Repaired TOF patients with the following criteria were enrolled: (1) ≥moderate PR, (2) two sequential cardiovascular magnetic resonance (CMR) examinations with at least a time gap of four years in between and (3) echocardiography within one year from each CMR examination. Patients were selected from previous CMR studies (7–9). Patients with a RV outflow tract intervention between the CMRs were excluded. Patients underwent a comprehensive clinical assessment, NT-proBNP measurements, cardiopulmonary exercise test, CMR examination and echocardiography, as previously reported (7). Percentage of predicted work load and percent predicted peak rate of oxygen consumption (VO₂) were variables used from the results of cardiopulmonary exercise tests. Local institutional ethics committee approved the study, and all patients provided written informed consent.

CMR images were acquired with clinical 1.5 T MRI scanners (GE Healthcare, Milwaukee, Wisconsin, USA). Included patients had undergone a CMR examination with at least a stack short-axis balanced steady-state free precession (bSSFP) cine images and phase contrast (PC) imaging of the pulmonary artery. Using a retrospective ECG-gated bSSFP sequence during mild end-expiratory breath-hold, ten to fourteen contiguous short-axis images were acquired covering the left ventricle (LV) and RV from basis to apex. Typical scan parameters were field of view(FOV) 30–38 cm, phase FOV 75%–100%, slice thickness of 6–8 mm with an interslice gap of 2–4 mm, matrix size 192 × 160 and 24 reconstructed phases per cardiac cycle (7). Breath-hold, retrospective ECG-gated through-plane PC imaging was planned perpendicular to the main pulmonary artery approximately 1 cm distal of the valve and before the pulmonary artery bifurcation. VENC was set at 180 cm/s and increased if necessary (7). Directly after the CMR examination, the exact same PC sequence parameters were used to scan a static gel phantom for phase off set error correction.

In Qmass (Medis, Leiden, The Netherlands), LV and RV endo- and epicardial contours were drawn on short axis CMR images during end-systole and end-diastole by one observer (SCSM) and were reviewed by a second observer (AH). Based on these contours, LV and RV end-diastolic volumes, end-systolic volumes and EF were calculated. RV mass was determined in end-systole {[epicardial volume–endocardial volume (in ml)] × 1.05 g/ml}. Volumes and mass were corrected for body surface area. With QFlow (Medis, Leiden, The Netherlands), PR fraction and regurgitation volume were quantified using PC imaging. Flow measurements were obtained after background subtraction with a static gel phantom.

RV diastolic pressure was derived based on the inferior vena cava diameter (≤20 mm or >20 mm) and percentage of collapse (≤50% or >50) on echocardiography (10). The maximum peak gradient across the pulmonary valve was determined using Doppler measurements and calculated using the modified Bernoulli equation [ΔP = 4 × (V_max)²].

Diastolic wall stress and strain were calculated using finite element (FE) analysis performed with Abaqus/Standard (Dassault Systèmes Simulia Corp., Providence, RI, USA). The stack of short axis CMR contours was interpolated to create three-dimensional biventricular models (Figure 1). The LV and RV walls were assumed to be hyperelastic, incompressible and anisotropic (11) with material parameters taken from Wang et al. (12). A smooth transmural variation of the fibre orientation from −70° at the epicardium to +80° at the endocardium was assumed as described by Wong et al. and implemented in the computational models (13).

Boundary and loading conditions were prescribed to the FE models as follows. Biventricular computational models were restrained from movement in the basal plane, simulating a stiff valvular plane. No constrains were applied to the apex that could move freely. Diastolic filling of the RV and LV was simulated by prescribing pressures on the ventricular endocardial surfaces. RV diastolic pressure estimate was obtained from echocardiography and assigned to the RV, while a physiological pressure of 9 mmHg was assigned to the LV.

These three-dimensional, mechanical FE models were used to simulate the passive filling during diastole. In total, twenty FE models were generated for ten patients at two timepoints. The end-diastolic wall stress (maximum principal wall stress) and distributions of stress in the RV free wall were computed. Maximum principal stress is the maximum values of wall stress in any direction. Mesh independency of the FE results was confirmed by mesh convergence analysis.

For the analysis, the septal part of the RV wall was excluded, as the ventricular septum is considered to belong morphologically to the LV (Figure 1). Because of regional variation of the magnitude of wall stress, a regional analysis approach was chosen. A standardized segmentation method for analysis of the LV was adapted and applied to the free wall of RV (Figure 1) (14), using a previously proposed method to create fifteen segments of approximately equal size. First, the RV free wall was divided into apical, mid and basal regions, which were further subdivided into four, five and six segments respectively (15) (Figure 1). These fifteen segments were also used to make a second division of the RV free wall into an anterior, lateral and posterior part (Figure 1). The mean and 95th-percentile peak wall stresses were calculated per region. The weighted mean, defined as the mean of the fifteen regional values, and peak RV wall stresses per biventricular model were also calculated.

