Predictors of Mortality in Patients with Bloodstream Infections Caused by Extended-Spectrum-β-Lactamase-Producing Enterobacteriaceae: Importance of Inadequate Initial Antimicrobial Treatment

Setting.This study was conducted at the Catholic University Hospital, a 1,700-bed academic medical center located in Rome, Italy. It offers a full range of clinical services and admits approximately 60,000 patients per year.

Study design and patients.The computerized database of our microbiology laboratory was searched to identify all adult inpatient BSI caused by ESBL-producing strains of E. coli, Klebsiella spp., or P. mirabilis diagnosed between January 1999 and December 2004. Recurrent infections were excluded, and only one episode per patient (the first) was included in our analysis. The data were collected from patients’ hospital charts and the laboratory database, which contains complete profiles for all patients with blood cultures positive for gram-negative bacteria since 1999. A retrospective cohort study design was used. The outcome measured was death within 21 days of the first positive blood culture. Survivor and nonsurvivor subgroups were compared to identify predictors of 21-day mortality.

Variables.The following variables were explored as possible predictors of 21-day mortality, For patient variables, we considered age; gender; Charlson comorbidity index (8); underlying diseases; immunosuppressive therapy; the duration of hospitalization; an intensive care unit (ICU) stay at the time of infection; and a history of previous hospitalization (during the 12 months preceding BSI onset), surgery (30 days preceding BSI onset), invasive procedures (including the insertion of central venous catheters, nasogastric tube, or Foley catheters), endoscopy, endoscopic retrograde cholangiopancreatography, bronchoscopy, parenteral nutrition, mechanical ventilation (during the 72 h preceding BSI onset), or prior exposure to antimicrobial therapy on admission and/or after admission before BSI onset. For infection variables, we considered acquisition type (health care acquired versus community acquired), presentation with septic shock, severity of illness, and source of infection. For treatment variables, we considered the inadequacy of initial antimicrobial therapy. For pathogen variables, we considered species, multiple ESBL production, and multidrug resistance.

Definitions.The following terms were defined prior to data analysis. An ESBL-BSI was defined as a bloodstream infection documented by blood culture positivity (at least one specimen) for an ESBL-producing strain of E. coli, Klebsiella spp., or P. mirabilis in a patient with the systemic inflammatory response syndrome (e.g., fever, tachycardia, tachypnea, and leukocytosis) (40). BSI onset was defined as the date of collection of the first blood culture yielding the study isolate. The source of the BSI was an infection at a distant site caused by a microbial strain identical to the bloodstream isolate. It was determined on the basis of microbiological results and physicians’ findings (16).

BSIs were classified as healthcare acquired when onset occurred >48 h after admission to the study hospital (16, 19). Earlier-onset infections, i.e., <48 h after hospital admission, were also considered healthcare acquired if the patients had been transferred directly from another hospital or nursing home or discharged from a hospital within the 30 days preceding admission to the study (19).

Septic shock was defined as sepsis associated with organ dysfunction and accompanied by persistent hypotension after volume replacement (40). Severity of illness at infection onset was expressed in terms of the acute physiology and chronic health evaluation (APACHE) III score (23). It was calculated on the basis of available clinical data relative to the first 24 h after BSI onset. Neutropenia was defined as an absolute neutrophil count <500 cells/mm³.

Inadequate initial antimicrobial therapy was defined as the initiation of treatment with active antimicrobial agents (identified as such based on in vitro susceptibility testing) >72 h after BSI onset.

Microbiology, ESBL identification, and antimicrobial susceptibilities.Bloodstream isolates were identified at the species level with the VITEK 2 (bioMérieux, Inc., Hazelwood, MO) and/or Phoenix (Becton Dickinson Microbiology Systems) systems. The ESBL status of each isolate was determined by phenotypic analysis with molecular identification of ESBL genes (44, 46, 47). The phenotypic assay consisted in disk diffusion screening with cefotaxime (30 μg) and ceftazidime (30 μg) disks, followed by confirmatory testing with disks containing cefotaxime and ceftazidime combined with clavulanic acid (CA; 10 μg) (from Becton Dickinson). Assays were performed and interpreted in accordance with the Clinical Laboratory Standards Institute (CLSI) guidelines, and K. pneumoniae ATCC 700603 and E. coli ATCC 25922 were used as positive and negative controls, respectively (9). The β-lactamase profiles of all isolates were determined with isoelectric focusing analysis, as described elsewhere (44, 47), and results were interpreted by using published criteria (http://www.lahey.org/studies/wetb.htm). Each isolate then underwent PCR amplification of bla_TEM, bla_SHV, bla_CTX-M, bla_PER, bla_OXA-10, and bla_OXA-2 genes and sequencing of both strands of the PCR products, as previously described (11, 12, 26, 27, 29, 37, 47).

