Effectiveness of initial methotrexate-based treatment approaches in early rheumatoid arthritis: an elicitation of rheumatologists’ beliefs

Abstract

Objectives

To quantify rheumatologists’ beliefs about the effectiveness of triple therapy (MTX + HCQ + SSZ) and other commonly used initial treatments for RA.

Methods

In a Bayesian belief elicitation exercise, 40 rheumatologists distributed 20 chips, each representing 5% of their total weight of belief on the probability that a typical patient with moderate–severe early RA would have an ACR50 response within 6 months with MTX (oral and s.c.), MTX + HCQ (dual therapy) and triple therapy. Parametric distributions were fit, and used to calculate pairwise median relative risks (RR), with 95% credible intervals, and estimate sample sizes for new trials to shift these beliefs.

Results

In the pooled analysis, triple therapy was perceived to be superior to MTX (RR 1.97; 1.35, 2.89) and dual therapy (RR 1.32; 1.03, 1.73). A pessimistic subgroup (n = 10) perceived all treatments to be similar, whereas an optimistic subgroup (n = 10) believed triple therapy to be most effective of all (RR 4.03; 2.22, 10.12). Similar variability was seen for the comparison between oral and s.c. MTX. Assuming triple therapy is truly more effective than MTX, a trial of 100 patients would be required to convince the pessimists; if triple therapy truly has no–modest effect (RR <1.5), a non-inferiority trial of 475 patients would be required to convince the optimists.

Conclusion

Rheumatologists’ beliefs regarding the effectiveness of triple therapy vary, which may partially explain the variability in its use. Owing to the strength of beliefs, some may be reluctant to shift, even with new evidence.

Rheumatology key messages

• Rheumatologists’ beliefs in the effectiveness of triple therapy and s.c. MTX varies widely.

• The strength of existing beliefs has implications for the design and uptake of future clinical trials.

Introduction

MTX is considered the preferred initial DMARD for patients with moderate–severe early RA [1, 2]. There is controversy, however, over how it should be administered (orally or s.c.) and whether it should be used alone or in combination with other conventional synthetic DMARDs (csDMARDs). Randomized controlled trial (RCT) evidence for MTX-based csDMARD combinations is largely confined to triple therapy: the combination of MTX, SSZ and HCQ [3–6]. The quality and generalizability of these trials, however, have been questioned [2]. Clinical practice guidelines by both the ACR and EULAR recommend MTX monotherapy over triple therapy, although the recommendation is conditional, with both regarding triple therapy as an effective option [1, 2]. RCT evidence for other commonly used treatments is very sparse. In a network meta-analysis [7, 8], only a single trial compared oral vs s.c. MTX for the primary outcome (ACR50 response) [9], and no trials directly compared MTX vs MTX plus HCQ, another commonly used csDMARD combination (‘dual therapy’).

In the absence of strong recommendations, it is perhaps not surprising to see considerable variability in practice patterns. In the Canadian Early Arthritis CoHort (CATCH), the use of combination csDMARDs as initial treatment varied between nine treatment centres from a low of 19% to a high of 78%, with the most common combinations being dual therapy or triple therapy [10]. The use of s.c. MTX as initial therapy varied from 0% to 96% [11]. Similar variability in practice patterns is common in other countries [12]. These may reflect perceived limitations in the clinical evidence, which vary from practice to practice. Rheumatologists’ views about the real-world effectiveness may be shaped by their knowledge and interpretation of the evidence, as well as their clinical experience.

In a Bayesian context, experts’ beliefs about an unknown parameter can be formally measured as probability distributions [13, 14]. These probability distributions are highly informative, as they quantify not only the expert’s belief regarding the average value of a parameter, but also their confidence in this belief. Various approaches exist for eliciting these distributions. In the ‘bin’ or ‘roulette’ method, each participant is given a certain number of chips that, when summed together, reflect the total weight of belief (e.g. 20 chips, each representing 5% of their total weight of belief) [15]. The participants place the chips on a grid to reflect their belief regarding the unknown parameter. If a rheumatologist is particularly confident in his or her belief, the distribution will be very narrow. Conversely, uncertainty is reflected in wider distributions.

The objective of this study was to quantify rheumatologists’ beliefs about the effectiveness of initial MTX-based treatment approaches for early RA. In doing so, we sought to understand how these beliefs compare to existing evidence from randomized trials, and furthermore, how these beliefs may impact the design of clinical trials.

