Rheumatoid arthritis (RA) is a chronic, inflammatory disease characterized by persistent synovitis and associated with pain, functional disability, and decreased quality of life as well as increased risk of death affecting an estimated 1.3 million Americans [1, 2]. The goal of therapy is to reduce disease activity and improve clinical outcomes. The current treatment paradigm is to first use conventional disease-modifying anti-rheumatic drugs (cDMARDs) followed by step-up to combination cDMARD therapy or initiation of a biologic [3–5]. Typically, a tumor necrosis factor inhibitor (TNFi) is the first biologic class initiated [6]. While this class of drugs is associated with improvement in the signs and symptoms of RA, it has been shown both in large randomized controlled trials (RCTs) and in everyday clinical practice that as many as 30–40% of patients develop an inadequate response to TNFis [7–10]. This inadequate response may be related to either primary nonresponse (lack of response after initiation) or a secondary nonresponse which is treatment failure due to drug resistance or intolerance.

However, there is conflicting information regarding which should be the next agent to manage a patient who has had an inadequate response to a TNFi. There have been inconsistent results regarding the benefits of changing mechanism of action in observational data as a general approach, or whether targeting a specific pathway after failure of a TNFi will optimize outcomes. For example, improved outcomes were demonstrated in comparisons of rituximab vs a subsequent TNFi [11–14] but not in abatacept (ABA) initiators vs a subsequent TNFi [15]. More recently a RCT found greater effectiveness with use of non-TNFi biologics as compared with a second anti-TNF drug in TNF inadequate responders [16].

Given the absence of head-to-head RCTs comparing the non-TNFi biologics in patients with inadequate response to an anti-TNF agent, comparative effectiveness studies using observational data from registries can be employed [17]. To address the limitations of observational studies, such as selection bias, propensity score methodology is commonly employed [15, 18, 19]. We used propensity score matching to compare the clinical effectiveness of tocilizumab (TCZ) vs ABA among RA patients with previous anti-TNF exposure in a large US cohort of RA patients using the Consortium of Rheumatology Researchers of North America (Corrona) registry. Specifically, we sought to compare change in disease activity, achievement of low disease activity (LDA), and change in function over 6 months.

Using data from Corrona, a large US-based RA registry, we compared the clinical effectiveness of TCZ vs ABA among TNFi-experienced patients with moderate-to-high disease utilizing patients from real-world rheumatology practices. We compared change in disease activity, achievement of LDA/remission as well as meaningful improvement in function in the two treatment groups. Both treatments were effective for the primary outcome because the minimally clinical important difference for change in CDAI is a reduction of 6 points for moderate disease activity and 12 points for high disease activity [24]. Furthermore, we conducted a post-hoc analysis to evaluate whether the dose of TCZ contributed to the results. Specifically, there was no difference in terms of change in CDAI, change in mHAQ, or achievement of LDA/remission among those on 4 mg/kg at the end of the study period as compared with those who had escalated to 8 mg/kg (n = 182, 69%). Additionally, the safety of the two agents was comparable over the study time period.

Our results are similar to an observational cohort from Japan [19], in which investigators examined outcomes in 102 matched pairs of patient treated with TCZ at 8 mg/kg every 4 weeks vs ABA dosed based on weight (500 mg for patients < 60 kg, 750 mg for patients 60–100 kg, and 1000 mg for patients > 100 kg) at weeks 0, 2, and 4 and then every 4 weeks after. They found no significant difference in retention rates between the two agents, which is consistent with our findings (Additional file 1: Table S1). When evaluating efficacy over 52 weeks using the SDAI they identified no differences between the two agents, with approximately 46–50% achieving LDA/remission at 6 months and 47–55% at 12 months. These rates are slightly higher than our study, but this may be in part be due to differences in the outcome measure used, patient characteristics, and study design. Of note, the patients in our cohort had higher baseline disease activity based on the CDAI. Furthermore, Corrona’s registry reflects real-world practice. Physicians may thus administer medications at doses and frequencies different from the package labeling (e.g., holding doses for procedures or hospitalizations). Additionally, we included patients in the analyses regardless of whether they switched to another biologic or discontinued the agent without starting a new biologic. With respect to functional impairment, both studies showed no difference between the treatment groups (ours using mHAQ scores and the Japanese cohort using HAQ-DI scores).

