COMPARING STATISTICS WITH MACHINE LEARNING MODELS TO PREDICT DOSE INCREASE OF INFLIXIMAB FOR RHEUMATOID ARTHRITIS PATIENTS

摘要

Rheumatoid arthritis (RA) is a chronic inflammatory joint disease that leads to irreversible joint destruction. To prevent this, new biological therapies, such as Infliximab, have been developed. The present analysis is based on an expanded access program in which 511 RA patients with chronic refractory disease were treated with Infliximab. They received a standard dose of 3 mg/kg on weeks 0, 6, 14 and 22. On week 22, the treating rheumatologist had to evaluate the progress of every patient and decide whether the current dose should be increased or not. This decision can be considered as a measure of insufficient response. In the present analysis, two machine learning classification techniques (support vector machines and multilayered perceptrons) are implemented to model the decision to give a dose increase. Their performance on increasingly multivariate real-life data will be studied and compared to classical statistics. Results show that both classical statistics, SVM and MLP - if configured well - show good classification performance. However, as the number of features increases, the performance decreases. SVMs suffer to a lesser degree from this curse of dimensionality.