Methods
Study aim
We aimed to determine the association between the use of frequently prescribed antibiotics and the corresponding resistance rates in E. coli and K. pneumoniae among the clinical departments of a tertiary care hospital.
Study design
We performed a retrospective observational study of antibiotic use and antibiotic resistance rates of E. coli and K. pneumoniae across 18 departments at the Bern University Hospital in Bern, Switzerland, between January 1st 2008 and December 31st 2016.
Hospital setting
Bern University Hospital is a 950-bed tertiary care teaching hospital that covers all medical specialties and includes a 30-bed mixed intensive care unit (ICU). There are around 40′000 admissions annually, resulting in 290′000 patient-days. The following 18 departments were included in the study: abdominal surgery and medicine, cardiology, cardiovascular surgery, critical care, dermatology, general internal medicine, gynecology and obstetrics, nephrology, neurology, neurosurgery, oncology and hematology, ophthalmology, orthopedics, otorhinolaryngology, plastic and hand surgery, rheumatology and clinical immunology, thoracic surgery and pulmonology, and urology.
Data collection
Data was extracted from Anresis [
6], the national surveillance program that collects data on antibiotic resistance and antibiotic consumption in all Swiss hospitals, including Bern University Hospital.
Antimicrobial susceptibility testing (AST) for the following frequently prescribed antibiotics active against wild type E. coli and K. pneumoniae were included: amoxicillin (only active against E. coli), amoxicillin-clavulanic acid, ceftriaxone, cefepime, piperacillin-tazobactam, meropenem, ciprofloxacin, norfloxacin, trimethoprim-sulfamethoxazole and gentamicin. Antibiotics mainly prescribed for perioperative prophylaxis (e.g. cefuroxime) were not included. All E. coli and K. pneumoniae isolates originated from routine clinical diagnostics. AST for isolates from urine was routinely performed only for amoxicillin, amoxicillin-clavulanic acid, quinolones and trimethoprim-sulfamethoxazole; all other isolates were tested for all listed antibiotics. Therefore, the number of tested samples varies for each of the antibiotics. AST was performed at the Institute for Infectious Diseases of the University of Bern and was based on breakpoints for inhibitory zone diameters (disk diffusion method) according to the guidelines of the Clinical and Laboratory Standards Institute (CLSI). Over the study period the approved version of the 9th, 10th and 11th CLSI edition were adopted by the laboratory. Intermediate susceptibility was recorded as resistance. Quinolone resistance was defined as resistance to at least one of the substances in this group (e.g. norfloxacin and/or ciprofloxacin). We included only the last sample per patient per calendar year. In the main analysis, we included only hospital-acquired samples, i.e. samples that had been isolated more than 2 days after hospital admission. In subsequent sensitivity analyses for community-acquired samples, only samples isolated during the first two hospitalization days were analysed. In addition, we repeated the analysis using a stricter definition of hospital-acquired infection including only the bacterial isolates that had been detected more than 5 days after hospital admission.
The hospital pharmacy provided data on antibiotic consumption in numbers of packages delivered to each department from 2008 to 2016. We assumed that deliveries to the departments reflected consumption as all unused antibiotics were returned to the pharmacy and subtracted for our analyses. All departments used the same delivery system without significant storage and without shifting of antibiotics from one department to another. In the Anresis database the aggregated data were converted into defined daily doses (DDD) using the ATC/DDD system promoted by the World Health Organization [
7]. The DDD is the assumed average maintenance dose per day for an antimicrobial used for its main indication in adults.
We obtained annual bed-days per department from the finance and controlling department of Bern University Hospital. In their calculation, the day of admission and the day of discharge were counted together as one bed-day. Annual Antibiotic use in DDD/100 bed-days was calculated for each of the 18 departments.
Statistical analyses
Antibiotic use across the departments was summarized by median, interquartile range (IQR, lower quartile-upper quartile) and total range (minimum-maximum) of the time-averaged data. The change in antibiotic use over time within each department was analysed with a linear regression model with year, department and the interaction of year and department as covariates. The marginal mean per department and the change per year and department are presented as DDD/100 bed-days with 95% confidence intervals (CI).
