vital information on cardiac size, left ventricular (LV) performance, and valvular function. In the setting of HF with preserved ejection fraction (EF) (HFpEF), Doppler imaging combined with 2D echocardiography provides information on LV filling pressure critical to diagnosis and optimal therapy regulation.
In 2016, the American Society of Echocardiography (ASE) and European Association of Cardiovascular Imaging (EACVI) introduced an algorithm incorporating multiple common echocardiographic variables to ascertain LV filling pressure status. 2 Studies evaluating the diagnostic accuracy of this algorithm in patients with normal EF, however, are limited and suggest poor or modest performance. [3][4][5][6] Left atrial reservoir strain (LAS r ) has been proposed as a novel, non-invasive diagnostic for grading diastolic dysfunction severity in patients with preserved EF. 7 In a recent study, Inoue et al. 8 demonstrated that LAS r and LA pump strain demonstrate stronger ability to identify elevated filling pressures than conventional echocardiographic measures in a large HF population. However, LAS r displayed generally poorer diagnostic performance in patients with preserved when compared with reduced LV systolic function in this study. Subsequently, the EACVI recommendations for multimodality imaging in HFpEF have suggested that incorporating LAS r into the ASE/EACVI diagnostic algorithm may improve its diagnostic performance. 9 With this background, we examined the additional diagnostic value offered by LAS r to the 2016 ASE/EACVI diastolic algorithm in a large,

Graphical Abstract
A. Venkateshvaran et al. dual-centre, haemodynamic database of patients with unexplained dyspnoea, and preserved LV EF employing invasive Pulmonary capillary wedge pressure (PCWP) as reference.

Study population
We retrospectively analysed all patients enrolled in the KARUM study, a dual-centre database of patients with unexplained dyspnoea undergoing near-simultaneous echocardiography and right heart catheterization (RHC). Details of the total KARUM cohort have been previously published. 10 Patients were evaluated at two referral centres in Sweden; Karolinska University Hospital in Stockholm between 2014 and 2018, and Norrlands University Hospital in Umeå between 2010 and 2015. All subjects were haemodynamically stable during assessment. Prior to inclusion, we excluded patients with acute coronary syndrome, valvular prosthesis, and left bundle branch block. Thereafter, we excluded specific conditions where diastolic function assessment using the ASE/EACVI algorithm has significant limitations. 2 This included patients with atrial fibrillation, pacemaker, hypertrophic or restrictive cardiomyopathy, and those that had undergone a heart transplantation or presented with >mild concomitant mitral valve disease. Finally, we excluded patients with EF <50% or poor echocardiographic image quality. Study protocol was approved by the local ethics committee at each centre (Karolinska: DNR 2008/1695-31 and Norrland: 07-092M, 2014-198-32M, and 2017-102-32M). All patients provided written informed consent.

Echocardiography
Comprehensive echocardiography was performed in both centres by experienced echocardiographers (A.V. and P.L.) employing commercial ultrasound systems (Vivid E9, GE Ultrasound, Horten, Norway) in keeping with current recommendations. 11 Pharmacological status was unaltered between echocardiography and catheterization. Digital loops were stored and analysed offline (EchoPAC PC, version 11.0.0.0 GE Ultrasound, Waukesha, Wisconsin) by experts (A.V. and P.L.) blinded to catheterization data. Mitral inflow interrogation was performed using pulsed-wave (PW) Doppler in the apical four-chamber view, placing the sample volume at the mitral leaflet tips. Transmitral early (E) and late diastolic velocities (A) were obtained and mitral E/A ratio computed. Doppler tissue imaging was utilized to measure early mitral annular velocities (e 0 ) at both septal and lateral walls and subsequently averaged to calculate E/e 0 mean . Maximal tricuspid regurgitation (TR) peak velocity was measured using continuous-wave (CW) Doppler. Left atrial (LA) volume was obtained from apical four-and two-chamber views and indexed for body surface area.
Assessment of diastolic function and filling pressure status was performed employing the recommended two-step ASE/EACVI approach. 2 In Step 1, diastolic function was evaluated considering recommended cutoffs provided for e 0 (septal and/or lateral), E/e 0 mean , TR peak velocity, and indexed LA volume and graded as normal (where <50% of the four echocardiographic parameters were positive), indeterminate (50% positive), or definite diastolic dysfunction (>50% positive). In Step 2, filling pressures were assessed as normal, indeterminate, or elevated among patients with diastolic dysfunction employing transmitral E-velocity and E/ A ratio and supplemented by E/e 0 mean , TR peak velocity and indexed LA as recommended. 2 LAS r was assessed using 2D speckle tracking echocardiography in keeping with expert recommendation. 12 In a nonforeshortened apical four-chamber view, the LA endocardial border was traced, taking care to exclude the appendage and pulmonary veins. The region of interest (ROI) was then visually inspected for quality of tracking, repeated if found inadequate and excluded if still suboptimal. Zero point was set at the onset of the QRS complex on the ECG. LAS r was defined as the maximal inflection point above the baseline during ventricular systole and assigned a positive value. Acquisition of LAS r using speckletracking echocardiography is illustrated in Figure 1.

