Background/Purpose:

For patients with rheumatoid arthritis (RA) who are refractory to tumor necrosis factor (TNF)-inhibitors, it is still not clear which determinants are associated with response when switching to rituximab (RTX). Prior studies have shown anti-CCP positivity and a lower number of previously-failed TNF inhibitors to be associated with good-to-moderate response. We developed a prediction model for response to rituximab with widely available clinical and laboratory parameters.

Methods:

RTX treated RA patients (n=115) were assessed prospectively for the following variables at baseline: demographic and disease characteristics, number of previous biologicals, co-morbidities, basic laboratory data, rheumatoid factor levels, anti-CCP levels and immunoglobulin levels (IgA, IgM and IgG). Clinical response was determined at 6 months after RTX initiation, using the European League Against Rheumatism (EULAR) response criteria for RA. Variables with p < 0.10 in univariate analyses were selected as candidate variables and entered into a forward and backward logistic regression analyses. A prediction model was constructed using backward logistic regression and an area under the curve (AUC) was calculated for the new prediction model.

Results: Mean disease activity score 28 (DAS28) was 5.34 at baseline and 4.44 at 6 months (p<0.001). A good-to-moderate response was achieved in 57 patients (49.6%). 101 patients (87.8%) used at least 1 biological prior to RTX treatment. RTX was re-administered in 70.9% of patients. The prediction model (see figure) consisted of age (in years), number of swollen joints, hydroxychloroquine use at baseline, steroid use at baseline and IgM level at baseline. The AUC (95% confidence interval) was 0.756 (0.663-0.850). When only using clinical variables (excluding IgM levels), the AUC was 0.714 (0.621-0.808).

Conclusion:

We developed a prediction model for response to rituximab with simple clinical and laboratory parameters. This model is easy to use and accurate in the prediction of response to treatment. Further validation of this model could make the implementation of this model cost-effective and patient friendly.

Figure. Area Under the Receiver Operating Characteristic (AUROC) of a predictive model of EULAR response at 6 months after RTX treatment (A) and the clinical determinants comprising the model (B)