Cardiac structural changes after transcatheter aortic valve replacement: systematic review and meta-analysis of cardiovascular magnetic resonance studies

We searched PubMed and Embase for original studies that reported CMR findings before and at least 1 month post-TAVR, from January 1, 2000 to August 23, 2018, using a combination of Medical Subject Heading (MeSH) terms and keywords, in accordance with Meta-analysis of Observational Studies in Epidemiology (MOOSE) guidelines (Supplementary Table 1) [13]. The MOOSE guidelines have been developed to help with standardized data design, data abstraction, meta-analysis, and reporting of observational studies. The guidelines also address some of the challenges of the observational studies such as the differences in study design, inherent biases (e.g. selection bias), and confounding factors such as varying disease severity [14].

All titles and abstracts of the retrieved studies were screened and where needed, the full manuscript text was reviewed for inclusion. We also searched the reference lists of included studies to identify additional potentially relevant studies. We included all cohort studies and case series that reported CMR findings before and at least 1 month after TAVR. The choice of the time interval was by authors’ consensus, so that there would be sufficient time for acute changes such as myocardial edema to resolve, and for reverse remodeling to occur [12, 15]. No language restriction nor sample size restriction were applied in selecting the articles. We excluded non-human studies, as well as ex-vivo heart assessments. Where the data from the same center were presented in multiple publications, we only included the results of the most comprehensive report from that center to avoid duplicate data.

For each study, the data screening and abstraction was performed by 2 independent investigators (GM, and PT). All discrepancies were resolved by consensus. In addition, the investigators of primary studies were contacted for clarifications or additional data, where relevant data were missing. Main factors of interest were LV end-diastolic volume index (LVEDVi), LV end-systolic volume index (LVESVi), LV mass index (LVMi), and left ventricular ejection fraction (LVEF).

Assessing the quality of included studies, especially the appropriate reporting of the methodological details is critical for interpretation of the meta-analysis results. We assessed the methodological quality of the included full-text studies using the Newcastle-Ottawa scale as recommended by Cochrane Collaboration [16]. The Newcastle-Ottawa scale scores each study to a maximum of 9, on 3 main areas of the selection of the study groups, comparability of the groups, and the ascertainment of the outcome of interest [17].

Standardized mean differences (SMD) from the included studies were pooled by conventional random effects meta-analytic techniques, to account for the variations across the studies. Random effects meta-analysis assumes that the estimated effects in different studies are not identical [18]. The degree of heterogeneity across studies was assessed by the I-squared (I²). The I² < 25% was considered as low heterogeneity, and I² > 75% as high heterogeneity [19]. We made a priori plans not to pool the data for variables that were reported by fewer than 3 studies. To account for possible differences in studies over time, we also performed a random effects meta-regression analysis, with year of publication and duration of follow-up as covariates, to verify the robustness of the main results. We used funnels plot and the Harbord’s regression modification of Egger’s test to assess potential publication bias for the small study effect [20]. A two-sided P-value of < 0.05 was considered as significant. We used STATA software, version 12.0 (StataCorp, College Station, Texas, USA) for all analyses.

CMR at follow-up showed a significant reduction in LVEDVi (9 cohorts; 242 patients; SMD: -0.25, 95% CI: − 0.43 to − 0.07, P = 0.006). No evidence of heterogeneity was observed between the included studies (I²: 0%, P_{heterogeneity} = 0.97) (Fig. 2a). The median study-level average reduction in LVEDVi was 4.0 ml/m² (interquartile range [IQR]: 3.1 to 8.2).

LVESVi was also reduced after TAVR (7 cohorts; 198 patients; SMD: -0.24, 95% CI: − 0.44 to − 0.05, P = 0.01) with no evidence of heterogeneity (I²: 0%, P_{heterogeneity} = 0.96) (Fig. 2b). The median study-level average reduction for LVESVi was 5.0 ml/m² (IQR: 3.0 to 6.0).

LVMi was significantly reduced after TAVR (9 cohorts; 246 patients; SMD: -0.82, 95% CI: − 1.0 to − 0.63, P < 0.001). Across the included studies, no evidence of heterogeneity was observed (I²: 0%, P_{heterogeneity} = 0.58) (Fig. 2c). The median average reduction for LVMi was 15.1 g/m² (IQR: 11.8 to 18.3).

LVEF was significantly increased after TAVR (12 cohorts; 305 patients; SMD: 22, 95% CI: 6 to 38%, P = 0.006). Among the included studies, there was no evidence of heterogeneity (I²: 0%, P_{heterogeneity} = 0.87) (Fig. 2d). Across the studies, the median average LVEF at baseline was 54.6% (IQR: 52.0 to 58.1%) and the median average increase was 3.4% (IQR 1.0 to 4.6%).

In a post-hoc sensitivity analysis, we repeated the analyses for LVEDVi, LVESVi, LVMi, and LVEF after excluding 3 studies that were presented as abstracts. Although the point estimates had modest changes, the results were consistent, suggestive of significant reduction in LVEDVi, LVESVi, and LVMi, and improvement in LVEF (Table 2, Supplementary Figure 1).

The quality of included studies was assessed using the Newcastle-Ottawa Scale. The total score varied between 6/9 and 8/9. The most frequent limitations were the lack of representation of the exposed cohort, and incomplete follow-up (Supplementary Table 2).

No publication bias was apparent from visual inspection of the funnel plots for LVEDVi, LVESVi, LVEF, and LVMi (Supplementary Figure 2) and this was confirmed quantitatively with the Harbord’s regression modification of Egger’s test (P > 0.25 for all).

Meta-regression analysis did not show a significant association between either the duration of follow-up post-TAVR or the year of study publication and any of LVEDVi, LVESVi, LVEF and LVMi measurements (P > 0.25 for all) (Supplementary Table 3).

Our study has several limitations. First, it is possible that some reports were not retrieved through our search strategy or were not indexed in PubMed. Second, it is clinically relevant to understand the pattern of reverse remodeling in certain patient subgroups, including those with concentric hypertrophy without LV dilation. In addition, mitral regurgitation is common in patients undergoing TAVR, and presence and severity of mitral regurgitation may also influence the remodeling [39]. However, detailed clinical and CMR data across the studies were not available to address this issue. Third, data regarding the valve type, and its potential effect on the cardiac remodeling was not available across the included studies. Fourth, the vast majority of included studies assessed the CMR characteristics 6 months post-TAVR. Additional studies are required to report post-TAVR CMR features in other time intervals. Fifth, there were a few (7/100) overlapping patients among 2 of the included studies. However, given the total number of patients in the pooled sample (N = 305), it is unlikely that such a modest overlap substantively impacted the results. Sixth, across the included studies, several patients were excluded from follow-up CMR post-TAVR due to permanent pacemaker implantation, death, or claustrophobia. It is possible that the patients who died or who had pacemakers have a different remodeling response to TAVR. This hypothesis is supported by subgroup analyses of prior relatively small TTE studies in patients with pacemakers, post-TAVR [40]. With the availability of CMR compatible pacemakers, as well as promising results of a recent investigation about the relative safety of CMR even in patients with traditional pacemaker devices [41], it is possible that future CMR studies further assess these subgroups.