Man-machine interaction-based phenotyping identifies pathophysiologically and prognostically informative clusters among patients with mitral regurgitation undergoing transcatheter edge-to-edge repair. (3rd October 2022)

• Record Type: Journal Article
• Authors: Lachmann, M; Trenkwalder, T; Covarrubias, H A A; Rippen, E; Schuermann, F; Presch, A; Ruff, C; Mayr, P N; Schunkert, H; Kastrati, A; Kupatt, C; Laugwitz, K L; Joner, M; Xhepa, E
• Is Part Of: European heart journal. Volume 43(2022)Supplement 2
• Publication Date: 2022-10-03
• Subjects: Cardiology -- Periodicals; Heart -- Diseases -- Periodicals; 616.12005
• DOI: 10.1093/eurheartj/ehac544.1568
• Languages: English
• ISSNs: 0195-668X

Abstract

Abstract: Background: Depending on etiology, extent of disease progression, and comorbidities, patients with severe mitral regurgitation (MR) typically present with considerable heterogeneity. Purpose: This study therefore sought to improve diagnostic and prognostic resolution in patients undergoing mitral valve transcatheter edge-to-edge repair (MV TEER) for severe MR by developing a man-machine interaction-based phenotyping approach. Methods: All 609 patients from this single-center registry underwent MV TEER for severe MR between 2009 and 2020. Unsupervised agglomerative clustering was applied to preprocedural echocardiography data, and an artificial neural network (ANN) was subsequently trained for future patient-to-cluster assignment. Primary outcome measure was postprocedural 5-year survival Results: Cluster analysis revealed four pathophysiologically and prognostically informative phenotypes: Cluster 1 was constituted by patients (n=188) presenting with preserved left ventricular ejection fraction (LVEF; 56.5±7.79%) and regular left ventricular end-systolic diameter (LVESD; 35.2±7.52 mm). 5-year survival in patients from cluster 1, hereinafter serving as a reference, was 60.9% (95% CI: 53.3–69.7%). Patients from cluster 2 (n=102) also presented with preserved LVEF (55.7±7.82%) and regular LVESD (34.9±7.68 mm), but showed the largest mitral valve effective regurgitant orifice area (0.623±0.360 cm 2 ) and highest systolic pulmonary artery pressures (68.4±16.2 mmHg). Consequently, their 5-year survival ranged at 43.7% (95% CI: 33.2–57.6%; p-value: 0.032). Patients from cluster 3 (n=270) were predominantly characterized by impaired left ventricular systolic function (LVEF: 31.0±10.4%) and dilated left ventricular diameters (LVESD: 53.2±10.9 mm), and their 5-year survival was reduced to 38.3% (95% CI: 31.9–46.1%; p-value: <0.001). Poorest 5-year survival (23.8% [95% CI: 12.8–44.3%]; p-value: <0.001) was observed in patients from cluster 4 (n=49) with biatrial dilatation (left atrial volume: 312±113 mL; right atrial area: 46.0±8.83 cm 2 ) although LVEF was only slightly reduced (51.5±11.0%). All patients from cluster 4 were diagnosed with atrial fibrillation. An ANN could precisely predict cluster assignment (accuracy: 85.2%), detecting patients from high-risk clusters 3 and 4 with excellent specificity (95.0% and 99.4%, respectively). Conclusion: Assigning patients to clusters using a multiparametric phenotypic approach can facilitate risk stratification in future clinical practice. Our unsupervised machine learning-based classification system differs from previous approaches for risk stratification, because we do neither hypothesize a linear sequence of accumulated pathologies caused by severe MR (potentially ignoring the aggravating impact of comorbidities), nor do we stratify patients into low- and high-risk cohorts in accordance with a single variable's dichotomy (prone to oversimplification). Funding Acknowledgement: Type of funding sources: None.