Community-onset bloodstream infection with multidrug-resistant organisms: a matched case-control study

Background Multidrug-resistant (MDR) organisms have been increasingly reported at hospital admission. Recognising the magnitude, trend and predictors for MDR organisms in community-onset bloodstream infections (COBSI) is crucial for guiding empiric antibiotic prescribing. Methods Positive blood culture isolates recovered from patients presenting to the emergency department during a ten-year period (1st Jan 2002-31st Dec 2011) were assessed. Trend analyses of MDR organisms were performed. Risk factors for COBSI caused by an MDR organism and predictors for 30-day mortality were also determined. Results A total of 1721 positive blood culture isolates were identified during the study period with a yearly incidence of 30-43 isolates/10 000 ED presentations. The proportion of MDR Escherichia coli causing COBSI increased from 9%-26% (P < 0.001), whilst methicillin-resistant Staphylococcus aureus remained at high levels (20%-30%). A total of 360 patients were included in a matched case-control (1:1) study, and residents in long-term care facilities (adjusted odds ratio [AOR], 4.9 [95% CI, 2.1-11.6]), home wound care (AOR, 5.5 [95% CI, 1.6-18.7]), underlying immunosuppression (AOR, 3.5 [95% CI, 1.6-7.7]), recent surgery (AOR, 3.5 [95% CI, 1.1-11.6]), and exposure to antibiotics within 3 months (AOR, 5.5 [95% CI, 2.8-10.6]) were independently associated with MDR COBSI. High risk source of COBSI, age and Pitt bacteraemia score were independent predictors for 30-day mortality. Conclusions A concerning trend in MDR organisms causing bloodstream infection from the community is occurring. Risk factors for MDR organisms have been identified to assist in empiric antibiotic prescribing for those presenting to hospital with sepsis.


Background
Antimicrobial resistance in common bacterial pathogens causing community-onset infections, including bloodstream infection (BSI), has increasingly been reported, most notably methicillin-resistant Staphylococcus aureus (MRSA) and various multidrug-resistant (MDR) gram-negative bacilli (GNB) [1,2]. This poses significant challenges to clinicians caring for patients presenting to hospital with suspected sepsis as empiric antibiotic prescribing is often targeting presumed, antibiotic-susceptible, community organisms. Identification of risk factors for MDR organisms in patients presenting from the community with sepsis is necessary to help optimise patient outcomes and minimise the use of broad-spectrum antibiotics.
Emergency departments (EDs) serve as an important interface between community and hospital care; the microbial ecology is often a combination of community-and healthcare-associated organisms, the latter being driven by healthcare re-admissions. Friedman et al. proposed that healthcare-associated (HCA) bloodstream infection should be treated with broad-spectrum antibiotics to cover for MDR organisms [3]. However, the broad nature of the HCA risk factors may lead to unnecessary use of last-line antibiotics, especially in settings with a low prevalence of MDR organisms [4]. Several studies have worked towards refining predictors for MDR organisms in communityonset pneumonia [5][6][7]; similar work to predict MDR organisms in community-onset bloodstream infections (COBSI) remains scant.
To date, the prevalence and trend of major MDR organisms causing COBSI remain poorly understood. Furthermore, data on risk factors and outcomes associated with COBSI due to MDR organisms are limited. Accordingly, we explored the longitudinal trend, risk factors and clinical outcomes associated with MDR organisms causing COBSI in a tertiary care institution. These data are critical in devising the most effective empiric antibiotics to maximise patient outcomes.

Study design and population
We performed an observational series and a matched case-control study at The Alfred, a tertiary referral hospital in metropolitan Melbourne, Australia. We studied all cases of bloodstream infection presenting from the community to The Alfred's ED. This ED provides specialised state-wide services for major emergency and trauma, burns, heart and lung transplantation, cystic fibrosis and neurosurgery patients. It attends to over 55 000 patients annually. The study was approved by the human research ethics committees of The Alfred and Monash University.

