Risk Factors for Colonization with Extended-Spectrum β-Lactamase–producing Bacteria and Intensive Care Unit Admission

Coexisting conditions and previous antimicrobial drug exposure predict colonization.

gram-negative bacteria are emerging pathogens. Clinicians, microbiologists, infection control practitioners, and hospital epidemiologists are concerned about ESBL-producing bacteria because of the increasing incidence of such infections, the limitations of effective antimicrobial drug therapy, and adverse patient outcomes (1)(2)(3)(4)(5).
Research conducted to date has focused on identifying risk factors for colonization with multidrug-resistant, gram-positive bacteria. In contrast, little research has been conducted to identify the risk factors for colonization with gram-negative multidrug-resistant bacteria in nonoutbreak settings. To our knowledge, no study of the magnitude of our study has been conducted, nor have any studies based in the United States sought to identify risk factors for colonization with ESBL-producing bacteria on admission to an intensive care unit (ICU).
The primary objective of our study was to identify factors predictive of colonization with ESBL-producing bacteria at admission to an intensive care unit (ICU). In addition, we identifi ed the percentage of patients colonized with ESBL-producing bacteria who had a subsequent positive clinical culture for the same species of ESBL-producing bacteria. Understanding risk factors for colonization is important for several reasons. First, understanding the potential causal mechanisms of colonization can lead to successful infection control, involving antimicrobial stewardship and public health interventions aimed at controlling the emergence of ESBL-producing bacteria. Second, such knowledge can help identify which patients should be receiving empiric ESBL-targeted antimicrobial therapy. Some hospitals have used active surveillance culturing for antimicrobial drug-resistant, gram-negative bacteria to help guide empiric therapy (6).

Study Population and Sample Collection
We conducted a prospective cohort study of patients admitted to either the surgical or medical ICU at the University of Maryland Medical Center from September 1, 2001, through June 1, 2005. Descriptions of the hospital and the ICUs are reported in other publications (7,8). During the study period, on average, 8.6 clinical cultures per month were positive for ESBL-producing bacteria. No outbreaks of ESBL-producing bacteria were found among clinical cultures based on control process charting. No additional infection control precautions were used for patients with ESBL-producing bacteria on clinical culture. ESBL surveillance culture results were not given to physicians or nurses. Contact isolation precautions were applied for patients with vancomycin-resistant enterococci or methicillin-resistant Staphylococcus aureus infections.
During the study period, nurses obtained perianal specimens for culture from all ICU patients within 48 hours of ICU admission. All patients who had admission culture results were included in this study. Patients with multiple admissions to either of the ICUs during the study period were allowed to enter the cohort of at-risk patients multiple times, as long as they were not positive for ESBLproducing bacteria on any prior admissions (because patients remain at risk for ESBL-producing bacteria on each subsequent admission). This study was approved by the Institutional Review Board of the University of Maryland, Baltimore. Informed consent was not required by the Institutional Review Board because perianal specimens were cultured as part of infection control quality improvement involving active surveillance culturing for vancomycin-resistant enterococci.

Microbiologic Methods
The perianal cultures were processed for ESBL-producing bacteria in real time as the specimens were collected. The perianal cultures were fi rst screened for potential ESBL-producing bacteria by plating onto MacConkey agar (Remel, Lenexa, KS, USA) with 2 μg/mL of ceftazidime added to the cooled agar before the plates were poured (9). Plates were incubated at 37°C for 24 to 48 hours. Lactosefermenting colonies growing on the ceftazidime-containing plates were identifi ed as Escherichia coli or Klebsiella species by using API 20E identifi cation strips (bioMérieux Vitek, Inc., Hazelwood, MO, USA). All E. coli and Klebsiella isolates underwent ESBL confi rmatory testing by disk diffusion for ceftazidime and cefotaxime with and without clavulanic acid as recommended by the Clinical Laboratory Standards Institute's guidelines (10).

Data Collection and Variables
For all patients included in the study, we collected data regarding the patient's previous hospital antimicrobial drug exposures, length of hospitalization before ICU admission, coexisting conditions, previous positive cultures, and other hospitalization-related and demographic information. Antimicrobial drug exposures were assessed in the period between hospital admission and ICU admission. Antimicrobial drugs were analyzed as binary variables; if an antimicrobial drug was received during the period defi ned above, it was classifi ed as having been received independent of the number of doses received. Duration of antimicrobial drug exposure was not analyzed. Coexisting conditions were assessed by the Charlson Comorbidity Index, the Chronic Disease Score (CDS), and the infectious disease-specifi c CDS (CDS-ID) (11)(12)(13).
Initial bivariable statistical comparisons were conducted by using the χ 2 test for categorical data and the Student t test or Wilcoxon test for continuous data. Continuous variables that were not normally distributed were categorized for the purpose of multivariable analyses. To identify patient characteristics associated with colonization by an ESBL-producing bacterium on ICU admission, we used multivariable logistic regression. Because patients were allowed to enter into the study multiple times, we also assessed the need to control for the correlated error structure of the data. All variables that were associated with ESBL colonization in the bivariable analysis at the p<0.1 level were included in the model-building stages of the multivariable analysis. A stepwise model building method was used. Variables were retained in the fi nal model if they were signifi cant at a p<0.05 level or if they were observed to have a confounding effect on the association between another predictor and ESBL colonization status. A confounding effect was defi ned as a change in the model coeffi cient by >10%. An additional bivariable statistical analysis was performed to identify risk factors for subsequent clinical culture positivity with the same species of ESBL-producing bacteria among the cohort of patients colonized with an ESBL-producing bacteria. We calculated the C statistic of the fi nal model. The C statistic reports values from 0.5 (indicating no predictive power) to 1.0 (indicating perfect prediction). In addition, we calculated the sensitivity, specifi city, positive predictive value, and negative predictive value for patients with or without all dichotomous variables in the fi nal model. Statistical analysis was performed with SAS Version 9.1 (SAS Institute, Cary, NC, USA).

