Background/Purpose: Systemic Lupus Erythematosus (SLE) is an autoimmune disease that can affect many parts of the body including skin, lungs, brain, heart, kidneys, joints, and blood vessels. How lupus looks in one patient is different than in another patient. Because of this, it is hard to diagnose a patient as having SLE. EHRs are widely used in healthcare settings and are a rich source of information about patients that can be mined for earlier diagnosis identification.

Methods: We identified 513 patients we knew had SLE from the Chicago Lupus Database (CLD) in the Northwestern Medicine Electronic Data Warehouse (NMEDW). We built an algorithm of SLICC classification criteria using ICD9/10 and labs to see if we are finding the same SLICC classification criteria that are in the CLD. According to SLICC classification rules, to have definite lupus you need at least 1 clinical criteria, at least 1 immunologic criteria, and a total of 4 or more criteria.

Results: As shown in Table 1, of 513 patients with SLE in the CLD, we detected the following SLICC classification criteria correctly in the NMEDW: clinical- chronic cutaneous 97%; acute cutaneous 98%; renal 65%; serositis 52%; arthritis 34%; neuro 29%; ulcers 16%; alopecia 3%; and labs – thrombocytopenia 99%; dsDNA 89%; hemolytic anemia 80%; complement 74%; leukopenia/ lymphopenia 73%; Anti-Sm Antibody 72%;  Antiphospholipid Antibodies 64%; Antinuclear Antibody 60%; Coombs 17%; Of the 513 patients with SLE in the CLD, all had at least 1 clinical criteria, 469 had at least 1 immunologic criteria, and 498 had 4 or more criteria. Using EHR data from the NMEDW and rules for the SLICC classification criteria, we categorized 467/513 (91%) patients as having definite lupus correctly.

Conclusion: Based the results, we are able to capture over 90% of patients in the NMEDW correctly. Future work includes implementing NLP on criteria like alopecia, oral ulcer, arthritis, and renal biopsy to improve identification of individual criteria in EHR data that ICD9/10 and labs missed. Once we develop an algorithm that is able to capture the same criteria in the NMEDW that are in the CLD, we can use that algorithm to identify patients that might have a missed diagnosis of lupus. If we are able to find patients who have lupus earlier in their disease progression, we can improve their quality of care and treat earlier with the goal of minimizing disease damage.

Table 1. SLICC Classification Criteria Identified in CLD and NMEDW