Right heart catheterization
RHC was performed immediately after echocardiography (within a 3h period) during the same visit by experts blinded to imaging results at each centre using fluid-filled 6F Swan-Ganz catheters employing jugular or femoral vein access. Transducers were zeroed at midthoracic level in each patient. PCWP was measured at mid-A wave and averaged from a minimum of five heart cardiac cycles. All pressure tracings were digitally stored and analysed offline using standard haemodynamic software package (WITT Series III, Witt Biomedical Corp., Melbourne, FL, USA).

Reproducibility analysis
Intra-and inter-observer variability analysis for LAS r was performed in 40 randomly selected patients. Measurements were performed first by the same operator on two subsequent days considering the same heart cycle, and then by independent readers blinded to each other in addition to invasive pressure readings. Measurement variability was expressed in coefficient of variation and intra-class coefficients.

Statistical analysis
Normality was tested using the Shapiro-Wilk test and visually reaffirmed using QQ plots. Continuous variables were expressed as mean ± standard deviation for parametric variables or median (interquartile range) for non-parametric variables. Categorical variables were expressed as numbers and percentage. Correlations between PCWP and individual echocardiographic parameters were performed using the Pearson's two-tailed test. Multiple linear regression models were generated to evaluate independent associations of echocardiographic variables with PCWP. Sensitivity, specificity, negative predictive value (NPV), positive predictive value (PPV), and overall accuracy were measured for each echocardiographic cut-off (as recommended by current guidelines) 2 using standard definitions. Inter-technique agreement between echocardiographic algorithms and invasive PCWP was tested using

Patient characteristics
A flow chart of patient selection in our study is presented in Figure 2.
Of 480 patients referred for catheterization at both centres, 210 met the inclusion criteria. Characteristics of the patient population are presented in Table 1. Seventy-four (35%) patients had arterial hypertension, 18 (9%) had diabetes, 38 (18%) had ischaemic heart disease,

Correlates of PCWP
Correlations of clinical and echocardiographic variables with invasive PCWP are presented in Supplementary data online, Table S1. Moderate significant correlations were observed between PCWP and mitral E-wave (r = 0.50), E/A ratio (r = 0.46), and E/e 0 mean ratio (r = 0.46, P < 0.05 for all). Indexed LA volume demonstrated a mild significant correlation (r = 0.36; P < 0.001). TR and e 0 velocities demonstrated no association with PCWP (P > 0.05 for all). LAS r demonstrated a significant inverse relationship with PCWP (r = -0.37,     Non-invasive assessment of left ventricular filling pressure P = 0.003) remained correlated with invasive PCWP (r 2 = 0.34; P < 0.001). In the second model, when LAS r was considered in addition to the above-listed variables, E/e 0 mean was no longer significantly associated with PCWP (P = 0.10). Here, mitral E-wave (standardized b coefficient 0.20; P = 0.02), E/A ratio (standardized b coefficient 0.24; P < 0.001), and LAS r (standardized b coefficient -0.26; P < 0.001) emerged as independent variables associated with PCWP (r 2 = 0.41; P < 0.001).