Continuous values were expressed as mean with standard deviation (SD) or, in case of skewed distribution, as median with a range or interquartile (IQR) range. Categorical data were summarized with frequencies and percentages. Regional wall stresses were log transformed because of a skewed distribution to calculate the geometric mean and geometric SD factor and are presented on the linear scale as geometric mean with range (geometric mean/geometric SD factor-geometric mean × geometric SD factor). Patients were divided in a group with a stable RVEF (RVEF follow-up minus RVEF baseline ≥0%) and a group with a decreasing RVEF (delta RVEF <0%) over time. RVEF was chosen because of its strong association with adverse clinical outcomes and exercise capacity in patients with repaired TOF (2, 16, 17). Continuous standard imaging parameters were compared with a Mann–Whitney test between groups and with a Wilcoxon one-sample test between baseline and follow-up.

Linear mixed models with a random intercept for each patient were used to investigate (1) local differences in wall stress (average and 95th percentile wall stress) between RV free wall regions (basal vs. mid vs. apical) and parts (anterior vs. lateral vs. posterior) (2) local change in wall stress between baseline and follow-up (3) local wall stress differences between patients with a stable RVEF and with a decreasing RVEF (4) associations of wall stress with clinical end-points [RVEF, NT-proBNP, exercise testing (peak VO₂ and percentage of predicted work)] and (5) associations of baseline PR and PR volume with change of wall stress over time. To account for spatial autocorrelation between the wall stress regions, a spatial Gaussian correlation structure was used in the models 1–3 and 5, based on the x-, y- and z-coordinates of regions. Standardized x-, y- and z-coordinates of regions were used for comparison between baseline and follow-up (model 2) and between patients (stable vs. decreasing RVEF, model 3) and. All models, except model 1, were corrected for time since initial surgery (follow-up duration), by adding this variable as a fixed effect.

All dependent parameters were transformed using log2 when the residuals plots showed deviation from normality. For visualization purposes, correlation plots were created showing the weighted mean wall stress values. For correlations between variables that include both baseline and follow-up measurements (repeated measurements), we controlled for within-individual association using a repeated measurement correlation (rmcorr package in R) (18). All analyses have been performed in R Statistical Software version 4.1.0 (R Foundation for Statistical Computing, Vienna, Austria). Two-tailed p-values below 0.05 were considered statistically significant.

Ten patients were included in the study. Median age was 24 years (IQR 17–28) at baseline and five (50%) were males (Table 1). Median follow-up period between the baseline and follow-up CMR study was 7.2 years (IQR 6.9–7.5). At baseline, median RV end-diastolic volume was 140 ml/m² (IQR 129–144), RVEF 46% (IQR 42–49), RV mass 15 g/m² (IQR 12–18), RV mass-to-volume ratio 0.11 g/ml (IQR 0.09–0.13) and PR 41% (IQR 36–45). All patients had ≤mild tricuspid regurgitation during the both visits. Diastolic echocardiography-based RV pressures were estimated at 3 mmHg (IQR 3–3) in all patients at both timepoints. At follow-up, RV end-diastolic volume was 150 ml/m² (IQR 144–151) and RVEF was 46% (IQR 44–48) (Table 2). Overall, none of the parameters changed significantly between baseline and follow-up (Table 2). In four patients, RVEF decreased between baseline and follow-up. In these patients, RV mass decreased more and a larger PR volume per beat was present at baseline (Table 2).

Figure 2 provides an overview of the regional wall stress distributions at baseline and follow-up per patient. Clear regional differences in wall stress patterns are observed within the RV, but also between baseline and follow-up and between patients. Weighted mean wall stress was 6.89 kPa (IQR 5.29–8.97) at baseline and 6.00 kPa (IQR 5.56–6.89) at follow-up and did not change significantly between baseline and follow-up (p = 0.460). Correlations between average wall stress, RV volumes and mass measured both at baseline and at follow-up are presented in Figure 3. Higher average wall stress was associated with lower RV mass (r_rm= −0.70, p = 0.017, Figure 3A) and lower mass-to-volume ratio (r_rm= −0.80, p = 0.003, Figure 3B). No associations were found between RV end-diastolic volume and mass and between RV end-diastolic volume and average wall stress (Figures 3C,D).

Overall, at baseline, regional analysis showed that wall stress in the mid region (7.24 ± 1.64 kPa, mean, SD) was higher compared to wall stress in the basal region {basal 5.40 ± 1.70 kPa, 0.37 difference [95% confidence interval (CI) 0.11, 0.63], p = 0.006} in log2 wall stress between mid and basal region), but not different from wall stress in the apical region [5.99 ± 1.70 kPa, 0.20 difference (95% CI −0.14, 0.53), p = 0.248] in log2 wall stress between mid and apical region) (Table 3). Furthermore, wall stress in the lateral part was significantly higher than wall stress in the posterior part [0.42 difference (95% CI 0.26, 0.58), p < 0.001] in log2 wall stress between lateral and posterior part), which was in turn significantly higher than wall stress in the anterior part [0.96 difference (95% CI 0.80, 1.12), p < 0.001] in log2 wall stress between lateral and posterior part) (Supplementary Table S1). The peak wall stress was also the highest in the mid region and lateral part (data not shown). Both at baseline and follow-up, no differences were observed in regional wall stress, RV end-diastolic volume, RV mass and RV mass-to-volume ratio between patients with a stable and decreasing RVEF (Tables 2, 3).

