Introduction

Escherichia coli is the most common Gram-negative pathogen in humans, and the leading cause of urinary tract infections (UTI) in children and bacteremia and meningitis in infants [1,2,3,4]. The incidence of antimicrobial-resistant E. coli infections is increasing globally [5,6,7] and contributing to poor outcomes [8,9,10] and increased healthcare costs [11]. The emergence of antimicrobial resistance in E. coli has increased faster than the development of new antibiotics, leading to a paucity of therapeutic agents for highly antimicrobial-resistant strains.

Despite the increasing prevalence of antimicrobial-resistant E. coli, risk factors for infection or colonization with antimicrobial-resistant E. coli and outcomes associated with such infections have not been well studied in children. While a few pediatric studies have evaluated infections with extended spectrum β-lactamase (ESBL)-producing strains [12], none have investigated associations with other resistance phenotypes, including resistance to multiple drug classes. Furthermore, these studies are limited by small sample sizes, restricted populations (e.g., neonates or oncology patients) [13] or infectious syndromes (bacteremia), and were published several years ago, prior to the emergence of the pandemic, multidrug-resistant (MDR) E. coli clone, ST131 [14]. Additionally, no recent study has addressed risk factors for community-associated, multidrug-resistant (CA MDR) E. coli infections in children.

Awareness of risk factors for MDR E. coli infections in children would help clinicians more quickly identify and effectively treat children with resistant infections, potentially improving outcomes. Therefore, we determined risk factors and outcomes of MDR E. coli infection among children seen at the Mayo Clinic.

Methods

We conducted a retrospective cohort study of children with E. coli infections who were evaluated at the Mayo Clinic Children’s Hospital in Rochester, MN, USA, in 2012. The Mayo Clinic Institutional Review Board approved this study. This article does not contain any new studies with human or animal subjects performed by any of the authors.

Study Population

Patients were identified by querying the Mayo clinical microbiology laboratory database for all cultures (including polymicrobial cultures) with growth of E. coli from extraintestinal sources (blood, sterile body fluids, urine, respiratory, wound) from children 0–18 years of age, obtained from January 1, 2012 through December 31, 2012. We included only children who had documented MN state research authorization (Minnesota Statute, section 144.295). We excluded children without state research authorization or whose E. coli isolates lacked antimicrobial susceptibility results. If a patient had multiple positive cultures during the study period, the first isolate was included, or sterile sources were prioritized over non-sterile sources (i.e. blood was included rather than urine).

Definitions

An isolate was classified as multidrug-resistant (MDR) if it was resistant to any antibiotic in 3 or more of the following 8 drugs/drug classes: (1) ampicillin/sulbactam, (2) piperacillin/tazobactam, (3) trimethoprim/sulfamethoxazole (TMP/SMX), (4) fluoroquinolones (ciprofloxacin or levofloxacin), (5) aminoglycosides (gentamicin, tobramycin, or amikacin), (6) 1st- or 2nd-generation cephalosporins (cefazolin, cephalothin, or cefuroxime), (7) extended-spectrum cephalosporins (ceftriaxone, cefotaxime, ceftazidime or cefepime), and (8) carbapenems (meropenem or ertapenem) [15]. A pan-susceptible isolate (PS) was susceptible to all of the drugs listed above.

We distinguished infection from colonization among isolates from respiratory, urine, or wound specimens. A urinary E. coli isolate represented a UTI when the patient exhibited all of the following: UTI symptoms such as fever, abdominal pain or dysuria, pyuria (defined as >5 white blood cells/hpf or a positive leukocyte esterase test), and monomicrobial growth of E. coli at >105 CFU per mL. Lower respiratory tract specimens represented infection (pneumonia, bronchitis or tracheitis) when patients had the following: fever, dyspnea or cough, and supportive objective findings (infiltrates on chest imaging or signs of airway inflammation on bronchoscopy). E. coli from wound cultures represented wound infections when patients had accompanying signs of ongoing inflammation from the culture site such as pain, discharge or fever. Patients who had urine, respiratory and wound E. coli isolates, but did not fulfill the above criteria for infection, were considered to be colonized.