Data collection and definitions
Data collection was broadly divided into two parts:

i) Prevalence and temporal trends of MDR organisms
Positive clinical isolates recovered from blood cultures of patients within 48 hours of ED admission during a ten-year period (1 st January 2002 and 31 st December 2011) were examined from a microbiology laboratory database. Repeat cultures from an individual patient growing the same organism within 14 days, cultures reported as normal flora or probable contaminants (eg. coagulase-negative staphylococci from a single blood culture bottle), and clinical specimens from patients aged < 18 years, were excluded. Species identification and antimicrobial susceptibility testing were performed using an automated system (Vitek®; bioMérieux). For the majority of the study period (Jan 2002-Jun 2010), susceptibility breakpoints were based on the British Society for Antimicrobial Chemotherapy (BSAC) guidelines [8]. These were changed to the European Committee on Antimicrobial Susceptibility Testing (EUCAST) guidelines from July of 2010 [9]. We observed no change in the proportion of resistant isolates at the time of the change in breakpoint definitions.
Multi-resistant MRSA was defined as MRSA resistant to three or more of the following antibiotics: erythromycin, ciprofloxacin, gentamicin, trimethoprim/sulfamethoxazole, tetracycline, rifampicin, fusidic acid and vancomycin [12]. The EMRSA-15-like strain was defined as MRSA resistant to ciprofloxacin +/-erythromycin only [12]. All other MRSA were non-multi-resistant MRSA. For the Streptococcus pneumoniae strains, the susceptibility interpretation was based on non-central nervous system breakpoints. ii) Risk factors and outcomes associated with MDR COBSI A matched case-control study was subsequently conducted for risk factor analysis. A case was defined as an adult (≥ 18 years) with COBSI (organism isolated from ≥ 1 blood culture within 48 hours of ED admission) due to a MDR organism. Only the first admission with COBSI due to a MDR organism was included in the analysis. Control patients were adults with COBSI caused by non-MDR organisms matched for the genus of the causative organism and the date of ED admission (± 1 year). If more than one control was identified, the control patient with the ED admission date closest to the case patient was chosen. Cases of polymicrobial bacteraemia, and patients who were directly transferred from other hospitals with a hospital stay > 48 hours were excluded. Information on patient demographics, sources of BSI, co-morbidities, recent bacterial infection and antibiotic use, invasive instrumentation in the preceding 48 hours, recent surgery, underlying immunosuppression, and the presence of healthcareassociated risk factors were reviewed. Sources of BSI were divided into low risk sources including urinary, skin and soft-tissue, and catheter-related, and high risk sources including respiratory, intra-abdominal, bone and joint, cardiovascular or unidentified [13]. Co-morbidities were quantified using the Charlson comorbidity index [14]. Presence of underlying immunosuppression was defined as the presence of neutropenia (absolute neutrophil count of < 500/mm 3 ), concurrent use of high dose oral corticosteroids (at least 10 mg of prednisolone or equivalent for minimum of 30 days) or long-term use of immunosuppressive agents (e.g. cyclosporine), active chemotherapy for malignancy, metastatic cancer, or acquired immunodeficiency syndrome. Healthcareassociated (HCA) BSI was defined according to Friedman et al. [3]. This included patients who had been hospitalised in an acute care hospital for ≥ 2 days in the preceding 90 days, residents of a nursing home or long-term care facility (LTCF), patients who received home intravenous (IV) therapy, specialised wound care, and those who underwent chronic haemodialysis or IV chemotherapy in the preceding 30 days. Information related to clinical presentation, management and outcomes of BSI was also collected, including severity of BSI according to the Pitt bacteraemia score [points assigned to specific clinical observations as (1)  , presence of severe sepsis or septic shock, types and duration of empiric antibiotic therapy, length of hospital stay and mortality within 30 days of COBSI. Empirical antibiotic therapy was considered active when an antibiotic was given during the first 48 hours that had in vitro activity against the causative organism.