Regional free wall stress of the RV decreased between baseline and follow-up [−0.54 change (95% CI −0.76, −0.31), p < 0.001, in log2 wall stress between baseline and follow-up]. While in patients with a decreasing RVEF, wall stress did not change significantly over time [β −0.20 (95% CI −0.81, 0.42), p = 0.530], in patients with a stable RVEF the wall stress decreased significantly [β −0.53 (95% CI −0.68, −0.39), p < 0.001]. The regional free wall stress decrease over time in patients with a stable RVEF was larger compared to that of patients with a decreasing RVEF [β −2.79 (95% CI −5.19, −0.38), p = 0.029] when corrected for follow-up period. The difference of the change of wall stress between the two groups was most prominent in the mid region [β −4.87 (95% CI −7.65, −2.08), p = 0.004] and anterior part [β −2.04 (95% CI −3.50, −0.58), p = 0.029]. Peak wall stress did not change over time, both in patients with a stable and with a decreasing RVEF.

Patients with a stable RVEF demonstrated a significant increase of RV mass and RV mass-to-volume ratio (both p = 0.031) between baseline and follow-up. These patients showed a larger RV mass increase during the follow-up period compared to patients with a decreasing RVEF (p = 0.010). Addionally, there was a tendency towards a larger RV mass-to-volume ratio decrease in patients with a decreasing RVEF compared to patients with a stable RVEF (p = 0.067) (Table 2). There were no significantly different changes in RV end-diastolic volume over time between patients with a decreasing RVEF and those with a stable RVEF.

In Figure 4A, the relationship between weighted mean wall stress of the free wall and RVEF is depicted. Higher wall stress was associated with a lower RVEF [β −1.16% RVEF per kPa increase in wall stress (95% CI −2.12, −0.20), p = 0.022]. When also corrected for LVEF, RV end-diastolic volume and time since surgery, average wall stress remained independently associated with RVEF [β −1.27% RVEF per kPa increase in wall stress (95% CI −2.36, −0.18), p = 0.029] (Supplementary Table S2). Mass-to-volume ratio was not associated with RVEF (Supplementary Table S2). There was also a tendency towards a negative association between weighted peak wall stresses and RVEF [β −0.41% RVEF per kPa increase in wall stress (95% CI −0.83, 0.01), p = 0.055] (Figure 4B).

Wall stress and peak wall stress at baseline were both not associated with the RVEF at follow-up (wall stress β −0.15 (95% CI −1.49, 1.20), p = 0.810, peak wall stress β 0.06 (95% CI −0.46, 0.58), p = 0.805), corrected for follow-up duration. This was also the case when corrected for follow-up duration and RVEF at baseline.

At follow-up, the median NT-proBNP level was 22 pmol/L (IQR 16–43) and the median percentage of predicted work load was 92% (IQR 88–110). Wall stress at baseline was not associated with NT-proBNP and percentage of predicted work load at follow-up when corrected for follow-up period. Peak VO₂ was available at follow-up for only three patients. Therefore, analyses of this parameter as end-point were not performed.

Finally, the relationship between PR and weighted mean wall stress was investigated (Figure 5). Patients with a higher PR fraction and a larger pulmonary regurgitant volume at baseline exhibited a larger increase of wall stress between baseline and follow-up when corrected for follow-up duration (increase of wall stress of 0.14 kPa per percentage increase in PR, p = 0.024, and 0.07 kPa increase of wall stress per ml PR volume increase, p = 0.040). In addition, patients with a decreasing RVEF had a larger PR volume than patients with a stable RVEF both at baseline and follow-up (Table 2). There was no significant change of PR and/or PR volume over time in patients with a stable or decreasing RVEF.

In accordance with the overall analysis, also baseline PR fraction was associated with regional changes in wall stress in the mid region β 0.21 (95% CI 0.04, 0.38), p = 0.025), anterior part β 0.11 (95% CI 0.03, 0.19), p = 0.015) and lateral part β 0.24 (95% CI 0.10, 0.39), p = 0.014) (Supplementary Figure S1). Also, PR volume at baseline was associated with regional wall stress changes in these regions (mid region β 0.11 (95% CI 0.04, 0.19), p = 0.010, anterior part β 0.05 (95% CI 0.01, 0.10), p = 0.033 and lateral part β 0.12 (95% CI 0.04, 0.19), p = 0.021).