Epidemiology/Health Services
Abstract Submission
Cameron Speyer, BA
Research Trainee
Division of Rheumatology, Immunology and Allergy, Brigham and Women's Hospital
Daniel Li, MBS
Research Trainee
Division of Rheumatology, Immunology and Allergy, Brigham and Women's Hospital
Hongshu Guan, PhD
Statistician
Division of Rheumatology, Immunology and Allergy, Brigham and Women's Hospital
Kazuki Yoshida, MD, ScD
Instructor in Medicine
Division of Rheumatology, Immunology and Allergy, Brigham and Women's Hospital
April Jorge, MD
Instructor in Medicine
Division of Rheumatology, Allergy, and Immunology, Massachusetts General Hospital
Harvard Medical School
Candace Feldman, MD, ScD
Assistant Professor of Medicine
Division of Rheumatology, Immunology and Allergy, Brigham and Women's Hospital
Brigham and Women's Hospital/Harvard Medical School
Karen Costenbader, MD, MPH
Director, Lupus Program
Division of Rheumatology, Immunology and Allergy, Brigham and Women's Hospital
Harvard Medical Schoool
Background : SLE severity is heterogeneous: some patients have mild disease with rashes and arthritis, while others have severe multi-organ system involvement. It is challenging to study SLE in administrative databases given this heterogeneity. Garris et al developed an administrative claims-based SLE severity algorithm derived from elements of the SLEDAI, SLAM and BILAG instruments (Garris, J Med Econ 2013). It employs ICD-9, CPT and NDC claims over a 1-year period and classifies patients as having mild, moderate or severe disease. We sought to validate this administrative algorithm in comparison to SLEDAI scores at clinical visits.
Methods : We identified 100 SLE patients followed at the Brigham and Women’s Hospital (BWH) Lupus Center (2008-2010) with SLEDAI-2K (Gladman, J Rheumatol 2002) data at each visit over a 1-year period per person. We also obtained ICD-9, CPT and NDC codes for the Garris algorithm items (e.g. codes for glucocorticoids, ICD-9 codes for pericarditis) for the same year per subject. We compared Garris SLE severity to the highest SLEDAI-2K in that year. We defined the SLEDAI-2K categories of mild <3, moderate 3-6, and severe >6 as in the literature (Polachek, Arthritis Care Res 2017). We compared classification in binary categories of mild vs. moderate/severe and mild/moderate vs. severe. For each, we calculated sensitivity, specificity, and C-statistics.
Results : We analyzed 377 SLEDAI-2K assessments on 100 subjects (mean 3.77 [SD 2.63]) in the BWH Lupus Cohort. For the Garris vs. highest SLEDAI-2K model, 56 of 100 subjects were classified similarly by Garris and highest SLEDAI-2K (23/36 mild, 22/34 moderate, and 11/36 severe by SLEDAI-2K). The performance characteristics compared to the highest SLEDAI-2K of the year were: C-statistics were 0.755 for mild/moderate vs. severe SLE severity and 0.740 for mild vs. moderate/severe (Table). Sensitivity of the Garris algorithm compared to the highest SLEDAI-2K were 63.9% for mild vs. moderate/severe and 94.3% for mild/moderate vs. severe. Specificity was 82.8% for mild vs. moderate/severe, but 36.7% for mild/moderate vs. severe.
Conclusions : The Garris algorithm, developed for use in administrative datasets, has acceptable performance for classifying SLE severity when compared to the gold standard of highest SLEDAI-2K assessment in 1 year in a Lupus Center. It may be used to classify patients in administrative datasets according to their SLE severity over 1 year.