Introduction

From the late 1990s, multidrug-resistant Enterobacteriaceae (mostly E. coli) that produce extended-spectrum beta lactamases (ESBLs) have emerged within the community setting as an important cause of urinary tract infections [1, 2], and among patients with health-care associated and nosocomial infections [3, 4], with a variety of clinical syndromes including bacteremia and sepsis [5]. The carbapenems are widely regarded as the drugs of choice for the treatment of severe infections caused by ESBL-producing Enterobacteriaceae, although comparative clinical trials are scarce [6–8].

The epidemiology of ESBL bacteria has evolved, and nowadays as many as 10% of E. coli isolated from the bloodstream produce ESBL, with increasing percentages in several countries [9, 10]. Moreover, there are reports in patients with cancer describing no significant differences in mortality between adequate and inadequate empiric antibiotic treatment [5]. When the issue of inappropriate treatment was thoroughly studied in P. aeruginosa bacteremias, for example, inappropriate empiric antimicrobial therapy was one of the independent risk factors for mortality in patients with high-risk sites of infection [11].

In this retrospective study we report our data from an 8-year period to quantify the effect of comorbidities and appropriate antibiotic treatment in affecting the mortality of patients with bacteremia caused by E. coli, K. pneumoniae and P. mirabilis producing ESBL in a single center.

Materials and methods

A retrospective study was performed with the aim of describing the effect of appropriate antimicrobial treatment and comorbidity on mortality in patients with bacteremia caused by isolates of E. coli, Klebsiella pneumoniae and Proteus mirabilis producing ESBL. The search was made through the computerized archive system of the Microbiology Laboratory from January 2000 to December 2007 with E. coli, K. pneumoniae and P. mirabilis producing ESBL from blood cultures. The study was conducted at the San Giovanni Battista University Hospital, a 1,200-bed Academic Center in Turin, Italy.

The identification of ESBL-producing bacteria was based on screening and confirmatory tests for ESBLs in the two automated systems for identification/susceptibility testing used in our laboratory (Phoenix Automated Microbiology System, BD Diagnostic Systems, Sparks, MD, and Microscan Walk Away, Siemens Healthcare Diagnostics, Deerfield, IL). The production of ESBL was suspected in strains with MIC values of aztreonam, cefotaxime, ceftazidime, ceftriaxone or cefpodoxime ≥1 μg/ml and was confirmed using the E test system (bioMerieux, Marcy l’Etoile, FR), a predefined stable gradient of antibiotic concentrations on a plastic strip to determine the on-scale MIC of various antibiotics against microrganisms and for detection of resistance mechanisms. Infact E test ESBL cefotaxime/cefotaxime + clavulanic acid (CT/CTL) and ceftazidime/ceftazidime + clavulanic acid (TZ/TZL) strips are designed to confirm the presence of clavulanic acid inhibiting ESBL enzymes in E. coli, K. pneumoniae, K. oxytoca and P. mirabilis.

ESBL-production organisms were defined from the follow MIC ratio: CT ≥ 0.5 μg/ml and CT/CTl ≥ 8 μg/ml or TZ ≥ 0.5 μg/ml and TZ/TZL ≥ 8 μg/ml, or a “phantom” zone or deformation of the CT or TZ ellipse.

The primary objective of our work was to describe the impact of appropriate treatment on mortality and the risk factors associated with mortality of ESBL-associated bacteremia. Comorbidities were quantified by the Charlson Index score [12], and adequate treatment was defined as the administration of antibiotics with documented in vitro sensitivity administered within 72 h after the first positive blood culture. The outcome measured was death within 21 days of the first positive blood culture. Other objectives were to detail the clinical impact of community-acquired (CA), health-care-associated (HC) or hospital-acquired (HA) infections according to the suggested criteria by the published literature [13]. CA infections were considered present in patients when positive blood culture occurred <48 h after hospital admission with no recent hospital admission (during the last 6 months), no intravenous antibiotic therapy during the last 30 days, not undergoing hemodyalisis, without bone marrow or solid organ transplants, and not transferred from a long-term care facility or admitted with a central venous catheter. HC infections were defined as the presence of at least one of these risk factors associated with a diagnosis of bacteremia within 5 days after hospital admission. HA infections were defined as when the onset of bacteremias occurred 48 h after the hospital admission if HC according to the definition was not present and after 5 days if the definition of HC infections was met. The source of the infection was an infection at a distant site caused by a microbial strain in the bloodstream isolate. It was determined on the basis of microbiological results and physicians’ findings.

Collected data included demographics and date of admission and discharge, time of diagnosis, comorbidities as calculated by the Charlson Index score [12] and inappropriate antimicrobial therapy (defined as the beginning of a treatment with active antimicrobial agents >72 h after bloodstream infection onset or with a resistant drug). The study was performed according to the Ethics Committee standards.

Statistical analysis was performed with Statview software 4.0 (Abacus Concept, Inc.). Unpaired t test, paired t test and contingency tables were used when appropriate. Variables with a significant association with mortality at univariate analysis were entered into a logistic regression model for multivariate analysis. A p value of ≤0.05 was used.

