Suspicion of respiratory tract infection with multidrug-resistant Enterobacteriaceae: epidemiology and risk factors from a Paediatric Intensive Care Unit

A retrospective, single-centre analysis was performed in ventilated patients admitted to the PICU of the University Children’s hospital in Tübingen during the period from 2005 to 2014. This 12-bed PICU cares for critically ill infants and children with around 860 admissions per year. The main reason for admission is requiring cardiac surgery (46%), followed by general paediatric surgery (28%) and paediatric medical conditions that require intensive care treatment (22%). About 40 patients per year (5%) are transferred from long-term care facilities and have chronic conditions that require long-term tracheostomy. The organisational structure of this ward remained unchanged during the study period. Foreign-born patients from Eastern Europe and Senegal admitted for surgical and cardiothoracic procedures are an increasing part of the patient population in this PICU. Mechanically ventilated patients aged below 18 years with a tracheal aspirate positive for Enterobacteriaceae were identified from the institution’s Microbiology Database (HyBase Database, Cymed) and included in the study. Long-term ventilated children with tracheostomy were excluded. Medical records of these children were reviewed and demographic, clinical and microbiological data were extracted. Data obtained included age, sex, gestational age, birth weight, weight, height and BMI on admission. As exposures, we collected the following underlying conditions: pulmonary, cardiosurgical, neurological, gastroenterological, haematooncological disease and immunodeficiency, ventilated days before infection, days on ECMO and presence and days of a central venous line in situ before infection, duration of catecholamine therapy and days of antibiotic therapy in the 30 days before the positive culture result. Diagnosis of ventilator-associated pneumonia was made at the time of the positive tracheal culture result and defined according to the CDC criteria [12]. Secondary outcomes after infection were total ventilation days, PICU length of stay and all-cause mortality after 6 months. Microbiological data included the isolated organism in tracheal aspirate and a resistogram.

On our PICU, children admitted from long-term care facilities, children with previous hospitalization within the last year or from high-prevalence countries are screened routinely for bacteria and resistances on admission. Screening includes a nasal swab for Methicillin-resistant Staphylococcus aureus (MRSA) since 2004 and a rectal swab for Vancomycin-resistant Enterococci (VRE) since 2012. During routine patient care, tracheal aspirates were taken from ventilated patients with clinical findings suggesting infection (including new onset of fever, rise in inflammatory markers or decline in oxygenation) or when tracheal aspirate had a purulent appearance. In case of isolation of an MDR organism, hygiene measures were set up according to the guidelines of the German Commission for Hospital Hygiene and Infectious Disease Prevention at the Robert Koch Institute and the local Hospital Hygiene Plan [22]. Isolation in a single room, contact and droplet precautions were performed to prevent transmission. To promote proper hand hygiene, an alcohol disinfectant is offered at each patient’s bedside. Patient to healthcare worker ratio was between 2:1 and 3:1.

The mechanical ventilation system including humidification was set up by a Paediatric Intensive Care Nurse following standard procedures. Ventilatory parameters were set according to physician’s orders, adjustments were guided by routinely obtained blood gas analysis. Open or closed suctioning was performed as clinically indicated.

Specimens were collected through deep suctioning with the catheter passing beyond the endotracheal tube tip into the trachea or bronchi. The aspirate was routinely analysed in our hospital Microbiology Laboratory. Analysis was performed according to the local guidelines of the hospital Microbiology Laboratory. Organisms from tracheal aspirates were grown on agar plates or with liquid culture technique and incubated at 37 °C. Growth of organisms was monitored according to the local protocol. Susceptibility testing was realized by disc diffusion technique or the automated rapid susceptibility test system VITEK 2 (bioMérieux). To identify the organisms, the isolates underwent MALDI-TOF analysis or biochemical identification with the automated VITEK 2 system. Clinical breakpoints recommended by the European Committee On Antimicrobial Susceptibility Testing (EUCAST) were used to define susceptibility and resistance [23].