The presence of quinolone resistance genes (gyrA, parC, qnrA, qnrB1, and qnrB2) in isolates with reduced susceptibility to ciprofloxacin (MIC of 0.5 to 1 μg/ml) was determined by PCR using primers and protocols described elsewhere (20, 48, 51).

The MICs of amoxicillin-CA, amikacin, cefoxitin, ciprofloxacin, cefepime, cefotaxime, ceftazidime, gentamicin, imipenem, levofloxacin, meropenem, piperacillin-tazobactam, and trimethoprim-sulfamethoxazole were determined with the E-test (AB Biodisc, Solna, Sweden), as previously described (44). E. coli ATCC 25922, E. coli ATCC 35218, and K. pneumoniae ATCC 700603 were included as quality control strains in all sessions. MICs were classified according to CLSI guidelines (9). In accordance with the same guidelines, all ESBL-producing isolates of E. coli, K. pneumoniae, and P. mirabilis were reported to be resistant to all penicillins, cephalosporins, and aztreonam, regardless of the MIC that emerged for these drugs. Multidrug resistance (MDR) was defined as in vitro resistance to at least one aminoglycoside (i.e., amikacin or gentamicin) or both, to the fluoroquinolones (levofloxacin and ciprofloxacin), and to trimethoprim-sulfamethoxazole.

Statistical analysis.Continuous variables were compared by the Student t test for normally distributed variables and the Mann-Whitney U test for non-normally distributed variables. Categorical variables were evaluated by using the chi-square or the two-tailed Fisher exact test. Odds ratios (ORs) and 95% confidence intervals (CIs) were calculated to evaluate the strength of any association that emerged. Values are expressed as means ± the standard deviation (continuous variables) or as percentages of the group from which they were derived (categorical variables). Two-tailed tests were used to determine statistical significance; a P value of <0.05 was considered significant.

Multivariate analysis was used to identify independent risk factors for 21-day mortality. For this analysis, we used logistic regression and incorporated variables found to be significant in univariate testing.

All statistical analyses were performed by using the Intercooled Stata program, version 8, for Windows (Stata Corp., College Station, TX).

Characteristics of ESBL-producing isolates.Isolated organisms included E. coli (n = 104, 55.9%), K. pneumoniae (n = 58, 31.2%), and P. mirabilis (n = 24, 12.9%). A total of 234 different ESBL genes were identified in these 186 isolates: 104 (44.4%) were bla_SHV, 83 (35.4%) were bla_TEM, and 47 (20%) were bla_CTX-M genes. Forty-six strains carried multiple ESBLs (Table 1).

As shown in Table 2, almost all of the isolates were inhibited by carbapenems and amikacin; 79.3% were susceptible to piperacillin-tazobactam, 55.9% were susceptible to amoxicillin-CA, 53.2% were susceptible to gentamicin, 43.5% were susceptible to trimethoprim-sulfamethoxazole, and 21% were susceptible to ciprofloxacin. Resistance to gentamicin was significantly more common among P. mirabilis and E. coli isolates than among K. pneumoniae isolates. The same was true of ciprofloxacin resistance. Almost half of the K. pneumoniae and E. coli isolates displayed resistance to amoxicillin-CA. Resistance to piperacillin-tazobactam was less common. None of the P. mirabilis isolates were resistant to either of these drug combinations. A total of 49 (26.3%) of the 186 isolates were classified as MDR: 35 E. coli, 6 K. pneumoniae, and 8 P. mirabilis isolates.

Susceptibility findings were reported to physicians 48 to 150 h after BSI onset (74 ± 24 h [mean ± the standard deviation]).

Patient characteristics and initial antimicrobial treatment.The characteristics of the patients with BSI are presented in Table 3. Over half of the patients were suffering from neoplastic disease. A total of 90% had healthcare-acquired infections, and 113 of 186 (60.7%) had histories of recent antimicrobial therapy. The most commonly identified primary infection site was the urinary tract, but the source of the BSI remained obscure in almost half of all cases.

Within a few hours after the index blood culture was drawn, all patients were empirically treated with currently recommended doses (4, 18) of the following antimicrobials: β-lactam-β-lactamase inhibitor combinations (amoxicillin-CA or piperacillin-tazobactam; n = 45; 24.2%), fluoroquinolones (ciprofloxacin or levofloxacin; n = 45; 24.2%), oxyimino cephalosporins (cefotaxime, ceftriaxone, ceftazidime, or cefepime; n = 38, 20.4%), aminoglycosides (amikacin or gentamicin; n = 30, 16.1%), and carbapenems (imipenem or meropenem) (n = 28; 15.1%). In most cases, the choice of the empirical regimen was made by the physician in charge of the patient, and protocols were not used.