Methods

Study population and recruitment

Rheumatologists were eligible if they were practicing rheumatologists who had seen at least one patient with newly diagnosed RA over the past year. We purposively sampled rheumatologists nationally to ensure representation by region, sex, years in practice and practice type (community/academic). The predetermined sample size was 40 rheumatologists, to allow separation into three to four subgroups of at least 10 participants. Rheumatologists were recruited in person or by e-mail invitation.

Bayesian belief elicitation

Scenario overview

Rheumatologists were asked to complete a Bayesian belief elicitation exercise. Each participant completed a one-on-one interview with a research assistant (P.M.H.), who had been trained and coached in interview techniques by researchers with prior experience in belief elicitation (S.R.J., G.T.). The interview was guided by a structured interview guide that was developed in an iterative process in discussions with methodologists (S.R.J., G.T.) and rheumatologists (see Supplementary Data S1, available at Rheumatology online).

In the scenario, rheumatologists were asked to consider a typical patient with moderate–severe early RA and no contraindications to MTX, SSZ or HCQ. We chose to group moderate to severe RA, as this aligns with treatment recommendations where MTX or MTX combination therapy are recommended [1, 2, 16]. We asked rheumatologists to consider the probability that that patient would have an ACR50 response within 3–6 months, with each of four treatments: oral MTX monotherapy, s.c. MTX monotherapy, MTX plus HCQ (dual therapy), or MTX plus HCQ and SSZ (triple therapy). For dual and triple therapy, rheumatologists were asked to first choose either oral or s.c. MTX (according to their usual practice) when conducting the exercises. Rheumatologists were asked to consider how they would typically use these treatments in practice, including their preferred dosing. We also stated that rheumatologists could co-prescribe CS and/or NSAIDs, again according to their typical practice.

Background information

Prior to the elicitation, participants received the same (scripted) background information. This included the purpose and motivation behind the project, a description of the task, a description of the patient, and the definition of an ACR50 response. In designing the survey, we were concerned that participants could consider the exercises as a quiz, which could influence their responses. We therefore included ‘cheap talk’ text [17], highlighting this potential bias, and clarifying that the goal of the study was to understand their opinions, and that there are many reasons why their beliefs may not match evidence summaries (see Supplementary Data S1, available at Rheumatology online).

Bayesian belief elicitation

The belief elicitation task for the four treatments proceeded in a step-wise fashion. Rheumatologists were first asked to rank the four treatments in their perceived effectiveness. These rankings were designed to provide an ordering for the task elicitation, as opposed to a set ranking. Participants could change their rankings at any time. The elicitation exercises then proceeded in the order of: 1 (least effective), 4 (most effective), 2 then 3. In the piloting, this was found helpful to provide ‘anchor’ points for the exercise.

The belief elicitation exercises were conducted using the ‘bin’ or ‘roulette’ method [15]. Participants were given 20 magnetic chips, each of which represented 5% of the total 100% probability. The participant was first asked to place a chip on what they believed to be the most likely estimate (probability of ACR50 response) for that treatment. They then placed chips at the upper and lower limits, which represented their highest and lowest plausible limits. They filled in the distribution with the remaining chips, such that the final histogram was a probability distribution that encompassed the range of their beliefs. Finally, the participant was asked to reflect on the shape of the distribution and make any changes he/she felt necessary.

The elicitation task was repeated for each of the four treatments. After each treatment, the research assistant drew an outline around the distribution in a coloured marker. This allowed the participant to view their responses to the other treatments when completing each subsequent elicitation exercise. Once the participant had completed all four elicitations, they were again asked to reflect on their distributions and make any changes as necessary.

Additional variables

Rheumatologists also completed an online survey, either before or after the interview. We collected variables to describe the participants, and additional questions related to their current treatment practices. The study was approved by the University of Calgary Conjoint Health Research Ethics Board (#REB16-0260). Signed informed consent was obtained from all participants.

Data analysis

Fitting of distributions

We fit parametric probability distributions for each of the four treatments for each participant. The analyses were conducted using R version 3.3.3, running package SHELF version 4.0 [18]. Using this package, we compared the sum of squared errors between each of six possible distributions (normal, Student’s t, gamma, log-normal, log Student’s t and beta), selecting the distribution with the best fit (which could be different for each of the four treatments).