Infection was defined as the recovery of an Enterobacteriaceae isolate from the usually sterile respiratory tract. In this setting it was only possible to use the term “infection” to refer to both the situation in which infection was suspected with Enterobacteriaceae recovered from tracheal aspirate as well as to a respiratory disease (e.g., mild inflammation of the airways, tracheitis, ventilator-associated pneumonia etc.) that was attributed to the isolated microbe [24]. Resistance status was determined for each isolate according to the definition proposed by the European Society of Clinical Microbiology and Infectious Diseases for interim standard definitions for acquired resistance [25]. MDR was defined as acquired non-susceptibility to at least one agent in three or more antimicrobial categories with respect to intrinsic resistances.

To avoid bias due to multiple isolates in one patient, only the first isolate per patient was included. Patient data were analyzed using IBM SPSS Statistics Version 22 for Windows. Missing data points or data that were not applicable to the analysis were excluded. Statistical analysis was performed in consultation with the Department of Statistics/Biometrics of the University of Tübingen. Categorical variables between groups were compared with the [chi]² test. Means of the two different groups were evaluated by two sample unpaired t-test if continuous variables were normally distributed. For intergroup comparison of continuous variables that were not normally distributed, the nonparametric Mann–Whitney U-Test was used. A p-value <0.05 was considered statistically significant. Results are presented as numbers for categorical variables. Normally and abnormally distributed quantitative variables are presented as mean ± standard deviation and median (minimum and maximum or interquartile range), respectively.

Various studies have been performed to study risk factors for infection with MDR organisms in PICU patients [26‐28]. Ten candidate risk factors, clinically relevant for infection with MDR organisms and available to study in our PICU setting, were selected from the literature [27‐30]. These were patient age, duration of mechanical ventilation before infection, days of antibiotic pre-exposure, duration of presence of a central venous line, duration of catecholamine therapy, pre-existing gastroenterological, cardiac and pulmonary disease, length of PICU stay before culture and days on ECMO. In order to select suitable predictor variables for a final multivariable logistic regression model, we undertook a stepwise approach: First, a classification and regression tree (CRT) was employed to select the main risk factors for infection with MDR Enterobacteriaceae. The CRT approach is particularly applicable to find specific subgroups, relationships and cut-offs of continuous variables in larger sets of predictor variables that might not be detected with more common methods, often used in similar analyses (e.g. multivariable regression equations). These cut-off points can then be used to transform continuous into categorical variables for further analysis [31, 32]. Hence, we used CRT to select the input set of variables and to potentially find optimal cut-off points for categorisation of continuous variables in the following analysis. Second, we validated the findings with the common method of simple logistic regression, since the CRT model is a relatively uncommon statistical method. Odds ratio (OR) and 95% confidence interval (CI) were determined for each of the 12 risk factor variables. Third, variables found by the CRT model as well as variables with a p-value of <0.2 in the univariate analyses were selected as candidates for a multivariable logistic regression model. A stepwise backward selection process was used. In each step the variable with the least significant effect was eliminated. A cut-off of α _crit   < 0.1 (“p-to-remove”) was set as a limit for removal of variables from the model. Statistics were all performed using SPSS (IBM SPSS Statistics Version 22 for Windows). A thorough explanation of the decision tree procedure, the model criteria and the command file for reproducing the classification tree are provided as Additional file 1 [33].

During the whole study period, we obtained 167 Enterobacteriaceae isolates from the lower respiratory tracts of 123 intubated patients. 116 (69%) isolates were susceptible and 51 (31%) of all isolates were identified as MDR Enterobacteriaceae. The spectrum of isolates is shown in Additional file 3. Enterobacter spp. were the most prevalent genera (51 isolates, 30.5%), followed by Escherichia coli (47 isolates, 28.1%) and Klebsiella spp. (46 isolates, 27.5%). Most MDR organisms were E.coli (26/47 isolates), followed by Klebsiella spp. (13/46 isolates), Enterobacter spp. (7/51 isolates) and Morganella (4/6 isolates). The total number of ESBL producing organisms was 13 (7.8%) of all isolates. 7 out of 47 (14.9%) E.coli and 3 out of 46 (6.5%) Klebsiella spp. were ESBL-producing isolates.

