Study on Rheumatoid Arthritis Prediction Model With Machine Learning and Ensemble Learning Methods, and Demographic, Clinical, Laboratory Characteristics of These Patients

The aim of this study is to develop a diagnostic and predictive model for rheumatoid arthritis (RA) with machine learning and ensemble learning methods using data from patients who applied to the rheumatology outpatient clinic with complaints of pain in the hand joints and to review the demographic, clinical and laboratory results of these RA patients. This study was conducted on demographic properties, clinical findings, and laboratory test results of a total of 421 patients, 260 with RA and 161 non-RA. Machine learning and ensemble learning algorithms were used for prediction. In this study, ensemble learning algorithms showed better performance than machine learning algorithms. With the Random Forest Classifier model 88% accuracy (95% confidence interval (CI), 81% - 94% ), 94% sensitivity (95% CI, 88% - 99%), 78% specificity (95% CI, 69% - 86%), 87% positive predictive value (PPV) (95% CI, 79% - 94%), 90% F1 score (95% CI, 84% - 96%), and 0.92 AUC (area under the curve) results were obtained. All three developed ensemble learning models showed similar performance in statistical metrics. These developed models can be improved with new multicenter studies with more patients and can be used in daily practice.