Article Text

Download PDFPDF

Concise report
Meta-analysis suggests that intensive non-biological combination therapy with step-down prednisolone (COBRA strategy) may also ‘disconnect’ disease activity and damage in rheumatoid arthritis
  1. Maarten Boers1,2,
  2. Lilian van Tuyl2,
  3. Marianne van den Broek3,
  4. Piet J Kostense1,
  5. Cornelia F Allaart3
  1. 1Department of Epidemiology & Biostatistics, VU University Medical Center, Amsterdam, The Netherlands
  2. 2Department of Rheumatology, VU University Medical Center, Amsterdam, The Netherlands
  3. 3Department of Rheumatology, University Medical Center Leiden, The Netherlands
  1. Correspondence to Professor Maarten Boers, Department of Epidemiology and Biostatistics, VU University Medical Center, PK 6Z 165, PO Box 7057, 1007 MB Amsterdam, The Netherlands; eb{at}


Background/objective Treatment of rheumatoid arthritis (RA) with tumour necrosis factor (TNF) antagonists changes the relationship between disease activity and progression of radiological joint damage (‘disconnect’): patients who have little or no response of disease activity still show reductions in damage progression. In early RA, the COBRA strategy (combination of methotrexate and sulfasalazine with step-down prednisolone) has been shown to be equivalent to high-dose methotrexate and infliximab in suppressing damage progression (BeSt trial). We investigated whether COBRA treatment can also ‘disconnect’ disease activity and damage.

Design A meta-analysis combined data from the COBRA trial (COBRA vs sulfasalazine monotherapy) with that of two arms of the BeSt trial (COBRA vs sequential monotherapy). Linear regression related 1-year progression of damage (Sharp van der Heijde score) as a dependent variable with disease activity (time-averaged Disease Activity Score in 44 joints (DAS44) or C-reactive protein (CRP)), treatment strategy (COBRA or control) and their interaction (indicator of a disconnect) as independent variables. The main outcome was the pooled interaction term.

Results Complete data from 60–100% of patients were available. Before pooling, disease activity was the only (strongly) significant independent factor related to damage progression. The pooled interaction term was (weakly) significant: time-averaged DAS44×treatment interaction, one-sided p=0.027; time-averaged CRP×treatment interaction, one-sided p=0.044.

Conclusions Changes in the relationship between disease activity and damage progression may not be limited to anti-TNF treatment, but a property of early, rapid and deep suppression of joint inflammation, also induced by conventional strategies that include glucocorticoids.

  • Rheumatoid Arthritis
  • Treatment
  • Disease Activity
  • Outcomes research
  • Corticosteroids

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.


Early and intensive treatment of rheumatoid arthritis (RA) has been found to dramatically reduce the impact of the disease, including the development of irreversible joint damage. It is well known that damage progression, as evidenced by radiography, is linked to disease activity, but in a highly individualised way.1 Some patients show progression, with little joint activity or even clinical remission, whereas others have no damage progression even in the presence of swollen joints.

With the introduction of biological agents, slowing, arrest or even reversal of radiographic progression has been shown for many of these agents. In addition, a claim has been made for infliximab, etanercept, adalimumab and, recently, tocilizumab that treatment with these agents uncouples the relationship between disease activity and damage, the so-called ‘disconnect’.2–6 In other words, damage progression is slowed or halted even in patients showing little clinical response—that is, maintaining moderate or even high levels of disease activity. In the case of infliximab, analysis of subgroups has methodological shortcomings3 ,7; to our knowledge, the adalimumab analysis has only been presented in abstract form; only the etanercept and tocilizumab analyses are fully published and can be considered robust enough to prove the disconnect.

