Introduction

During the past 15 years, major paradigm shifts have occurred in the treatment of rheumatoid arthritis (RA), related in large part to the introduction of biologic disease-modifying anti-rheumatic drugs (bDMARDs) and earlier initiation of conventional synthetic DMARDs, which have demonstrated substantial benefits [1,2,3,4,5].

RA is a chronic and debilitating disease which negatively impacts patients’ functional and overall health status, and reduces quality of life [6, 7]. In some patients, the promising therapeutic benefits of currently available bDMARDs, demonstrated in randomized controlled trials (RCTs), may be undermined by poor adherence and/or early discontinuation of treatment in clinical practice [8]. Frequent interruptions of treatment and switching of therapies makes longer-term clinical management of RA more challenging, results in greater costs for healthcare systems, and requires clinicians to develop sequential treatment strategies based on limited clinical and economic evidence [5, 9,10,11,12,13,14].

The extent to which patients with RA in routine clinical practice settings are cycling through bDMARDs has been characterized by some European registries, which have reported 1-year discontinuation rates of 25–54% for tumor necrosis factor antagonists [15] and 25–55% for other bDMARDs [16]. Two-year discontinuation rates of 35–50% for TNF antagonists have also been reported [15, 17, 18]. There are currently no RCTs fulfilling inclusion criteria for switching between bDMARDs [19].

In the US, few studies have examined the continuation of bDMARDs in patients with RA [8, 20,21,22,23,24,25]. The majority have focused on patients from large US private health plans or managed care organizations, which generally exclude patients who are very poor, sick, disabled, or old [26]. Therefore, there is a need for further investigation into the rates and reasons for bDMARD discontinuations.

The Consortium of Rheumatology Researchers of North America (Corrona) RA Registry is an independent database of patients with RA established in 2001, which includes clinical, laboratory, imaging, medication, and safety data [27, 28]. To date, Corrona has collected over 90,000 patient-years of data from over 600 participating rheumatologists throughout the US (both academic and private) [28].

This observational study was designed to examine initiation of bDMARDs in patients with RA within the Corrona database, and to characterize reasons for treatment discontinuations.

Methods

A study protocol was developed in line with the Agency for Healthcare Research and Quality (AHRQ) guidelines for observational studies [29].

Study Population

Figure 1 shows the patient population of the Corrona database. Data were collected from patients and their rheumatologists using questionnaires that gathered information on: disease duration, prognostic information, physician- and patient-determined standardized disease severity and activity measures, medical comorbidities, use of medications (including DMARDs), laboratory values, and adverse events (AEs). Follow-up assessments were requested at 4-month intervals and were completed during routine clinical consultations.

Fig. 1
figure 1

Corrona patient population. *bDMARD not started and dropped between visits. †At start of treatment. bDMARD biologic disease-modifying anti-rheumatic drug, RA rheumatoid arthritis

Full size image

Approvals for participation in Corrona were obtained from the respective institutional review boards of the participating academic sites and from a central institutional review board for private practice sites. Researchers had access only to de-identified patient data, and patient anonymity and confidentiality were safeguarded in compliance with the Health Insurance Portability and Accountability Act (HIPAA).

Study Design

Patients with RA who joined Corrona during the calendar years 2002–2011 were included in this study. Inclusion criteria were: adult patients (aged ≥18 years); adult-onset RA (aged ≥16 years); first bDMARD on-study; ≥6 months’ follow-up available after initiation of first or subsequent bDMARD therapy (defined as a visit ≥180 days after the initiation of bDMARD). The bDMARDs included TNF antagonists (adalimumab, certolizumab pegol, etanercept, golimumab, and infliximab) and non-TNF antagonists (abatacept, anakinra, rituximab, and tocilizumab).

The primary outcome was time to discontinuation of therapy, defined as the first report of stopping initial therapy or initiation of a new bDMARD at/between visits using a follow-up physician questionnaire. As many as three reasons for discontinuation were captured, which were not mutually exclusive. Short-term (<12 months after induction) and long-term (≥12 months after induction) discontinuation were also investigated.

Reasons for discontinuation reported by the physician were summarized for the overall population and for cohorts defined by follow-up duration, bDMARD class initiated, history of bDMARD treatment, and concurrent DMARD treatment. Reasons were categorized as follows: efficacy (loss of efficacy, disease flare, inadequate initial response, failure to maintain initial response); safety (toxicity, lymphoma/malignancy, serious AE, infection, minor AE); physician preference (patient doing well, treatment no longer required); patient preference (patient preference, frequency of administration, fear of future AE); access (formulary restriction, lack of insurance, cost/insurance); other (recent journal report, recent meeting report, recent lecture, withdrawn by US Food and Drug Administration, peer suggestion, other reason).

