Journal of Infection (2007) 55, 254e259

www.elsevierhealth.com/journals/jinf

Mortality following bacteraemic infection caused by extended spectrum beta-lactamase (ESBL) producing E. coli compared to non-ESBL producing E. coli* Mark Melzer a,*, Irene Petersen b a b

Department of Microbiology, Queen’s Hospital, Barking, Havering and Redbridge Trust, Romford, Essex RM7 OAG, UK Royal Free and University College Medical School, Royal Free Hospital, Rowland Hill Street, London NW3 2PF, UK

Accepted 18 April 2007 Available online 14 June 2007

KEYWORDS ESBL producing E. coli; Bacteraemia; Mortality; Death

Summary Objectives: To determine the differences in mortality and length of hospital stay in patients with bacteraemic infection caused by ESBL and non-ESBL producing Escherichia coli. Main outcome measures were mortality, time from bacteraemia to death and length of inpatient stay. Methods: From June 2003 to November 2005, we prospectively collected clinical and microbiological data on all adult patients with E. coli bacteraemia. Results: ESBL producing E. coli caused 16/242 (6.6%) community-acquired and 30/112 (26.8%) hospital-acquired bacteraemic infections. The most common sites of infection were urine 239/ 354 (67.5%) and bile 41/354 (11.6%). All ESBL producers were resistant to cephalosporins. Resistance to ciprofloxacin, trimethoprim, gentamicin and amikacin were 42/46 (91.3%), 39/46 (84.8%), 14/46 (30.4%) and 2/46 (4.3%), respectively. A significantly higher proportion of patients died following a bacteraemic infection caused by ESBL producing E. coli, 28/46 (60.8%), compared to non-ESBL producing E. coli, 73/308 (23.7%). The adjusted odds ratio for death was 3.57 (95% CI 1.48e8.60, p < 0.005). Delay in initiating an appropriate antibiotic was significantly associated with death and ESBL production. There was no significant difference between time from bacteraemia to death (median time 7 days (ESBL þve group) vs 5 days (ESBL ve group); p Z 0.69) and, in those who survived, length of inpatient stay (median time 9 days (ESBL þve group) vs 12 days (ESBL ve group); p Z 0.111).

*

Funding: Research and Development Department, Barking, Havering and Redbridge Trust. * Corresponding author. Tel.: þ44 845 130 4204x3756/6251; fax: þ44 20 8970 5784. E-mail address: [email protected] (M. Melzer).

0163-4453/$30 ª 2007 The British Infection Society. Published by Elsevier Ltd. All rights reserved. doi:10.1016/j.jinf.2007.04.007

Mortality following bacteraemic infection

255

Conclusions: Mortality following bacteraemic infection caused by ESBL producing E. coli was significantly higher than non-ESBL producing E. coli. These findings have serious implications for antibiotic prescription, as cephalosporins are ineffective treatment for many E. coli infections. ª 2007 The British Infection Society. Published by Elsevier Ltd. All rights reserved.

Introduction Escherichia coli is a common cause of bacteraemia. It is often precipitated by bile and urinary tract infections, sterile fluids in close proximity to the gastrointestinal tract. A small number of serotypes cause disease and virulence factors (e.g. antiphagocytic capsules, adhesions, invasins, lipopolysaccharide and toxins) are well described.1 Clinical studies of patients with E. coli infection and bacteraemia have demonstrated a mortality rate of 10e30% with chances of survival increasing with early administration of an effective antibiotic.2 Cephalosporins are the mainstay of empirical antibiotic therapy and, until now, have effectively treated most E. coli infections. In 2003, the UK Health Protection Agency (HPA) confirmed reports that extended-spectrum beta-lactamase (ESBL) producing E. coli was causing infections in different parts of the UK. These were predominantly of the type CTX-M-15. Most strains were resistant to beta-lactams and other classes of antibiotics and, in some cases, only carbapenems and aminoglycosides (eg. amikacin) could reliably be used to treat infections.3e5 The mechanism of multiple-drug resistance via ESBL production is well elucidated6 and recently, the HPA, in collaboration with the British Society for Antimicrobial Chemotherapy, issued laboratory guidelines for detection of ESBL producers.7 In patients with E. coli infection and bacteraemia, quinolone resistance has been reported to be associated with a higher incidence of death.8 The effect of ESBL production on clinical outcomes is less well described9 although one small Scandinavian study reported a high death rate in bacteraemic patients treated with ceftazidime.10 Smaller studies have examined the effect of ESBL production in several Enterobacteriaceae species together,11e18 but rarely in E. coli where one ESBL predominates.19e22 The primary aim of this study was to compare mortality in those with infection and bacteraemia caused by ESBL and non-ESBL producing E. coli. Secondary aims were to determine time to death following bacteraemia and, in those who survived, length of inpatient stay.

