Immunophenotyping and machine learning-based stratification reveal prognostic immune signatures in Behçet’s disease

Objective The aim was to characterize peripheral blood lymphocyte subsets in BD patients, and to stratify patient subgroups and identify prognostic biomarkers. Method Immunophenotyping of 11 immune cell subsets was performed using flow cytometry in peripheral blood samples from treatment-naïve BD patients(n = 71) and age- and sex-matched healthy controls (HCs)(n = 71). XGBoost, PLS-DA and k-means clustering were employed to analyze the immune profiles and stratify BD patients based on their immunological subtypes and clinical parameters. Kaplan-Meier analysis and log-rank tests were used to compare relapse-free survival across BD patient clusters. Cox regression analysis was used to assess the clinical and immunological predictors of relapse. Results Our study reveals a T cell dominant immune dysregulation in BD patients compared with HCs. Machine learning integration of immunophenotyping and clinical parameters identified two BD subgroups. Cluster 2 exhibited hyperactivated T cells(increased memory CD4+ T cells, CD8+DR+ T cells, and CD8+CD38+ T cells), higher inflammatory markers (hsCRP, ESR, NLR and PLR), higher disease severity scores and relapse risk, Multivariate cox regression analysis revealed that high hsCRP and CD8+DR+T cells proportion were independent predictors of relapse. The HR-weighted combinations of CD8+DR+T%, hsCRP and disease severity scores exhibited superior relapse predictive performance (AUC = 0.815). Conclusion This study provided a comprehensive immunophenotypic profile of treatment-naïve BD patients. Machine learning integration of lymphocyte subsets and clinical parameters enabled the identification of subgroups at high risk of relapse at an early stage, providing a rationale for personalized treatment strategies.