Topological Data Analysis Identified Prognostically-Distinct Phenotypes in Transcatheter Edge-to-Edge Repair Patients

Figures

• 

  Figure 1. The topological graph representing the TEER dataset. (A) The distribution of the 2 clusters; cluster 0 on the left (green-colored) and cluster 1 (red-colored) on the right. There were transitional nodes that contained patients of both clusters (orange or light-orange colored). (B) The distribution of the STS MV Replace risk score. Nodes with high-risk patients (red-colored) were distributed in both clusters. (C) The distribution of mortality rate. Nodes with higher-mortality rate patients (orange-red colored) were mainly located in the area of cluster 1 (right side). The color of each node represents the average value of the interested parameter of all the patients included in the node. Please refer to the Supplemental Method for detailed interpretation of the topological graphs.

• 

  Figure 2. SHAP analysis plots for model interpretation. (A) Histogram of feature importance at the global level showed the input features in the order of relative importance: TR, MV prolapse, MV tethering, AFib or Aflutter, MV leaflet calcification, mitral annulus calcification, MV flail, prior MI, functional MR, degenerative MR and other etiologies. The clusters were color-coded (cluster 0: blue and cluster 1: red). Higher absolute mean SHAP value indicates more importance of the variable in contributing to the model output. (B) cluster 0 was characterized with higher proportion patients of none or mild TR, posterior MV prolapse, no MV tethering, no Afib or flutter, no leaflet calcification, no annulus calcification, posterior MV flail, and no previous MI. Cluster 1 had the opposite characters as found in (C)

• 

  Figure 3. Kaplan-Meier survival analysis of the 2 clusters. (A) Cluster 0 (blue line) had significantly better survival performance compared to cluster 1 (red line) in the development cohort (log-rank P<.001). (B) Similar survival difference was observed in the prospective validation cohort (log-rank P=.04).

• 

  Figure 4. The topological graph represents the MV morphology variables. The low-risk cluster (cluster 0) nodes were on the left side and the high-risk cluster nodes were on the right side, as shown in Figure 1. (A) Cluster 1 had higher proportions of anterior leaflet or bileaflet prolapse. (B) Cluster 1 had more patients with MV tethering, and cases with anterior (orange colored) or bileaflet tethering (red-colored) were mostly belong to cluster 1. (C) Cluster 0 predominantly had posterior leaflet flail cases (light green to white colored) compared to cluster 1. (D and E) Cluster 1 had higher proportions of patients with mitral annular and leaflet calcifications.

• 

  Supplemental Figure S1.

• 

  Supplemental Figure S2.

• 

  Supplemental Figure S3.

• 

  Supplemental Figure S4.