Calculating relative treatment effects

For each participant, we calculated pairwise treatment effects [relative risks (RRs)] by sampling values from the fitted distributions. The treatment effect for dual and triple therapy was calculated relative to each participant’s chosen preferred route of administering MTX (oral or s.c.). The treatment effect for s.c. vs oral MTX was calculated separately.

It is likely that probability distributions for the four treatments are correlated within individuals. For example, if we were to tell an individual, after the task, that the true effect for MTX monotherapy is actually at the upper end of their distribution, they would also likely shift their distributions upwards for the other treatments. Capturing this correlation is challenging. It can be elicited from experts [19], but we found this to be cognitively challenging and not feasible in the context of this study. For the base case, we assumed moderate positive correlation (0.7) but conducted sensitivity analyses with a range of correlation values from low correlation (0.5) to strong positive correlation (0.9). We then sampled values from a multivariate distribution of the best fitting univariate distribution for each treatment across the range of correlation values. As our main comparison of interest was to MTX monotherapy, we also added an analysis where we set the effect of oral MTX as the elicited median value, instead of sampling from its probability distribution; this obviated the need to specify a multivariate distribution.

Some of the best fitting distributions did not confine the sampled values to between 0 and 100%. We removed these from the analyses. If the proportion of out-of-range samples was >1% of the total, we chose the next best fitting distribution. For each rheumatologist, we also calculated the median standard deviation across their four elicited distributions. We used this as a measure of the overall strength in their beliefs.

Average and subgroup distributions

We determined an average distribution by combining the distributions for the relative risks across all participants, reporting the median and 95% credible interval (CrI). We then divided participants into pessimistic and optimistic groups, according to the bottom and top quartiles for the effect of triple therapy relative to oral MTX, and for the effect of s.c. MTX relative to oral MTX.

Associations between characteristics and beliefs

We evaluated whether there was an association between rheumatologists’ characteristics and: (i) their belief regarding the effectiveness of triple therapy relative to MTX monotherapy; (ii) the average strength in their beliefs across the four treatments. These associations were evaluated using univariate regression models using the median effect for triple therapy relative to MTX or the strength in beliefs (median standard deviation across the distributions) as the dependent variables, and the following characteristics as the independent variables: gender, years in practice, full-time equivalent clinical workload, practice type (academic vs community/both), proportion of full-time equivalent in clinical practice. We considered a P-value of <0.05 to be statistically significant.

Impact of beliefs on future clinical trials and evidence syntheses

We used the distributions in two exploratory analyses. First, we simulated multiple clinical trials of triple therapy vs MTX with varying sample sizes (1:1 randomization) and calculated the expected RR given the elicited priors. We conducted these analyses for two possible hypotheses: (i) that triple therapy was truly more effective than MTX effective, with an ACR50 response rate of 60 vs 40% with MTX (RR 1.5), based on results from our network meta-analysis (NMA) [7, 8]; and (ii) that triple therapy was truly no more effective than MTX monotherapy (RR 1). These analyses were conducted separately for the overall group, and optimistic and pessimistic subgroups. To be consistent with our second exploratory analysis (described below), we placed the prior on the log-odds of the treatment effect, which was assumed to be normally distributed. The posterior distribution was then converted to the RR.

Next, we conducted an analysis to illustrate how elicited priors could also be used as informative priors in an evidence synthesis. In our prior NMA [7, 8], we had used uninformative priors, which assumes there is no knowledge of the treatments prior to the analyses. We repeated our analysis, instead using the elicited distributions relative to MTX monotherapy as informative priors on the log-odds of each treatment effect, and compared the results. In these analyses, we recognized that the elicited priors are not true priors, as they have likely been informed by the same clinical trials included in our NMA. They were therefore viewed as exploratory.

Results

Demographics of rheumatologists

Of the 71 rheumatologists who were approached and expressed interest, 40 ultimately completed the belief elicitation. Just over half were female (55%), and most rheumatologists (68%) had been in practice for >10 years (Table 1). The majority worked full-time, with similar numbers in academic and community practice. There was a range of experiences related to the number of patients seen each week and involvement in RA guidelines and clinical trials (Table 1).

Table 1.