Statistical analysis
Microbiological data were available from 2002 and denominator data on the number of ED presentations per year were available from 2004 onwards. A rate was calculated based on the number of isolates per 10 000 ED presentations per year. Bi-yearly trends in distribution of major organisms and antimicrobial resistance patterns were assessed using the chi-squared test for trend or linear regression. Risk factor analysis was performed using conditional logistic regression analyses, firstly by univariate, followed by multivariate analysis. Variables with a P < 0.1 on univariate analysis were eligible for entry into the multivariable model. The variables of 'invasive instrument in-situ' , 'underlying immunosuppression' and 'prior exposure to antibiotic' were considered for inclusion in the multivariable model as binary variables, ignoring constituent variables within the categories. For correlated variables (pairwise correlation coefficient > 0.7), only one of the covariates was selected for inclusion into the candidate models on the basis of the strength of association. Independent variables in the final multivariable model were selected using a stepwise forward process. Model fit was assessed against the number of variables used, quantified using the Akaike information criterion to prevent overfitting. Risk factor analysis for COBSI caused by any MDR organism was performed first followed by a risk factor analysis for specific MDR organisms (MDR GNB and MRSA).
In addition, categorical and continuous variables associated with 30-day all cause mortality were compared between the survival and non-survival groups using univariate and then multivariate logistic regression analyses. All tests were 2-tailed, and P-values of < 0.05 were considered statistically significant. Data analyses were performed using IBM SPSS Statistics version 20 (IBM, New York, USA).

Temporal trend of MDR organisms causing COBSI
A total of 1721 positive blood culture isolates were identified over a 10-year period from patients presenting to the ED. From 2004 onwards, there was a yearly incidence of 30-43 BSI per 10 000 ED presentations (Table 1). Overall, gram-negative organisms (n = 941, 55%) predominated (Table 1). When adjusted for the number of ED presentations over time, the yearly incidence rate of grampositive isolates declined over time (P = 0.02 for trend), whilst gram-negative isolates remained relatively constant (P > 0.05 for trend).
The proportion of MDR organisms according to genus is shown in Figure 1. Overall, Escherichia coli was the most common causative gram-negative organism. The proportion of MDR E. coli (n = 82) increased significantly during the study period from 9% (2002-2003) to 26% (2010-2011) (P < 0.001 for trend). Increasing resistance to important first-line agents was also observed from years 2002-2003 to 2010-2011; ceftriaxone (1% to 6%, P < 0.01), gentamicin (1% to 12%, P < 0.001) and ciprofloxacin (3% to 22%, P < 0.01). In contrast, the incidence of MDR Klebsiella spp. (n = 11) and other Enterobacteriaceae (n = 20) did not show a significant increase over time; however this may be partly explained by the small number of isolates. As expected, BSI caused by non-fermenting GNB were less common; however there were still 51 and 12 cases of Pseudomonas aeruginosa and Acinetobacter spp. in COBSI, respectively. MDR cases were also observed (2/51, [4%] for P. aeruginosa and 3/12, [25%] for Acinetobacter spp.).

Risk factors for COBSI due to MDR organisms
There were 194 MDR blood culture isolates identified from 2002-2011. Fourteen isolates were excluded; polymicrobial  Table 2. Sources of COBSI due to MDR organisms composed primarily of urinary sites (n = 60, 33%) and unidentified sources (n = 53, 29%), followed by catheter-related (n = 18, 10%), skin and soft tissue (n = 16, 9%), bone and joint (n = 13, 7%), intra-abdominal (n = 10, 6%), and respiratory sites (n = 10, 6%). Overall, 82% of the case patients met the Friedman et al. criteria [3] for HCA BSI, however 40% of the control patients also had at least one of these defined risk factors. Multivariable analysis showed that residents in LTCF, home wound care, underlying immunosuppression, recent surgery and prior exposure to antibiotics were independent risk factors for COBSI caused by a MDR organism (Table 2). Subgroup analyses for specific risk factors for MDRGNB and MRSA are shown in Table 3. In addition to being a resident in a LTCF and prior exposure to antibiotics, home IV therapy and younger age were also independently associated with COBSI due to MDR GNB. Seven of 18 (39%) MDR GNB case patients with an age < 40 years had no identified predisposing risk factors and all but one were COBSI associated with a urinary source. On the other hand, hospitalisation in the last three months, residing in LTCF, receiving home wound care and underlying immunosuppression were independently associated with COBSI caused by MRSA. A significantly higher proportion of COBSI cases with MRSA (63/67, 94%) compared to cases with MDR GNB (81/109, 74%) fulfilled at least one Friedman et al. criteria (P = 0.001).

