Predicting therapeutic outcomes in Rheumatoid Arthritis using real-world pharmacogenetic and clinical data

OBJECTIVE: To develop a pharmacogenetic and clinical model to predict effectiveness outcomes in a cohort of patients diagnosed with rheumatoid arthritis (RA) treated with methotrexate. METHODS: Five machine learning methods were tested on the patient data set with previous preprocessing for dataset cleaning and feature selection: Logistic regression, decision trees, random forests, Support Vector Machines (SVM) and Artificial Neural Networks (ANN). The models were applied in a cohort of 155 patients treated with MTX which was derived in a training (124 patients) and a test cohort (31 patients). Both clinical variables and genetic variations were included. The chosen outcome was the therapy response measured as a DAS 28 <3.2. The performance evaluation criterion was the area (AUC) under the receiver operating characteristics (ROC) curve. RESULTS: The algorithms with the highest predictive power were SVM (AUC: 0.81, 95% CI: 0.74 - 0.90) and ANN (AUC: 0.82, 95% CI: 0.75 - 0.89). For the MTX cohort, the main selected variables were age, follow-up time, functional class, and genotypes of the rs9344, rs4148396, rs4673993, rs1801133 and rs7279445 variants. CONCLUSIONS: A prognostic model with high predictive power was developed in the cohort of patients treated with MTX, which is able to identify patients less likely not to respond to treatment.