Characteristics of rheumatologists

Characteristic	N (%)
Female	22 (55)
Years in practice
>20	19 (48)
11–20	8 (20)
5–10	9 (23)
<5	4 (10)
Type of practice
Academic	22 (55)
Community	8 (20)
Both	10 (25)
FTE
1.0	29 (73)
0.5–0.9	8 (20)
<0.5	3 (8)
Proportion of FTE in clinical work
>75%	14 (35)
50–75%	16 (40)
<50%	10 (25)
Patients with RA seen in a typical weeka
>25	14 (36)
11–25	16 (41)
5–10	9 (23)
Involved with RA guidelines	8 (20)
Involved with RA clinical trials	21 (53)
Patient recruitment	9 (23)
Trial design	8 (20)
Trial oversight	3 (8)
Other	3 (8)

^aMissing for one respondent. FTE: full-time equivalent.

Current stated treatment practices

Treatment practices varied considerably. In the scenario of a typical patient with moderate–severe early RA and no contra-indications to therapy, 13 rheumatologists stated they would never prescribe triple therapy and 10 would never prescribe MTX monotherapy (Table 2). When asked to rate the importance of various considerations when choosing an initial treatment from 1 (very unimportant) to 5 (very important), effectiveness had the highest mean rating (4.55, s.d. = 1.1), followed by tolerability (4.23, s.d. = 1.0), dosing/administration (3.98, s.d. = 1) and cost (2.98, s.d. = 1.2).

Table 2.

Stated practice patterns among rheumatologists for initial therapy in patients with moderate–severe RA

	Frequency of use, n (%)
	Never (0%)	1–25%	26–50%	51–75%	75–99%	Always (100%)
Treatment
MTX monotherapy	10 (25)	8 (20)	10 (25)	4 (10)	6 (15)	2 (5)
Dual therapy	3 (8)	12 (30)	8 (20)	5 (13)	9 (23)	3 (8)
Triple therapy	13 (33)	20 (50)	5 (13)	0 (0)	2 (5)	0 (0)
Route of MTX
Oral	3 (8)	9 (23)	9 (23)	7 (18)	6 (15)	6 (15)
S.c.	6 (15)	6 (15)	9 (23)	7 (18)	9 (23)	3 (8)

Relative treatment effects

Triple therapy, dual therapy and MTX monotherapy

Photographs of an example belief elicitation are presented in supplementary Fig. S1, available at Rheumatology online. Individual beliefs for the RR (median, 95% CrI) of triple therapy vs MTX monotherapy ranged from 0.53 (0.08–1.10) to 7.29 (2.79–38.06) (supplementary Table S1, available at Rheumatology online). When the treatment effects were pooled across all participants, both triple therapy and dual therapy were superior to MTX monotherapy, with the 95% CrI excluding the null effect (Table 3). Triple therapy was also statistically superior to dual therapy, although the effect was modest and the 95% CrI just narrowly excluded the null effect (RR 1.32; 1.03, 1.73). For the subgroup that was optimistic towards triple therapy, the relative risk for both triple therapy and dual therapy demonstrated a large beneficial effect (RR >3, Table 3). In the triple therapy pessimistic subgroup, the treatment effects between MTX monotherapy, dual therapy and triple therapy were all similar, including the null effect in all pairwise comparisons (Table 3).

Table 3.

Pairwise treatment effects from the elicited beliefs for triple therapy, dual therapy and MTX monotherapy

	Relative risk, median (95% credible interval)
	MTX monotherapy	Dual therapy	Triple therapy
Overall group
MTX monotherapy	1	–	–
Dual therapy	1.62 (1.13, 2.70)	1	–
Triple therapy	1.97 (1.35, 2.89)	1.32 (1.03, 1.73)	1
Triple therapy optimistic subgroup
MTX monotherapy	1	–	–
Dual therapy	3.51 (1.99, 8.36)	1	–
Triple therapy	4.03 (2.22, 10.12)	1.54 (1.31, 2.61)	1
Triple therapy pessimistic subgroup
MTX monotherapy	1	–	–
Dual therapy	1.20 (0.91, 1.67)	1	–
Triple therapy	1.29 (0.88, 1.73)	1.11 (0.83, 1.42)	1

Subcutaneous vs oral MTX

In the overall group, the 95% CrI for the pooled elicited treatment effect between s.c. MTX monotherapy and oral MTX monotherapy crossed the null effect (Table 4). However, similar to the analyses for triple therapy, the elicited beliefs for optimistic subgroup showed a sizable beneficial effect (RR 2.39), with the null value excluded from the 95% CrI (Table 4). Individual beliefs also showed similar variability as the effect of triple therapy vs MTX monotherapy (supplementary Table S1, available at Rheumatology online).

