Objective: To develop Systemic Lupus International Collaborating Clinics/American College of Rheumatology Disease Damage Index (SLICC/ACR-DI) item weightings that improve the ability of the measure to predict patient mortality in systemic lupus erythematosus (SLE).
Methods: Disease damage was measured for 738 patients followed at the University of Toronto Lupus Clinic since diagnosis. Using Rasch analysis, item weightings were determined and tested for their ability to predict death in a logistic regression model. Receiver operating characteristic (ROC) curves were produced to compare the original and weighted scales' ability to discriminate patients that died during the followup period from those who remained alive.
Results: The average SLICC/ACR-DI score per patient was 1.66. In total, 138 of the patients died during a mean followup of 9.2 years. A Rasch analysis derived weighting scheme using weighted domain scores (SLICC/ACR-DI-weighted) was the best weighted scale, with item reliability = 94%, model mean square infit = 1.01 (STD = 0.05); model mean square outfit = 0.99 (STD = 0.3), separation 4.08. The SLICC/ACR-DI-weighted was modestly better than the SLICC/ACR-DI in discriminating patients who died from those who remained alive. Using standardized scores for comparability, the SLICC/ACR-DI-weighted was better in predicting patient death than the unweighted SLICC/ACR-DI [OR(death)(SLICC/ACR-DI-weighted) = 1.7 vs OR(death)(SLICC-ACR-DI) = 1.4; p < 0.005]. ROC curve analysis supports that the SLICC/ACR-DI-weighted was somewhat superior to the SLICC/ACR-DI for predicting mortality.
Conclusion: In this test set, the SLICC/ACR-DI-weighted was modestly better in predicting death than the traditional unweighted SLICC/ACR-DI. However, the SLICC/ACR-DI-weighted is more difficult to apply and the weightings appear not to have provided a clinically relevant improvement of the SLICC-ACR-DI.