POS0734 EXTRAPOLATION OF LONG-TERM OUTCOMES IN SYSTEMIC LUPUS ERYTHEMATOSUS: REPLICATING A HOPKINS LUPUS COHORT ANALYSIS WITH THE SYSTEMIC LUPUS INTERNATIONAL COLLABORATING CLINICS (SLICC) INCEPTION COHORT

Abstract

Background:A disease model of systemic lupus erythematosus (SLE) that predicts short-term outcomes (disease activity and prednisone use) and links them to long-term outcomes (accrual of organ damage and mortality) was previously developed in a single center SLE cohort (Johns Hopkins [JH]) to support health economic analyses (Watson 2015), which has not been comprehensively replicated in other cohorts or contexts.Objectives:As part of an effort to develop and refine this existing disease model, the aim of this study was to replicate the previously estimated network of risk equations for short- and long-term outcomes in the SLICC Inception Cohort, an international cohort of patients (33 centers,11 countries).Methods:The SLICC Inception Cohort enrolled patients fulfilling ACR Classification Criteria for SLE within 15 months of diagnosis from 1999-2011 with annual follow-up through April 2020. The network of risk equations included two linear random effects models to predict (1) change in annual average Systemic Lupus Disease Activity Index (SLEDAI) score based on patient characteristics and the presence of renal, hematological, and immunological involvement in the prior year and (2) average annual prednisone dose based on SLEDAI score in the same year. These equations were then linked to parametric survival models that predicted time to the occurrence of organ damage (system-specific based on the ACR/SLICC Damage Index) and mortality. We compared model performance between the SLICC Cohort and the original analysis from the JH Cohort.Results:In comparison to the JH cohort (N=1354), the SLICC cohort (N=1697) had a smaller fraction of patients of African descent (39% vs 17%) and shorter disease duration at entry (4.8 vs 0.5 years). In the first equation predicting change in annual SLEDAI score, predictors were generally aligned with the same direction and significance, with the exception of renal involvement in the prior period, which had a positive association with change in SLEDAI in the SLICC cohort but was negatively associated in the JH cohort (Table 1). The second equation predicting prednisone dose was also consistent with the original analysis showing a significant positive association between higher disease activity and prednisone use. In all of the parametric survival analyses (individual organ damage and mortality models), coefficients were generally in the same direction and magnitude, though some were no longer significant in the SLICC cohort.Conclusion:The relationships identified in the original analysis were broadly replicated in the SLICC Inception Cohort. Observed differences may reflect differences in the patient populations, structure of the two cohorts (prevalent vs inception), and frequency of visits (quarterly visits in the JH cohort vs annual visits with the SLICC cohort may more closely capture a decrease in SLEDAI associated with treatment specifically related to renal involvement). Additional analyses relaxing the requirement to completely align with the original structure are underway to further assess the predictive accuracy of these models.