Table 4.

Treatment effects for s.c. vs oral MTX monotherapy from the elicited beliefs

	Relative risk median (95% credible interval)
Overall group	1.38 (0.95, 2.40)
Subcutaneous MTX optimistic subgroup	2.39 (1.61, 4.73)
Subcutaneous MTX pessimistic subgroup	1.14 (0.84, 1.55)

Sensitivity analyses

In sensitivity analyses, where we varied the strength of the multivariate correlation, the median effect was unchanged, but the credible interval narrowed somewhat as the correlation we specified increased (supplementary Tables S2 and S3, available at Rheumatology online).

Associations between characteristics and beliefs

In regression analyses, spending >75% of time doing clinical work was associated with a lower strength of belief (greater median standard deviation across the four distributions), although the confidence interval of this effect just narrowly excluded the null (Table 5). No other rheumatologist characteristic was associated with either the magnitude of the treatment effect for triple therapy relative to MTX monotherapy, or with the strength of belief (Table 5).

Table 5.

Association between rheumatologists’ characteristics and beliefs

Characteristic	Relative risk of triple therapy versus MTX monotherapy		Median standard deviation across the elicited distributions
	Estimate (95% CI)	P-value	Estimate (95% CI)	P-value
Female	0.47 (−0.54, 1.49)	0.35	1.35 (−0.46, 3.16)	0.14
Years in practice (per 10 years)	0.29 (−0.10, 0.70)	0.15	0.036 (−0.40, 1.10)	0.33
FTE (0–1)	0.46 (−0.67, 1.60)	0.41	0.14 (−1.94, 2.22)	0.89
Academic practice	0.45 (−0.56, 1.47)	0.37	−0.77 (−2.62, 1.08)	0.40
>75% of FTE in clinical practice	−0.21 (−1.28, 0.86)	0.69	1.90 (0.06, 3.74)	0.04

FTE: full-time equivalent.

Impact of beliefs on future clinical trials and evidence syntheses

The expected impact of a new clinical trial of triple therapy vs MTX monotherapy is shown in Fig. 1. If triple therapy is truly more effective than MTX monotherapy, an RCT with 100 patients would be expected to convince the pessimistic subgroup of benefit; that is, to provide sufficient data such that the lower bound of the 95% CrI for this pessimistic subgroup excludes the null effect (Fig. 1A). The overall group and the optimistic subgroup are already convinced of its benefit; further RCTs would only narrow the credible intervals. If the true treatment effect for triple therapy relative to MTX is 1 (i.e. they are the same), a non-inferiority trial of 420 patients would be required to convince the overall group of no benefit; that is, to provide sufficient data such that the upper bound of the 95% CrI for the overall group was below a modest effect (RR 1.5) (Fig. 1B). The optimistic group would require an RCT with 475 patients, and the pessimistic group would require a trial with 280 patients (Fig. 1B). Using the elicited priors as informative priors in the prior NMA resulted in little change to any of the pairwise treatment effects, suggesting the data from the NMA overwhelmed the prior information (supplementary Table S4, available at Rheumatology online).

Fig. 1.

Predicted posterior distributions of triple therapy vs MTX for trials of varying size

(A) Superiority RCT, assuming triple therapy is truly more effective than MTX (relative risk 1.5); and (B) Non-inferiority RCT, assuming triple therapy is not more effective than MTX (relative risk 1). Black dashed line in (B) reflects the chosen non-inferiority margin. RR: relative risk.

Discussion

Through a Bayesian belief elicitation, we found that, on average, rheumatologists in our study viewed triple therapy to be superior to MTX monotherapy, but similar to dual therapy, for which little RCT evidence exists. We also quantitatively demonstrated considerable variability in rheumatologists’ beliefs. A pessimistic group were unconvinced of any substantial added benefit of triple therapy, while others believed it to be highly effective. Similarly, there were optimists and pessimists regarding s.c. vs oral MTX. These differences may partially explain the variability in prescribing patterns seen in this study and reported in observational cohorts in early-stage RA [11, 12, 20]. By quantifying these beliefs, we demonstrated how, in a Bayesian context, we might expect them to shift in the face of new evidence.

