Machine learning identifies right index finger tenderness as key signal of DAS28- CRP based psoriatic arthritis activity

Abstract

Abstract Background Psoriatic arthritis (PsA) is a chronic inflammatory systemic disease that is often categorized based on the Disease Activity Score 28 (DAS-28 CRP). However, since DAS28-CRP was originally designed for rheumatoid arthritis, it may not perfectly reflect PsA, and periodic re-evaluation has been recommended. Methods A cohort of 80 PsA patients (44 women and 36 men, aged 56.3 ± 12 years) with a range of disease activity from remission to moderate was analyzed using unsupervised and supervised methods applied to the DAS28-CRP components. Results Machine learning-based permutation importance identified tenderness in the metacarpophalangeal joint of the right index finger as the most informative item for PsA activity staging. This symptom alone allowed a random forest classifier to identify PsA remission with 67% balanced accuracy in new cases. Projection of the DAS28-CRP data onto an emergent self-organizing map of artificial neurons identified outliers, who following enhancement of group sizes by generative artificial intelligence (AI) could be defined as subgroups particularly characterized by either joint tenderness or swelling. Conclusions AI-assisted re-evaluation of the DAS28-CRP for PsA has narrowed the score items to a most relevant symptom, and generative AI has been useful for identifying and characterizing small subgroups of patients whose symptom patterns differ from the majority. These findings represent an important step toward precision medicine that can address outliers.