This single-center retrospective study included 55 patients who underwent CMR preoperatively before LVAD implantation between 2018 and 2024. CMR was performed 55 days (15–403 days) before implantation. Patients had to be older than 18 years old and not suffer from congenital cardiac abnormalities. Patients with insufficient image quality, for example, artefacts affecting strain analysis, were excluded. Baseline clinical characteristics of the included and excluded patients are presented in Supplementary Table S1. Ethical approval was obtained from the local ethics committee, and written consent was waived.

According to the 2020 Mechanical Circulatory Support-Academic Research Consortium (MCS-ARC) guidelines, patients requiring a tRVAD implantation during the operation after LVAD implantation are defined as having early, acute right heart failure (eaRHF) (6). A CentriMag Pump (Levitronix LLC, Waltham, MA, USA) was used to assist the RV temporarily.

Clinical data were collected from the patients using the electronic medical record (EMR). All patients underwent a trans-thoracic echocardiogram (TTE) before surgery. 35 patients received an RHC preoperatively, and 5 patients did not due to hemodynamic instability.

CMR imaging was acquired using routine clinical scanners with a magnetic field strength of 1.5 or 3 Tesla. A dedicated cardiac phased-array coil was used for signal reception. All examinations were performed following the CMR imaging guidelines of the Society for Cardiovascular Resonance Imaging (7). All routine examinations include steady-state free-precession (SSFP) 4-chamber (3 slices) and short-axis (SAX) views covering the entire area of the right ventricle (RV) (12–16 slices). Each CMR examination had 25 to 40 phases per cardiac cycle. Participants were evaluated while lying in the supine position. A vector electrocardiogram was used to carry out acquisitions triggered by cardiac activity.

All statistical analyses were performed using R software (R Foundation for Statistical Computing, Vienna, Austria, version 4.4.2). Due to the small sample size, data are expressed as medians with 25th and 75th percentiles. Differences between the two groups were assessed using the Mann–Whitney U test, with statistical significance set at a p-value <0.05. Categorical variables were compared between groups using Fisher's exact test. Receiver Operating Characteristic (ROC) curve analysis was performed to identify markers predictive of eaRHF. The area under the curve (AUC), 95% confidence intervals (using DeLong's method, with additional bootstrap resampling of 2,000 replicates), optimal cut-off values (Youden index), as well as sensitivity and specificity, were calculated using the pROC package in R. Data handling was performed using the readxl and openxlsx packages. Intraobserver reproducibility of CMR-derived strain parameters (GRS, GCS, and FWS) was assessed using intraclass correlation coefficients (ICC) and Bland–Altman analyses.

Of the 55 patients who were included, 40 (49.2 ± 15.2 years, 83% male) had sufficient image quality for offline strain analysis. 10 patients in the study group developed eaRHF. These patients had a higher preoperative bilirubin level [1.1 (0.6–1.4) mg/dl vs. 1.6 (1.3–1.9) mg/dl, p = 0.010]. Otherwise, there was no statistically significant difference between the groups. Baseline clinical characteristics are shown in Table 1.

The echocardiographic examination was performed 7.5 (3–12) days before LVAD implantation. TAPSE tended to be lower in patients with eaRHF [12 (11.8–14.3) mm vs. 18 (12–19) mm, p = 0.080]. Otherwise, there were no differences between the groups (Table 2). RHC was performed 6 (2–11) days before implantation. Patients with eaRHF had a lower RV-SWI [13.9 (11.2–15.1) g·m/m² vs. 22.9 (20–28.8) g·m/m², p < 0.001]. Furthermore, these patients had lower CO [2.3 (2.2–3) L/min vs. 4.1 (3–4.9) L/min, p = 0.003] and CI [1.2 (1–1.5) L/min/m² vs. 1.9 (1.8–2.5) L/min/m², p = 0.004] as well as a tendency toward lower PAPi. The results of the RHC are shown in Table 3.