We used mixed logistic regression to model antibiotic resistance in dependence of consumption. The main analysis was cross-sectional (i.e. consumption and resistance were assessed at the same time) and was restricted to nosocomial strains. We fitted antibiotic-specific models that only considered resistance towards and use of a specific antibiotic and a combined model that considered data from all antibiotics and treated the antibiotic as random effect. The antibiotic-specific model included a fixed effect for antibiotic use and random intercepts for department and year (nested within department). The combined model included a fixed effect for antibiotic use and random intercepts for antibiotic, department (nested within antibiotic), and year (nested within antibiotic and department). Overall resistance proportions and corresponding 95% CI were obtained by marginal predictions. The association of consumption and resistance is expressed as an odds ratio (OR) per each step of 5 DDD/100 bed-days (i.e. referring to the incremental change in the odds for resistance for an increase in consumption by 5 DDD/100 bed-days) with corresponding 95% CI.
We performed four sensitivity analyses. First, resistance was regressed on the antibiotic use of the preceding year to investigate a potential temporal relation. Second, we repeated the analysis for community-acquired bacterial strains to check for a potential spillover of the (hospital) consumption effect. The strength of the association was compared with that found for nosocomial strains by fitting models using both nosocomial and community data with an additional random intercept for the mode of acquisition. We compared models with and without the interaction of antibiotic use and the site of acquisition using likelihood ratio tests; the p-value is reported as p-value for interaction. Third, we performed the analysis using a stricter definition of hospital-acquired infection including only the strains that were obtained more than 5 days after hospital admission in the analysis. Fourth, we performed separate analyses for isolates from urine, blood and other sites of origin. We calculated a p-value for interaction from likelihood ratio tests of models with and without an interaction for consumption and the origin of the sample. All analyses were done with Stata version 15 (StataCorp. 2017. Stata Statistical Software: Release 15. College Station, TX.)
Discussion
Our analysis showed an association between use and resistance for amoxicillin-clavulanic acid, piperacillin-tazobactam, quinolones and trimethoprim-sulfamethoxazole in E. coli and for amoxicillin-clavulanic acid and trimethoprim-sulfamethoxazole in K. pneumoniae across different hospital departments.
It is intuitive that antibiotic resistance emerges predominantly in environments with high antibiotic consumption. This has been shown for individual countries [
8] and individual hospitals [
9‐
11]. It is however conceivable that even antibiotic prescription within individual departments of the same hospital may affect bacterial resistance in the patients sharing the same environment, as bacteria can be transmitted from patient to patient and antibiotic resistance can spread between bacteria. In this regard, the results of our study confirmed the main findings reported by Willemsen et al. demonstrating that a higher use of amoxicillin-clavulanic acid and quinolones was associated with higher antimicrobial resistance in
E. coli.
Interestingly, we only found associations between antibiotic use and resistance for some antibiotics and different associations between E. coli and K. pneumoniae.
A plausible explanation is the considerable difference in the quantitative use of the antibiotics studied across our hospital’s departments. Considering the substantial use of amoxicillin-clavulanic acid at our institution, it is not surprising that its use was associated with resistance in both microorganisms analysed. A wide quantitative range of antibiotic use across different departments may also facilitate the detection of an association, as observed for piperacillin/tazobactam, where the median consumption was low, but the range was very wide.
However, the resistance barrier of the individual antibiotics may also be relevant for the emergence of resistance. For cefepime, which is recommended in our internal antibiotic guidelines [
12] as first-line antibiotic for several nosocomial infections (e.g. pneumonia and fever in neutropenia), we did not find an association between use and resistance, even though it was the second most frequently used antibiotic overall. Cefepime is a robust fourth generation cephalosporin, which is relatively stable against β-lactamases and maintains activity even against difficult organisms. In contrast,
Enterobacteriaceae may acquire resistance to quinolones and trimethoprim-sulfamethoxazole more readily and we indeed observed significant associations between their use and resistance in
E. coli.
Results for K. pneumoniae were less clear than for E. coli, mainly because the sample size was much smaller. In particular, for ceftriaxone, cefepime, piperacillin-tazobactam and gentamicin the number of susceptibility tests performed for K. pneumoniae (n = 369) was substantially lower than for E. coli (n = 1085).
Our first sensitivity analysis examined the potential temporal relationship between the consumption of antibiotics and the development of resistance a year later. However, the last year’s use did not have a stronger influence than the current use. Possible explanations are the relatively stable antibiotic use over time in our setting or a different timing of the emergence of resistance. Even a reverse causation is possible — the reduction in use due to the emergence of resistance.
