Routine biomarker profile for the prediction of clinical phenotypes of adult-onset Still's disease using unsupervised clustering algorithm.

Abstract

This study addresses the challenge of predicting the course of Adult-onset Still's disease (AoSD), a rare systemic autoinflammatory disorder of unknown origin. Precise prediction is crucial for effective clinical management, especially in the absence of specific laboratory indicators. We assessed the effectiveness of combining traditional biomarkers with the k-medoids unsupervised clustering algorithm in forecasting the various clinical courses of AoSD-monocyclic, polycyclic, or chronic articular. This approach represents an innovative strategy in predicting the disease's course. The analysis led to the identification of distinct patient profiles based on accessible biomarkers. Specifically, patients with elevated ferritin levels at diagnosis were more likely to experience a monocyclic disease course, while those with lower erythrocyte sedimentation rate could present with any of the clinical courses, monocyclic, polycyclic, or chronic articular, during follow-up. The study demonstrates the potential of integrating traditional biomarkers with unsupervised clustering algorithms in understanding the heterogeneity of AoSD. These findings suggest new avenues for developing personalized treatment strategies, though further validation in larger, prospective studies is necessary.