Abstract 1407 Machine learning-based phenotyping captures clinical complexity and improves prognostic resolution in patients undergoing transcatheter edge-to-edge repair for severe mitral regurgitation

Background: Transcatheter edge-to-edge repair (TEER) has emerged as new therapy for patients with severe mitral regurgitation (MR) and increased peri-operative risk. Depending on the extent of disease progression, comorbidities, and demographics, patients present with significant heterogeneity in structural and functional disease conditions.

Objective: This study sought to improve diagnostic and prognostic resolution in patients undergoing TEER for severe MR by unravelling clinically meaningful phenotypes using unsupervised agglomerative clustering (UAC) in conjunction with an artificial neural network (ANN).

Methods: 521 patients undergoing TEER for severe MR between 2009 and 2020 with technical success and significant reduction of MR were included in the analysis. Unsupervised agglomerative clustering was applied to pre-procedural echocardiography data. Primary outcome measure was post-procedural 5-year mortality.

Results: Cluster analysis revealed three distinct phenotypes:
Cluster 1 (296 patients [56.8%]) was constituted by patients presenting with preserved left ventricular ejection fraction (LVEF; 55.5 [95% CI: 54.5-56.4] %) and regular left ventricular end-systolic diameter (LVESD; 35.4 ± 7.75 [95%CI: 34.5-36.4] mm). A flail leaflet as marker of primary MR was diagnosed in 164/296 patients (55.4% vs. 1.9% in cluster 2 and 5.7% in cluster 3, p-value: <2x10^-16). 5-year survival in patients from cluster 1 was 51.0% (95% CI: 44.7-58.2%).
Patients from cluster 2 (102 patients [19.8%]) presented with reduced LVEF (30.1 [95% CI: 28.4-31.8] %) and enlarged LVESD (51.8 ± 9.41 [49.7-53.9] mm). Surprisingly, 5-year survival was 54.5% (95% CI: 44.2-67.1%) with no statistical difference to cluster 1 (p-value: 0.972).
Advanced stages of cardiac remodeling including biventricular dysfunction and biatrial dilatation were found in patients from cluster 3 (122 patients [23.4%]), and 5-year survival was significantly reduced (27.1% [95% CI: 18.5-39.8%], p-value: <0.001).
After randomly dividing the study population into derivation and validation cohorts, an ANN could precisely predict cluster assignment (accuracy: 91.0%), detecting patients from high-risk cluster 3 with good sensitivity (75.6%) and excellent specificity (99.1%).

Conclusion: Owing to distinct etiologies and varying degrees of disease progression, UAC identified three distinct phenotypes among patients with severe MR beyond the conventional classification of primary and secondary MR. Assigning patients to clusters can thus facilitate a more sophisticated risk stratification in future clinical practice.

Clin Res Cardiol (2022). https://doi.org/10.1007/s00392-022-02002-5
Machine learning-based phenotyping captures clinical complexity and improves prognostic resolution in patients undergoing transcatheter edge-to-edge repair for severe mitral regurgitation
T. Trenkwalder¹, M. Lachmann², H. A. Alvarez Covarrubias¹, E. Rippen², F. Schürmann¹, A. Presch³, M. von Scheidt¹, C. Ruff¹, P. Mayr⁴, I. Ott⁵, T. Schuster⁶, G. Harmsen⁷, S. Yuasa⁸, S. Kufner¹, P. Hoppmann², C. Kupatt², H. Schunkert¹, A. Kastrati¹, K.-L. Laugwitz², M. Joner⁹, E. Xhepa¹
¹Klinik für Herz- und Kreislauferkrankungen, Deutsches Herzzentrum München, München; ²Klinik und Poliklinik für Innere Medizin I, Klinikum rechts der Isar der Technischen Universität München, München; ³Ambulanz Erwachsenenkardiologie, Deutsches Herzzentrum München, München; ⁴Institut für Anästhesiologie, Technische Universität München, München; ⁵Medizinische Klinik I, Kardiologie, Helios Klinikum Pforzheim, Pforzheim; ⁶McGill University, Montreal, CA; ⁷University of Johannesburg, Johannesburg, ZA; ⁸Keio University School of Medicine, Tokyo, JP; ⁹Deutsches Herzzentrum München, München;
Background: Transcatheter edge-to-edge repair (TEER) has emerged as new therapy for patients with severe mitral regurgitation (MR) and increased peri-operative risk. Depending on the extent of disease progression, comorbidities, and demographics, patients present with significant heterogeneity in structural and functional disease conditions. Objective: This study sought to improve diagnostic and prognostic resolution in patients undergoing TEER for severe MR by unravelling clinically meaningful phenotypes using unsupervised agglomerative clustering (UAC) in conjunction with an artificial neural network (ANN). Methods: 521 patients undergoing TEER for severe MR between 2009 and 2020 with technical success and significant reduction of MR were included in the analysis. Unsupervised agglomerative clustering was applied to pre-procedural echocardiography data. Primary outcome measure was post-procedural 5-year mortality. Results: Cluster analysis revealed three distinct phenotypes: Cluster 1 (296 patients [56.8%]) was constituted by patients presenting with preserved left ventricular ejection fraction (LVEF; 55.5 [95% CI: 54.5-56.4] %) and regular left ventricular end-systolic diameter (LVESD; 35.4 ± 7.75 [95%CI: 34.5-36.4] mm). A flail leaflet as marker of primary MR was diagnosed in 164/296 patients (55.4% vs. 1.9% in cluster 2 and 5.7% in cluster 3, p-value: <2x10^-16). 5-year survival in patients from cluster 1 was 51.0% (95% CI: 44.7-58.2%). Patients from cluster 2 (102 patients [19.8%]) presented with reduced LVEF (30.1 [95% CI: 28.4-31.8] %) and enlarged LVESD (51.8 ± 9.41 [49.7-53.9] mm). Surprisingly, 5-year survival was 54.5% (95% CI: 44.2-67.1%) with no statistical difference to cluster 1 (p-value: 0.972). Advanced stages of cardiac remodeling including biventricular dysfunction and biatrial dilatation were found in patients from cluster 3 (122 patients [23.4%]), and 5-year survival was significantly reduced (27.1% [95% CI: 18.5-39.8%], p-value: <0.001). After randomly dividing the study population into derivation and validation cohorts, an ANN could precisely predict cluster assignment (accuracy: 91.0%), detecting patients from high-risk cluster 3 with good sensitivity (75.6%) and excellent specificity (99.1%). Conclusion: Owing to distinct etiologies and varying degrees of disease progression, UAC identified three distinct phenotypes among patients with severe MR beyond the conventional classification of primary and secondary MR. Assigning patients to clusters can thus facilitate a more sophisticated risk stratification in future clinical practice.
https://dgk.org/kongress_programme/jt2022/aP1608.html