Icon Legend

This session is not in your schedule.

This session is in your schedule. Click again to remove it.

Bioinformatics, Computational and Systems Biology

Bioinformatics, Computational and Systems Biology - Poster Session C

Poster A4 - Identifying Phenotypic Differences and Risk Factors in Cardiac Sarcoidosis Patients Using Unsupervised Machine Learning

Friday, October 25, 2024

10:00 AM - 11:00 AM EST

Location: Exhibit Hall E, F & G

Presenting Author(s)

Thoya Raman

Biomedical Engineering Undergraduate Student
Johns Hopkins University
Fremont, California, United States

Co-Author(s)

AW

Anjali Wagle

Medical Student
Johns Hopkins University School of Medicine, United States
Ishan Vatsaraj, MS

PhD Student
Johns Hopkins University, United States
SL

Shane Loeffler, PhD

Postdoctoral Fellow
Johns Hopkins University
Baltimore, Maryland, United States

Last Author(s)

NG

Nisha Gilotra

Associate Professor of Medicine
Johns Hopkins University School of Medicine, United States

Primary Investigator(s)

NT

Natalia Trayanova, PhD (she/her/hers)

Professor of Biomedical Engineering
Johns Hopkins University
Baltimore, Maryland, United States

Introduction: Sarcoidosis is a systemic, inflammatory disease that affects multiple organs, including the heart. Cardiac involvement in sarcoidosis is linked to ventricular arrhythmias (VAs), heart failure (HF), and atrioventricular block (AVB), but the causes and progression of the disease to these conditions remain poorly understood. Previous statistical studies identified sex- and race-based differences in the presentation, treatment, and outcomes of cardiac sarcoidosis (CS) patients. In this work we apply unsupervised learning to CS clinical data to better understand underlying patterns in CS manifestation and its outcomes. This aims to provide clinicians with better insights for early identification and improved management of the disease.

Materials and

Methods: This study used clinical data from 249 patients at the Johns Hopkins Hospital diagnosed with CS. The patient data consisted of demographic information, cardiac data at presentation, extra-cardiac sarcoidosis pathology, echocardiogram data at initial, 6-month, and 12-month intervals, immunosuppression treatment data, and patient outcomes. The analysis included data preprocessing (missing value imputation, scaling), dimensionality reduction, clustering, and cluster analysis (Figure 1). Principal component analysis (PCA) was then performed, retaining components that explained at least 80% of the variance in the data. Subsequently, the K-Means clustering algorithm was applied, and clusters were visualized in 2D space. A random forest model was used to identify top features that differentiated the clusters. Finally, cluster feature and outcome statistics were calculated to better understand key differences between the clusters.

Results, Conclusions, and Discussions: Following data preprocessing, analysis was performed on 206 CS patients. The algorithm successfully clustered the patients into two distinct groups (Figure 2). Cluster 1 consisted of 88 patients, while cluster 2 consisted of 118 patients. Very few patients in cluster 1 had clinical presentation with extra-cardiac sarcoidosis, but many received an implantable cardioverter-defibrillator (ICD) and anti-tachycardia pacing (ATP) or cardioversion shocks (delivered by their ICD) to treat ventricular tachycardias (VT). In contrast, most patients in cluster 2 presented with extra-cardiac sarcoidosis but few received an ICD or treatments delivered by it. Additionally, patients in cluster 1 exhibited high rates of new sustained VT occurrences after starting immunosuppression treatment when compared to those in cluster 2 (Table 1).

Further analysis revealed a strong, positive correlation between new occurrences of sustained VT and receiving pacing or cardioversion treatments in patients in cluster 1. A significantly higher rate of new sustained VT was observed in patients who received both an ICD and pacing or cardioversion than in those who only received an ICD but never had treatment delivered by it. Additionally, the rate of heart failure was also higher in patients who received pacing or cardioversion treatment than in those who did not (Table 2).

These results suggest that there are 2 distinct phenotypes found in patients with CS. Receiving ATP or cardioversion shock treatment delivered by an ICD was identified as a potential risk factor for new sustained VT occurrences after initiation of immunosuppression treatment in CS. This highlights the importance of close monitoring and potentially preemptive interventions for patients with these risk factors. Additionally, the higher rate of heart failure in patients receiving ATP or cardioversion treatments suggests a more severe disease progression in this subgroup, highlighting the potential need for more aggressive management strategies in this subgroup.

Acknowledgements (Optional):