Statistical Analysis

For the analysis of time-to-change in therapy, Kaplan–Meier estimates of time to discontinuation were conducted for all initiators; an unadjusted Kaplan–Meier product limit was used to estimate durability rates at 6 and 12 months. The influence of baseline covariates of clinical interest on risk-of-treatment change and duration of response was investigated using Cox proportional hazards regression models. Univariate hazard ratios were used to explore the association between discontinuation and initiator characteristics. Hazard ratios for continuous characteristics are for a 1-unit increase.

Results

Patient Demographics

A total of 6209 patients who enrolled in the Corrona database between 2002 and 2011 met the inclusion criteria and were included in this analysis. Mean duration of follow-up was 37.2 months. Patient demographic and disease characteristics at initiation of therapy are detailed in Table 1. Overall, mean age was 57.6 years, 78.3% of patients were women, mean duration of RA was 10.7 years, mean Clinical Disease Activity Index (CDAI) was 16, mean number of swollen and tender joints was 4.9, and mean modified Health Assessment Questionnaire (mHAQ) was 0.5. Patients in the non-TNF antagonist cohort were older and with longer disease duration and greater disease activity scores compared with patients initiating TNF antagonists.

Table 1 Patient characteristics at treatment initiation
Full size table

Treatment Initiation

Overall, 80.7% of patients initiated bDMARD treatment with TNF antagonists; 19.3% with non-TNF antagonists (Table 1). Etanercept, adalimumab, and infliximab accounted for >95% of TNF antagonist use, while abatacept and rituximab comprised almost 90% of non-TNF antagonist use. At initiation, 43.4% of patients were bDMARD-naïve and 42.3% in the non-TNF antagonist cohort had received ≥2 prior bDMARDs.

Durability of Response

Median time to discontinuation or change of treatment was 25.1 months (Fig. 2a). Patients initiating non-TNF antagonist had lower durability rates versus those initiating TNF antagonists (log-rank p < 0.0001); median time to discontinuation was 26.5 months with TNF antagonists (N = 5010) versus 20.5 months (N = 1199) with non-TNF antagonists (Fig. 2b). Patients who were bDMARD-naïve (N = 2693) had numerically greater durability of response to the first bDMARD treatment compared with patients previously treated with bDMARDs, although this difference was not significant (log-rank p = 0.0601; data not shown). Durability curves were not significantly different between patients receiving bDMARDs as monotherapy and those receiving bDMARDs in combination with other DMARDs (data not shown).

Fig. 2
figure 2

Kaplan–Meier estimates of durability of response to first treatment (a) in the overall study population and (b) by bDMARD class initiated. CI confidence interval

Full size image

Treatment Discontinuation

Overall, 82.2, 67.3, and 51.1% of patients remained on initiated treatment at 6, 12, and 24 months, respectively (Fig. 2a). Proportions of patients receiving TNF antagonists remaining on initiated treatment at 6, 12, and 24 months were 82.5, 68.2, and 52.2%, respectively, compared with 80.9, 63.4, and 46.0% for patients receiving non-TNF antagonists.

Reasons for treatment discontinuation were captured for 49.2% of patients (1763/3584). Loss of efficacy was the most common reason for discontinuation (35.8%), followed by physician preference (27.8%), safety (20.1%), patient preference (17.9%), and lack of access to treatment (9.0%) (Fig. 3a). Patients with longer follow-up were more likely to give reasons for discontinuation, compared with those with shorter follow-up (Fig. 3b). Reasons for discontinuation between the TNF antagonist and non-TNF antagonist cohorts, between those who had previously received bDMARD treatment, and between those who were and were not receiving concurrent DMARD treatment were generally similar (Fig. 3c–e); however, the non-TNF antagonist group was more likely to discontinue due to loss of efficacy compared with the TNF antagonist group. In addition, patients who had previously received bDMARD treatment were more likely to discontinue due to loss of efficacy than those who were bDMARD-naïve.

Fig. 3
figure 3

Reasons given for discontinuation of treatment: a in overall population; b by follow-up duration; c by bDMARD class initiated; d by history of bDMARD treatment; e by concurrent DMARD treatment. bDMARD biologic disease-modifying anti-rheumatic drug

Full size image

Table 2 presents univariate hazard ratios exploring the association between discontinuation and baseline characteristics. Among the strongest predictors for discontinuation were mHAQ score, patient-reported anxiety/depression, year of bDMARD initiation (2007–2010 vs. 2002–2003), CDAI score, Disease Activity Score (DAS) 28, and number of prior DMARDs received.

