1657. Real-life implementation of a machine learning based algorithm for urinary tract infection antibiotic resistance management in the community setting

Abstract Background At outpatient visits, one of the most common reasons for antibiotics administration is Urinary Tract Infection (UTI), where selection of the appropriate empiric antibiotic requires consideration of multiple factors. The aim of this study was to assess the implementation and effectiveness of a machine learning based model for UTI antibiotic resistance, embedded in the physician's medical encounter workflow. Methods A longitudinal, nationwide prospective study, conducted by Maccabi Healthcare Services, a health organization in Israel serving 2.5 million members, with about 120,000 medical encounters with a UTI diagnosis per year. We adopted a machine-learning algorithm to predict antibiotic susceptibility using personal electronic health record data. Then, a designated order set was incorporated within our electronic health record interface, named 'UTI Smart-set' (UTIS). When a UTI diagnosis code is selected by the physician, a pop up screen emerges with the recommended antibiotics. The recommendation is based on the personal predicted susceptibility model developed, followed by the optimal narrow-spectrum antibiotic as recommended by the organizational clinical guidelines. . We analyzed the antibiotic prescriptions that were recommended by UTIS, the actual prescriptions given, and compared them to the culture results and the pathogen antibiotic resistance profile. Results From July 1st 2021 to April 1st 2022, among 73,148 medical encounters with a UTI diagnosis, 34,659 antibiotic prescriptions were provided. In 21,924 (63.2%) encounters the UTIS recommendation was accepted whereas in 12,736 (36.7%) it was declined. Use of the UTIS recommendation resulted in a 31.8% reduction of the measured resistance to antibiotics prescribed to patients, 10.9% (2,224) of cases who followed the UTIS recommendation for antibiotics versus 16.0% (1,879) case who did not. A 47% cumulative improvement in adherence to clinical guidelines was also observed using the tool. Conclusion This unique, point of care, antibiotic stewardship tool, based on a combination of clinical guidelines and personalized medicine, led to a major improvement in the selection of appropriate antibiotics. Disclosures Shirley Shapiro Ben David, MD, pfizer: Grant/Research Support.