In vitro susceptibility testing revealed that the initial therapy was inadequate in 47.8% of the cases (89 of 186). Histories of prior exposure to antimicrobial therapy were significantly more common in this subgroup (67 of 89 [75.3%] versus 46 of 97 [47.4%] of those treated adequately from the outset; OR = 3.37; 95% CI = 1.73 to 6.64; P < 0.001), and in roughly one-third of these cases (23 of 67 [34.3%]), the drug prescribed for the BSI came from the same class of antibiotics previously administered. The ineffective drug was an oxyimino cephalosporin in 38 of 89 (42.7%) of the cases; 29 (32.6%) were treated with fluoroquinolones, 12 (13.5%) were treated with aminoglycosides, and 10 (11.2%) were treated with β-lactam-β-lactamase inhibitor combinations.

After a mean interval of 76 h (range, 72 to 120 h) from the index blood culture, 75 of the 89 patients on inadequate initial antimicrobial regimens were placed on active drugs; the remaining 14 of 89 had died before switching therapy. All decisions on definitive therapy were made with the aid of an infectious disease specialist. Of these 75 patients, 33 (44%) were changed to carbapenems; 22 (29.3%) were switched to a β-lactam-β-lactamase inhibitor combination; 10 (13.3%) were placed on an aminoglycoside; and 9 (12%) were changed to a fluoroquinolone.

21-day mortality rates.Twenty-one days after onset of ESBL-BSI, 71 of 186 (38.2%) of the patients had died: 31.7% (33 of 104) of those with E. coli BSI, 48.3% (28 of 58) with K. pneumoniae BSI, and 41.7% (10 of 24) with P. mirabilis BSI.

Univariate analysis revealed significant differences between the survivor and nonsurvivor subgroups (Table 4). A significantly higher percentage of the latter group had liver disease and histories of previous hospitalization. Nonsurvivors also had significantly higher mean APACHE III scores at infection onset and BSI presentation that included septic shock. Their infections were more frequently caused by K. pneumoniae and organisms that were MDR or carried multiple ESBL genes, the source of infection was more likely to remain obscure, and a higher percentage of the patients were initially treated with an inadequate antimicrobial regimen. In logistic regression analysis, the only two variables independently associated with 21-day mortality were inadequate initial antimicrobial therapy (OR = 6.28; 95% CI = 3.18 to 12.42; P < 0.001) and unidentified primary infection site (OR = 2.69; 95% CI = 1.38 to 5.27; P = 0.004).

In fact, the group that initially received inadequate treatment had a 21-day mortality rate (59.5% [53 of 89]) that was roughly three times higher than that of the group treated from the outset with an active drugs (18.5% [18 of 97]) (OR = 2.38; 95% CI = 1.76 to 3.22; P < 0.001). Within the inadequate-treatment group, the mortality rate among those treated with an oxyimino cephalosporin (22 of 38 patients [57.9%]) was not significantly different from that observed among patients placed on other ineffective antimicrobial regimens (31 of 51 patients [60.8%]) (P = 0.78). Within the former subgroup, higher mortality was observed when the MIC of the oxyimino cephalosporin used for treatment was ≥16 μg/ml than when the MIC was indicative of susceptibility (i.e., ≤8 μg/ml) (60% [21 of 35] versus 33.3% [1 of 3]) (P = 0.36). The three patients in the latter category included one treated with ceftazidime (MIC of 8 μg/ml), who died. The remaining two, who survived, were treated with cefotaxime and cefepime, which displayed MICs of 2 μg/ml for the infecting pathogens, and subsequently they were switched to carbapenems.

For the subgroup of patients whose inadequate initial treatment was corrected based on in vitro susceptibility data, the 21-day mortality rate was 52% (39 of 75), as opposed to only 18% in the subgroup that received adequate treatment within a few hours of BSI onset (OR = 2.18; 95% CI = 1.58 to 3.01; P < 0.001). Analysis of the 97 patients in the adequate treatment group revealed that the 21-day mortality rate was highest among the 16 patients treated from the start with fluoroquinolones (44.4%; OR = 4.05; 95% CI = 1.89 to 8.65; P < 0.001), and the lowest rate was associated with carbapenem therapy (5.5%; OR = 0.14; 95% CI = 0.02 to 1.03; P = 0.01) (Table 5). The eight patients treated with a fluoroquinolone (ciprofloxacin in all cases) who died were infected by pathogens with ciprofloxacin MICs ranging from 0.5 to 1 μg/ml. Four of the eight isolates had a mutation at codon 83 of gyrA, and the qnrB gene was detected in the other four.