Abstract
Background In previous validation work, the SLICC 2012 SLE classification criteria were more sensitive than the revised ACR-11 criteria, while both criteria had similar agreement with physician diagnoses. Both of these classification rules count each SLE manifestation equally. Our objective was to derive and test a classification rule which differentially weights the variable used in the SLICC classification rule. We also compared this rule to a recently proposed EULAR/ACR classification rule that also uses a weighted approach. [Costenbader KH, Johnson S, Aringer M. EULAR/ACR Classification Criteria Update for SLE. Presented at the 2017 ACR/ARHP Annual Meeting, San Diego CA, November 4–8, 2017].
Methods The physician-rated patient scenarios used to develop the 2012 SLICC classification criteria were re-employed to devise a weighted classification rule. A multiple linear regression model was constructed with the 2012 SLICC criteria variables as predictors and the binary outcome (physician classification of SLE) as the outcome. Weights for each criteria were generated by multiplying each criteria’s coefficient by 100 and rounding to the nearest integer. The ‘Direct Coombs’ criteria was deleted for simplicity. Weights for remaining manifestations were: acute cutaneous (26), chronic cutaneous (12), oral ulcers (16), arthritis (9), serositis (16), renal without biopsy (9), neurologic (9), hemolytic anemia (1), leukopenia or lymphopenia (14), thrombocytopenia (15), alopecia (9), ANA (17), anti-dsDNA (19), anti-Sm (16), antiphospholipid antibodies (8), low complement (11). Classification cutoff was the score that maximized overall agreement (i.e., the sum of sensitivity and specificity) of the new weighted criteria with physician diagnosis. Patients with lupus nephritis or the new weighted classification rule of 56 or more with at least one clinical component and one immunologic component were classified as SLE. We evaluated the performance of this revised SLICC criteria on an independent set of patient scenarios and compared this to the performance of the older revised ACR criteria, the previous SLICC 2012 criteria, and the newly proposed EULAR/ACR criteria.
Results Table 1 shows the performance of the four classification rules. There was no statistically significant difference between any pair of rules with respect to overall agreement with the physician diagnosis.
Conclusions The two newly derived weighted classification rules did not perform better than the existing list-based rules in terms of over-all agreement. Since the list-based rules are easy to calculate, they may be preferred in most clinical settings.
Acknowledgements Presented on behalf of SLICC. The Hopkins Lupus Cohort is funded by NIH AR 43727 and NIH AR 69572.