Empiric antibiotics were defined as those administered prior to availability of E. coli antimicrobial susceptibility results. An empiric antibiotic was considered effective if it was active against the isolate in vitro and was clinically appropriate for the site of infection (e.g., an oral antibiotic was considered inappropriate for a bloodstream infection). Nosocomial, healthcare-associated (HA) and community-associated (CA) infection were defined as previously described [16, 17]. A complex infection was defined as E. coli infection in any non-urine site or in a child who had any of the following: immune compromise, GU abnormality, upper urinary tract infection or hospitalization. A non-complex infection was defined as E. coli UTI in an immunocompetent child without GU abnormalities (except neurogenic bladder) or upper urinary tract infection, and who was managed as an outpatient [18].

Variables that were abstracted from the electronic medical record as potential risk factors and outcomes (e.g., age, location of care at time of specimen collection, etc.) are shown in Tables 1 and 5.

Table 1 Demographic and clinical characteristics of children with E. coli infection or colonization at Mayo Clinic Children’s Hospital, 2012
Full size table

Statistical Analysis

Continuous variables were summarized with medians and interquartile ranges (IQRs), while categorical features were summarized with counts and percentages. The ability of each feature to distinguish between MDR isolates and non-MDR isolates was evaluated using a univariable logistic regression model. A multivariable model to predict MDR isolates was developed using stepwise selection, with the P value for a feature to enter or leave the model set to 0.05.

Results

Demographic Characteristics

During the study period, 368 children each contributed 1 E. coli isolate. The median patient age was 7 years. Most isolates were community-associated (82%) and collected from outpatients (86%) and females (86%). The predominant specimen type was urine (90%). Infection versus colonization could be evaluated for 346 children; of these, 318 (92%) had infection and 28 (8%) had colonization. Thirty-seven (10%) had 1 or more invasive devices at the time of culture collection (Table 1).

Antimicrobial Susceptibilities

Among all isolates, 194 (53%) were PS, 128 (35%) were resistant to 1–2 drug classes and 46 (13%) were MDR (Table 2). The majority of the MDR isolates were from urine (83%), with the remainder from respiratory (9%), blood (7%) and intraabdominal (1%) sources. Among the 46 MDR isolates, 23 (50%) were resistant to 3 drug classes, most commonly ampicillin/sulbactam, TMP/SMX, and a 1st-generation cephalosporin. Six isolates were resistant to 6 drug classes and 1 isolate was resistant to all 8 drug classes (Table 2). Ten isolates (22%) displayed resistance to extended spectrum cephalosporins.

Table 2 Antimicrobial susceptibility of E. coli isolates from 368 pediatric patients at Mayo Clinic Children’s Hospital, 2012
Full size table

Risk Factors for MDR E. coli Infection or Colonization

Univariable analysis of children with MDR versus non-MDR isolates identified 10 variables that were significantly associated with colonization or infection with MDR E. coli: healthcare acquisition (OR 3.13, 95% CI 1.50–6.55), nosocomial acquisition (OR 3.86, 95% CI 1.14–13.11), cerebrospinal anomaly (OR 3.56, 95% CI 1.45–8.73), genitourinary (GU) tract anomaly (OR 4.68, 95% CI 2.27–9.63), receipt of biologics in the previous 30 days (OR 10.13, 95% CI 2.19–46.82), surgery in the previous 90 days (OR 3.79, 95% CI 1.66–8.6)), presence of an invasive device at the time of culture collection (OR 7.14, 95% CI 3.34–15.30), any hospitalization in the preceding 1 year (OR 3.69, 95% CI 1.94–7.00), and antibiotic use at the time of specimen collection (OR 5.22, 95% CI 2.47–11.06) or within the preceding 3 months (OR 2.25, 95% CI 1.20–4.21) (Table 3). In multivariable analysis, the following variables were independently associated with MDR E. coli infection or colonization: presence of a GU tract anomaly (OR 2.42, 95% CI 1.03–5.68), presence of an invasive device (OR 3.48, 95% CI 1.37–8.83) and antibiotic use at the time of specimen collection (OR 2.62, 95% CI 1.06–6.47) (Table 4).

