One hundred patients with severe AS (aortic valve area <1 cm² or indexed AVA <0.6 cm²/m²) undergoing TAVI between 2008 and 2019 were retrospectively identified from the prospective AS registry of the University Heart Center Zurich. Patients were included when a comprehensive echocardiographic examination was available within three months prior to TAVI allowing complete strain analysis of both ventricles. A flow diagram (Supplementary Figure S1) illustrates how patients were enrolled for the study. Ethical committee approval and informed consent were obtained prior to patient inclusion.

Transthoracic echocardiographic (TTE) examinations were performed using commercially available equipment (iE33 or Epiq 7, Philips Medical Systems, The Netherlands; E9 or E95, GE Healthcare, USA). Echocardiographic measurements were made by certified specialists according to current recommendations (26–28). Calculation of LVEF was based on Simpson's biplane method.

TomTec ImageArena Cardiac Performance Analysis (Version 4.6) was used for offline STE measurements of both ventricles according to current recommendations (26, 27). Endocardial tracing was performed manually excluding sigmoid septal hypertrophy, papillary muscles, and trabeculations. Heart cycle timing was identified from the M-mode.

Apical 2- (A2C), 3- (A3C) and 4-chamber (A4C) views were used for measuring LVGLS. End-diastole was set at the last frame before mitral valve closure and end-systole at the last frame before aortic valve closure. LVGLS is indicated as average peak systolic strain based on the 16-segment model. Focused RV views were used for measuring RV strain (29). End-diastole was defined as the last frame before tricuspid valve closure and end-systole as the smallest ventricular systolic dimension, respectively. RVGLS is indicated as average peak systolic strain based on the 6-segment model and RV free wall strain (RVFWS) on the 3 segments of the free wall, respectively.

Reproducibility was tested on 15 echocardiographic examinations by two observers to investigate inter-observer agreement and repeated by the main observer with a difference of 3 months between the first and the second measurement to determine intra-observer agreement. Concordance correlation coefficient was used for assessing reproducibility (inter-observer agreement) and repeatability (intra-observer agreement). These results are reported in Supplementary Table S1. There is strong inter- and intra-observer agreement for LV and RV strain measurements.

The date of the echocardiography examination before TAVI (i.e., within 3 months prior to procedure) marks the date of study inclusion. The date of the TAVI procedure indicates the start of follow-up. All-cause mortality was defined as the primary endpoint. Patient survival status was evaluated through patient records and/or phone calls.

The Shapiro–Wilk test was used to assess normal distribution, with most variables displaying a non-normal distribution. Continuous variables are given as median [interquartile range, IQR] and categorical variables as absolute number (percentage). Continuous variables were compared with the Mann–Whitney–Wilcoxon test, categorical variables with Fisher's exact test. Receiver operating characteristic (ROC) curve analysis was used to determine the optimal cutpoint value for distinguishing survivors from non-survivors, and model discrimination was summarised by area under the curve (AUC). The AUC from models was compared with the DeLong method using pROC (version 1.18.0) package. Kaplan–Meier survival curves and log-rank tests of time-to-event data were analysed with the survminer (version 0.4.9) package; variables were either dichotomised at their optimal cutpoint according to the ROC curve or according to the literature. For some analyses, variables were divided into tertiles or quartiles. Association with all-cause mortality was analysed in uni- and multivariable Cox regression models. Proportional hazard assumptions were assessed for all models using the scaled Schoenfeld residuals. Variables with clinical relevance such as age, sex, EuroSCORE II, STS score, AS severity, LVEF, RVFAC were included in multivariable models regardless of their significance level in univariable models and tested for sensitivity. A further sensitivity analysis was performed excluding periprocedural deaths from the total number of events. Cox regression analysis of variance (Cox-ANOVA) was used to test model fit. The chi-squared (χ²) log-likelihood ratio and Harrell's C-statistic were used to examine the incremental value of predictors in the multivariable model compared to the nested univariable model. Collinearity between variables in the regression models was tested using Spearman's correlation and variance inflation factor tests. Standard mean difference (SMD) was used for determining the representativeness of the study cohort compared to the overall registry population. SMD values of 0.2, 0.5, and 0.8 are considered to represent small, medium, and large differences, respectively (30, 31). Statistical analyses were performed using MedCalc® version 19.6.4 and R version 4.1.3. Statistical significance was considered at a two-sided P value <0.05.

Tables 1, 2 summarise the clinical and echocardiographic baseline characteristics, respectively. Almost all patients exhibited normal size and systolic function of both ventricles as determined by RV enddiastolic area index (RVEDAI), RVFAC, TAPSE, left ventricular enddiastolic volume index (LVEDVI), and LVEF. There was little difference between the baseline characteristics of the study cohort and those of the overall registry population (N = 1,467; SMD 0.0–0.4; Supplementary Table S2). The study cohort was slightly younger and exhibited a lower EuroSCORE II, while mean transaortic pressure gradient (MTPG) was slightly higher, AVA was similar, and RVFAC, TAPSE, LVEF were marginally higher (Supplementary Table S2).