Table 2 Association between discontinuation and baseline characteristics (univariate proportional hazards models)
Full size table

Discussion

This study was a comprehensive evaluation of the initiation of bDMARDs for treatment of patients with RA, and associated rates and reasons for treatment discontinuation from patients within the US Corrona database. For the period of time included in this study (2002–2011), a large proportion of patients did not remain on their initiated bDMARD for >24 months. Approximately one-third of patients discontinued the initiated bDMARD treatment by 12 months, and half by 24 months. Median time to discontinuation was significantly longer for patients who received TNF antagonists compared with those who received non-TNF antagonists; however, recorded reasons for treatment discontinuation were similar between the two cohorts. Physicians frequently reported more than one reason for discontinuation. Overall, the reported reasons were multifactorial and reflected expectations on the part of both patients and physicians. Loss of efficacy was the most common reason for discontinuation and likely reflected switching of therapies.

Discontinuation rates, such as those shown in this study, present a challenge for the management of RA. Patients had received a bDMARD ≥6 months, presumably indicating an initial response to treatment; therefore, this practice does not appear to reflect the empiricism required for selecting bDMARD therapies, as no biomarkers are available to facilitate selection of specific treatment classes. Reasons for discontinuation were varied, and frequently multiple. It was decided to focus these analyses on discontinuations—not differentiating between patients who switched treatment versus those who discontinued treatment altogether. Nonetheless, it is reasonable to assume additional healthcare resources may be required to manage such transitions and treatment changes.

The decision regarding how to manage patients who do not have a durable response to the initiated bDMARD treatment is challenging. Rheumatologists may consider several alternative approaches, including switching to a different agent in the same class (e.g., from one TNF antagonist to another), switching to a different class of bDMARD, or modifying concomitant DMARD treatment. Several professional associations, including the American College of Rheumatology [30] and the Consensus Group on Advances in Targeted Therapy [31], recommend switching between TNF antagonists when the first agent is associated with an inadequate response or poor tolerability, based on evidence derived primarily from observational studies. However, the likelihood of response to subsequent bDMARDs decreases as the number of prior treatments increases [32]. Changing bDMARD class is also an option in patients who discontinue due to lack of efficacy. Data from published RCTs have confirmed that non-TNF antagonists are effective in patients with inadequate responses to >1 TNF antagonist [33,34,35,36].

Results of this study are consistent with recent findings from national registries and other longitudinal observational studies, which indicate that 12-month survival rates for use of bDMARDs, such as TNF antagonists, are 65–83% [5]. In this current study, reasons for discontinuation could be determined in approximately 50% of patients, whereas previous US-based studies have been unable to ascertain these reasons. Factors with the strongest correlations to treatment discontinuation were higher CDAI and mHAQ scores, as well as patient reports of anxiety/depression. These reasons may be linked indirectly (patients with more active disease may be harder to treat and may be more likely to be depressed) [37], or directly (patients may experience depression as a result of ineffective treatment) to treatment discontinuation.

Of interest, patients who initiated bDMARD treatment in the period 2007–2010 were significantly more likely to discontinue treatment, compared with those in the base period (2002–2003). This may be explained either by increased availability of bDMARD treatment alternatives, or by increased patient and physician experience with these therapies in 2007–2010. Although few patients initiated treatment with certolizumab pegol, golimumab, or tocilizumab, more recent availability of these bDMARDs may have impacted decisions to switch therapies.

Several limitations need to be considered when interpreting the results of this study. Corrona is a voluntary database and therefore patients with RA at any one site, or patients treated consecutively, are not enrolled automatically. Nonetheless, a recent study compared the characteristics of patients participating in the Corrona registry with Medicare patients of the same age range and demonstrated that Corrona participants are representative of the patient population with RA in the US. The demographics and comorbidity profiles of Corrona participants and Medicare patients are very similar and, although Corrona participants are somewhat more likely to receive bDMARDs or DMARDs compared with Medicare patients, this difference is small (85% for Corrona participants vs. 73% for Medicare patients) [38]. As this study focused on time to discontinuation as the primary outcome, data related to dose or dose frequency adjustments were not evaluated. Furthermore, the length of time that patients discontinued bDMARD treatment could not be distinguished. However, treatment holidays were an exception to the treatment paradigm at the time, so discontinuation of one medication would typically result in an immediate switch to another.

While reasons for discontinuation (loss of efficacy, physician preference, safety, patient preference, and no access to treatment) have been provided here, evaluation of the future management of patients, including analyses of treatment switching data, is beyond the scope of the present report. Previous studies have demonstrated that patients with RA who experience treatment failures with one bDMARD (TNF antagonist) frequently switch to another [10, 39].

Conclusions

In conclusion, this study examined initiation of bDMARD therapies within the US Corrona database between 2002 and 2011, and characterized reasons for bDMARD discontinuation. Results highlight significant discontinuation rates during the investigated time period, with the most common reason being loss of bDMARD efficacy.