Several non-biological strategies have shown effects similar to early biological treatment in countering disease activity and damage.8–12 In particular, the COBRA strategy of step-down prednisolone, methotrexate and sulfasalazine has proven to be equally efficacious to high-dose methotrexate and infliximab.10 ,13 In the original COBRA trial in patients with early RA, the strategy proved better than sulfasalazine alone for signs, symptoms and progression of damage. The clinical advantages were mostly lost in the second half year of the trial when prednisolone and methotrexate were discontinued, but the radiological benefits persisted, with evidence of permanent disease modification after 5 and 11 years.14 ,15

Therefore, in a post hoc meta-analysis of two early RA trials, we investigated whether COBRA therapy also uncouples the relationship between disease activity and damage.


A meta-analysis combined data of the COBRA trial (COBRA vs sulfasalazine monotherapy; original n=76 vs 79) with that of two arms of the BeSt trial (COBRA vs sequential monotherapy; original n=133 vs 126). We performed linear regression analyses with the 1-year progression of damage (Sharp van der Heijde score) as dependent variable (radiographs at baseline and 1 year). Independent variables were disease activity (measured every 3 months), treatment strategy (COBRA or control) and their interaction. We performed separate analyses with either 1-year time-averaged Disease Activity Score for 44 joints (ta-DAS44) or C-reactive protein (ta-CRP) as a measure of disease activity. CRP was chosen for analysis because this variable has a strong relationship with damage progression.1 Time averaging was carried out by calculating the average of all values in the interval, weighted for the length of each interval.

We set out to document a disconnect, in our case meaning that the relationship between disease activity and damage progression (ie, more activity results in more damage) is less prominent in COBRA patients than in monotherapy patients. For example, in monotherapy, a mean DAS44 over the year of 1 point higher could result in 5 Sharp units extra damage. If in COBRA patients the same DAS44 increase results in only 3 Sharp units extra damage, this would document the presence of a disconnect. In a regression model predicting damage progression that includes disease activity and treatment group/strategy as predictive factors, such a disconnect is suggested when the interaction between these two factors (disease activity and treatment strategy) becomes a significant predictive factor. Thus this interaction was the primary focus of our analyses.

We performed a standard meta-analysis by pooling the interaction terms of the model analyses in the two datasets: we calculated an overall mean interaction term by averaging the individual interaction terms weighted by their inverse variance. We used one-sided significance tests because of the directionality of the hypothesis and the relative insensitivity of regression for the detection of interaction.


In both source trials, at randomisation, groups were very similar with regard to their baseline characteristics: high disease activity, high CRP and, with short duration, very low baseline damage.8 ,13 For the present study, 60–100% of the original trial patients supplied complete data. For the original COBRA trial, radiographs and DAS44 data were available for all, and CRP data for 143, patients (71 monotherapy, 72 COBRA). The BeSt trial had more missing data, especially for radiographs (DAS44 dataset: monotherapy 112, COBRA 74; CRP dataset: monotherapy 102, COBRA 69). However, as documented in the original trial report,13 the patients with missing data were not found to differ much from the other patients, and missing data were distributed equally among the treatment groups. In both trials, COBRA treatment resulted in significantly lower disease activity and radiographic progression compared with monotherapy.8 ,13

When we analysed the trials separately, time-averaged DAS44 and CRP were the only (strongly) significant independent factors related to damage progression. Neither treatment group nor the interaction between treatment and disease activity entered as significant factors into the statistical model, although the individual regression coefficients for interaction pointed in the same direction.

Pooling of these coefficients resulted in a (weakly) significant test for interaction for both measures of disease activity (table 1). The numerical value of the interaction term expressed in the pooled regression coefficients (β values) appears clinically relevant, as illustrated by figure 1 where the effect is shown at clinically relevant cut-off points. Specifically, patients with moderate to high disease activity receiving COBRA therapy appear to have less damage progression than patients with similar activity levels receiving monotherapy. However, figure 1 should be interpreted with caution: power is lost when data are grouped, and some of the subgroups with a very low or very high mean DAS44 are very small, resulting in an unreliable estimate of the mean effect.

