Background EULAR and ACR are supporting multi-phase development of SLE classification criteria based on weighted criteria and a continuous probability scale. Prior steps included criteria generation, criteria reduction through Delphi and Nominal Group Technique exercises, literature review for sensitivity/specificity of candidate criteria, and organization of candidate criteria into seven clinical and three immunologic domains.
Objectives To refine definitions of candidate criteria, determine relative weights using multicriteria decision analysis, and determine a threshold score for SLE classification.
Methods An SLE Expert Panel (9 North American, 8 European) submitted 167 unique cases with a range of SLE probability. Experts scored 20 representative cases using the candidate criteria and rank-ordered them. In a 2-day meeting, experts reviewed inter-rater reliability of scoring, refined criteria definitions, and participated in a multicriteria decision analysis (MCDA) exercise using 1000MindsTM software. Experts were presented a series of decisions between two cases, each with different criteria from two domains (e.g. oral ulcers [cutaneous] and acute pericarditis [serositis] vs. alopecia [cutaneous] and pleural effusion [serositis]) and anonymously voted for the case more likely to be classified as SLE. Votes were discussed until consensus was reached for each decision. Using the consensus decisions, 1000Minds™ calculated criteria weights, assigned a total score to each of remaining 147 cases and rank-ordered the cases. Experts voted on whether each case should be classified as SLE. MCDA was repeated for criteria whose calculated weights were inconsistent with expert opinion until group consensus was achieved. 1000MindsTM then re-calculated criteria weights and re-ranked cases once. The score of the last case for which expert consensus was achieved was the threshold score.
Results Inter-rater reliability was good; human data entry error, not following instructions, and differing interpretations of criteria definitions accounted for discrepancies. Arthritis and pericarditis definitions were modified through group discussion. The MCDA involved 74 pairwise decisions. Cranial neuropathy and Class VI lupus nephritis were removed as they added little to SLE classification. MCDA was repeated for the arthritis and cutaneous domains as initial weights did not match expert opinion. After criteria weights and scores were re-calculated once, experts reached consensus for SLE classification for case score >83.
Conclusions Using an iterative process, the expert panel refined definitions, weighted candidate criteria and determined a threshold score of >83 for SLE classification, which will undergo validation.
Acknowledgements Joint support from EULAR and ACR
Disclosure of Interest None declared