186. Risk Factors for Extended Spectrum β-Lactamase Bacteremia and External Application of a Clinical Prediction Tool

BACKGROUND: Extended spectrum β-lactamase (ESBL) bacteria are resistant to many antibiotics, which increases the risk of inadequate early antibiotic therapy. A previous single-center study had created a prediction tool to assist clinicians in identifying patients at risk for ESBL bloodstream infections. The purpose of our research project was to assess validity of this tool while also identifying risk factors for ESBL bacteremia within our own institution, which would allow for assessment of alternative prediction tools. METHODS: We performed a retrospective chart review of adult patients admitted to an urban university hospital who were found to have bacteremia with Escherichia coli, Klebsiella pneumoniae, and/or Klebsiella oxytoca between October 2016 and April 2018. Demographics and comorbidities were assessed, along with other potential risk factors including exposure to antibiotics and hospitalizations within the past 6 months. RESULTS: A total of 214 instances of bacteremia were identified and 14% were due to ESBL organisms. Risk factors for ESBL bacteremia in our cohort included history of positive culture for ESBL (RR = 5.9) or MRSA (RR = 3.5) and antibiotic usage in the past 6 months (RR = 2.3). Patients with ESBL bacteremia were hospitalized longer (mean 16 days vs. 6 days for non-ESBL), received longer durations of antibiotic therapy (11.7 days vs. 5.3 days), and were exposed to greater numbers of different antibiotics (1.9 vs. 0.7) in the previous 6 months. Multivariate logistic regression showed that history of prior ESBL infection (OR 14.7, CI 1.8–120) and increasing number of different antibiotic classes administered in the prior 6 months (OR 4.3, CI 1.7–11.2) were significant risk factors for ESBL bacteremia. The previously created prediction tool did not sufficiently differentiate higher and lower risk for ESBL bacteremia in our cohort. CONCLUSION: Although risk factors were similar, the previously derived stepwise prediction tool did not predict ESBL bacteremia in our external cohort. Point-based prediction modeling might better assess risk across institutions. Additionally, the number of different antibiotics received was associated with risk for ESBL bacteremia and should be investigated further. DISCLOSURES: All authors: No reported disclosures.