Table 3 Univariable analysis of risk factors for multidrug-resistant E. coli colonization or infection
Full size table
Table 4 Multivariable analysis of risk factors for colonization or infection with MDR E. coli
Full size table

Similar results were obtained on univariable analysis when children with colonization were excluded, when ampicillin was included as a drug class in the definition of MDR, and when children with MDR E. coli were compared with those with PS E. coli (data not shown). When the subset of patients with CA infections (n = 268) was analyzed, there were 23 (8.6%) MDR specimens and 245 (91.4%) non-MDR specimens. When children with CA MDR infections were compared to those with CA non-MDR infections, hospitalization in the preceding 1 year (OR 3.15, 95% CI 1.25–7.99) was the only risk factor significantly associated with MDR on univariable analysis. When children with CA MDR E. coli infections were compared with those with CA PS E. coli infections, multivariable analysis revealed that GU tract anomaly (OR 5.23, 1.49–18.32) and hospitalization in the preceding 1 year (OR 2.66, 1.06–6.69) were significantly associated with CA MDR E. coli infection.

Outcomes of Infection with MDR E. coli

Compared with children with non-MDR infections, children with MDR infections were more likely to have a complex infection (35% vs. 17%, P = 0.026), less likely to receive effective empiric antibiotics (47% vs. 74%, P < 0.001) and had significantly longer time from culture collection to receipt of effective antibiotics (median 19.2 vs. 0.6 h, P < 0.001). The duration of antibiotic treatment was slightly longer for children with MDR infections compared with those with non-MDR infections (11 vs. 10 days, P = 0.027). The hospital length of stay was longer for children with MDR infections compared with those with non-MDR infections (10 vs. 4 days, P = 0.029). Cure rates did not differ significantly between the groups although children with MDR strains tended to have more recurrences than those with non-MDR strains (35% vs. 19%, P = 0.28) (Table 5).

Table 5 Comparison of outcomes among children with MDR and non-MDR E. coli infections
Full size table

Discussion

We determined risk factors associated with MDR E. coli causing a variety of infectious syndromes by evaluating a contemporary cohort of children, and demonstrated that children with MDR E. coli infections have worse outcomes than those with more drug-susceptible E. coli infections.

We found three independent risk factors associated with MDR E. coli colonization or infection: presence of a GU tract anomaly, presence of an invasive device, and antibiotic use at the time of specimen collection. Our finding that a child with a pre-existing GU tract anomaly is approximately 2.5 times more likely to be colonized or infected with an MDR E. coli strain than a non-MDR strain is consistent with the findings of Topaloglu et al. in their study of risk factors for community-acquired ESBL UTIs (caused by E. coli or Klebsiella species) in Turkish children [19]. Children with GU tract anomalies often require surgical interventions, intermittent urinary catheterization, and antibiotics for prophylaxis or treatment of frequent UTIs. These factors may explain why these children are at high risk for colonization or infection with MDR E. coli. We also found that children with any type of invasive device present at the time of specimen collection were 3.5 times more likely to be colonized or infected with MDR than non-MDR E. coli. These children may have healthcare-acquisition of MDR strains, as invasive devices reflect severe illness and extensive healthcare contact. Additionally, they may be heavily colonized with MDR E. coli strains due to biofilm formation and disruption of the host microbiome [20, 21], severe illness, and antibiotic exposure. Although the association of drug-resistant E. coli and vascular or urinary catheter use has been documented in adult studies [22, 23], this has not been consistently observed in pediatric studies. Zaoutis et al. [12], in a study of children with ESBL-producing Gram-negative bloodstream infections, found that the presence of a central venous or urinary catheter was not a risk factor for ESBL infection. Our inclusion of all types of invasive devices (central venous catheters, ventriculo-peritoneal shunts, etc.) may explain why we found an association between invasive devices and MDR E. coli infection. Lastly, previous broad-spectrum antibiotic use is a well-described risk factor for infection with drug-resistant Gram-negative infections in adults and children [9, 12, 13, 24], and our study provides further evidence for this association.