Additional diagnostic contribution of LAS r to ASE/EACVI algorithm
Next, we studied the additional value LAS r adds seperately to Step 1 (that identifies diastolic dysfunction) and Step 2 (that identifies elevated filling pressures) of the conventional diastolic algorithm.
Step 1 could be assessed in 175 patients (83%), where all four variables (septal or lateral e 0 , E/e 0 , TR velocity, and indexed LA volume) were available. In this analysis, 70 patients (40%) were classified as having normal diastolic function, 60 patients (34%) as indeterminate, and 45 (26%) as having diastolic dysfunction. Among those with indeterminate diastolic dysfunction, LAS r cut-offs identified elevated PCWP with reasonable to good sensitivity (63% to 88% depending on cutoff employed) and modest accuracy (Supplementary data online, Table S4). Further, LAS r improved overall accuracy of Step 1 in the ASE/EACVI algorithm, irrespective of cut-off chosen ( Table 4).
Estimation of LV filling pressures employing Step 2 of the ASE/ EACVI algorithm was feasible in 187 patients (89%). Assessment of transmitral E velocity and mitral E/A ratio classified 26 patients (12%) as having normal, 167 (80%) as requiring additional criteria (E/e 0 mean , TR velocity, and indexed LA volume) and 17 (8%) as having elevated LV filling pressures. When additional parameters were also assessed, Step 2 of the ASE/EACVI algorithm classified 110 patients (52%) as having normal LV filling pressures, 23 patients (11%) as having indeterminate filling pressures, and 77 patients (37%) having elevated filling pressures.
We studied the potential value LAS r adds to the conventional diastolic algorithm to assess LV filling pressures using three different models. Salient findings of this analysis have been presented in Figure 5 and the Graphical Abstract. In Model 1, we substituted TR peak velocity >2.8 m/s (which demonstrated lowest feasibility and poor overall accuracy to detect elevated filling pressures) with LAS r <18%. This approach (sensitivity 69%, specificity 84%) demonstrated higher feasibility (91% vs. 89%), overall accuracy (81% vs. 68%), and greater agreement with invasive measurements (OE coefficient 0.48 vs. 0.30) when compared with the ASE/EACVI approach ( Table 5). Comparison of ROC curves demonstrated higher diagnostic performance to identify elevated filling pressures using Model 1 when compared with the conventional algorithm (AUC = 0.77 vs. 0.69, P = 0.001). In Model 2, in keeping with the EACVI recommendations for multimodality imaging in HFpEF, 9 when one of the additional variables (E/e 0 , TR velocity, or indexed LA volume) were missing and the other two were conflicting, the missing variable was replaced with LAS r . This approach (sensitivity 71%, specificity 70%) demonstrated higher feasibility than the ASE/EACVI algorithm (98% vs. 89%), in addition to marginally higher specificity (70% vs. 68%), accuracy (70% vs. 68%), and greater agreement with invasive measurements (OE coefficient 0.32 vs. 0.30). Finally, in Model 3, we considered LAS r in addition to conventional parameters used in the ASE/EACVI algorithm. To clarify, after first evaluating filling pressures based on transmitral velocities and E/A ratio, we added LAS r to the additional criteria (E/e 0 , TR velocity, and indexed LA volume). Using this approach, an elevated filling pressure status was assigned if >50% of the additional criteria (3 or 4) were found positive, and non-elevated if < _50% (1 or 2) were found positive. This approach also demonstrated higher specificity (84% vs. 68%), accuracy (80% vs. 68%), agreement with invasive assessment (OE coefficient 0.47 vs. 0.30), and a tendency to higher diagnostic performance on ROC analysis (AUC = 0.75 vs. 0.69; P = 0.06) when compared with the ASE/EACVI algorithm ( Table 5).

