A model for predicting nosocomial carbapenem-resistant Klebsiella pneumoniae infection

Mortality associated with infections due to carbapenem-resistant Klebsiella pneumoniae (CR-KP) is high and the infections need to be predicted early. The risk factors for CR-KP infection are heterogeneous. The aim of the present study was to construct a model allowing for the early prediction of CR-KP infection. Nosocomial infections due to K. pneumoniae were evaluated retrospectively over a 2-year period. The case cohort consisted of 370 inpatients with CR-KP infection. For each case enrolled, two matched controls with no CR-KP infection during their hospitalization were randomly selected. Matching involved month of admission, ward, as well as interval days. The Vitek 2 system was used for identification of isolates and antimicrobial susceptibility testing. General linear model with logistic regression was used to identify possible risk factors. The predicted power of the model was expressed as the area under the receiver-operating characteristic curve. Age, male gender, with cardiovascular disease, hospital stay, recent admission to intensive care unit, indwelling urinary catheter, mechanical ventilation, recent β-lactam-β-lactamase inhibitors, fourth-generation cephalosporins and/or carbapenems therapy were independent risk factors for CR-KP infection. Models predicting CR-KP infection developed by cumulative risk factors exhibited good power, with areas under the receiver-operating characteristic curves of 0.902 [95% confidence interval (CI), 0.883–0.920; P<0.001] and 0.899 (95% CI, 0.877–0.921; P<0.001) after filtering by age (≥70 years). The Yonden index was at the maximum when the cumulative risk factors were ≥3 in the two prediction models. The results show that the prediction model developed in the present study might be useful for controlling infections caused by CR-KP strains.