Background: The timely identification of carbapenem resistance is essential in the management of patients with Klebsiella pneumoniae bloodstream infection (BSI). An algorithm using electronic medical record (EMR) data to quickly predict resistance could potentially help guide therapy until more definitive resistance testing results are available.
Methods: All cases of Klebsiella pneumoniae BSI at Mount Sinai Hospital from 9/2012 through 9/2016 were identified. Cases of persistent BSI or recurrent BSI within two weeks were included only once. Patients with recurrent BSI after more than two weeks of negative blood cultures were considered distinct cases and included more than once. Carbapenem resistance was defined as an imipenem minimum inhibitory concentration of ≥2 μg/mL. Extensive EMR data for each patient was compiled into a relational database using SQLite. Possible risk factors for carbapenem resistance were queried from the database and analyzed via univariate methods. Significant factors were then entered into a multiple logistic regression model in a forward stepwise approach using SPSS.
Results: 613 cases of Klebsiella pneumoniae BSI were identified in 540 unique patients. The overall incidence of imipenem resistance was 10% (61 cases). Significant markers of resistance included in the final model were (1) prior colonization with imipenem-resistant Klebsiella pneumoniae (2) hospital unit (defined as high risk unit, low risk unit, and emergency department), (3) total inpatient days in the previous five years, (4) total days of oral or parenteral antibiotics in the past two years, and (5) age >60 years old (Figure 1). The model generated a receiver operating characteristic curve with an area under the curve of 0.75 (Figure 2). At a cut point of 0.083 the model correctly predicted 72% of imipenem-resistant cases while incorrectly labeling 32% of susceptible cases as resistant (Sn=72%, Sp=63%, Figure 3).
Conclusion: A multiple logistic regression model using EMR data can generate immediate, clinically-useful predictions of carbapenem resistance in patients with Klebsiella pneumoniae BSI. Larger data sets are needed to improve and validate these findings.
Figure 1: Algorithm variables
Figure 2: Receiver operating characteristic curve
Figure 3: Classification table