Results

During the 8-year study period (2000–2007) there were 176 patients with infection by ESBL-producing organisms, and the study sample included 128 episodes of patients that had medical records available for review and had positive blood cultures; the single positive cultures from removed central venous catheter tips were excluded.

The 128 samples positive for ESBL bacteria included 80 caused by E. coli (62.5%), 28 by K. pneumoniae (21.9%) and 20 by P. mirabilis (15.6%). There was a trend to increased numbers per year, from 7 cases in 2000, 11 cases in 2001 and 2002, 10 cases in 2003, 14 cases in 2004, 21 cases in 2005, 23 cases in 2006 to 31 cases in 2007.

The antibiotic sensitivity in vitro is shown in Table 1. Resistance to piperacillin-tazobactam was significantly more common among E. coli and K. pneumoniae than among P. mirabilis isolates, while resistance to amikacin was significantly more common among E. coli and P. mirabilis than among K. pneumoniae isolates.

Table 1 Antimicrobial susceptibility test results for 128 bloodstream isolates of ESBL-producing E. coli, K. pneumoniae and P. mirabilis
Full size table

Patient characteristics and initial antimicrobial treatment

The characteristics of the patients are presented in Table 2. The mean age was 63.5 ± 17.4 years (male = 90; female = 38). The mean Charlson Comorbidity Index score was 3.0 ± 2.1, and there was a trend toward higher scores during the years of observations, from 2.4 ± 2.1 in 2000 to 3.8 ± 3.0 in 2007. There were 71 (55.5%) and 35 (27.3%) cases in surgical and general internal medicine wards.

Table 2 Characteristics of the 128 patients with bloodstream infection by ESBL-producing organisms
Full size table

The mean duration of hospital stay before the diagnosis was 18.3 ± 27.8 days overall, and it was 13.1 ± 18.2 days for E. coli, 37.8 ± 45.5 days for K. pneumoniae and 11.7 ± 19.8 days for P. mirabilis. The majority (89.0%) of patients had hospital-acquired (HA) or health-care associated (HC) infections. The source of ESBL bacteremia was attributed to the biliary tract in 55 patients (45%) and to the urinary tract in 17 cases (14%).

Data on antibiotic treatment and antibiotic sensitivity were available for 106 patients. The empiric monotherapy was appropriate in 74 (69.8%) out of 106 episodes: imipenem or meropenem and other antibiotics were given in 54 (73%) and 20 (27.0%) cases, respectively. Three additional patients were given carbapenems in combination with other antibiotics. Among the 20 patients with monotherapy with antibiotics other than carbapenems but effective in vitro, five (4.7%) were treated with amikacin, eight (7.5%) with β-lactam-β-lactamase inhibitor combinations (six with piperacillin-tazobactam and two with amoxicillin/clavulanate), six (5.7%) with a fluoroquinolone (ciprofloxacin or levofloxacin) and one (0.9%) with trimethoprim-sulfamethoxazole.

The 74 patients with appropriate empiric treatment had a mean length of stay of 18.4 ± 24.2 days, and it has to be noted that in 30 (40.5%) patients the diagnosis of bacteremia was made within 5 days from hospitalization. The etiology included E. coli in 49 (66.2%) patients, K. pneumoniae in 15 (20.3%) patients and P. mirabilis in 10 (13.5%) patients. The mean Charlson Comorbidity Index score was 3.1 ± 2.2.

The initial antibiotic therapy was inadequate in 32 (30.2%) of 106 patients, either because of absence of activity in vitro (21 cases, 65.6%) or because of treatment delay (11 cases, 34.4%). Of the former group, five (23.8%) patients were treated with a cephalosporin, seven (33.3%) with a fluoroquinolone, eight (38.1%) with a β-lactam-β-lactamase inhibitor combination and one (4.8%) with an aminoglycoside. In 14 out of 32 patients (43.8%) with inadequate initial antimicrobial regimens, carbapenems were administered after a mean of 5.6 ± 3.3 days.

Twenty-one-day mortality rates

Twenty-one days after onset of bloodstream infection by ESBL-producing organisms, 22 of 128 (17.2%) of the patients had died: 15% (12/80) infected by E. coli, 21.4% (6/28) infected by K. pneumoniae and 20% (4/20) by P. mirabilis. In Table 3 the differences between the survivor and death subgroups are shown. At univariate analysis significant differences were found only for previous positive blood cultures (p = 0.004) and presence of septic shock at diagnosis (p = 0.006), with a p value for mortality with and without appropriate treatment of 0.051 (data at 21 days after diagnosis available for 91 patients). There were no significant associations with mortality for age, sex, Charlson score, etiology, HC and HA infections, days after admission or source of bacteremia. At multivariate analysis significant differences were confirmed for previous blood cultures (p = 0.013) and septic shock (p = 0.016) (Table 4).

