Methods: Electronic medical records were examined for clinical and microbiological data. The first episodes of mono-microbial bacteremia in patients at the Houston VA with positive blood cultures that grew either E. coli or Klebsiella species during 2016 were included. The JHH-CDT was used to predict whether or not the isolate would be ESBL+. Empiric initial antibiotic selection was also collected.
Results: Eighty-seven cases occurred during the study period; 95% were in men. In veterans at the VA in Houston compared to patients at JHH, respectively, the JHH-CDT demonstrated lower sensitivity (35.7% vs 51%), positive predictive value (83.3% vs. 90.8%), negative predictive value (88.8% vs. 91.9%) but similar specificity (98.6% vs. 99.1%). Of note, of the five questions in the JHH-CDT, only one was applicable to the Veteran population: history of ESBL colonization or infection in the prior 6 months. Two other CDT questions did not apply to the VA population (no Veterans had these conditions): hospitalization for >=1 day in an ESBL high-burden in the prior 6 months and age <43 years old. Standard of care led to carbapenems being empirically prescribed for 4/14 (28.6%) ESBL+ bloodstream infections and for 3/73 (4.1%) of non-ESBL bloodstream infections.
Conclusion: In this VA population, the JHH-CDT had low sensitivity because two decision nodes did not apply to our older population with little international travel. Standard of care empiric choice of antibiotics also had low sensitivity, covering only 28.6% of ESLB infections appropriately. These findings highlight the importance of developing and validating population-specific predictive stewardship tools.
L. Zechiedrich, None
B. W. Trautner, Paratek: Consultant , Consulting fee . Zambon: Consultant , Consulting fee and Research grant .
See more of: Poster Abstract Session