the time of bacteraemia, was defined as less than 90/ 60 mmHg and ‘appropriate treatment’ defined as an antimicrobial agent to which E. coli was susceptible by BSAC disc diffusion methods. Delay in appropriate treatment was defined as time from bacteraemia to administration of an appropriate antibiotic and dichotomised as more or less than 24 h. Length of inpatient stay was calculated as time from bacteraemia to discharge home or transfer to a rehabilitation ward. Over the study period, hospital doctors were advised to empirically treat suspected urinary or biliary tract infections with intravenous (i.v.) cefuroxime 1.5 g three times a day (tds) or (i.v.) co-amoxiclav 1.2 g tds. Additional metronidazole (i.v.) 500 mg tds was used for bilary tract infection. Where patients were hypotensive, an additional once daily dose of (i.v.) gentamicin, 7 mg/kg, or (i.v.) amikacin, 15 mg/kg in two divided daily doses, was advised. Where non-ESBL producing E. coli were isolated, treatment was switched to an appropriate antibiotic depending upon susceptibility profiles. All patients with confirmed ESBL producing E. coli infections were treated with (i.v.) imipenem 500 mg four times a day. Patients with bacteraemia secondary to bilary tract infection were referred for ultrasound scan and drainage of infected bile if required. Where infection was secondary to the obstructed flow of urine, a nephrostomy tube was inserted or a blocked urinary catheter changed after (i.v.) gentamicin, 120 mg single dose. Blood culture was analysed using an automated system (MBBacT, biomerieux, France) and E. coli blood stream isolates identified to species level using api20E (biomerieux, France). ESBL production was screened for by cefpodoxime resistance and confirmed by a ‘double leaf appearance’ between appropriately spaced ceftazidime, co-amoxiclav and cefotaxime antibiotic discs. Although some ESBL producing isolates appeared sensitive to cefotaxime and ceftazidime, all cephalosporins were reported resistant.7 Where isolates appeared sensitive to co-amoxiclav and piperacillin/tazobactam resistance was inferred because of reports of variable susceptibility in other ESBL producing coliforms23,24 and our own experience locally. Minimum inhibitory concentrations (MICs) were not performed, in accordance with HPA guidance.7

Methods

Data analyses

The study was undertaken at King George Hospital in Essex, part of Barking, Havering and Redbridge Trust, which serves an ethnically diverse population of 300 000. From June 2003 to November 2005, all patients, aged 16 years and above, with E. coli bacteraemia were seen in the ward by two clinical microbiologists and managed in conjunction with the patients’ medical or surgical team. Demographic and clinical data were prospectively collected on each patient. Hospital-acquired infection was defined as infection acquired more than 48 h after hospital admission. Hypotension, at

Unvariable and multivariable logistic regression was used to estimate risk of death for patients with ESBL and non-ESBL producing E. coli bacteraemia. The multivariable regression model was adjusted for age, sex, site of infection, hospital or community-acquired infection, presence in intensive care, hypotension at the time of bacteraemia, malignancy and neutropenia. Age was divided into quintiles and treated as a categorical variable. The Wald test was used to assess whether death was associated with increasing age. Adjustment for delay in appropriate treatment was not made as

256

M. Melzer, I. Petersen

this was considered to be on the causal pathway.25 The difference in delay in appropriate treatment was determined between ESBL and non-ESBL groups by chi-squared test. We also determined the association between delay and risk of death, irrespective of ESBL production by multivariable logistic regression adjusted for factors mentioned above and ESBL production. Time to death and time to discharge were compared using two-sample Wilcoxon rank-sum tests.

