This review concluded that the specificity of the American college of rheumatology 1987 criteria in early rheumatoid arthritis was low, but sensitivity and specificity in established rheumatoid arthritis were higher. There was considerable heterogeneity among studies and details on the included studies were lacking and so these conclusions are unlikely to be reliable.

To evaluate the ability of ACR (American College of Rheumatology) criteria to diagnose rheumatoid arthritis compared with expert opinion according to disease duration.

MEDLINE via PubMed (1988 to December 2007) and EMBASE (1988 to April 2007) were searched. Search terms were reported and included a diagnostic filter. References of retrieved studies and abstracts from relevant international congresses between 2005 and 2007 were screened. The review was restricted to studies published in English, French and Spanish.

Studies that evaluated accuracy of the 1987 ACR criteria in adults compared to the reference standard of expert opinion and that reported sufficient data to allow calculation of sensitivity and specificity were eligible for inclusion.

Cases in the studies had rheumatoid arthritis that ranged in duration of from less than three months to 16 years. Controls had osteoarthritis, systemic autoimmune disease, undifferentiated arthritis, spondyloarthritis, other diagnoses such as fibromyalgia, septic or viral arthritis and crystal-induced arthropathies.

The authors stated neither how the papers were selected for the review nor how many reviewers performed the selection.

The authors did not state that they assessed study quality.

One reviewer extracted data as 2x2 tables of test performance.

Data were pooled using a summary receiver operating characteristic (SROC) approach. Q*, the point of equal sensitivity and specificity, was estimated. Sensitivity and specificity were pooled using the arcsin of transformed proportions. Heterogeneity was assessed using the Q statistic. A random-effects model was used where heterogeneity was present. Studies were stratified according to duration of rheumatoid arthritis (<1 years versus >1 year) and whether ACR criteria were used in list or tree format. A regression model was used to assess the effect of disease duration on accuracy. Publication bias was assessed with a funnel plot.

Nineteen studies were included in the review (n=7,438).

Early rheumatoid arthritis: Sensitivity/specificity of the ACR criteria in list format (six studies) was 77% (95% CI 68% to 84%) for both and in tree format (two studies) sensitivity was 80% (95% CI 72% to 88%) and specificity 33% (95% CI 24% to 43%).

Established rheumatoid arthritis: Sensitivity/specificity of the ACR criteria in list format (15 studies) was 79% (95% CI 71% to 85%) for sensitivity and 90% (84% to 94%) for specificity and in tree format (six studies) was 80% (95% CI 71% to 85%) for sensitivity and 93% (95% CI 86% to 97%) for specificity.

There was significant heterogeneity (p<0.001) for all analyses with the exception of early rheumatoid arthritis in tree format which was only assessed in two studies.

Based on the regression model, diagnostic performance was better for establish than for early disease (p=0.003).

The funnel plot suggested some evidence of publication bias.

Specificity of ACR 1987 criteria in early rheumatoid arthritis was low and these criteria should not be used as diagnostic tools. Sensitivity and specificity in established rheumatoid arthritis were higher, which reflected their use as classification criteria gold standard.

The review addressed a focused question supported by clearly defined inclusion criteria. The literature search involved two relevant databases, but use of a diagnostic filter and restriction of the review to studies published in certain languages meant that relevant studies may have been missed. It was unclear whether appropriate steps were taken to minimise bias and errors in the selection of studies; no such steps were taken for data extraction. Study quality was not assessed and details on individual studies were lacking, especially in relation to study design; therefore, validity and generalisability of the findings were unclear. Methods used to pool studies were appropriate and results were clearly presented with the aid of graphical displays. However, there was considerably heterogeneity in pooled estimates and so these should be interpreted with extreme caution; further investigation of differences between studies may have helped determine the true accuracy of these criteria. The methods used to investigate publication bias were not appropriate for diagnostic accuracy data and were unlikely to produce reliable results. The review councluded that ACR criteria had greater overall accuracy in established than with early rheumatoid arthritis; however, the summary estimates of sensitivity were similar for early and established rheumatoid arthritis. The authors' conclusions were based on very heterogeneous pooled data from studies of unknown validity, which combined with a lack of details on the included studies meant that the authors' conclusions are unlikely to be reliable.

Practice: The authors stated 1987 ACR criteria should not be used as diagnosis tools in early rheumatoid arthritis. The authors also suggested that ACR criteria may be better in tree format than in list format.

Research: The authors stated that a cohort of patients with early arthritis should be followed up for at least two years to determine the final diagnosis to allow the diagnostic value of each clinical, radiological, biological and exclusion criterion to be assessed.

Not stated.

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.