Identification of Distinct Cardiac Phenotypes in Autoimmune Rheumatic Disease Patients Using CMR-Based Clustering: Associations with Adverse Outcomes

Background: Clustering-based methods have the unique potential to uncover novel phenotypes that might be overlooked by traditional statistical techniques. This is particularly valuable for characterizing understudied populations, such as patients with autoimmune rheumatic diseases (ARD). In this study, we utilized clustering in combination with cardiovascular magnetic resonance (CMR) imaging features to identify distinct cardiac phenotypes within a large cohort of ARD patients. Additionally, we examined the association of these phenotypes with adverse events and developed a CMR-based algorithm for phenotypic classification.

Methods: In this retrospective study, we identified patients with ARD from the Yale CMR registry who underwent contrast CMR between July 2011 and December 2023. Left ventricular (LV) and right ventricular (RV) volumes and ejection fraction (EF) were calculated. Definite late gadolinium enhancement (LGE) was defined as any LGE after excluding nonspecific insertion zone LGE. K-means clustering (SPSS 31.0.0.0) was used to identify distinct CMR phenotypes. The predictive value of cluster membership was evaluated using Kaplan-Meier survival analysis for a combined outcome of death, heart failure hospitalization, myocardial infarction or revascularization (MACE).

Results: Retrospective chart review identified a total of 699 patients with ARD (mean age: 59±15 years, 69% female) with the most common conditions being rheumatoid arthritis (n=244) and systemic lupus erythematosus (n=91). Clustering analysis identified four unique phenotypes: one cluster with a high presence and extent of LGE, a cluster with predominantly impaired LV function, a cluster with dilated RV and RV failure and a cluster with normal function and cavity size (Fig1A). Notably, patients with distinct ARD were dispersed amongst the CMR phenotypes. Event-free survival rates varied between clusters with the LV cluster showing highest adverse event rates, followed by the LGE then RV clusters (Fig1B). A decision tree was developed based on the driving CMR features of distinct clusters (Fig1C), and the decision-tree groups showed similar separation for MACE compared to the k-means cluster-defined groups (Fig1D).

Conclusion: CMR feature-based clustering successfully identified four distinct cardiac phenotypes within a large cohort of ARD patients, which were associated with adverse outcomes. Additionally, a decision tree was developed based on the dominant CMR characteristics of these clusters to facilitate identification of patients at increased risk for MACE. Future research should focus on validating these findings in other ARD cohorts and exploring their potential for guiding personalized treatment strategies.

Figure 1. Cardiac magnetic resonance (CMR)-based clustering in patients with autoimmune rheumatic diseases (ARD). 1/A Radar plot represents the unique features of four CMR phenotypes identified by K-means clustering. 1/B CMR-based ARD phenotypes associated with adverse cardiovascular outcomes. 1/C CMR-based decision tree enables simple classification of ARD patients using routinely measured CMR parameters 1/D The Kaplan–Meier curves using the decision tree-based separation were similar to the actual k-means clustering-derived groups. Abbreviations: CI= cardiac index; HR= hazard ratio; LGE= late gadolinium enhancement; LVEDVi= left ventricular end diastolic volume index; LVEF= ventricular ejection fraction; LV GLS= left ventricular global longitudinal strain; LVESVi= left ventricular end systolic volume index; LV massi= left ventricular mass index; LVSVi= left ventricular stroke volume index; MR=mitral regurgitation; RVEDVi= right ventricular end diastolic volume index; RVEF= right ventricular ejection fraction; RVESVi= right ventricular end systolic volume index; RVSVi= right ventricular stroke volume index; PA= pulmonary artery; TR= tricuspid regurgitation