We combined the EULAR/ACR derivation and validation cohorts for a total of 1197 systemic lupus erythematosus (SLE) and n=1074 non-SLE patients with a variety of conditions mimicking SLE, such as other autoimmune diseases, and calculated the sensitivity and specificity for antinuclear antibodies (ANA) and the 23 specific criteria items. We also tested performance omitting the EULAR/ACR criteria attribution rule, which defines that items are only counted if not more likely explained by a cause other than SLE.
Positive ANA, the new entry criterion, was 99.5% sensitive, but only 19.4% specific, against a non-SLE population that included other inflammatory rheumatic, infectious, malignant and metabolic diseases. The specific criteria items were highly variable in sensitivity (from 0.42% for delirium and 1.84% for psychosis to 75.6% for antibodies to double-stranded DNA), but their specificity was uniformly high, with low C3 or C4 (83.0%) and leucopenia <4.000/mm³ (83.8%) at the lowest end. Unexplained fever was 95.3% specific in this cohort. Applying the attribution rule improved specificity, particularly for joint involvement.
Changing the position of the highly sensitive, non-specific ANA to an entry criterion and the attribution rule resulted in a specificity of >80% for all items, explaining the higher overall specificity of the criteria set.