Unsupervised phenomapping unveils fibrosis-driven phenotypes and arrhythmic risk in DSP-cardiomyopathy

Thursday, February 5, 2026

4:13 PM - 4:20 PM

Location: Galapagos / Aruba / Ilhabela (2nd Lower Level)

Presenting Author(s)

VS

Vincenzo Scialo', MD

Doctor, PhD student
Humanitas Research Hospital; Department of Radiology, Mayo Clinic, Rochester MN USA
Milan, Lombardia, Italy

Primary Author(s)

VS

Vincenzo Scialo', MD

Doctor, PhD student
Humanitas Research Hospital; Department of Radiology, Mayo Clinic, Rochester MN USA
Milan, Lombardia, Italy

Co-Author(s)

FE

Fatima M. Ezzeddine, MD

MD, MS
Department of Cardiovascular Medicine, Mayo Clinic, Rochester, MN, USA, United States
JG

John Giudicessi, MD, PhD

MD, PhD
Department of Cardiovascular Medicine, Mayo Clinic, Rochester, MN, USA, United States
MA

Michael Ackerman, MD, PhD

MD, PhD
Department of Cardiovascular Medicine, Mayo Clinic, Rochester, MN, USA, United States
SA

Samuel Asirvatham, MD

MD
Department of Cardiovascular Medicine, Mayo Clinic, Rochester, MN, USA, United States
LC

Leslie T. Cooper, Jr., MD

Chair
Mayo Clinic in Florida
Jacksonville, Florida, United States
EL

Esther Lutgens, MD, PhD

MD PhD
Department of Immunology, Mayo Clinic, Rochester, MN, USA, United States
TL

Tim Leiner, MD, PhD

Professor of Radiology
Mayo Clinic
Rochester, Minnesota, United States

Background: Desmoplakin-associated arrhythmogenic cardiomyopathy (DSP-ACM) is a genetic disorder characterized by progressive myocardial fibrosis. Risk stratification remains challenging due to phenotypic heterogeneity [1,2]. Current risk stratification guidelines for ACM continue to demonstrate suboptimal performance [3,4].
To overcome the limitations of the current risk stratification models, we applied an unsupervised cardiac magnetic resonance (CMR)-based phenomapping approach to identify distinct fibrosis-driven phenotypes in DSP-ACM and evaluate their associations with adverse cardiovascular outcomes.

Methods: We retrospectively studied 80 consecutive patients with pathogenic/likely pathogenic DSP variants who underwent CMR imaging. Sixteen late gadolinium enhancement (LGE) variables were used for Partitioning Around Medoids clustering with Gower distance. Cluster stability was assessed via silhouette and consensus analysis. ROC curves determined optimal LGE% thresholds for clustering. Associations with arrhythmic and heart failure outcomes were assessed using logistic regression analysis.

Results: Eighty patients with DSP-ACM were included (median age 39 years (IQR: 30), 44 (55%) female, and 50 (62.5%) with positive family history). The 3 clusters identified were: Cluster 1 (severe fibrotic phenotype (n=18), median LGE 34.1% (IQR: 26.8)), Cluster 2 (intermediate fibrotic phenotype (n=24), median LGE 15.3% (IQR:10)), and Cluster 3 (mild fibrotic phenotype (n=38), median LGE 0.0% (IQR: 1.3)). LGE% thresholds ( >32%, >6.6%) discriminated clusters with an AUC of 0.918 and 0.725, respectively (Figure 1). Cluster 1 had higher odds of sudden cardiac arrest (OR 4.87, 95 %CI 1.20–20.25, p=0.012) and heart transplantation (OR 5.42, 95 %CI 1.01–31.42, p=0.023), while Cluster 3 had lower odds of arrhythmic events (OR 0.18, 95 %CI 0.06–0.50, p< 0.001) and heart failure (HF) hospitalization (OR 0.28, 95 %CI 0.08–0.88, p=0.024) (Figure 2). In multivariate analysis, indexed left ventricular end diastolic volume (OR 1.048, 95 %CI 1.1–1.11, p=0.044) and circumferential LGE (OR 6.78, 95 %CI 1.70–31.90, p=0.009) independently predicted ventricular arrhythmic events (Table 1).

Conclusion: CMR-based phenomapping uncovered three unique phenotypes defined by distinct fibrosis burden and outcome patterns. By incorporating CMR features into phenotypic classification and risk stratification, we provide a new framework for refining risk stratification in patients with pathogenic/likely pathogenic DSP variants.

Multivariate analysis assessing predictors of arrhythmic events

Univariate analysis Multivariate analysis

Predictors	OR (95% CI)	P-Values	OR (95% CI)	P-Values

LVEDV/BSA	1.03 (1.01–1.05)	0.0004	1.048 (1.1–1.11)	0.0436*
RVEDV/BSA	1.03 (1.00–1.06)	0.0078	0.979 (0.928–1.029)	0.4201
LV WMA	3.50 (1.41–9.07)	0.0066	0.136 (0.014–0.888)	0.0567
LV Circumferential LGE	8.00 (3.03–22.91)	< 0.0001	6.785 (1.7–31.9)	0.009*
RV LGE	6.36 (2.00–24.64)	0.0013	4.632 (0.856–30.779)	0.0872
LV GLS	1.17 (1.05–1.32)	0.0037	1.144 (0.995–1.336)	0.0683

Abbreviations: BSA: body surface area; GLS: global longitudinal strain; LGE: late gadolinium enhancement; LV: left ventricle; LVEDV: left ventricular end diastolic volume; RV: right ventricle; RVEDV: right ventricular end diastolic volume; WMA: wall motion abnormalities.

Figure 1. ROC curves showing the performance of LGE% in discriminating each phenotypic cluster from the rest of the cohort in patients with DSP-ACM. The red curve represents Cluster 1 vs Clusters 2 and 3, with high discriminative capacity (AUC = 0.918), and an optimal threshold of >32%. The green curve represents Cluster 2 vs Clusters 1 and 3, with moderate performance (AUC = 0.725), and a cut-off of >6.6%. The blue curve represents Cluster 3 vs Clusters 1 and 2, which showed poor discriminative ability (AUC = 0.018). Dotted vertical lines indicate the optimal cut-off points derived from Youden’s Index for each comparison. The diagonal line represents the line of no discrimination (AUC = 0.5). Abbreviations: AUC = area under the curve; DSP-ACM = desmoplakin-associated arrhythmogenic cardiomyopathy; LGE = late gadolinium enhancement; ROC = receiver operating characteristic
Figure 2. Each cell shows the OR (95% CI) for the outcome in each cluster versus all others, with p-values (p < 0.05 marked with *). Color intensity is based on log₂(OR) scale, with white at OR = 1 and equal shading for proportional increases and decreases (e.g., OR = 2 vs. 0.5). Red indicates higher odds, blue lower odds. Abbreviations: HF: heart failure; ICD: implantable cardioverter defibrillator; SCA: sudden cardiac arrest.