Patient characteristics and the clinical outcomes of 123 intubated children with Enterobacteriaceae infection are shown in Tables 1 and 2. Patients with MDR organisms did not differ from patients infected with susceptible organisms in sex, age, percentage of infants, gestational age, birth weight, BMI, presence and days with CVC in situ and days of antibiotic exposure during the last 4 weeks prior to the tracheal aspirate. Underlying diseases and conditions were similar between both groups with the exception of immunodeficiency, which was significantly more frequent in patients infected with an MDR organism. The clinical outcome was similar in children infected with an MDR versus a susceptible strain: children infected with Enterobacteriaceae had a median total ventilation time of 6 [IQR 3–19] and 8 [IQR 2–16] days, respectively. PICU length of stay was also similar in patients with MDR and susceptible strains (14 and 14.5 days). VAP occurred in 18 cases, resulting in a VAP incidence rate of 14%. With 9 VAP cases in both groups, this results in a VAP incidence rate of 11% among patients who were infected with a susceptible organism. Incidence rate of infection (21%) and all-cause mortality after 6 months (20%) were almost doubled in patients infected with MDR organisms. However this difference did not reach significance (p = 0.15, p = 0.22) (Tables 1 and 2).

The CRT analysis (Fig. 1) revealed that length of antibiotic pre-exposure and presence of gastrointestinal comorbidity were the most relevant predictors for infection with an MDR strain. In this study, 62% of Enterobacteriaceae isolates were MDR if the duration of antibiotic pre-exposure was ≥7 days. Furthermore, the analysis revealed gastrointestinal comorbidity as a second predictive factor for MDR: In individuals infected with Enterobacteriaceae and a short duration of antibiotic exposure (<7 days), isolation of MDR strains was more likely if gastrointestinal comorbidity was present (40%). Conversely, isolation of susceptible strains was most likely in individuals without gastrointestinal comorbidity and with a short duration of antibiotic pre-exposure (<7 days) (85%). When evaluating the entire CRT model with both predictive parameters, the overall percentage of correct prediction was 71%, the risk of misclassification was 29%. Classification of susceptible strains was far more accurate (86% classified correctly) than classification of MDR strains alone (42% classified correctly). Normalised importance of the factors revealed days of antibiotic pre-exposure as the most relevant factor for MDR (100%), followed by gastrointestinal comorbidity (47%) (Additional file 4). These most relevant factors were included into a multivariable logistic regression analysis for further validation. According to the results of the CRT, antibiotic exposure was transformed into a categorical variable (antibiotic exposure for ≥7 days) for logistic regression analysis. Simple logistic regression analysis revealed antibiotic exposure for ≥7 days as a significant risk factor for infection with an MDR organism (OR 4.25; 95% CI 1.62-11.14). Gastrointestinal comorbidity did not reach significance as a potential risk factor for MDR in this analysis (OR 1.97; 95% CI 0.93–4.18) (Additional file 5). Variables found by the CRT model and variables with a p-value of <0.2 in the univariate analyses were selected as candidates for a multivariable logistic regression model. Candidates included antibiotic exposure for ≥7 days, gastrointestinal comorbidity and days on ECMO. After a backward elimination process, antibiotic exposure for ≥7 days and gastrointestinal comorbidity remained the most important risk factors for isolation of MDR Enterobacteriaceae. However, when entered into a multivariable logistic regression model (Table 3), gastrointestinal comorbidity did not reach significance as a predictive factor (OR 2.3; 95% CI 0.92–5.77). Antibiotic exposure for ≥7 days remained the only significant factor predicting MDR status in infected patients (OR 4.56; 95% CI 1.69–12.30).