Antimicrobial resistance (AMR) is considered one of the greatest global health threats [1]. Multidrug-resistant infections are associated with longer hospital stays, higher treatment costs, and increased morbidity and mortality. In Europe alone, over 33,000 annual deaths are attributed to resistant infections [2,3]. Inappropriate and incorrect prescribing is a major driver for resistance and is associated with poorer outcomes [4–6]. In response to the increase in resistance, antimicrobial stewardship (AMS) programs have been widely implemented, aiming to optimize therapies and outcomes. Recent metanalyses showed that AMS programs were associated with enhanced appropriate use of antimicrobials, reduced incidences of AMR and Clostridioides difficile infections (CDIs), shorter hospital stays, and lower therapy costs [7–9].

Institution-specific guidelines (ISGs) are considered an important part of AMS programs, as they offer locally-tailored decision support for empirical therapies while reflecting local pathogen and resistance epidemiology as well as national and other societal recommendations [10,11]. Several studies showed that local guidelines can lead to behavioral changes among physicians, yet few investigated the influence on antimicrobial consumption and resistance [12–16]. The Leipzig University Hospital first introduced ISGs in 2014 as a printed pocket guide; an electronic application was added in 2017 [17,18]. In a previous, survey-based study we found that the ISGs were successfully implemented, regularly used across disciplines and had a perceived influence on physicians’ prescribing behavior [17].

In this study we aimed to assess the effects of ISGs on antimicrobial consumption and subsequently the incidence of AMR and CDIs. We hypothesized that, after the implementation of the ISGs, the use of antimicrobial agents that were commonly prescribed prior to the intervention but not often recommended in the ISGs (i.e., ciprofloxacin, sultamicillin, oral cefuroxime, ceftriaxone; Table 1) would be reduced. Subsequently, we expected a decrease in resistance to the respective antibiotics and a decrease in CDI. Conversely, we expected an increased use of antibiotic agents that were not used frequently before the introduction of the ISGs but were recommended frequently, such as cefotaxime, or oral clarithromycin (combination partner for the treatment of community-acquired pneumonia). Furthermore, we aimed to assess whether there would be secondary effects on the proportion of oral and parenteral antibiotics use, as well as on the proportions of Access, Watch and Reserve antibiotics as defined by the World Health Organization (WHO) AWaRe classification [19].

We conducted an observational interrupted time series study with antibiotic consumption, resistance and CDI data to assess the effect of the ISGs. Simple and segmented linear regression models were used to analyze trends and changes thereof. The study is reported in accordance with the STROBE-AMS recommendations of the EQUATOR network [20,21].

The Leipzig University Hospital (LUH) is a 1,451-bed tertiary-care facility with 29 departments and clinics of all specialties in Leipzig, Germany. Since 2012, several interdisciplinary AMS interventions have been implemented, including regular ward visits by the AMS team, intense training of staff, restriction of selected reserve antibiotics, surveillance of antibiotic consumption and resistance and the implementation of professional ISGs, as described before [17,22–24].

The ISGs were developed in cooperation with experts from the represented disciplines and implemented in June 2014. They cover a wide range of infections for many clinical disciplines and provide information on empirical treatment, diagnostics and prophylaxis as well as dose adjustments. The current 4th edition of the ISGs (in German) can be accessed online [18].

All inpatient departments, except pediatrics and psychiatry, were included in the analyses. Psychiatric units were excluded as very few antimicrobials are prescribed there, and pediatric units as they prescribe at small absolute doses. Their inclusion could have led to an underestimation of the antibiotic consumption in the hospital overall. Outpatient departments were excluded since the ISGs are specifically designed for the inpatient setting.

The microbiological analysis focused on isolates from Klebsiella pneumoniae, Escherichia coli, Staphylococcus aureus, S. epidermidis, Enterococcus faecalis, E. faecium and Pseudomonas aeruginosa collected 2012 to 2020. All diagnostic samples from the entire hospital with the exception of routine screening material were included in the analysis. In addition, all CDIs confirmed by positive cultures were recorded.

