Association between left ventricular mechanics and diffuse myocardial fibrosis in patients with repaired Tetralogy of Fallot: a cross-sectional study

The study protocol was approved by the institutional review board at The Children’s Hospital of Philadelphia, and informed consent for research was obtained from all patients. Between December 2012 and July 2015, a modified Look-Locker (MOLLI) T1 mapping sequence [15, 16] and spiral cine Displacement Encoding with Stimulated Echoes (DENSE) sequence [17‐19] were both acquired in 43 patients with repaired TOF undergoing a clinical CMR evaluation.

All images were acquired using a 1.5-T Siemens Avanto (Siemens Healthineers, Erlangen, Germany) CMR system. For each subject, the imaging protocol included cine balanced steady-state free precession (bSSFP) images, 2D phase contrast velocity, phase-sensitive inversion recovery, spiral cine DENSE, and MOLLI acquired both before and 15–20 min after administration of gadolinium contrast (gadopentetate dimeglumine, 0.2 mmole/kg, Bayer Health Care, Whippany New Jersey, USA). The MOLLI and DENSE acquisitions were co-localized at 3 ventricular short-axis locations (basal, mid-ventricular, and apical), such that cardiac mechanics and diffuse myocardial fibrosis could be interrogated over the same regions and directly compared. Details for the acquisition and post-processing for each specific sequence are as follows.

Segmentation of the MOLLI T1 images was repeated 2 months following the initial analysis by the primary analyst, and independently by a second analyst to quantify intra- and inter-observer reproducibility, respectively. The mean coefficient of variation (CoV) and the 95% limits of agreement [26] were quantified. The mean CoV was computed as:

Analysis was performed in R (version 3.4.0; R Foundation for Statistical Computing, Vienna, Austria) [27]. Data are presented as mean ± standard deviation. Z-scores for ventricular volumes, mass, and EF were computed based on population data assumed from Alfakih et al. [28] Pearson correlation was used to detect simple associations between strain/T1 derived measures and ventricular volumes, LV mass, and LV/RV EF. Linear mixed models [29] were fit to capture the mean association of regional (base, mid-ventricle, or apex) or segmental LV mechanics with T1-derived measures, while accounting for effects of sex, age, measurement region or segment, and within-subjects repeated measures (e.g., multiple regions or segments per patient). Repeated measures for the primary regional analysis were modeled using an unstructured correlation matrix. For the segmental analysis, a compound symmetric correlation matrix was assumed to ensure model convergence. Continuous outcome and independent variables were standardized for model fitting to report standardized β coefficients. Outcome variables were log transformed, as appropriate, to adjust for heteroscedasticity or non-normality. The statistical significance level was set at 0.05, with a Benjamini-Hochberg adjustment for multiple comparisons of each outcome variable [30]. Preliminary power analysis [31], assuming a linear correlation coefficient of 0.30, demonstrated that 84 samples were required to achieve 80% power.

From the original 43 patients, two were excluded for either poor DENSE image quality (n = 1) or errors with the reconstruction of T1 images (n = 1). An additional patient was excluded because of DENSE and MOLLI image slice mis-registration, leaving a total of 40 subjects in the final study group. An additional eight DENSE slices (4 base, 4 apex) from seven different subjects were also excluded for poor quality. Because of the need to compute a ventricular reference curve for the dyssynchrony analysis, the dyssynchrony index was not computed for these seven subjects. However, the remaining data for these subjects were included. Demographic details as well as ventricular volume indices, EF, and LV mass indices for these subjects are provided in Table 1. The RVs on average were severely dilated (Z = 4.6) with reduced EF (Z = −3.1), while the LVs had mostly normal chamber volume (Z = −0.6) and myocardial mass (Z = −0.7), but impaired EF (Z = −2.0). These structural and global functional characteristics are generally consistent with contemporary findings from the INDICATOR study [3].

Most patients in this study (62.5%) were negative for LGE (Table 1). Of those with LGE reported, the predominant finding (in 73%) was a small region of enhancement at the inferior RV hinge point.