Table 1

Interaction between treatment group and disease activity, pooled over two trials investigating COBRA versus monotherapy

Figure 1

Illustration of the disconnect between disease activity and damage progression induced by COBRA therapy compared with monotherapy. The progression after 1 year (1-y) of treatment is compared in subgroups defined according to clinically relevant cut-offs for Disease Activity Score in 44 joints (DAS44) and C-reactive protein (CRP). In both trials (COBRA and BeSt), COBRA-treated patients (closed circles) showed little or no damage progression (SvdH, Sharp-van der Heijde total score) compared with patients treated with monotherapy (open circles). The left panels show that this difference between treatments is apparent in subgroups with a mean (time-averaged) DAS44 over the year at remission or minimal disease activity levels, but also in subgroups with mean disease activity at moderate to high levels. Likewise, the right panels show that the difference in progression between COBRA and monotherapy is present at low and intermediate, but also at high, levels of mean (time-averaged) CRP. To optimise the vertical scale, means and upper or lower half of the 95% CIs are depicted. This analysis is for illustration only and is suboptimal for analysis, as the use of clinically relevant cut-offs resulted in subgroups of unequal, and sometimes small, size. For example, the subgroup with a mean DAS44 below 1.6 contains only a few patients, resulting in large CIs.


Our meta-analysis of two trials suggests that the relationship between disease activity and damage progression may be altered by early aggressive treatment with the COBRA strategy. In other words, the ‘disconnect’ may not be limited to treatment with tumour necrosis factor and interleukin 6 antagonists. This makes biological sense, as prednisolone acts on many, if not most, of the mechanisms targeted by biological therapy. However, the effect was not as strong as that shown for the combination of etanercept and methotrexate or tocilizumab. Several reasons for this apparent discrepancy can be suggested. The first is that the original COBRA schedule is less powerful than biological therapy. This is probably intrinsic to the original COBRA dosing schedule: after the step-down phase, prednisolone is continued until 6 months and then mandatorily tapered and stopped in a few weeks. In addition, methotrexate is started at a low dose and either maintained at that dose (original COBRA) or increased only if disease activity remains high (BeSt); it is then stopped after 9 months (in BeSt only if DAS was 2.4 or less). In the two trials, the rules allowing reintroduction of either or both of the drugs differed in strictness, but, in both, most patients were at least temporarily receiving monotherapy, potentially reducing the efficacy compared with the continuous treatment with high-dose methotrexate and biological agents in their respective trials.2–6 The second reason is the lack of 6-month radiographs in the BeSt trial, which prevented us from pooling data at 6 months, where the effects of dosing reductions would not apply. In the original COBRA dataset, the interaction appeared to be stronger at 6 months than at 1 year (data not shown).

The third reason is related to weaknesses inherent in this post hoc analysis: meta-analysis of available data and lack of power for interaction in regression, compounded by the well-known skewness in the distribution of radiographic results. A further limitation of this study is that both trials studied patients with early RA, and there is limited damage progression in year 1, particularly in the BeSt study because of DAS=<2.4 targeted treatment.

In conclusion, this study suggests that the COBRA regimen should be added to the treatments that can slow radiographic progression even in patients with little response of clinical disease activity. It strengthens the case for timely (and cost-effective!) intensive combination therapy in early RA, but should not detract from proper disease management that includes tight control and treatment switches when disease activity persists.


View Abstract


  • Contributors MB and LvT contributed COBRA data; MvdB and CFA contributed BeSt data. PJK and MB performed the analyses. All authors contributed to the design of the study, the interpretation of the data, and the writing of the manuscript.

  • Funding None.

  • Competing interests MB has received support (consultancy fees) from Mundipharma, Horizon, Roche, GlaxoSmithKline, Novartis and UCB. CFA has received speaker's fees of less than US$5000 from UCB, Abbott and MSD. The BeSt study was partly sponsored by Schering-Plough and Centocor.

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • Data sharing statement The full COBRA dataset is available for interested researchers.