Table 3 Risk factors associated with 21-day mortality
Full size table
Table 4 Mortality rates (21 days) in 74 patients with bloodstream infection by ESBL-producing organisms treated with appropriate antimicrobial agents
Full size table

The mortality for patients adequately treated was 14.9% (11/74), and it was observed within 5 days in six patients (54.5%) treated with carbapenem (imipenem in 2/3 cases), piperacillin-tazobactam in two cases and levofloxacin in one case (per os). Among the descriptive reasons for early mortality in appropriately treated patients, there was no attributable effect for the Charlson score (4 ± 2 vs. 4 ± 3), but there were three patients with early mortality treated with lower dosages of antibiotics. The rate of septic shock was 5/6 (83.3%) in patients dead within 5 days versus 2/5 (40%).

The mortality was 14% (8/57) with carbapenem and 15% (3/20) for those adequately treated with another effective antibiotic. Among the latter 20 patients, the mortality rate was 33.3% (2/6) with β-lactam-β-lactamase inhibitors, although the dosage was not adequate in some patients, and 20% (1/5) with a fluoroquinolone. No deaths were reported with empiric aminoglycoside given in five patients and trimethoprim-sulfamethoxazole in one patient.

The mortality for patients inadequately treated was 35.2% (6/17) and was due to a delay in the administration of effective antibiotics in four patients and to the presence of severe sepsis or septic shock.

Discussion

Sepsis caused by ESBL-producing bacteria such as E. coli, K. pneumoniae and P. mirabilis is a contemporary issue of antibiotic resistance in patients with nosocomial and health-care associated infections. There is wide agreement that carbapenems are the first choice in patients with such infections, but there is a low level of evidence for this recommendation. For example, the data from Paterson et al. [6] and Tumbarello et al. [7] showed lower mortality when they were used as first-line treatment in the setting of a retrospective study or of a high percentage of cancer patients [14]. A wide administration of carbapenems in the empiric treatment of complex patients has been the practical consequence, although the rate of appropriate empiric therapy with carbapenems has not reached 100% in any of the published case series. For example, it was only 27.8% when compared with 79.6% of controls (OR paired = 0.053, p = 0.001) [15], and it was around 50% in an Italian series [7] and 52.3% in our series. Whether other agents with in vitro sensitivity can be administered remains a matter of debate, especially considering that some microbiology laboratories have been suggested to report piperacillin-tazobactam, ceftriaxon, cefepime and cefotaxime as resistant antibiotics when an ESBL-producing microrganism has been isolated.

Another issue has been recently added by Gudiol et al. [5], who compared bacteremias by ESBL-producing and -non-producing E. coli in cancer patients and did not find a significant difference in mortality between adequate and inadequate empiric treatment. Thus, the comorbidity and efficacy of carbapenem or other antibiotics still need to be clarified in patients with bacteremias caused by ESBL-producing microrganisms. In our study the underlying comorbidities according to the Charlson score [12] were not associated with worse outcome, and data on antibiotic treatment show that other antibiotics are as effective as carbapenems.

The clinical problem of ESBL is further complicated by the rising importance of HC-associated infections and its various definitions, as well as the increasing importance of cross resistance to fluoroquinolones. The importance of HC-associated infections is well recognized in patients with pneumonia, where higher rates of resistant pathogens have been documented [13]; however, caution should be employed in the frequent administration of wide-spectrum antibiotic treatment early after admission in such patients because of the potential for further fueling antibiotic resistance [16]. The concern for cross resistance of fluoroquinolones has already been reported, with therapeutic failure reported in as many as 50% of patients empirically treated with quinolones even when in vitro sensitivity was documented [7]. The low incidence of quinolone sensitivity is noteworthy in our study (23.8% overall), as in the study by Gudiol et al. [5] (35%). However, in our study, notwithstanding the low number of patients treated with ciprofloxacin with in vitro activity, the mortality associated with empiric treatment was only 16.6% (1 out of 6 patients). Our study has some limitations, such as the retrospective nature and the absence of full medical and/or microbiological records available for some patients because of a change from full paper charts to an electronic medical records archive, which made tracking some patients impossible. The same flaw affected our ability to have full microbiological data on all isolates because of the changes in the electronic microbiology system over the years. A further issue is that we considered all agents with in vitro sensitivity as appropriate treatment, as in other papers [7], even if carbapenems are widely regarded as the treatment of choice. Since EUCAST is currently suggesting to report microbiological data “as found,” independently of ESBL production, we should be ready, if possible, not to rely only upon carbapenems for patients with risk factors for infections caused by ESBL-producing bacteria. However, further studies are needed since we lack strong clinical evidence in favor of the effectiveness of other antibiotics different from carbapenems in this special setting [17].

From a practical point of view, owing to the wide administration of carbapenems in the setting of empiric treatment of complicated nosocomial infections, we can conclude that antibiotics other than carbapenems, such as piperacillin-tazobactam, amikacin and perhaps trimethoprim-sulfamethoxazole or fluoroquinolones, may still have a major role to play in the treatment of bacteremias caused by ESBL-producing bacteria. Such antibiotics may indeed be very useful even after empiric carbapenem when there is a need to streamline the empiric wide-spectrum antimicrobial treatment after in vitro results are available.