Risk factors and outcome in extended spectrum beta-lactamase producing Escherichia coli and Klebsiella pneumoniae bacteremia.

Given our interest in the epidemiology of nosocomial infection, and the impact of multidrug resistance, I read with great interest the study by Dr. Memon et al1 regarding risk factors and mortality associated with bacteremia caused by extended-spectrum ß-lactamase producing (ESBL) Escherichia coli and Klebsiella pneumonia. In this study, the only variable identified as a significant risk factor for ESBL involvement was nosocomial infection. Because hospitals are recognized as epicenters of antibiotic consumption this finding is in line with the expectations. In univariate analysis, no difference in mortality was found between patients with bacteremia caused by ESBL or non-ESBL pathogens. Indeed, on the condition of early initiation of appropriate antimicrobial therapy, the involvement of multidrug resistance does not necessarily worsen the prognosis of patients with Enterobacteriaceae bacteremia.2-5 However, I have some concerns on the results obtained by the multivariable regression model, performed to identify independent risk factors for death. Based on this analysis, the variable ‘nosocomial infection’ was identified as a risk factor for mortality, whereas ESBL involvement was not. Given the strong relationship between nosocomial infection and ESBL involvement, it seems like there exists an important collinearity, which potentially confounds the results of the multivariable regression analysis. It is unlikely that ‘nosocomial infection’ as such would increase the risk of death, but ESBL involvement clearly is a risk factor for inappropriate antimicrobial therapy, and as a consequence may cause bad outcome. Therefore, I wonder how the regression model would turn out if the variable ‘nosocomial infection’ would be replaced by the (closely related) variable ‘ESBL-producing pathogen’. It would also be worthwhile to explore the impact of early initiated appropriate therapy in the logistic regression model of this specific cohort. Furthermore, I assume there might as well exist a substantial collinearity between the variables ‘septic shock’ and ‘intensive care unit care’, which are both recognized as independent risk factors for mortality. I would greatly appreciate if the authors could elaborate on these comments and suggestions.