RVFAC [40.5% (37.8–44.0)] and LVEF [58.5% (53.0–64.3)] were preserved among the study population. In contrast, longitudinal deformation of both ventricles was impaired (RVGLS [−16.5% (−19.8 to −13.8)]; RVFWS (−17.6% [−21.6 to −14.3]; LVGLS [−10.5% (−13.0 to −8.4); Figure 1].

The concordance correlation coefficient for assessing intra- and inter-observer variability was each 0.8 for LVGLS, each 0.9 for RVGLS, and 0.9 and 0.8 for RVFWS, respectively (Supplementary Table S1).

During a median follow-up time of 1,367 [959–2,123] days, 33 patients (33%) died, of which 23 (23%) due to a cardiovascular cause. There was no significant difference in survival between women and men (Table 1). Among the STE-derived parameters, RVGLS and RVFWS were significantly lower in non-survivors (−13.9% [−16.4 to −12.9]; P = 0.001 and −15.7% [−18.3 to −12.8]; P = 0.002, respectively) compared to survivors (−17.1% [−20.2 to −15.2]; P = 0.001 and −18.7% [−22.6 to −15.9]; P = 0.002, respectively), while LVGLS did not differ (P = 0.303; Table 2). The cutpoint values for RVGLS (≥−14.6%; sensitivity 61%; specificity 79%; ROC AUC 70%; P < 0.001) and RVFWS (≥−18.3%; sensitivity 79%; specificity 54%; ROC AUC 69%; P = 0.001) differentiated survivors from non-survivors, while that for LVGLS did not (P = 0.243). Kaplan–Meier analyses indicated a higher survival probability when the population was dichotomised according to the cutpoint value for RVGLS (P < 0.001) and for RVFWS (P = 0.012; Figure 2), but not for LVGLS (P = 0.580).

Univariable Cox regression analysis associated a lower RVGLS [HR 1.13 (95% CI 1.04–1.23); P = 0.003; ANOVA χ² 9.27; χ² P = 0.002] and RVFWS (data not shown) with an increased mortality risk, while this was not the case for LVGLS [HR 1.06 (0.95–1.17); P = 0.300; ANOVA χ² 1.09; χ² P = 0.297] or the conventional echocardiographic parameters (Table 3). Bivariable Cox regression analysis demonstrated that the association of RVGLS and RVFWS with mortality remained independent of LVGLS or other clinically relevant covariables such as RVFAC, LVEF, and EuroSCORE II (Table 3; RVFWS data not shown). Likelihood ratios showed that inclusion of RVGLS or RVFWS to these models significantly improved their fitness with potential incremental value for association with mortality, while LVGLS did not (Table 3; Figure 3; RVFWS data not shown). The greatest C-index was observed for the univariable RVGLS model and the bivariable model containing RVGLS and LVGLS [both 0.69 (0.58–0.80)].

A multivariable Cox regression analysis was performed to test for a possible interaction between LVGLS and RVGLS or RVFWS, respectively, with inclusion of EuroSCORE II to account for possible confounders (Table 4). RVGLS [1.15 (1.04–1.26); P = 0.005] and RVFWS [1.14 (1.05–1.25); P = 0.003] were associated with mortality independent of LVGLS and EuroSCORE II. There was a significant interaction between RVGLS and LVGLS regarding the association with mortality [0.97 (0.95–1.00); P = 0.043]. In contrast, RVFWS and LVGLS did not interact significantly in terms of association with mortality [0.98 (0.96–1.01); P = 0.210]. The highest C-index [0.69 (0.58–0.80)] of the multivariable models was observed for the model containing RVFWS, LVGLS, their interaction, and the EuroSCORE II. However, this is no higher than the greatest univariable C-index and there were only small differences in C-index between the multivariable models. Finally, a sensitivity analysis was performed by replacing EuroSCORE II with either STS score, age and sex, or parameters defining AS severity (i.e., MTPG, AVA), revealing most compelling results for models including EuroSCORE II (data not shown).

Assessment of RV systolic function is an integral part of every echocardiographic examination. RV strain is obtained from the RV focused apical view with little additional effort. However, care must be taken in the real-wold setting to ensure that all methodological requirements are met in order to perform a reliable deformation analysis. Impaired RV longitudinal strain indicates increased mortality in patients with severe AS after valve replacement and emerges as an important parameter to improve the prognostic understanding of these patients. Hence, RV longitudinal strain should be measured in patients with severe AS and its reduced values should promote the decision to replace the aortic valve for avoiding persistent cardiac remodelling and reduced outcome after intervention.

This study is limited by its retrospective single-center design; however, the cohort is representative of a typical TAVI population and the number of patients included, although limited, is reasonable to address the research question. A selection bias cannot be excluded because inclusion criteria involved the quality of echocardiographic examinations as well as several additional parameters. Although multivariable models were generated, it cannot be excluded completely that unobserved confounding factors affected the results. The predictive performance, as measured by AUC, was greater for the parameters of interest than for EuroSCORE II, which contains several predictors important to understanding individual prognosis. Due to the limited amount of data, the EuroSCORE II was used as a single predictor and its individual components were not re-estimated with the data at hand, tipping the comparison in favor of the new parameters. More data on this population is needed to develop true clinical prediction models with limited risk of overfitting and performance overoptimism.