The results of our sample size calculations provide a demonstration as to how the strength of a prior belief can impact the uptake of new evidence. Additionally, any new evidence also needs to be regarded as trustworthy by rheumatologists [21]. If a rheumatologist has strong beliefs regarding a treatment, they may be more likely to engage in confirmation bias [22]; that is, to judge new trial information that conflicts with strongly held beliefs as biased or not generalizable to their clinical practice or population. This is not to say that some of these critiques may not be without merit. RCTs in RA are often highly exploratory in nature, with strict protocols and procedures that do not reflect routine clinical practice [23, 24]. Understanding experts’ degree of scepticism prior to a trial may help inform its design, particularly for commonly used treatments where strong beliefs may have formed. Engaging these sceptics may help ensure that the trial is more pragmatic; that is, designed to reflect real-world clinical practice. It may also help inform the selection of the appropriate comparator(s). In our study, we found that experts’ beliefs regarding dual therapy were quite similar to triple therapy. Including dual therapy as a comparator would help provide evidence for this commonly used treatment, as there is currently little RCT evidence to support the used of dual therapy [7, 8].

Elicited beliefs have been used in the design of actual clinical trials [25–27]. Powering a trial to elicited priors may help ensure the trial is powered to provide meaningful results that shift clinicians’ or policy-makers’ existing beliefs. Another common advantage of using prior information, elicited from experts or derived from existing evidence, is a reduction in sample size. Although, as we demonstrated, if the experts are sceptical regarding the primary hypothesis of the trial, it may actually increase the sample size. A criticism of using prior information, of course, is that improperly specified priors may result in biased posterior estimates. Trialists therefore need to carefully consider the context of their trial, and whether (and which) prior information should be included [28]. Further, beliefs may be shaped by factors external to a trial (e.g. clinical experience, guidelines) that can change over time, so trialists would need to consider whether priors elicited before a trial would still hold after the trial is completed, but before results are available.

We believe this is the first study to quantify experts’ beliefs regarding RA treatments. Strengths include using a validated approach for belief elicitation, sampling a range of both community and academic rheumatologists, and having a sample size above a recommendation of at least 30 participants for a belief elicitation [26]. The demographics of our sample were similar to a recent workforce survey of Canadian rheumatologists for gender (55 vs 50% female), years in practice (48 vs 42% >20 years in practice) and practice type (58 vs 60% academic) [29]. In our exercises, we tried to allow a ‘pragmatic’ approach to the treatment scenarios, allowing flexibility in the preferred route of delivery of MTX and steroid and NSAID use. While we did not ask about patters of CS use in our study, CS use is common in patients with moderate–severe early RA in Canada. In a survey of Canadian rheumatologists, 90% reported using CS as bridging therapy in patients with moderate–severe RA [30]. Dosing of MTX in Canada is also typically quite high, with a large cohort study showing mean starting doses of 17 mg/week and 22 mg/week for oral and s.c. MTX, respectively [11].

A limitation is that our belief elicitation was only focussed on treatment benefits. We recognize that a hesitation to use triple therapy is also based on concerns over side effects, although rheumatologists must first be convinced of its benefit. When asked in our survey, treatment benefits were the most important consideration to rheumatologists, followed closely by side effects. This aligns with the preferences of patients with early RA, who tend to weigh treatment benefits higher than other considerations [31, 32]. Our belief elicitation focussed on ACR50 response, as we were interested in comparing the elicited distributions to the network meta-analysis, and using them as prior distributions. While this is not commonly used in clinical practice, it is a well-known outcome measure. We found rheumatologists had no difficulty in formulating their beliefs around this outcome in the elicitation. While we internationally designed the elicitations to be pragmatic and reflective of clinical practice, rheumatologists may have been conceptualizing the outcome in a more explanatory context as it would be measured in clinical trials. Finally, it is possible that some rheumatologists did not understand the elicitation exercise, although we provided a carefully worded script, and conducted all elicitations in person, to ensure understanding as much as possible.

In conclusion, our study demonstrates substantial variability in rheumatologists’ beliefs about the effectiveness of the most commonly used initial treatments for moderate–severe early RA. The strength of these beliefs may impact the uptake of a treatment, and therefore understanding these beliefs may be useful for informing the design of new clinical trials.