Despite the fact that multiple studies have revealed characteristics related to effective treatment outcomes in rheumatoid arthritis (RA), there was a dearth of accurate prediction models for patients on biologic drugs who were in remission. Apart from logistic regression (LR), no machine learning (ML) techniques have been tested on the class of issues to the best of knowledge. Patients with RA who began a biological disease-modifying antirheumatic medication (bDMARD) in a tertiary care institution were studied in the longitudinal research. At the therapy baseline, 12-month, and 24-month follow-up, demographic and clinical information was gathered. An attribute core set was determined using a wrapper feature selection approach. Four alternative machine learning algorithms, LR, random forest, K-nearest neighbors, and extreme gradient boosting, were then trained and validated with 10-fold cross-validation to predict 24-month sustained DAS28 (Disease Activity Score on 28 joints) remission. The algorithms’ performance was then compared, with accuracy, precision, and recall being evaluated.

The study comprised 367 patients (male 323/367, 88%) with a mean age SD of 53.7+12.5 at the time of bDMARD initiation. About 175 (47.2%) of 367 patients experienced long-term DAS28 remission. Acute-phase reactant levels, Clinical Disease Activity Index, Health Assessment Questionnaire–Disability Index, and other clinical features were among the attribute core set utilized to train algorithms. Extreme gradient boosting outperformed random forest (accuracy, 65.9%; precision, 65.6%; recall, 59.3%), LR (accuracy, 64.9%; precision, 62.6%; recall, 61.9%), and K-nearest neighbors (accuracy, 64.9%; precision, 62.6%; recall, 61.9%). (accuracy, 63%; precision, 61.5%; recall, 54.8%).

Researchers demonstrated that machine learning models might be used to predict long-term remission in patients with RA on bDMARDs. Furthermore, the technique was based on only a few simple patient characteristics. The findings were encouraging, but they needed to be validated in larger cohort studies.

Reference:journals.lww.com/jclinrheum/Abstract/2022/03000/A_Machine_Learning_Approach_for_Predicting.13.aspx