This review used a mixed treatment comparison meta-analysis to compare treatments for rheumatoid arthritis. The authors concluded that treatment with tumour necrosis factor alpha (TNFα) antagonists is more effective than anakinra, and that there is no difference between TNFα antagonists. The authors' conclusions appear reliable, but their results should be treated as exploratory in the absence of direct trial evidence.

To compare the efficacy of cytokine antagonists in patients with rheumatoid arthritis (RA) using statistical techniques for indirect comparisons.

MEDLINE, EMBASE, the Cochrane Library, DARE and the Scientific Citation Index were searched from 1990 to January 2005; the search terms were reported. In addition, proceedings from the American College of Rheumatology and European Congress of Rheumatology meetings (2001 to 2004) and the reference lists from identified reports were screened. Submissions to the Food and Drug Administration were searched for more information on trial results.

Randomised controlled trials (RCTs) that compared cytokine antagonists with placebo or methotrexate (MTX) in patients with a clinical diagnosis of RA were eligible for inclusion. Studies had to have a minimum follow-up period of 6 months and provide sufficient data to enable the calculation of odds ratios (ORs) of an ACR20 or ACR50 response.

The primary review outcome was the American College of Rheumatology (ACR) response criteria (ACR20 or ACR50). ACR20 was defined as a reduction of 20% or more in the number of swollen or tender joints plus improvement in at least three of five specified outcome measures; ACR50 was defined as a 50% or more improvement.

The included studies evaluated three tumour necrosis factor alpha (TNFα) antagonists (etanercept, infliximab and adalimumab) and one interleukin 1 (IL-1) antagonist (anakinra). Most of the studies compared several different doses of the agents, either alone or in combination with MTX, with placebo or MTX (treatment regimens were reported in the tables). The studies included patients with varying characteristics at baseline: the mean disease duration ranged from 1 to 13 years, the mean age per treatment arm ranged from 23 to 57 years, the mean number of previous disease-modifying antirheumatic drugs ranged from 0 to 4, and the mean Health Assessment Questionnaire (HAQ) disability score ranged from 1.3 to 1.9. The included studies reported outcomes at 6 and/or 12 months.

Two reviewers independently selected the studies and resolved any disagreements through discussion.

The authors did not state that they assessed validity.

Where required, ACR responses were estimated from graphs. Where possible, for each study, the percentage of patients with an ACR20 or ACR50 response was reported for each treatment arm at 6 and 12 months and ORs were calculated.

Two reviewers independently extracted the data on an intention-to-treat basis.

The data were pooled using a mixed treatment comparison meta-analysis, which uses evidence from all treatment comparisons and not just direct comparisons of treatment with the same comparator. Two types of models were used: a random-effects analysis, which accounted for the fact that there were multiple treatment arms within some studies; and a validation model, which assumed each treatment arm was independent of the study and that MTX treatment effects were the same for each arm (this was used as a sensitivity analysis).

Meta-regression analyses were used to explore heterogeneity between the studies. Random-effects models were used to assess the effects of disease duration and baseline HAQ disability on the log OR of disease response. Estimated ORs for each treatment alone, and in combination, compared with placebo after adjustment for disease duration and baseline HAQ were presented. Separate analyses were performed for ACR20 and ACR50 responses, and models were fitted using WinBUGS software.

Thirteen RCTs (n=6,694) were included.

All four treatments were associated with increased odds of an ACR20 and an ACR50 response compared with placebo. Differences were statistically significant apart from the effects of anakinra on ACR20; the ORs ranged from 1.70 to 4.21. The efficacy of etanercept, infliximab and adalimumab were similar. The combination of MTX and a TNFα antagonist was more efficacious than each treatment alone.

The results of the mixed treatment comparison model, which compared the treatments against each other, showed no evidence of any differences between most of the different TNFα antagonists. Only comparisons involving anakinra were statistically significant (p<0.05): anakinra was more effective than placebo for ACR20 (OR 2.134, 95% CI: 1.269, 4.219), while TNF antagonists were more effective than anakinra for ACR20 (OR 1.956, 95% CI: 1.039, 4.001) and ACR50 (OR 1.928, 95% CI: 1.057, 3.493).

The mean disease duration and mean disability index at baseline accounted for 67% and 91%, respectively, of the heterogeneity in the log ORs of ACR20 and ACR50 responses. An increase in disease duration was associated with an increase in the odds of a positive ACR50 response at 6 months, but the odds of response were decreased for patients with greater disability at baseline.

Sensitivity analyses using the validation model showed similar results.

For patients with RA, treatment with TNFα antagonists is more effective than the IL-1 antagonist anakinra and anakinra is better than placebo. There is no difference between TNFα antagonists. Adding MTX to all drug treatments improves response rates.

The review question was stated clearly. Several relevant sources were searched and attempts were made to minimise publication bias; no attempts to minimise language bias were reported Appropriate methods were used to minimise reviewer error and bias during the review process. Only RCTs were included but validity was not assessed. Potential factors accounting for differences between the studies were examined, but details of all the characteristics explored were not listed and only the two most influential factors were included in the analyses. Complex statistical models were used to allow indirect comparisons to be made for the different treatments. Conclusions about the relative effect of different treatments were based on comparisons formed from all the data, rather than from direct comparisons within trials. Plots showing the results from the original trials and the predicted results from the model showed that the model predictions seemed reliable; a sensitivity analysis showed similar results. The authors discussed the limitations of their analysis, particularly those involving assumptions made in the meta-regression. The authors' conclusions appear reliable, but the use of mixed treatment comparisons and meta-regression mean that they should be treated as exploratory in the absence of trials directly comparing these drugs.

Practice: The authors stated that, for policy makers, clinicians and patients, review results suggest that etanercept, infliximab and adalimumab are equally effective for ACR response.

Research: The authors stated that further evidence synthesis and uncertainty analysis is needed to take account of radiographic progression, adverse events, quality of life and costs. The results of this current analysis can also be updated as new trial data are published.

Not externally funded.

This is a critical abstract of a systematic review that meets the criteria for inclusion on DARE. Each critical abstract contains a brief summary of the review methods, results and conclusions followed by a detailed critical assessment on the reliability of the review and the conclusions drawn.