Representative T1 and DENSE images are shown in Fig. 1, and summary measures are presented in Table 2. Figure 2 represents the measured regional values of ECV, DI, and circumferential and radial strains, as well as global (“mean”) values. LV ECV (Fig. 2a) was generally within normal ranges, although mean values for seven patients (18%) were greater than 0.28, which has previously been reported as the upper bound of the normal range [10]. LV ECV was moderately associated with age (r = 0.35, p = 0.03), but not with LV or RV EF, LV mass index, LV or RV end-diastolic volume indices, or pulmonary regurgitant fraction (data not shown). Additionally, mean ECV was not different between patients based on the presence or absence of LGE (data not shown).

Peak LV circumferential strain (Fig. 2b) demonstrated a gradient of increasing magnitude from base to apex, which is consistent with normal LV function [32]. However, mean values were at the low (magnitude) end of the normal range, with 12 patients (31%) having strain magnitudes below the normal range (i.e., > −14.4%). Only two of these 12 patients also had an elevated mean LV ECV. This moderate impairment in LV circumferential strain mirrored the impairment in LV EF, as the two measures were significantly correlated (r = −0.54, p < 0.001), but mean LV circumferential strain was not associated with LV mass, LV/RV volumes, RV EF, or pulmonary regurgitation, and did not differ between patients with or without LGE. LV dyssynchrony was moderately elevated from the normal mean (DI = 16.1 ± 8 ms; Fig. 2c), and seven patients (21% with measured DI) had a value above the normal range (>21.1 ms) [4]. Only three of these seven patients also had reduced mean LV circumferential strain. Finally, mean LV radial strains were generally within the normal range (Fig. 2d), and did not differ between patients with or without LGE (data not shown).

At the segmental level, Fig. 3 presents the values for LV ECV, LV circumferential strain, and LV radial strain for the standard 16-segment LV model. LV ECV values tended to be higher in anterior and septal segments than the posterior and free wall, although the overall spatial variation was low (standard deviation of the segment means, SD = 0.02). LV circumferential strains in the mid-ventricular and apical free wall segments were preserved; however, strain in other segments were considerably lower (SD = 2.0%). LV radial strain also demonstrated segmental heterogeneity (SD = 4.4%), with higher strains in the basal and mid-ventricular free wall compared with the apex and mid-ventricular septum.

From mixed model analyses, T1-derived measures had the strongest association with regional LV DI (Table 3), as λ_Gd, ECV, and post-contrast T1 time were all significantly associated with increased regional LV DI after adjusting for age, sex, and regional differences. In each case, the β coefficients of these associations ranged between 0.33 (absolute) and 0.67, with ECV having the largest individual coefficient (Fig. 4a). Moreover, four of the seven patients with elevated mean ECV also had an abnormal mean LV DI. All T1-derived measures were also significantly associated with reduced regional LV radial strain, with ECV again having the largest magnitude coefficient (β = −0.36; Fig. 4b). The other statistically significant models were post-contrast T1 being related with regional LV systolic (β = −0.30) and (log-transformed) diastolic circumferential strain rates (β = 0.37; Fig. 4c). There was also a trend with post-contrast T1 and regional LV circumferential strain (p < 0.05; Table 3), but this did not meet statistical significance thresholds after adjustment for multiple hypothesis testing. No associations were observed at the segmental level (Table 4).

Consistent with previous studies [10], T1-derived measures were highly reproducible at a regional level on both intra- and inter-observer bases (Fig. 5). In all cases, the mean CoV was less than or equal to 3%, denoting tight limits of agreement relative to measurement means. The reproducibility of segmental-level values was similarly strong, with mean CoVs ≤5% (not shown).

From our findings, the overall LV myocardial burden of diffuse fibrosis in patients with repaired TOF was generally low, based on data from healthy controls in other studies [34, 35]. This finding is consistent with the results of Chen et al., who similarly reported a mean LV ECV of 0.24 ± 0.03 in a cohort of 84 TOF patients of similar mean age [10]. However, seven patients (18%) had an elevated ECV (>0.28), which was previously associated with increased arrhythmic risk [10]. Continued follow-up of our patients is needed to determine if a similar risk association is observed. Our findings do conflict with those of Broberg et al., who reported a slightly higher LV ECV (0.29 ± 0.03) in a smaller (n = 17) patient group [9]. The patients in that study were older (34 ± 12 years), which may explain this difference given the positive association we observed between ECV and age.

