Study design and data source
We undertook an observational cross-sectional study to analyse carbapenem non-susceptibility of K. pneumoniae isolates in hospitals in Germany.
Data on antimicrobial susceptibility testing are obtained from ARS. Participating laboratories share their data on routine antimicrobial susceptibility testing (AST) of microbiological samples from hospitals and medical practices . Participation is voluntary for the laboratories and can change over time. The sample of hospitals and medical practices providing data to ARS is organised by laboratory clusters. The geographical distribution of the hospitals contributing data to the study is shown in Additional file 1: Appendix S2. The identity of the hospitals and medical practices is kept confidential. Data on patients is anonymised.
The laboratories identify bacteria from specimens sent in from hospitals or medical practices and determine the zone diameters or minimum inhibitory concentrations (MIC) of routinely used antibiotics (e.g. with microdilution, gradient or disk diffusion). Based on international guidelines (e.g. by the European Committee on Antimicrobial Susceptibility Testing (EUCAST)), the zone diameter or MIC are used for the interpretative results “susceptible” (S), “intermediate” (I), or “resistant” (R) against the tested antibiotic.
All participating laboratories are accredited to perform pathogen identification and antimicrobial susceptibility testing. Data are checked for plausibility during the data transmission process and are validated by the laboratories annually for completeness and consistency.
Outcomes and covariates
The main outcome of the study is the proportion of carbapenem non-susceptible K. pneumoniae isolates in relation to all K. pneumoniae isolates tested for carbapenem resistance. An isolate is considered non-susceptible against an antibiotic if the susceptibility test results are interpreted as “resistant” (R) or “intermediate” (I). Age was converted into a categorical variable for the analysis (< 1, 1–19, 20–39, 40–49, 50–59, 60–69, 70–79, and 80+ years). The specimen types were grouped as follows: swabs (swabs from eye, nose, throat, ear, tongue, and urogenital sites as well as intraoperative swabs and other/unspecified swabs), blood (blood cultures), puncture (tissue biopsy, cerebrospinal fluid, and aspirate from pleural cavity, abscess, ascites, or joint puncture, other punctures), urine (urine samples), wound (swabs from wounds and abscesses), respiratory (bronchial lavage, bronchial secretions, sputum, tracheal secretion, other respiratory samples), other (dialysate, ejaculate, catheters, other). To analyse for seasonality, a categorical variable was created according to the month in which the isolate was obtained: January – March, April – June, July – September, October – December. The geographic regions were grouped as follows: Northwest (Bremen, Hamburg, Lower Saxony, Schleswig-Holstein), West (North Rhine-Westphalia), Southwest (Baden-Wuerttemberg, Hesse, Rhineland-Palatinate, Saarland), Southeast (Bavaria, Saxony, Thuringia), and Northeast (Berlin, Brandenburg, Mecklenburg-West Pomerania, Saxony-Anhalt). Several variables on the county level of the hospital were also included in the analysis: counties were divided into “rural” or “city” based on a list from the Federal Agency for Cartography and Geodesy [12, 13]. Moreover, the social deprivation index per county was derived from the German Index of Socioeconomic Deprivation (GISD) [14, 15]. The GISD uses nine indicators from publicly available administrative datasets. It is based on factor analysis for indexing and weighting of the indicators to the three latent dimensions education, occupation and income. For the analysis, a categorical variable was created dividing the counties by social deprivation index into quintiles with 1 indicating the lowest deprivation and 5 the highest. In addition, the density of hospital beds per 10.000 inhabitants was also included on a county level. For the analysis a categorical variable was created dividing the counties by hospital bed densities into quartiles.
Analysed risk factors include age and sex of the patient, hospital care level (Secondary Care, Tertiary Care, Specialist Care, Prevention and Rehabilitation Care, other), type of care (intensive, normal hospital ward, other), clinical speciality (surgery and related, internal medicine, other), specimen type, region, county type (rural, city), social deprivation index of the county where the hospital is located, hospital beds per 10.000 inhabitants in the county where the hospital is located as well as year and quarter when the isolate was obtained.
In- and exclusion criteria of K. pneumoniae isolates
In the analysis, we focused on materials that are most likely derived from clinical samples, so isolates labelled as screening samples, anal swabs, and stool samples were excluded. We excluded isolates without susceptibility testing. In order to avoid bias from repeated testing, we only included the first isolate per patient per quarter in the analysis irrespective of the specimen type. If several isolates from one patient were tested on the same day we selected the most relevant isolate for the analysis according to this priority: isolate tested non-susceptible against at least one carbapenem > isolate tested against at least one carbapenem > isolate not tested against at least one carbapenem.
The distribution of baseline characteristics of the K. pneumoniae isolates was analysed using percentages and 95% confidence intervals (95% CI) for categorical variables accounting for clustering on the hospital level. Continuous variables were analysed as means with standard deviations if normally distributed and as median with interquartile ranges if non-normally distributed.
Carbapenem non-susceptibility was defined as the proportion of non-susceptible isolates in relation to all tested isolates in the analysis and expressed in percentage and 95% confidence intervals accounting for clustering on the hospital level. An isolate was considered non-susceptible to at least one carbapenem if it was tested as intermediate or resistant against meropenem, imipenem, or ertapenem.
Risk factors for carbapenem non-susceptibility were analysed using univariable and multivariable multilevel (hierarchical) mixed-effects logistic regression models with random intercepts, accounting for clustering on the county and hospital level. Mixed models allow calculating intraclass correlation coefficients for the random intercepts, to quantify variance on different levels . P-values were calculated using Wald tests. For the multivariable model, year, quarter, age and sex of patient, hospital care level, type of care, clinical speciality, specimen type, region, county type, county deprivation index, and hospital beds per 10,000 inhabitants were included. Isolates with missing data in any of these variables were not included in the uni- and multivariable regression models (full case analysis).
For the analysis of carbapenem non-susceptibility over time, a univariable multilevel (hierarchical) mixed-effects logistic regression model was calculated with year as a continuous variable. Only isolates from hospitals that continuously contributed data from 2011 to 2016 were included in the analysis. The p-value was derived from a Wald test.
Not all isolates have been tested for resistance against all three carbapenems (meropenem, imipenem, and ertapenem) and we conducted a sensitivity analysis restricted to isolates that were tested against all three carbapenems.
Differences in EUCAST and CLSI breakpoints might lead to different interpretations of carbapenem non-susceptibility. Consequently, in some cases an isolate might have been categorized as “sensitive” according to one standard and as “intermediate” (i.e. non-susceptible) according to the other standard. Since the use of standards changes over time, this could have affected our results over time. To address this issue, we performed sensitivity analyses for changes in carbapenem non-susceptibility over time: 1) restricted to isolates evaluated according to EUCAST since this is the most commonly used standard in our surveillance data; 2) excluding isolates categorized as “intermediate” to one or more carbapenems and not “resistant” to any carbapenem because for some isolates the classification as “sensitive” or “intermediate” might depend on the standard used.
Resistance against some antibiotics is not routinely tested (e.g. colistin), but only when resistance against other antibiotics is found. Since this can introduce bias into the analysis, we conducted a sensitivity analysis for these antibiotics by restricting the analysis to isolates from laboratories that routinely test ≥90% of all isolates against the respective antibiotic.