A multi-parameter response prediction model for rituximab in rheumatoid arthritis

Objectives: To validate the IFN response gene (IRG) set for the prediction of non-response to rituximab in rheumatoid arthritis (RA) and assess the predictive performance upon combination of this gene set with clinical parameters. Methods: In two independent cohorts of 93 (cohort I) and 133 (cohort II) rituximab-starting RA patients, baseline peripheral blood expression of eight IRGs was determined, and averaged into an IFN score. Predictive performance of IFN score and clinical parameters was assessed by logistic regression. A multivariate prediction model was developed using a forward stepwise selection procedure. Patients with a decrease in disease activity score (Delta DAS28) >= 1.8 after 6 months of therapy were considered responders. Results: The mean IFN score was higher in non-responders compared to responders in both cohorts, but this difference was most pronounced in patients who did not use prednisone, as described before. Univariate analysis in cohort I showed that baseline DAS28, IFN score, DMARD use and negativity for IgMRF and/or ACPA were associated with rituximab non-response. The multivariate model consisted of DAS28, IFN score and DMARD use, which showed an area under the curve (AUC) of 0.82. In cohort II, this model revealed a comparable AUC in PREDN-negative patients (0.78), but AUC in PREDN-positive patients was significantly lower (0.63), which seemed due to effect modification of the IFN score by prednisone. Conclusions: Combination of predictive parameters provided a promising model for the prediction of non-response to rituximab, with possibilities for optimization via definition of the exact interfering effect of prednisone on IFN score. (c) 2017 Published by Elsevier Masson SAS on behalf of Societe francaise de rhumatologie.