In the subset of children with community-acquired infections, the above risk factors were not significantly associated with MDR E. coli, likely because our study was underpowered to detect small differences between groups. However, when we excluded children with isolates that were resistant to 1 or 2 agents (and potentially pre-MDR) from this analysis and compared children with CA MDR E. coli to those with CA, pan-susceptible E. coli, GU tract anomalies and prior hospitalization were identified as significant risk factors for MDR infection, likely because there were greater differences in these variables between high-risk and low-risk groups. This suggests that in children, even in community settings, MDR E. coli infections tend to occur in those with comorbidities and healthcare exposure, but are not common among relatively healthy subjects.

Our analyses of outcomes of infection with MDR versus non-MDR isolates revealed that children with MDR infections were more likely to have a complex infection, but that the final outcome of infection did not differ between groups. This is similar to findings by Zaoutis et al. who found no mortality differences between children with ESBL and non-ESBL E. coli and Klebsiella bloodstream infections [12]. In contrast, a study in Taiwan by Tsai et al. [13] of neonates with MDR Gram-negative bacteremia showed higher mortality among those with MDR than those with non-MDR strains. Similarly, a study in Korea by Kim et al. [9] found higher mortality for children with ESBL E. coli and Klebsiella bloodstream infections compared to those with non-ESBL-producing strains. The higher mortality observed in these studies compared with our study may be due to more severe infection in the Asian cohorts (which focused on bloodstream infections and included a large proportion of immunocompromised patients), or geographic differences in antimicrobial resistance. It is also likely that our study was underpowered to detect a difference in mortality between the two groups, given that death is an uncommon outcome of E. coli infections in children in the United States. This is in contrast to studies of adults with E. coli infections which report higher mortalities. We hypothesize that this difference is due to higher prevalence of drug-resistant E. coli in the adult population. This higher prevalence is likely a consequence of more frequent antibiotic exposure and healthcare contact in adults compared to children, a higher number of comorbidities; and inclusion of elderly adults resident in chronic care facilities.

In our study, children with MDR isolates were less likely than children with non-MDR isolates to receive effective empiric antibiotics and had a significantly longer time to receipt of effective antibiotics. While this has been reported among neonates in Taiwan with MDR Gram-negative bloodstream infections, this is the first US study to demonstrate a longer time to appropriate antibiotic therapy among children with MDR E. coli infections. Clinicians may not anticipate MDR E. coli when selecting empiric antibiotics, especially among outpatients, which made up the bulk of our cohort. The longer duration of antibiotic therapy and length of hospital stay noted in children with MDR E. coli infections have also not been demonstrated in other pediatric studies, and may be explained by more severe or complicated infections, or delayed effective therapy among patients with MDR strains.

Our study has the following limitations. We collected data retrospectively which may have led to selection and misclassification biases. Our study was conducted in 2012 and, although our hospital antibiogram has not indicated a change in resistance patterns to suggest that MDR E. coli has expanded more broadly in the community, more recent studies to confirm or refute this are necessary. Urine was the predominant source of E. coli isolates in this study, so our findings may not be applicable to infections or colonization at other body sites. We had few MDR and ESBL isolates and may have had insufficient power to detect differences between some risk factors and outcomes, especially in the subgroup analysis of CA infections. We did not collect data about diet, pets, and household contacts with infection, all of which have been previously implicated in CA E. coli infections [25,26,27,28], as medical records regarding these exposures were incomplete. Although we captured the bulk of antibiotic exposures in our cohort, we may have missed antimicrobials that were prescribed outside of Mayo Clinic. Lastly, our study was conducted in a single geographic region and limited to individuals who provided MN research authorization. Therefore, findings may not be generalizable to other communities or institutions with different patient populations and antibiotic resistance phenotypes.

Conclusion

In conclusion, in our region of the US Upper Midwest, MDR E. coli infections were associated with genitourinary tract anomaly, history of recent invasive devices, and recent antibiotic use, and poor outcomes, including delay in initiation of effective antibiotic therapy and longer hospital stay. Our findings demonstrate the need for strategies to predict, prevent and treat drug-resistant E. coli infections, particularly those caused by MDR strains. Such strategies might include implementation of inpatient and outpatient antibiotic stewardship programs, improved infection control practices, provider education about the epidemiology, outcomes, and risk factors for MDR E. coli infections, and the development of novel diagnostics that enable rapid detection of MDR E. coli.