RA and RV morphology and function were assessed using CMR. When both groups were compared, there were no differences in RA strain [RHF: 7.9 (3.6–22.7) vs. no RHF: 6.7 (5.3–15.2), p = 1.000]. No differences were seen in the other RA parameters measured with CMR (Table 4). When looking at RV function, patients with eaRHF tended to have a higher ESV [165.1(93.5–231.8) ml vs. 226.5 (161.8–260.2) ml, p = 0.061] and ESVi [85.9 (51.4–117.7) ml/m² vs. 123.6 (87.7–134.8) ml/m², p = 0.080], but there was no statistically significant difference. The EDV [216.7 (143.4–255.1) ml vs. 288.4 (192.3–307.7) ml, p = 0.046] and EDVi [104.7 (76.9–122.6) ml/m² vs. 135.4 (114–163.9) ml/m², p = 0.003], however, were statistically higher in patients with eaRHF compared to patients without eaRHF. Furthermore, RV-GRS tended to be lower in patients with eaRHF [9.2 (5.4–14.8) % vs. 4.4 (3.7–7.4) %, p = 0.065]. Table 5 shows the results of RV parameters measured with CMR.

To assess intraobserver variability of RV GRS, GCS, and FWS, ICC and Bland–Altman analyses were performed. The reproducibility was excellent for all parameters with ICC (2,1) values of 0.98 (95% CI: 0.94–0.99) for GRS, 0.94 (95% CI: 0.86–0.98) for GCS, and 0.92 (95% CI: 0.81–0.97) for FWS. Bland–Altman analyses revealed small mean differences between the repeated measurements (GRS −0.18 ± 1.95%; GCS +0.27 ± 1.61%; FWS −0.38 ± 2.69%), with no systematic bias observed (all paired comparisons: p > 0.45).

ROC curve analysis was conducted to evaluate the predictive ability of CMR, echocardiographic and RHC parameters (Table 6). Among CMR parameters, RV EDVi showed moderate predictive capability (AUC = 0.73, p = 0.040), achieving a sensitivity of 70% and a specificity of 80% at a threshold of 129.7 ml/m². RV GRS also demonstrated moderate predictive performance (AUC = 0.70, sensitivity = 70%, specificity = 80%, threshold = 5.25%, p = 0.044). Echocardiographic-derived TAPSE exhibited moderate predictive value (AUC = 0.71, sensitivity = 75%, specificity = 67%, threshold = 14.0 mm, p = 0.022). Parameters derived from RHC had the strongest predictive performance, with CO yielding an excellent prediction (AUC = 0.90, sensitivity = 100%, specificity = 72%, threshold = 3.31 L/min, p < 0.001). CI also demonstrated high predictive accuracy (AUC = 0.88, sensitivity = 83%, specificity = 83%, threshold = 1.55 L/min/m², p < 0.001). Additionally, RV SWI demonstrated substantial predictive strength (AUC = 0.86, sensitivity = 83%, specificity = 86%, threshold = 3.43 g·m/m², p < 0.001). Other CMR-derived parameters exhibited trends toward predictive capability but did not reach statistical significance. ROC curves representing the most predictive parameters are presented in Figure 2.

This study's limitations warrant consideration. Primarily, the retrospective design and small sample size limit generalizability and introduce potential selection bias. Pre-LVAD-implant CMR examinations in advanced heart failure patients are rare, as many patients have devices that preclude MRI. Additionally, there was significant variability in the time interval between CMR and LVAD implantation, with some patients undergoing CMR more than a year prior to surgery. Image quality issues also led to the exclusion of several patients, further introducing potential bias. Only TAPSE was included from echocardiographic measurements. Other routinely available echocardiographic parameters of RV function (e.g., RV S′, RV FAC) were not measured in all patients and could not be incorporated into the analysis. Importantly, the operational definition of early acute RHF as intraoperative temporary RVAD implantation may only identify the most severe cases. As a result, this may underestimate the overall incidence of RHF and restrict generalizability to broader RHF phenotypes. Finally, while understandable given the limited sample size, the absence of multivariable modelling restricts assessment of whether CMR parameters independently predict RHF.