Key points are not available for this paper at this time.
Background: Sustained remission is the primary therapeutic goal for patients with rheumatoid arthritis (RA). Although several studies have identified factors associated with treatment response in rheumatoid arthritis (RA), there are no precise predictive models for sustained remission in patients treated with biologic therapies. Objectives: Evaluating the performance of machine learning to predict short-term sustained remission in rheumatoid arthritis patients under biological disease modifying antireumatic drugs (bDMARDS). Methods: We included patients with rheumatoid arthritis, who started biological disease modifying antireumatic drugs, from the Moroccan registry of biological therapies, with a follow-up of at least 1 year and an evaluation in at least two consecutive visits with an interval of 6 months. Sustained remission was defined as DAS28 Results: We included 130 patients (88.5% female); mean age ± SD was 51.52 ± 10.86 years. 38 patients (29.2%) achieved sustained remission. The set of variables used to train the algorithms included initial DAS 28, rheumatoid factor and anti-citrullinated peptides antibodies positivity, Health Assessment Questionnaire-Disability Index, type and number of biotherapies, Methotrexate use, as well as several clinical features. Adaboost showed the best performance (accuracy, 75%; precision, 75%; recall, 100%), outperforming Random Forest (accuracy, 57%; precision, 75%; recall, 60%), K-nearest neighbors (accuracy, 57%; precision, 33%; recall, 100%) and logistic regression (accuracy, 29%; precision, 33%; recall, 25%). Conclusion: Machine-learning models can be used to predict sustained remission in rheumatoid arthritis patients on biotherapy. REFERENCES: NIL. Acknowledgements: NIL. Disclosure of Interests: None declared.
Aharrane et al. (Sat,) studied this question.