Feasibility and reproducibility
Measurement of echocardiographic surrogates of diastolic dysfunction in the ASE/EACVI algorithm were highly feasible (Supplementary data online, Table S2). TR peak velocity could be measured in 89% of cases. Measurement of LAS r could be performed in 94% of patients. Double measurements in 40 randomly selected patients Non-invasive assessment of left ventricular filling pressure demonstrated a coefficient of variation of 10% with an intra-class correlation coefficient of 0.91 (95% CI 0.73-0.96). Test-retest analysis for LAS r yielded a coefficient of variation of 12.8%.

Discussion
In a large, dual-centre, haemodynamic database of patients with unexplained dyspnoea and preserved EF, LAS r was independently associated with invasive PCWP and demonstrated strong diagnostic ability to identify elevated LV filling pressure. Applying LAS r to the 2016 ASE/EACVI diastolic algorithm enhanced feasibility of the recommended approach, in addition to overall accuracy and agreement with invasive filling pressure. Our findings support an important role for LAS r in routine echocardiographic assessment of LV filling pressure status.

Diagnostic performance of ASE/EACVI guidelines
Since their introduction in 2016, the ASE/EACVI recommendations for evaluation of diastolic function have been validated in multiple cohorts but have shown conflicting results. Initial studies performed in patients with a wide spectrum of cardiac disease including both reduced and preserved EF suggest generally good diagnostic performance. [14][15][16][17] Most of these studies included patients with suspected coronary artery disease referred for left heart catheterization 14,17 and, by design, excluded patients with pulmonary disease. Further, even in these studies, a generally poorer sensitivity was observed when patients with preserved EF were exclusively considered. 14 Recently, in a smaller cohort (n = 63) of patients with pulmonary hypertension of which 44% had left heart disease, the ASE/ EACVI algorithm demonstrated high sensitivity and specificity (84% and 80%, respectively) to identify elevated filling pressures, defined as PCWP >12 mmHg. 18 The study, however, did not exclude all cardiac conditions where recommendations suggest that the ASE/EACVI algorithm is less reliable 2 and this, in addition to a lowered PCWP cut-off, may have influenced sensitivity analysis. We chose to strictly mirror selection to exclude challenging sub-populations in keeping with recommendations. 2 Despite careful patient selection, e 0 (irrespective of whether septal or lateral velocities were considered) and TR peak velocity demonstrated no significant association with invasive PCWP and poor ability to distinguish elevated filling pressure in our cohort. Lack of association of these specific variables with PCWP has been recently reported in another stringently selected cohort and may be attributed to the inclusion of patients with pulmonary disorders in both our and the above-mentioned cohort. 6 Further, E/e 0 , which is widely utilized and has the most evidence as per recent systematic reviews 3 demonstrated only a modest correlation with PCWP and showed no significant association with PCWP when LAS r was introduced into our multivariable regression model.
Differentiating HFpEF from pulmonary disease is a common clinical conundrum in patients presenting with unexplained breathlessness. While our findings may not be applicable to a general HFpEF population, they are consistent with what is seen in clinical practice at specialist PH centres and provide a real-world context. Furthermore, our findings are consistent with other studies in patients with preserved EF that suggest that individual variables demonstrate poor ability to represent invasive pressures, and advocate a multiparametric approach to filling pressure assessment. 3,5,6 Incremental value of LAS r to filling pressure assessment Studies showcasing modest ASE/EACVI algorithm discriminatory performance suggest that complementary scores and/or more advanced investigative approaches may improve diagnostic accuracy and reduce indeterminate classification. 6 In a recent EACVI survey to study adoption of the ASE/EACVI guidelines in routine practice, LAS r was chosen by 34% of respondents as a method of choice in challenging diagnostic scenarios. 19 Additionally, LAS r has been incorporated in a revised algorithm in the recent EACVI recommendations for multimodality imaging in HFpEF. 9 LA phasic function can be evaluated during the reservoir, conduit, or contractile phase and changes in strain has been shown to be independent of LA volume in HFpEF. 20 We chose to evaluate deformation during the reservoir phase, LAS r , based on high feasibility, wider utilization, and greater agreement with worsening diastolic dysfunction when compared with other LA phasic strain measures. 7 LAS r is reduced in diastolic dysfunction, demonstrates better agreement with invasive LV filling pressure when compared with conventional algorithms when EF is preserved, 21 and accurately grades diastolic dysfunction severity. 7 Further, LAS r has previously been shown to correlate well with elevated filling pressure in HF during rest, 22 and has demonstrated strong ability to discern disproportional pressure rise during exercise. 23 Our data suggest that incorporating LAS r into the current diastolic algorithm further enhances its feasibility, accuracy, and diagnostic performance and, hence, may be a promising clinical tool. Enhanced accuracy seen in models considering LAS r in this study was driven predominantly by improved specificity of the revised approaches to determine elevated filling pressures. The stronger ability of these modified echocardiographic approaches to 'rule-in' diastolic dysfunction by better identification of those with non-elevated filling pressure may further enhance utility to screen and triage patients with unexplained breathlessness. Given that LA filling corresponds with LV long-axis shortening during systole, further studies are necessary to establish additional information LAS r provides over what LV longitudinal strain already offers.