The antibiotic use varied widely between the different departments and within the individual departments; there were significant variations in antibiotic use over time. Notably, the antibiotic use was high in predominantly surgical departments like urology, plastic and hand surgery, and otorhinolaryngology. Even though cefuroxime, which is recommended for perioperative prophylaxis, was not included in the analysis the high antibiotic use of these departments is probably due to the frequent use of amoxicillin-clavulanic acid and quinolones (urology) for perioperative prophylaxis that in some cases is given longer than suggested by international guidelines. During the study period, there were no specific antimicrobial stewardship interventions in any hospital department. Improving perioperative prophylaxis in the mentioned surgical departments will be a priority in the future.
To address the specificity of the association we compared nosocomial and community strains, given that the latter was not exposed to antibiotic use in the hospital. As expected, the association between antibiotic consumption and resistance tended to be systematically stronger for nosocomial strains than for community strains in
E. coli even though we did not always find significant interactions. This result is consistent with another Swiss study showing that strains isolated ≥48 h after admission were less susceptible than those stemming from the community [
13]. The same pattern was not found for
K. pneumoniae, mainly because we did not observe many clear associations in the first place. Trimethoprim-sulfamethoxazole, for which the association persisted in community-acquired
K. pneumoniae strains, is also frequently prescribed in the outpatient setting [
14]. Furthermore, we could not distinguish patients with previous contacts with the healthcare system and some community-acquired infections could actually be healthcare-associated.
The accuracy of distinguishing community- from hospital-acquired pathogens plays an important role in our analyses. We chose a 2-day cut-off according to the Centers for Disease Control and Prevention (CDC) definition [
15]. A stricter definition for hospital-acquired samples (i.e. a cut-off of five instead of 2 days after hospital admission) did not change any of the main findings, as the vast majority of samples (79% for
E. coli and 81% for
K. pneumoniae) was obtained after 5 days of hospital admission.
In recent years, several studies have analyzed the relationship between antibiotic use and development of resistance in
Enterobacteriaceae. Consistent with our findings, several studies have described an association between in-hospital antibiotic use and resistance to quinolones in
E. coli [
16‐
18]
. This association may be particularly relevant for Switzerland with its comparatively higher quinolone consumption than other European countries [
19].
Two other studies have reported associations in
E. coli between amoxicillin-clavulanic acid [
5,
20] and piperacillin-tazobactam [
21,
22]. The main discrepancy between our results and those of other studies is the absence of an association between the use of ceftriaxone and the emergence of third-generation cephalosporin resistance in our study [
20,
23,
24]. This may be due to a relatively low median use of ceftriaxone of 2.6 DDD/100 patient days compared to other studies.
Our study has several limitations. First, the number of AST available differed considerably among the antibiotics and the sample size was low for some. The majority of samples (65% of E. coli, and 51% of K. pneumoniae) originated from urine and a complete AST of these samples is only performed if routine testing shows several resistances. Therefore, the resistance data for cefepime, ceftriaxone, piperacillin-tazobactam and gentamicin was only available in one sixth of the urine samples. The low sample size reduces the power of the analysis, which was particularly problematic for K. pneumoniae where it severely hampered the conclusions. Moreover testing some antibiotics only in higher resistant isolates may lead to an overestimation of resistance in urinary samples against these antibiotics, due to sampling bias. Second, this was a single-centre study, therefore our results may not be generalizable to other contexts with potential differences in case-mix and epidemiologic characteristics. Third, we could not take into account individual risk factors for infections with resistant microorganisms (such as underlying diseases or previous antimicrobial therapy), because of the anonymization of the samples. Fourth, we did not analyse the association between the use of one particular antibiotic and the resistance to a second antibiotic. Fifth, the antibiotic consumption expressed in DDD per 100 bed-days does not reflect the adequacy of treatment in terms of proper dosage and duration. Sixth, we calculated antibiotic use based on deliveries to the hospital departments rather than administration to the patients. The departments of our hospital are advised to order only those antibiotics that are needed for daily use. Antibiotics that are delivered to the departments but are for some reason not administrated to the patients must be returned to the hospital pharmacy. We subtracted the returned antibiotics from the deliveries and are confident that our data reflect the real antibiotic administration to the patients.
However, to the best of our knowledge, our study is the first comprehensive analysis of the use of the most important antibiotics for the treatment of gram-negative bacteria and corresponding antimicrobial resistance across multiple departments within one institution. We included all clinical E. coli and K. pneumoniae samples isolated across 18 departments at our institution over the study period of 9 years and used the well-established, standardized ATC/DDD system. The study therefore has important relevance for clinical practice.