Results
During the study period, 5,209 (84%) admitted patients had results of admission perianal cultures and were included in this study, 4,398 patients had 1 ICU admission, and 618 patients had repeat admissions. Ninety-one percent of the surveillance cultures were obtained within the fi rst 12 hours of ICU admission. The cross-sectional patient cohort consisted of 2,096 (40%) admissions to the medical ICU and 3,113 (60%) admissions to the surgical ICU. The mean age of the patients was 55 years. The mean comorbidity score as measured by the CDS-ID was 2.73 and 2.42 as measured by the Charlson Comorbidity Index. Based upon International Classifi cation of Diseases, 9th revision (ICD-9) codes, 1,285 (25%) had diabetes, 1,344 (26%) had cancer, and 193 (4%) were HIV positive; 1,594 (31%) of patients had been transferred from another healthcare facility, and 1,693 (33%) had been previously admitted to the same hospital within the past year.
We examined patient characteristics, coexisting conditions, and previous antimicrobial drug exposures to identify factors potentially associated with colonization by an ESBL-producing bacterium on ICU admission (  34-3.31) were all found to be independently associated with colonization by an ESBL-producing bacterium on admission to an ICU. No other antimicrobial drug was found to have a signifi cant (p<0.05) effect in the fi nal multivariable model. Note that we did not adjust for the correlated error structure of the data in the fi nal analysis; because the correlation was low, this adjustment had little effect on our estimates (data not shown). The C statistic of the fi nal model was 0.69. Patients categorized on the basis of the presence of all of the following dichotomous predictors of the fi nal model (zosyn, vancomycin and age >60) yielded a sensitivity of 9.4%, specifi city of 97.3%, positive predictive value of 7.3%, and negative predictive value of 97.9%.
For the 117 patients identifi ed as colonized with ESBL-producing bacteria, we assessed their history of culture positivity with ESBL-producing bacteria as well as other antimicrobial drug-resistant bacteria (Table 3). Of the ESBL-colonized patients, 6 (5%) had positive clinical cultures for ESBL-producing bacteria during the same hospital admission but before ICU admission, and 29 (25%) had a subsequent ESBL-positive clinical culture from the time an ICU admission surveillance specimen was obtained for culture to the date of hospital discharge. The only risk factor that predicted subsequent positive ESBL clinical culture was the amount of time in the hospital between positive surveillance culture and hospital discharge (OR 1.03 per additional day, 95% CI 1.01-1.06). These 29 patients had 56 clinical cultures with ESBL-producing bacteria. The sources of the 56 clinical cultures positive for ESBLproducing bacteria were the following: 9 blood cultures, 17 sputum or bronchoscopy specimens, 10 urine cultures, 12 wound cultures, and 8 miscellaneous sources. Of 117 ESBL-colonized patients, 41 (35%) were known to have

