Deep hierarchical subtyping of multi-organ systemic sclerosis trajectories - a EUSTAR study

Area of Science:

• Immunology
• Rheumatology
• Computational Biology

Background:

• Systemic sclerosis (SSc) is a complex autoimmune disease characterized by multi-organ damage.
• Current SSc classification relies on skin involvement (limited vs. diffuse), but organ-specific variability suggests more subtypes exist.

Purpose of the Study:

• To develop a semi-supervised deep learning framework for SSc patient stratification.
• To identify novel, clinically meaningful SSc subtypes based on organ involvement and severity.

Main Methods:

• A generative deep learning model was applied to the European Scleroderma Trials and Research (EUSTAR) database (14,000 patients, 67,000 visits).
• Expert-defined organ involvement criteria were leveraged to model disease trajectories.
• Model performance, robustness to missing data, and clinical interpretability were systematically evaluated.

Main Results:

• Five distinct patient clusters were identified based on the degree of organ involvement.
• A subset of patients with limited skin SSc exhibited significant risks for lung and heart complications.
• The data-driven approach identified clinically relevant patient stratification beyond traditional classification.

Conclusions:

• Deep learning models can uncover novel SSc subtypes by integrating multi-organ data.
• This approach enhances patient stratification and prognosis, particularly for identifying high-risk individuals.
• Multi-organ modeling is crucial for complementing clinical insights in SSc management.