Abstract
Background SLE is difficult to diagnose given diverse manifestations that occur over time and across care sites. Electronic health records (EHR) are now used in a majority of health care settings throughout the country, and present a rich source of information about patients which can be mined for earlier diagnosis identification, to improve quality of care, or enable clinical studies. To identify SLE patients in EHR data, we developed a rules-based algorithm based on the SLICC classification criteria and compared against a gold standard SLE patient registry data set. Methods We identified 513 patients in the Chicago Lupus Database (CLD) fulfilling 4 or more of the ACR classification criteria for SLE who also had records in the Northwestern Medicine Electronic Data Warehouse (NMEDW). ICD-9/10 codes were used to identify clinical SLICC SLE classification criteria items. Laboratory results were identified using lab test names in combination with threshold numeric values in order to determine whether patients met the SLICC lab test classification criteria requirements. Results As shown in table 1, of 513 patients with SLE in the CLD, we detected the following SLICC classification criteria, in the NMEDW: clinical- chronic cutaneous 97%; acute cutaneous 98%; renal 65%; serositis 52%; arthritis 34%; neuro 29%; ulcers 16%; alopecia 1%; and labs- dsDNA 89%; hemolytic anemia 80%; complement 74%; leukopenia/lymphopenia 73%; APL 64%; ANA 52%; thrombocytopenia 22%; Coombs 17%; Sm 0%. Of 513 patients with SLE in the CLD based on ACR criteria, 513 had at least 1 clinical criteria, 466 had at least 1 immunologic criteria, and 471 had 4 or more criteria. Using EHR data from the NMEDW, and rules for the SLICC classification criteria that were based on ICD9/10 codes and labs and required identification of at least one clinical and one immunological criteria, we categorized 450/513 (88%) patients as having definite lupus. Conclusions Query of patient EHR data with ICD-9/10 codes and lab tests for specific SLICC classification criteria items requires refinement to improve identification of some criteria. Using the SLICC classification rule for definite SLE, we were able to identify 88% of those with definite SLE by the ACR criteria, using ICD9/10 codes and labs. Text searching of notes (by simple string matching or natural language processing) may improve identification of individual SLICC criteria (e.g. renal biopsy) and may be critical for mining physicians’ notes for criteria that are not well documented with diagnosis codes or lab results. Acknowledgements Funding-NIH/NIAMS R21AR074081.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have