Antibiotic consumption was measured in monthly defined daily doses (DDD) per 100 bed days (DDD/100BD). We focused on antibiotic agents and antibiotic combinations that met at least one of the following criteria: 1) they were among the 15 most frequently dispensed antibiotics in the year before the intervention; 2) they were frequently recommended as first-line therapy in the ISGs; 3) they received broad recommendation for at least one very common infection (Table 1). The analysis was done separately for parenteral and oral application. Furthermore, we analyzed relevant commonly used groups (Table 2).

Resistance outcomes were measured annually in percent of total isolates. We used the 2021 EUCAST breakpoints (www.eucast.org) to determine annual resistance rates [25]. Klicken oder tippen Sie hier, um Text einzugeben. Intermediate (susceptible, increased exposure) minimum inhibitory concentrations (MICs) were not considered resistant. The absolute number of positive C. difficile routine cultures was used to analyze CDI activity.

The hospital pharmacy provided data on antimicrobials dispensed to hospital cost centers (e.g. wards) for the period from 1 January 2012 to 31 December 2020. The data was accessed via the hospital benchmarking software IQVIA Premax (www.iqvia.com). Monthly bed days (BD) of all cost centers were received from the controlling department. Annual MIC data of pathogen isolates and the number of CDIs were provided by the Institute for Medical Microbiology and Virology. German antibiotic resistance data collected in the European Antimicrobial Resistance Network (EARS-Net) was accessed through the ECDC Surveillance Atlas for AMR (www.ecdc.europa.eu/en/antimicrobial-resistance/surveillance-and-disease-data/data-ecdc) [26].

We considered regular ward rounds by the AMS team and interventions by other departments as strong potential confounders. To address this, we ran all analyses again with a subset that did not include wards receiving regular AMS rounds (i.e., intensive care units, pneumology, gastroenterology, and two surgical wards) and hematologic oncology wards, as these wards have their own antimicrobial therapy recommendations.

A stockout of ampicillin/sulbactam caused a drastic reduction of its use between September 2015 and February 2016; to avoid this as a potential confounder in the analysis we excluded the respective data points in this timeframe [27]. A change in the record keeping system of BD in 2013 caused a large increase in BD (about 40%) without any relevant change of activity in the hospital. To avoid a confounding effect on the outcome variable (DDD/100BD), we adjusted the BD between January 2012 and July 2013 with a correcting factor (1.38) that was determined by comparing the monthly averages of a five-month period that was fully documented in both records (August to December 2013). Old, new, and adjusted monthly BD are plotted in S1 Fig.

Hospital resistance rates are partly determined by external factors such as treatment in other facilities or the spread of resistant strains in the population, and observed changes in resistance rates are therefore not necessarily due to changes in prescribing within the hospital. We therefore also compared our data to selected resistance rates collected in the EARS-Net, to control for broad changes in resistance rates within the German patient population [26]. The documented S. aureus resistance rates in the LUH were extremely high in 2014, likely due to a reporting or documentation error, resulting in a strong negative trend in the post-intervention phase. To avoid misleading results, we did not include 2014 S. aureus resistance rates in the analysis.

A segmented linear regression model of interrupted time series was used to analyze the changes in antibiotic consumption, as described before [28–31]. This method allows for the evaluation of immediate changes in level, as well as long-term effects due to trend changes. A time series is a sequence of observations of a certain outcome, which is aggregated for each time point (e.g. DDD/100BD per month). The time series is divided (interrupted) into pre- and post-intervention segments. Least square regression lines are fitted to each segment of the time series and together expressed by the following model [28]: Yt=β0+β1*timet+β2*interventiont+β3*timeafterinterventiont+et.

In this model, Y_t is the outcome at time point t, in this case DDD/100BD in each month. Time is a continuous variable counting the number of months of the whole timeframe, intervention is an indicator for the time before (intervention = 0) and after (intervention = 1) the intervention, and time after intervention is a continuous variable counting the time after the intervention. β0 estimates the baseline level at time point zero (y intercept), β1 estimates the baseline trend, β2 estimates the level change between the pre- and the post-intervention period, and β3 estimates the change in trend after the intervention. The sum of β1 and β3 estimates the post-intervention slope. The error e_t represents the variability not described by the model.

