Prediction of hidden patterns in rheumatoid arthritis patients records using data mining

Rheumatoid Arthritis (RA) disease is an inflammatory disease, which is characterized by persistent synovitis and autoantibodies that eventually lead to joint damage and reduced quality of life. This paper implements two data mining methodologies to explore the most important attributes that correlated with RA disease activity: (1) Feature selection algorithms to be used by Association rules (Apriori and Predictive) or Classification algorithms (J48 and J48 Consolidated), (2) Predictive rules (Rule Induction), Feature weight (Information Gain) and Trees algorithms (CHAID). This study experiments a pre-collected dataset consists of 260 patient records with a confirmed diagnosis of RA. The experimented algorithms are measured in terms of F-Measure, Accuracy, and the output tree. The accuracy of the J48 classification algorithm result was 79.18%. Many new rules were found by using the Predictive- Apriori technique from the association rules algorithms. By using the Information Gain algorithm, the most important attributes that highly correlated with the disease discovered were identified. This study revealed a model that validates the previous RA studies and includes new parameters that include both non-pharmacologic measures (No smoking, physical exercise and patient compliance) and pharmacologic therapies (MTX dose above 20 mg /week, prednisone dose > 5 mg/day as add-on therapy and biologic DMARDs (adalimumab, preferred in our study) and Hb > 10.8 g/dl). The model would help RA patients to have will controlled and low disease activity.