Machine Learning for Predicting Treatment Response in Rheumatoid Arthritis Patients Receiving Rituximab: A DAS28-Based Approach

Abstract

Rheumatoid Arthritis (RA) is a chronic, debilitating autoimmune disease characterized by chronic inflammation of the joints, leading to progressive joint damage and functional disability [1], [2]. Despite significant advancements in pharmacological therapies, including conventional synthetic disease-modifying antirheumatic drugs (csDMARDs) and biological DMARDs (bDMARDs), not all patients achieve remission or low disease activity [3], [4]. Rituximab, a B-cell depleting agent, represents a crucial bDMARD option for patients with RA, particularly those who have responded inadequately to other therapies [5]. However, response to rituximab is heterogeneous, and predicting which patients will benefit most remains a significant challenge, leading to empirical treatment choices, potential side effects, and delayed access to effective therapy for non-responders [7], [9]. The Disease Activity Score 28 (DAS28) is a widely accepted composite measure for assessing RA disease activity and treatment response [26], [27]. This article investigates the application of various machine learning (ML) approaches to predict the DAS28 score after rituximab treatment in RA patients, leveraging a comprehensive set of clinical and genetic variables. By analyzing complex interactions within patient data, ML models offer the potential to identify subtle patterns indicative of future treatment response, thereby facilitating personalized medicine and optimizing therapeutic strategies. Our methodology involved collecting pre-treatment patient characteristics, including demographics, disease activity markers, prior treatment history, and relevant genetic polymorphisms. Several ML algorithms were trained and evaluated to predict DAS28 scores at specific post-treatment time points. The results highlight the superior predictive capabilities of ML models compared to traditional clinical prognostication, offering a promising tool for clinicians to make more informed treatment decisions, improve patient outcomes, and reduce healthcare costs by avoiding ineffective therapies.