Clinical management and outcomes of COBSI
Community-onset BSI caused by MDR organisms was significantly more likely to be treated with inactive empirical antibiotic therapy (42% vs. 1%, P < 0.001). The median length of hospital stay for those infected with MDR organisms was 11 days (interquartile range [IQR] 6-20) compared with 9 days (IQR 5-17) (P < 0.05) for those with non-MDR organisms. The overall 30-day mortality observed in both groups was 13% (46 / 360), with 16% (n = 28) in those infected with MDR organisms as opposed to 10% (n = 18) in non-MDR infections (P = 0.127). The prognostic factors for 30-day mortality are shown in Table 4. Older age, higher Pitt bacteraemia score, and high risk sources of bacteraemia were independent predictors for mortality within 30 days of COBSI. Community-onset BSI caused by a MDR organism, receipt of inactive empiric antibiotic therapy and COBSI defined as HCA BSI were not significant predictors.

Discussion
This study has shown that MDR GNB have joined MRSA as an important cause of bloodstream infection presenting from the community, highlighting the evolving challenges of empiric antibiotic prescribing for severe infections at the front-line of hospital admission. We identified differences in risk factors for MDR GNB and MRSA that are in keeping with the changing epidemiology of these organisms. The increase in MDR GNB was largely attributable to MDR E. coli, whereas MDR in non-fermenting GNB remained relatively uncommon. Traditional risk factors for infection with an MDR organism such as HCA risk factors proposed by Friedman et al. [3], appeared predictive for MRSA COBSI but were less predictive for MDR GNB infection. Patients with COBSI due to an MDR organism were more likely to be prescribed inactive empiric therapy and were associated with longer hospital stays; nonetheless, COBSI with an MDR organism was not related to significantly higher 30-day mortality.
In the last decade, emergence of MDR GNB at hospital admission has been reported [1,15,16]. The most recent Australian surveillance data revealed an emerging trend of MDR E. coli in urinary isolates from the community, rising from 4.5% in 2008 to 7.2% in 2010 [17].
Apart from the known increase in MDR GNB causing community urinary tract infections, fewer studies have focused on the trends in more severe community infections like bloodstream infection [1]. Our data showed that a staggering 26% of E. coli bloodstream infection presenting to ED was MDR. Our standard first-line empiric therapy to cover GNB for suspected sepsis is often an aminoglycoside or a third-generation cephalosporin [18]; however 6% and 12% of blood culture isolates were resistant to these agents, respectively. Furthermore, 22% were resistant to ciprofloxacin. Identifying risk factors to help determine who should receive broader-spectrum antibiotics for patients presenting to ED with suspected BSI is needed.
In contrast to previous studies [16,19], older age was not shown to be a predictor for acquiring MDR organisms; on the contrary, older patients were less likely to have a COBSI with MDR GNB. Our data suggest a potential shift in the epidemiology of MDR GNB to a younger patient population, and this was not limited to patients with underlying medical conditions or predisposing healthcareassociated risk factors. This trend warrants further investigation into other potential community-acquired risk factors in this patient group such as travel and close contacts. A study by Kennedy and Collignon shows that approximately 50% of Australian travelers returning from high incidence countries such as India and China were colonised with drug-resistant E. coli and can remain colonised for months [20]. As such, specific questioning about travel is now recommended for all patients with presumed infections at ED admission, with modification of empiric antibiotic prescribing if positive.