Results Of 354 patient episodes of E. coli bacteraemia, 46 (13.0%) were caused by ESBL producers. Sixteen (6.6%) of 242 were community-acquired and 30 (26.8%) of 112 were hospitalacquired infections. Among the ESBL producers, resistance to ciprofloxacin, trimethoprim, gentamicin and amikacin were 42/46 (91.3%), 39/46 (84.8%), 14/46 (30.4%) and 2/46 (4.3%), respectively. All isolates were sensitive to imipenem. The differences in ESBL and non-ESBL susceptibility profiles are illustrated in Fig. 1. Patient characteristics are summarised in Table 1. The proportion of patients who died within 30 days following a bacteraemic illness caused by ESBL producing E. coli was 28/46 (60.9%), significantly higher than non-ESBL producing E. coli, 73/308 (23.7%); OR 5.01 (95% CI 2.62e9.57, p < 0.001) (Table 2). After adjustment for age, sex, community or hospital infection, malignancy, neutropenia, site of infection, stay in intensive care and hypotension at the time of bacteraemia, the OR for death for ESBL producers was reduced to 3.57 (95% CI 1.48e8.60, p < 0.005) (Table 2). Patients with hypotension at the time of bacteraemia were more likely to die and the risk of death significantly increased with age (Table 2). Risk of death was fourfold higher among patients with undefined sites of infection compared to patients with urinary tract infection (UTI) (Table 2). In patients with UTIs (n Z 239) the adjusted OR for death for ESBL producers was 6.32 (95% CI 1.99e20.09, p Z 0.002). Over three-quarters of patients with ESBL producing E. coli, 36/46 (78.3%), experienced a delay of one or more days in receiving appropriate antibiotic treatment compared to 54/308 (17.5%) of patients with non-ESBL producing E. coli (c2 Z 77.84, df Z 1, p < 0.001). Delay in appropriate

Ampicillin Trimethoprim Gentamicin Ciprofloxacin Cefuroxime Cefotaxime Ceftazidine Amikacin 0

20

40

60

80

100

% % of isolates resistant to antimicrobials

Figure 1 ESBL and non-ESBL E. coli susceptibility profiles. Black bars: non-ESBL E. coli, white bars: ESBL E. coli.

antibiotic treatment was associated with death (OR 3.36; 95% CI 1.59e7.09, p Z 0.001) and the effect of ESBL production on death was not significant (OR 1.81; 95% CI 0.69e3.52, p Z 0.23). This suggests that ESBL production is likely to be associated with increased risk of death via delay in appropriate treatment. There was no significant difference in the median time from bacteraemia to death (ESBL group 7 days vs non-ESBL group 5 days; p Z 0.69) and, in those who survived, median length of inpatient stay (ESBL group 9 days vs non-ESBL group 12 days; p Z 0.11).

Discussion Since the emergence of ESBL producing E. coli of the CTX-M type, this is the first study within the UK to compare clinical outcomes in bacteraemic patients with E. coli infection. After adjustment for host factors, risk of death was significantly higher in the ESBL cohort. Delay in initiating appropriate antibiotic treatment was associated with increased risk of death, and delay in appropriate treatment was more common in the ESBL group. A surprising observation was the apparent lack of difference in length of inpatient stay between the ESBL and non-ESBL patients. Other studies have shown a longer length of stay in ESBL infected cohorts and this finding was counterintuitive. A possible explanation is the high proportion of deaths in the ESBL group, making comparisons of length of stay unrepresentative. There are three explanations for the observed difference in clinical outcomes. A difference in bacterial virulence is one possibility. In the 1980s, in SE London, an outbreak of a multidrug resistant E. coli occurred and the majority of these isolates were typed as serogroup O15.26,27 Infections were predominantly severe and community-acquired but the epidemic strain subsided spontaneously and the source of infection never determined. More recently, two small studies from France28,29 and one from Canada30 investigated the presence of virulence factors in ESBL producing E. coli of the CTX-M type compared to three control groups; CTX-M non-producers, ESBL producers of the TEM and SHV type and non-ESBL producers. Overall, virulence factors in ESBL producers of the CTX-M type were similar or fewer than control isolates. In the UK, serotyping of ESBL producing E. coli of the CTX-M type has demonstrated some clonality,3 predominantly serogroup O25. More work is required to determine whether these isolates are more virulent than non-ESBL producers, in particular, non-O25 isolates. The second explanation is the difference in host factors and co-morbidities. The higher death rate in the ESBL cohort might partly be explained by patient’s greater vulnerability, resulting in more hospital admissions, frequent exposure to antibiotics and increased likelihood of colonisation and infection with ESBL producing E. coli. Thirdly, a delay in administering an appropriate antibiotic due to unanticipated resistance is the final explanation. We demonstrated that delay in appropriate treatment increased the risk of death irrespective of ESBL production. We also observed that more patients in the ESBL cohort experienced longer delays in receiving appropriate antibiotic treatment compared to the non-ESBL cohort. This suggests that delay in appropriate treatment is a likely