We additionally examine the impact of concomitant RA medications on our findings. This included an evaluation as to whether there were differences in the use of prednisone (both initiation and dose escalation) and cDMARDs (initiation and discontinuation) between the two groups. No such differences were found (Additional file 1: Tables S2 and S3). We also explored whether our selection approach influenced the results. We found similar outcomes in ABA patients who were not selected for inclusion into the study (because of receipt of the medication prior to the availability of TCA) to the ABA patients who were included.

We believe that comparative effectiveness studies of non-TNF biologics are needed for practicing rheumatologists when trying to decide which agents to prescribe in their patients with prior TNFi exposure. Rheumatologists want to prescribe the right drug to the right patient at the right time. Providing comparative data with accompanying safety information is essential for rheumatologists in order to facilitate patients achieving the best clinical outcomes. Given the lack of head-to-head RCTs, observational studies are the best approach for evaluating comparative treatment outcomes. Using “best practice” approaches, propensity score methodology was used specifically due to selection bias. Based on these findings, both agents were associated with substantial improvement in RA symptoms. The individual choice of which agent to prescribe will likely be informed based on patient characteristics, pharmacy benefit plan, and patient preference.

This study had several strengths. It is the largest known comparative effectiveness study of ABA vs TCZ in the United States and included a nationwide sampling of patients with RA. The all-comers study design recruited individuals from multiple rheumatology centers, resulting in a range of patients with real-world disease activity and comorbidities not often seen in RCTs. We also evaluated effectiveness by examining response in all patients regardless of continuation of initial treatment. This has been shown to provide a conservative estimate [15]. Additionally, we examined comparative effectiveness broadly with evaluation of response using the CDAI, function using the mHAQ, and safety between the two drug-exposure cohorts because all three elements are needed for clinical decision-making.

This study also had some limitations related to the challenges of operationalizing available, real-world data, and applying analytical methods conservatively. As with any registry, there is a concern that patient enrollment may not reflect the type of patients observed elsewhere in general practice. However, a previous study [25] demonstrated that Corrona RA patients enrolled in Medicare shared similar demographic and clinical characteristics to the national cohort of RA patients based on Medicare rheumatology-based claims. Of note, we only had sufficient sample size to explore 6-month outcomes. Potentially a longer time frame may have demonstrated differences. Also, as in any observational study, bias is a concern because physicians prescribe therapies based on the patient’s profile and treatment selection is not random. To overcome this limitation, we matched by propensity score and stratified by prior TNFi. However, this method does not address unmeasured confounders. We did include patients in the analyses who discontinued ABA or TCZ (<4% in each group) without initiating a subsequent biologic because the reason for discontinuation was not clear (e.g., drug holiday due to a good response to the medication, side effects, out-of-pocket costs, etc.). Because there was no significant difference between groups in discontinuation rates, however, the outcome of the study (i.e., no difference in clinical response between the two drugs) would not be altered if we excluded these patients from the analyses. Because of sample size considerations, sensitivity analyses were not performed based on reasons for discontinuation of the prior TNFi, although other studies have shown that this can influence treatment response [26].

Among patients with prior exposure to TNFi, use of ABA and TCZ both resulted in improved disease activity based on the MCID of the CDAI and function as measured with the mHAQ. Furthermore, the safety profiles of the two treatments were similar over the 6-month time frame. Based on these findings, the results suggest that both agents would be appropriate choices as the next therapeutic agent in patients with moderately-to-severely active RA with prior exposure to TNFis. Further analyses are necessary to better identify the patient subsets likely to respond to a particular agent.