Clinical relevance
Our study has important clinical implications in the management of unexplained dyspnoea, as accurate identification of the underlying cause has direct consequences on choice of therapy. This is even more relevant in the setting of normal EF, where focused assessment of LV size and systolic function may not suitably explain the patient's symptoms. While the accuracy of LA pump strain to identify normal filling pressure in preserved EF has recently been suggested, 8 our data add that LAS r may also offer additional value when complementing the 2016 ASE/EACVI recommendation as suggested recently by the EACVI recommendations for multimodality imaging in HFpEF. 9 Limitations Important limitations of the current study are the low proportion of patients with elevated PCWP and inclusion of patients with pulmonary parenchymal or vascular disorders. This may bias our results, which include sensitivity analysis in the proposed models incorporating LAS r . Further validation of these findings is necessary in additional large-sample, multicentre studies. Our study cohort comprised patients with unexplained breathlessness referred to two tertiary-care centres and may not reflect a general HFpEF population. Nevertheless, the study presents a real-world scenario in PH specialist centres where distinction between HFpEF and pulmonary disorders is often challenging and stronger screening algorithms to rule-in diastolic dysfunction are needed. We employed fluid-filled catheters to obtain PCWP measurements rather than high-fidelity microcatheters and this may be considered a limitation. Inter-operator and inter-evaluator variability may be considered a limitation in our study, given that we did not employ a core-lab approach to echocardiographic image analysis. However, a standard international acquisition and analysis protocol was followed by two experienced echocardiographers with over 15 years' experience. Further, we evaluated the contribution of LAS r to diastolic assessment in the same cohort rather than employ a retrospective derivation and independent prospective validation cohort, which may have strengthened our study design. Absence of information on LA pump strain in this cohort may also be considered a limitation, given that high LA pump strain has demonstrated strong ability to identify patients with normal filling pressure when EF is preserved. 8 However, we chose to focus on LAS r which is more robust, and in keeping with the recent expert consensus algorithm. 9 Finally, diastolic stress tests were not performed in all our patients. Although stress testing may have improved the diagnostic accuracy of the ASE/EACVI algorithm, this was not the aim of this study.

Conclusion
In the setting of preserved EF, incorporation of LAS r into the 2016 ASE/EACVI recommendation-based assessment improves detection of elevated filling pressure.

Supplementary data
Supplementary data are available at European Heart Journal -Cardiovascular Imaging online. reports personal fees from Merck, grants and personal fees from Vifor-Fresenius, grants and personal fees from AstraZeneca, personal fees from Bayer, grants from Boston Scientific, personal fees from Pharmacosmos, personal fees from Abbott, personal fees from Medscape, personal fees from Myokardia, grants and personal fees from Boehringer Ingelheim, grants and personal fees from Novartis, personal fees from Sanofi, personal fees from Lexicon, personal fees from Radcliffe cardiology, outside the submitted work; P.L. reports fees from Pfizer.

Data availability
The data underlying this article cannot be shared publicly due to the privacy of individuals that participated in the study and GDPR regulations.