Discussion:
The average 2D echocardiographic volume and pressure dimensions, including EDV, ESV, LAVi, and RVSP, all significantly decreased after Mitraclip placement, consistent with other studies. However, the LVEF and GLS, as more discrete markers of LV function, demonstrate more complex findings. Given that mean LVEF was 52% and only 25% of patients had LVEF<40%, many of our patients did not have signifanct LV dysfunction detectable by LVEF. LVEF showed less change post clip than standard 2D estimating variation (about +/- 8%27). GLS, which detects earlier dysfunction and is less load dependent28, likely offers amore sensitive assessment of LV dysfunction in our patients , and our data suggests more pronounced GLS associations at more extreme GLS values, particularly severely impaired GLS.
Our population consisted largely of patients with abnormal pre-GLS, 51/57 (89%), abnormal as defined by ASE as GLS <|18%|29. Notably, GLS <│7%│ was recently showed to demonstrated worse all-cause mortality outcomes than GLS>│7%│ among functional MR patients28, and only six of our study patients were below │7%│. In our study, the curvi-linear graph (Figure 2) shows that among the lower, more abnormal │pre-GLS│ values on the left-side of the graph, a higher │pre-GLS│ is associated with greater reductions in EDV as depicted by the greater curvature. That pattern changes at higher, more normal │pre-GLS│ values. If studied in a more abnormal GLS population of <│7%│, the trend seen in the more left-sided part of our curve may be more apparent. The left-sided curvilinear trend is supported by findings in subgroups, where higher │pre-GLS│ sample averages were noted in patients with at least 10% reduction in EDV compared to those without, as seen in Table 4, though not statistically significant odd ratios as seen in Table 3. In such, our data’s association among more severely abnormal GLS values, although not robust, may be more clinically useful.
In a study of 41 FMR patients in Italy, worse GLS was shown to predict lack of reverse remodeling, as defined by 10% reduction in ESV at 6 months. GLS was the only independent correlate of reverse remodeling (<p=0.01), and a GLS cut off of -9.25% (<0.01) was associated with reverse remodeling on a ROC curve, 81% sensitivity and 74% specificity. 16 Our results likely differ because our population was more heterogenous including FMR, DMR and mixed. This seemingly prominent impact of etiology on GLS predictive ability may reflect the inherent pathophysiologic relationship of functional MR with LV dysfunction. Combining MR types may therefore be clinically impractical. Also, our study allowed for a longer follow up, which may have underestimated our observed effect as over time other cardiac impairments could develop. Furthermore, we used EDV, where this study used ESV. Our average LVEF and │pre-GLS│ values were higher at 52% vs 34.4% (SD5.4) LVEF and 12.5% (SD 4.2) vs 11.3% (SD 3.9) │pre-GLS│.16 The early, left-sided portion of our curvilinear graph shows similar findings seen in this Italian study which addressed patients with more severe LV dysfunction. Our finding that EDV was stronger than ESV in the multivariable analysis supports our use of EDV change as the volume dimension for evaluating reverse remodeling in our population. To minimize type I error, we only looked at reverse remodeling in terms of change of EDV but perhaps evaluating other markers of reverse remodeling for correlation to GLS such as ESV, LAVi, RVSP, or EF could have detected significant predictability.
Our study supports other studies in finding automated GLS highly reproducible, despite needing practitioner edits. Inter and intra reliability in our study, r=0.90 and r=0.95, respectively was similar other mainstream studies (0.89 inter, 0.93 intra observer reproducibility)28Limitation and Hypothesis-generating considerations
A sample size of 641 patients would have been required to confidently state GLS fails to predict reverse remodeling (avoiding type II error), rendering our study significantly underpowered. Many of our limitations stem from small sample size and our efforts to reduce type I error risk. Our sample only had 7 FMR patients, and we therefore chose to combine the clinically similar FMR and mixed MR subgroups. The combined group better matches our larger proportion of DMR when performing statistical comparisons.
With only 57 patients, the adjusted model significantly reduces the power allotted to any regression co-efficient, and may explain the loss of significance in GLS. Since ESV and EDV correlate clinically as heart size changes and the crude regression values are very similar, and multicollinearity likely explains the ESV change from negative to positive regression coefficients seen in Table 5.
To be most generalizable, we looked at all MitraClip patients, rather than selecting just those that demonstrated evidence of pre-clip remodeling. Perhaps we would have seen more reverse remodeling if we targeted just those with pre-clip remodeling. This is difficult to identify clinically without a clear gold standard, particularly in FMR patients where LV dilation may be due to reasons other than MR.
Our graph depicts four patients with particularly large EDV changes, and while these patients are all DMR with at least moderately enlarged pre-EDV (all >200ml) and │pre-GLS│ greater than │9.25│, discrete factors affording specifically them major improvements are unknown and warrant further study.
Tomtech software is unable to accurately calculate strain with heart rate variability >10%, such as in atrial fibrillation or arrythmia. We did not exclude patients based on heart rate or arrythmia as we felt the GLS autostrain values were not grossly unexpected compared to clinician visual estimates, and we favored including all patients to increase sample size and generalizability. We avoided additional adjusting as that further reduces power. However, the precision in GLS measurement may have been compromised slightly.