Abstract:Introduction Severe tricuspid valve regurgitation (TR) is associated with increased 1-year morbidity and mortality. Characterization by valve etiology (primary, secondary, and lead-associated), a classification borrowed from mitral valve disease, has not been universally shown to correlate with outcomes. Purpose Among a large, racially diverse cohort with newly identified severe TR, we aimed to 1) characterize outcomes of severe TR by etiology, and 2) assess whether unsupervised phenoclustering or supervised outcome-driven prediction trees were more effective in establishing subgroups of TR with differential clinical risk profiles. Methods We retrospectively analyzed outcomes of 5-year all-cause death and a composite of death or heart failure hospitalization (HFH) among adult patients with new severe TR identified by echocardiography between 2007 to 2018 at a large academic tertiary referral center in the United States. Patients were initially categorized by etiology of TR, including primary, secondary, and lead-associated. Second, we separately applied unsupervised hierarchical clustering to identify distinct clusters using demographics, clinical, and echo data at the time of diagnosis. Third, we applied a supervised recursive partitioning algorithm (survival trees) by each outcome to identify distinct TR subgroups. We estimated the cumulative incidence of death and composite death or HFH over 5 years by 1) etiology of TR, 2) distinct clusters, and 3) groups identified by supervised learning (prediction trees). Results Among 2,379 consecutive patients with newly identified severe TR, the median age was 70 years, 61% were female, and 40% were Black. Event rates (95% CI) were 30.9 (29.0 to 32.8) events/100 PY for death and 49.0 (45.9 to 52.2) events/100 PY for composite death or HFH over median follow-up of 1.6 years. Event rates were similarly high across TR etiology groups for both death and composite death or HFH (Figure 1). Multiple methods of unsupervised clustering did not yield distinct clusters by patient demographic and imaging characteristics. After applying supervised survival tree modeling, four phenoclusters with distinct clinical prognoses were separately identified for death and composite death or HFH (Figure 2). Variables identified to partition the cohort to discriminate both death and composite death or HFH were age, albumin, blood urea nitrogen, right ventricular contractility, and right ventricular systolic pressure (all p<0.05). Conclusions Five-year cumulative incidence of adverse events among patients with newly diagnosed severe TR was 69% for death and 80% for composite death or HFH. TR etiology did not stratify prognosis, while supervised survival tree models identified phenoclusters with distinct clinical risk. The identified subgroups of severe TR with differential outcomes offer insights towards enrichment in clinical trials of TR and risk/benefit analysis in patients undergoing TR therapies. Funding Acknowledgement Type of funding sources: Private company. Main funding source(s): Abbott

Clinical Phenotypes and Outcomes of Patients with Left Ventricular Thrombus: An Unsupervised Cluster Analysis

Clinical Characteristics and Prognosis of Patients with Left Ventricular Thrombus in East China

Time Trends of Etiology, Treatment, and Long-Term Outcomes Among Patients with Left Ventricular Thrombus

Presence of Residual Cardiac Thrombus Predicts Poor Outcome in Cardioembolic Stroke after Reperfusion Therapy.

AI-based cluster analysis enables outcomes prediction among patients with increased LVM

Predictors and Prognosis of Left Ventricular Thrombus in Post-Myocardial Infarction Patients with Left Ventricular Dysfunction after Percutaneous Coronary Intervention.

The 5 Phenotypes of Tricuspid Regurgitation: Insight From Cluster Analysis of Clinical and Echocardiographic Variables

Clinical outcomes among distinct groups of patients with severe tricuspid valve regurgitation

Left ventricular thrombus after acute ST-segment elevation myocardial infarction: multi-parametric cardiac magnetic resonance imaging with long-term outcomes

Left ventricular reverse remodelling as a promising strategy for resolving left ventricular thrombus

Phenotypes and outcomes in non-anticoagulated patients with atrial fibrillation: An unsupervised cluster analysis

Left Ventricular Diastolic Dysfunction Is Associated with Poor Functional Outcomes after Endovascular Thrombectomy

Unsupervised clustering of venous thromboembolism patients by clinical features at presentation identifies novel endotypes that improve prognostic stratification

Contemporary incidence and predictors of left ventricular thrombus in patients with anterior acute myocardial infarction

Clinical phenotypes of heart failure patients with supranormal ejection fraction

Morphology and location of thrombus and sludge in patients with non-valvular atrial fibrillation

Factors predicting resolution of left ventricular thrombus in different time windows after myocardial infarction

Left ventricular thrombus in patient with nonischemic cardiomyopathy: A case report

Prevalence and clinical significance of left ventricular thrombus discovered by pathological examination in explanted hearts of patients with end-stage cardiomyopathy

Prognostic significance and natural history of left ventricular thrombi in patients with acute anterior myocardial infarction: a two-dimensional echocardiographic study

Identifying clinical phenotype clusters for coronary artery disease: a novel approach using clustering by unsupervised machine learning