Discussion
In this study, we identifi ed risk factors for colonization with ESBL-producing E. coli and Klebsiella spp. at ICU admission. We identifi ed age >60 years, comorbidity as measured by the CDS-ID, previous in-hospital piperacillin-tazobactam use (current admission), and previous present admission in-hospital vancomycin use (current admission) as independent risk factors. We also quantifi ed the ESBL colonization/clinical culture positivity rate among these patients and addressed the question of whether patients colonized with ESBL had a history of colonization with MRSA and VRE.
The risk factors identifi ed are potentially important because they can help determine which patients may need empiric antimicrobial drug therapy targeted to the ESBLproducing bacteria. Carbapenem antimicrobial agents may be preferred as empiric choice for patients at risk for ESBL-producing bacteria (2). We are not recommending that all patients with these risk factors receive empiric antimicrobial drug therapy targeted to ESBL-producing bacteria. However, among particular patients with the identifi ed risk factors and levels of severity of infection that require empiric therapy, a choice of empiric therapy that includes coverage of ESBL-producing bacteria may be warranted. Thus, we recommend that for patients in ICUs with similar characteristics to the units in this study, physicians consider using antimicrobial agents targeted against ESBL-producing bacteria. These ESBL-targeted drugs should be considered when the physician chooses to prescribe an antimicrobial drug for situations such as fever of unknown origin, suspected pneumonia, or suspected bacteremia. In addition, these risk factors identifi ed may be of use to hospital antimicrobial drug stewardship programs and pharmacy and therapeutics committees.
We hope that our risk factor study and other risk factor studies in the area of antimicrobial drug resistance will be used in future antimicrobial agent stewardship intervention studies and future infection control intervention studies. Previous risk factor studies have led to antimicrobial agent stewardship intervention studies aimed at controlling ESBLproducing bacteria (14,15). In the areas of pneumonia and neutropenia patients with fever, risk factors studies have successfully led to intervention studies that have affected national guidelines (16)(17)(18). Well-designed intervention studies, based on risk factor studies of antimicrobial drug resistance, can lead to more appropriate antimicrobial drug use, which will improve patient outcomes and decrease the emergence of antimicrobial drug resistance (19,20).
The risk factors identifi ed may be causally related to the outcome of ESBL-colonization or may only be statistically associated. Age >60 years and the presence of coexisting conditions have validity and biologic plausibility for a causal association with colonization status (1,9,21). The identifi cation of piperacillin-tazobactam and vancomycin as risk factors is more intriguing. Vancomycin and piperacillin-tazobactam are widely used at our tertiary-care hospital, the University of Maryland Medical Center, and thus may just be markers of ICU patients who require broadspectrum antimicrobial coverage. However, understanding intestinal ecology and antimicrobial drug resistance is still in nascent stages. Vancomycin and piperacillin-tazobactam may be true causal risk factors for colonization with ESBLproducing bacteria. Piperacillin-tazobactam is believed to be effective against ESBL-producing bacteria only when the inoculum is low (22). Thus, with regard to the intestinal fl ora, piperacillin-tazobactam may not be effective at eradicating ESBL-producing bacteria due to inoculum effects and low intestinal concentration of piperacillin-tazobactam. Additionally, we were surprised by the identifi cation of piperacillin-tazobactam as a risk factor as some hospitals have adopted antimicrobial drug stewardship policies that have limited the prescribing of cephalosporins and increased the use of antimicrobial drugs, including piperacillin-tazobactam, in an effort to control ESBL-producing bacteria (15,23). Vancomycin may be a risk factor through relative decolonization of the normal fl ora through vancomycin exposure and then subsequent colonization with ESBL strains through horizontal transmission before ICU admission (24,25). We found a ratio of colonization to clinical culture positivity that was the same order of magnitude as for VRE and MRSA (26)(27)(28)(29). In addition, only 35% of patients with ESBL-colonization were previously known to be VRE or MRSA positive. These numbers and the local prevalence rate of ESBL-producing bacteria are important parameters in assessing the cost-effectiveness of active surveillance for ESBL-producing bacteria. Further work, including costeffectiveness studies, needs to address whether active surveillance is benefi cial for ESBL-producing bacteria.
Several studies. performed worldwide, have analyzed risk factors for colonization with multidrug resistant Enterobacteriaceae. Many studies have not analyzed the specifi c antimicrobial drug resistance mechanism and thus are not directly comparable to our study. A study from Canada determined that several antimicrobial drugs were risk factors for multidrug resistant Enterobacteriaceae (30). In contrast to our study, most of their isolates had AmpC as a resistance mechanism, and thus their study did not determine risk factors for ESBL-producing bacteria. A 4-year cohort study done in France determined the ESBL-producing bacteria colonization rate in 2 ICUs to be 0.97% and thus concluded that, in their setting, active surveillance was unlikely to be cost-effective (31). A study in Israel identifi ed 26 (10.8%) of 241 patients tested by active surveillance as colonized with ESBL-producing bacteria. Risk factors identifi ed in multivariable analysis were poor functional status, current antimicrobial drug use, chronic renal insuffi ciency, liver disease, and the use of histamine 2 receptor antagonists (32).
A limitation of our study is that we did not have access to records of the antimicrobial drugs that patients may have received as outpatients before their hospital admission. However, relevant to the question of empiric therapy, most intensive care clinicians do not have access to records of outpatient antimicrobial drug use when they are empirically choosing antimicrobial agents. Another limitation of the study is that we did not have access to the subsequent ESBLpositive clinical isolates and thus were unable to compare them by molecular epidemiologic methods, such as pulsedfi eld gel electrophoresis, to see whether they were identical to the ESBL-colonizing isolates identifi ed previously. We did not perform chart review; thus, the subsequent clinical cultures with ESBL-producing bacteria could have represented either clinical infection or colonization, based on defi nitions from the Centers for Disease Control and Prevention (33). The use of ceftazidime in the screening agar may have caused the CTX-M family of β-lactamases to be missed. However, no CTX-M enzymes were detected in a sample of clinical isolates from the University of Maryland Medical School and the adjacent Veterans Affairs Medical Center from 2001 to 2002 (34).
In this study, we identifi ed risk factors for ESBL-producing bacterial colonization among ICU patients. These data may be useful for identifying which patients may warrant empiric ESBL-targeted antimicrobial drug therapy. We also demonstrate that subsequent infections with ESBL-producing bacteria develop in a large percentage of ICU patients colonized with ESBL-producing bacteria.