Background/Purpose: Response to biologic agents approved for RA may be associated with patient reported factors that are not related to disease and usually not included in studies attempting to predict response to therapy. The objective was to evaluate whether such factors can influence the probability of response to biologic agents and persistency of therapy.

Methods: Patients with RA initiating a biologic agent while in moderate or high disease activity (based on CDAI levels) were eligible for participation in the Corrona-CERTAIN comparative effectiveness study and were included in the analysis. ENSEMBLE Minimum Data Set (MDS), which is comprised of 10 patient (pt) reported scales not specific to RA, (such as income levels, depression, perceived stress and depression, social support, education) was used to evaluate prediction of response to biologic agents in combination with factors collected in the Corrona registry (“traditional factors”). ENSEMBLE components were measured at time of initiation of biologic. Response to biologic therapy was evaluated at 6 months as achievement of low disease activity (LDA). Persistency was estimated using the Kaplan-Meier method. Associations of response to therapy and persistency with baseline values of ENSEMBLE components were assessed based on logistic and Cox regression, respectively. To determine whether ENSEMBLE components added value to “traditional” disease related factors in predicting response, Akaike information criterion (AIC) and Bayesian information criterion (BIC) were evaluated for the three models: mix of “traditional” and ENSEMBLE variables; traditional variables only; and emphasis of ENSEMBLE over traditional variables. The resulting response and persistency models with lowest AIC and BIC were identified and further evaluated for discriminatory power.

Results: Analysis included 2152 biologic initiations in moderate or high disease activity. At 6 months, 682 patients had achieved LDA (responders) and 1470 had not (non-responders). Of the responders, 58% were treated with a TNF inhibitor and 37% were biologic naïve prior to enrollment in CERTAIN, vs 49% and 30% of non-responders respectively.  Of the three response models, the best AIC and BIC values corresponded to the logistic model that emphasized ENSEMBLE over traditional variables and the discriminatory power for this model was comparable (72.6%) to the other two models (72% and 72.5%).%) in predicting response to therapy. For time to discontinuation of biologic, the emphasized ENSEMBLE model was again the best in terms of lowest AIC and BIC and discriminatory power (C-statistic) was comparable (60.8% to 59.7% and 61.7%).

Conclusion: The generic, non-disease specific set of scales of ENSEMBLE added value to models predicting response to therapy of RA with biologic agents. Such factors may allow a refined assessment of patient heterogeneity beyond traditionally used disease specific measures and have a similar performance in predicting therapy response and persistency with traditional factors.