994. A Novel Informatics Tool to Detect Antibiotic Allergies in Patients Undergoing CIED Procedures

Abstract

Abstract Background Antibiotics are one of the leading causes of emergency room visits for adverse drug events, yet surveillance for antimicrobial allergy adverse events is limited and identifying true cases is challenging. As part of a larger study to improve antimicrobial use, we sought to develop and validate a tool for near real-time measurement of antimicrobial allergy adverse events. Methods An existing cohort of patients undergoing cardiac device procedures with known antimicrobial exposure was split into a development and validation set. Candidate triggers for identifying allergic reactionswere identified a priori, using disease phenotype codes “phecodes”, allergy documentation on allergy module of the electronic medical record (EMR), and keyword searches applied to clinical notes (e.g., “anaphylaxis,” “rash”), medication administration (e.g, corticosteroids alone or with antihistamines) and administrative codes (ICD-10 codes and phecodes). Cases were reviewed for presence of a true event, and the tool was iteratively updated based on chart review findings. The tool was then applied to the validation cohort and a sample of trigger-flagged and unflagged cases underwent manual review. Data were analyzed in SAS and model triggers were selected using a LASSO technique. Results Among 34,703 patients, N=431 cases underwent manual review (350 development; 120 validation), and 104 true allergy adverse events were identified. Among chart reviewed cases, the most frequently detected flags were keywords in unstructured clinical notes (35%), phecodes (26%), corticosteroid administration (15%), observed allergy documentation in EMR (14%) and reported allergy documentation in EMR (13%). The final model contained 7 triggers and had an AUC of 0.95, and a positive predictive value of 67% (Figure). The strongest predictors of true adverse events were the allergy health factors (aOR 358, 95% CI 76.3-999) and specific Phecodes (Table1). Conclusion We developed an antibiotic allergy measurement tool using structured and unstructured data that can be applied to detect antimicrobial adverse events in near-real time. This model may be applied to provide near real-time feedback to clinicians about antimicrobial allergy adverse events and may be useful for antimicrobial stewardship programs. Disclosures Westyn Branch-Elliman, MD, MMSc, DLA Piper,LLC/Medtronic: Advisor/Consultant|Gilead Pharmaceuticals: Grant/Research Support.