Mortality following bacteraemic infection Table 1

257

Characteristics of patients with E. coli bacteraemia

Characteristics

Non-ESBL producing E. coli (n Z 308)

ESBL producing E. coli (n Z 46)

Total (n Z 354)

n (%)

n (%)

n (%)

Age 16e29 30e49 50e70 >70

10 36 62 200

0 5 15 26

10 41 77 226

Sex Male Female

107 (34.7) 201 (65.3)

28 (60.9) 18 (39.1)

135 (38.1) 219 (61.9)

Presence of malignancy

38 (12.3)

10 (21.7)

48 (13.6)

Presence of neutropenia

7 (2.3)

2 (4.4)

9 (2.5)

16 (34.8) 30 (65.2)

242 (68.4) 112 (31.6)

9 (2.9)

9 (19.6)

18 (5.1)

79 (25.7)

23 (50.0)

102 (28.8)

26 7 11 2

239 41 58 16

(3.3) (11.7) (20.1) (64.9)

Community or hospital-acquired bacteraemia Community acquired 226 (73.4) Hospital-acquired 82 (26.6) Patient in intensive care at the time of bacteraemia Hypotension at the time of bacteraemia Sites of infection Urine Bile Not defined Other

213 34 47 14

(69.1) (11.0) (15.3) (4.6)

explanation for the increased risk of death following infection and bacteraemia caused by ESBL producing E. coli. A number of studies have examined whether ESBL production by infecting organisms has an adverse effect on clinical outcomes. They have produced variable results. One study, similar to ours, demonstrated an overall mortality four times greater than that of patients infected with non-ESBL producing organisms.13 Others have found no overall difference. These discordant results might be due to small or underpowered studies, failure to adequately adjust for severity of underlying disease, failure to focus on one Enterobacteriaceae species and differences in antimicrobials used empirically. Our study is the largest to date to focus solely on E. coli whereas others have examined E. coli, Klebsiella pneumoniae or several Enterobacteriaceae together. It was also performed in an area where one ESBL, CTX-M-15, predominates. Unlike some other studies, we defined community and hospital-acquired infections, sites of infection and examined the relationship between delays in administering appropriate antibiotic treatment and clinical outcomes. Outcomes were compared to a suitably selected control group. Our case mix is similar to other district general hospitals so these results are likely to be applicable to other hospital settings where bacteraemic infections caused by ESBL producing E. coli are occurring. There were some limitations to this study. The authors were not blinded to ESBL status, so time from bacteraemia

(0.0) (10.9) (32.6) (56.5)

(56.5) (15.2) (23.9) (4.4)

(2.8) (11.6) (21.8) (63.8)

(67.5) (11.6) (16.4) (4.5)

to death and time from bacteraemia to discharge were used as the main outcome measures to avoid observer bias. Although we adjusted for hypotension and presence on ICU, markers of disease severity at the time of bacteraemia, better validated measures of disease severity, such as APACHE or PITT scores could have been used. Also, further adjustments for co-morbidities might have reduced the odds ratio for mortality between the two groups. Modification of community and hospital antibiotic prescribing for UTIs should be considered as cephalosporins are ineffective treatments for many E. coli infections. For patients with lower UTIs empirical first line treatment with nitrofurantoin rather than trimethoprim or cephalexin might be more appropriate as most uropathogens, including ESBL producing E. coli, are sensitive to nitrofurantoin. Advantages of this strategy include cheaper cost compared to cephalexin and the unlikelihood of selection for ESBL producing coliforms. Disadvantages include failure to treat proteeae, contraindication in renal impairment and failure to adequately treat upper urinary tract infections. Another strategy could be the use of cefepime, a fourth-generation cephalosporin with greater stability than third-generation cephalosporins against many ESBL-producing bacteria. The literature on the use of cefepime against ESBL-producing pathogens is limited31 and there is controversy whether this agent should be used against these organisms. Uncontrolled data suggest that fofomycin may be a suitable and cheap alternative in the treatment of ESBL producing

258 Table 2

M. Melzer, I. Petersen Results of univariable and multivariable logistic regression with death as outcome Univariable regression

Multivariable regression

OR

95% CI

p

OR

95% CI

p

ESBL producer Agea

5.01 e

2.62e9.57 e