Intensive care unit-acquired pneumonia caused by Klebsiella pneumoniae in China: Risk factors and prediction model of mortality

Abstract

Intensive care unit (ICU)-acquired pneumonia (ICUAP) is a major concern owing to its associated high mortality rate. Few studies have focused on ICUAP caused by Klebsiella pneumoniae (KP). This study aimed to investigate the risk factors for ICUAP-associated death due to KP and to develop a mortality prediction model. Patients with KP-associated ICUAP at Renji Hospital were enrolled from January 2012 to December 2017. The patients were registered from the ICU units of the Surgery, Gynecology and Obstetrics, Neurosurgery, Emergency and Geriatric Departments, and were followed for 30 days. A multivariate analysis was performed to analyze the differences between 30-day survivors and nonsurvivors, and to determine the independent risk factors. Receiver operator characteristic (ROC) curves were also used to determine the predictive power of the model. Among the 285 patients with KP-associated ICUAP, the median age was 70.55 years, and 61.6% were men. Fifty patients died. The nonsurvivors had a lower Glasgow coma score (GCS), platelet count, and albumin concentrations, but higher lactate concentrations, than the survivors. The nonsurvivors were also more likely to be admitted to the ICU for respiratory failure and surgery, and they received less appropriate empirical antimicrobial therapy than the survivors. A lower GCS (odds ratio [OR] = 0.836, 95% confidence interval [CI]: 0.770–0.907), lower albumin concentrations (OR = 0.836, 95% CI: 0.770–0.907), higher lactate concentrations (OR = 1.167, 95% CI: 1.0013–1.344) and inappropriate empirical treatment (OR = 2.559, 95% CI: 1.080–6.065) were independent risk factors for mortality in patients with KP-associated ICUAP. ROC curve analysis showed that the risk of death was higher in patients with 2 or more independent risk factors. The predictive model was effective, with an area under the ROC curve of 0.823 (95% CI: 0.773–0.865). The number of independent risk factors is positively correlated with the risk of death. Our model shows excellent predictive performance.