In a prospective study of 201 patients with Behçet syndrome, unsupervised machine learning integrating flow-cytometry immune-cell subsets and clinical features identified 4 distinct subtypes with different organ involvement, remission rates, and treatment responses. The study also linked each cluster to distinct transcriptomic pathways, including TNF, JAK-STAT, coagulation/MAPK, and NF-κB signatures.

Why It Matters To Your Practice

• Behçet syndrome is clinically heterogeneous, and this analysis suggests immune-cell profiling may help explain why patients with similar diagnoses behave differently in clinic.

• The 4 clusters separated patients by patterns clinicians recognize: isolated mucocutaneous disease, arthritis-predominant disease, cardiovascular involvement, and neurological involvement.

• One-year follow-up showed these groups also differed in remission and drug response, raising the possibility of more tailored treatment selection.

Clinical Implications

• Cluster 1 had isolated mucocutaneous lesions, lower inflammation, and higher remission rates.

• Cluster 2 featured arthritis and higher inflammatory markers and responded well to TNF-α inhibitors.

• Cluster 3 included cardiovascular involvement, reduced CLA+ Tregs, and also responded well to TNF-α inhibitors.

• Cluster 4 showed neurological involvement, elevated CD161+ Tregs, lower remission, and better response to mycophenolate mofetil.

Insights

• The model combined peripheral immune phenotyping with clinical manifestations rather than relying on symptoms alone.

• RNA sequencing suggested biologic plausibility for each subtype: IFN-γ/IL-6/JAK-STAT in Cluster 1, TNF and B-cell activation in Cluster 2, coagulation/platelet activation/MAPK in Cluster 3, and T-cell activation/NF-κB in Cluster 4.

• For clinicians interested in AI, this is a practical example of unsupervised learning finding disease structure that could support phenotype-driven care.

The Bottom Line

• Machine learning classified 201 Behçet syndrome patients into 4 immune-clinical subtypes tied to prognosis and treatment response.

• It is not ready to replace bedside judgment, but it points toward a future in which immune profiling and AI help match Behçet patients to more personalized therapy.