Clustering of rheumatoid arthritis patients: an unsupervised machine learning approach for characterizing the TNFi response.

Abstract

This study aimed to identifydistinct profiles of response to Tumor Necrosis Factor inhibitors (TNFi) in rheumatoid arthritis (RA) patients using unsupervised machine learning to integrate clinical, demographic, and genetic data. A cohort of 294 RA patients was analyzed using hierarchical clustering techniques. Data collected included body mass index (BMI), adherence to physical activity, prevalence of comorbidities, seropositivity, Health Assessment Questionnaire (HAQ) scores, and genetic polymorphisms in TNF pathway. Responders to TNFi demonstrated a more favorable clinical profile, with a lower BMI (25.6 vs. 27.0; p = 0.03), higher physical activity adherence (56.6% vs. 33.3%; p < 0.01), reduced prevalence of comorbidities (69.8% vs. 82.4%; p = 0.02), and improved functionality (HAQ: 0.8 vs. 1.7; p < 0.01) compared with non-responders. The C allele of rs767455 was significantly more frequent among responders (74.4% vs. 62.4%; p = 0.04). Clustering analysis identified three distinct subgroups: a better prognosis group characterized by a 73.5% response rate, minimal use of combination therapies, high adherence to physical activity, and male predominance; a poorer therapeutic response group with 82.9% therapeutic failure, intensive medication use, and low exercise adherence; and an intermediate group. TNFi response in RA is multifactorial. Integrating clinical, demographic, and genetic data enables the identification of distinct patient profiles, supporting the development of personalized treatment strategies and the advancement of precision medicine in RA management.