Findings of reduced systolic LV function, including impaired myocardial strain, are common in patients with repaired TOF [4, 5, 36‐39]; the fact that mean LV circumferential strain was moderately impaired in this study was not surprising. However, none of the T1-derived measures were associated with circumferential strain at either the regional or segmental levels, which was unexpected given that such associations were reported by Kuruvilla et al. [12] and Donekal et al. [14] in acquired heart diseases. This difference suggests a distinct etiology of impaired circumferential strain in the setting of repaired TOF, which is not fibrosis-related. Instead, these changes likely result from other mediating factors, such as surgical techniques [2] or septal shifting secondary to abnormal ventricular-ventricular interactions [36, 38]. Alternatively, the pattern of fibrosis may be such that myofibril contraction is affected to a much greater extent in the radial direction than in the circumferential direction.

The functional factor that demonstrated the strongest relationship with T1-derived measures of fibrosis was the LV dyssynchrony index. Findings of dyssynchrony in TOF are also common [4, 36, 40], and have been thought to be evidence of inter-ventricular interaction [4]. Our results suggest that diffuse fibrosis may be a primary mediator of the development of LV dyssynchrony in these patients. This linkage may help to explain the association between elevated LV ECV and increased risk of arrhythmia [10] through mechanisms such as electromechanical coupling [36, 41], but further study is warranted to definitively establish such connections. More generally, the prognostic significance of ventricular dyssynchrony in TOF has not been fully evaluated. A recent study found that dyssynchrony derived from CMR feature tracking did not predict longitudinal changes in LV or RV EF [5], although endpoints such as arrhythmias and mortality were not considered.

The remaining regional mechanical factors found to depend on T1-derived measures of fibrosis were LV radial strains and LV circumferential strain rates. Interestingly, the mean values for these measures were comparable to data for healthy controls from other studies [42‐44], suggesting that they may serve as a compensatory mechanism to prevent more severe reductions in EF. The dependence on ECV, particularly for radial strain, is thus a potential cause for concern if the fibrotic expansion or cell atrophy [10] weakens that compensatory ability. Longitudinal follow-up, particularly with continued increase in ECV expansion, is needed to further evaluate these relationships.

This was a small cross-sectional study, so the ability to firmly establish causal relationships or long-term significance was limited. However, the use of CMR to measure all endpoints did ensure excellent reproducibility and reduced variance to enhance statistical power [45].

We did not quantify fibrosis or cardiac mechanics for the RV, which would be of obvious interest in repaired TOF, because of the generally insufficient resolution of the images for the thin wall of the RV. Future studies will leverage newer, higher resolution T1 mapping techniques [46] and higher resolution DENSE to better resolve the RV [44]. However, given the fact that LV diffuse fibrosis [10] and LV systolic dysfunction [3] have been implicated as predictors of poor outcomes in TOF, our focus on the LV is highly relevant.

Hematocrit data were not systematically available for all patients, necessitating the use of a “synthetic” hematocrit estimation to define ECV. This model may have produced inaccurate estimates for some patients, particularly younger individuals. While not the current gold standard for ECV estimation, it is important to note that the synthetic hematocrit model was derived and validated using a MOLLI sequence on 1.5 T Siemens scanners (inclusive of the Avanto model used for the present study), and demonstrated excellent agreement with the conventional measurement (95% limits of agreement of approximately ±4%) [20]. Given these findings, we feel its use, in the absence of hematocrit data, is justified. Moreover, associations of ECV and mechanics were generally similar to those of other T1-based endpoints, which did not require the synthetic hematocrit assumption, demonstrating the appropriateness of the ECV estimations.

No normal reference range for ECV was available from our laboratory, as is recommended for clinical reporting [47]. Differences in acquisition and analysis techniques may thus confound comparisons of absolute values to published reference ranges. However, as the primary objective of the study was to quantify associations with cardiac mechanics, this limitation had minimal effect on our primary findings.

Finally, no long-axis DENSE or T1 data were acquired to allow for the quantification of LV longitudinal strain. Previous studies have shown impairments in longitudinal strain in patients with TOF [36], so future studies should additionally explore the relationship of such impairments with fibrosis.