Since the ISGs were implemented in June 2014, July 2014 was chosen as the first time point of the post-intervention period. A second intervention was added in two constellations. First, on 26 June 2016 the microbiology laboratory introduced routine real-time polymerase chain reaction (PCRs) for the diagnosis of pneumocystis pneumonia (PCP). This led to a higher detection rate, and subsequently more prescriptions of trimethoprim/sulfamethoxazole. We considered this a second intervention for trimethoprim/sulfamethoxazole, and so July 2016 was set as the first post-intervention data point. Second, two warning letters (“Rote-Hand-Briefe”) had been issued by the German Federal Institute for Drugs and Medical Devices (BfArM) in October 2018 and April 2019, warning of the serious adverse effects of fluoroquinolones [32,33]. These warnings were actively conveyed to hospital physicians by the AMS team and reflected in their recommendations. November 2018 was chosen as the first post-intervention data point.

MIC data was only available in annual aggregates which means that there were not enough data points to run ITSA. We calculated simple linear regressions to determine the trends of resistance rates over the whole study period. To assess national trends of AMR we used annual resistance rates from 2012 to 2019, as data for 2020 was not uploaded at the time of submission [26]. P values of <0.05 were considered statistically significant.

All calculations were carried out using RStudio [34]. The R-code will be shared on request.

The ethics committee of the Medical Faculty of Leipzig University reviewed and approved the study on January 22, 2019 (registry number 011/19-ek). All data is anonymous.

Between 2012 and 2020 the annual antibiotic consumption per 100 BD fell by 14%, from 543 DDD/100BD to 468 DDD/100BD. This reduction mainly occurred prior to the study intervention (Fig 1).

Table 2 shows the results of the ITSA of total antibiotic consumption as well as of different subgroups and antibiotic agents.

Changes in the proportions of AWaRe groups were minimal (Fig 2).

The proportion of Access antibiotics increased by 2%, from 36% in 2012 to 38% in 2020. The intervention was associated with an initial reduction of -1.8 (-2.9 to -0.7, p<0.01) followed by a positive trend with a trend change of +0.13 per month (95%CI 0.08 to 0.19, p<0.001). The introduction of ISGs was associated with an increased use in parenteral antibiotics. With a trend change of +0.09 per month (95%CI 0.04 to 0.14; p<0.001), the proportion of total antibiotic consumption in DDD/100BD increased from 53% in 2012 to 72% in 2020 (Fig 3).

During the study period fluoroquinolone consumption fell by 67%, from 104 DDD/100BD in 2012 to 35 DDD/100BD in 2020. The introduction of ISGs was associated with a large, immediate reduction of ciprofloxacin consumption (-1.1 DDD/100BD; 95%CI -1.6 to -0.6; p<0.001) (Fig 4). A further immediate decrease affecting all fluoroquinolones occurred when the BfArM issued specific warning letters (“Rote-Hand-Briefe”) in 2018 and 2019 (-1.7 DDD/100BD, 95%CI -2.5 to -0.9, p<0.001) [32,33].

For the consumption of cephalosporins, the introduction of ISGs was associated with a positive slope change of +0.08 DDD/100BD (95%CI 0.04 to 0.12, p<0.001). While overall cephalosporin consumption changed only moderately, there were major changes within the antibiotic class (Fig 5): cefuroxime and ceftriaxone prescriptions decreased continuously, with a significant level change for ceftriaxone of -0.3 DDD/100BD (95%CI -0.5 to -0.1, p<0.001) following the ISGs’ implementation. Conversely, for cefotaxime, the intervention was associated with a significant positive trend change of +0.06 DDD/100BD per month (95%CI 0.04 to 0.07, p<0.001).

Total penicillin consumption increased by 14% from 112 DDD/100BD in 2012 to 128 DDD/100BD in 2020, without the ISGs having a significant impact, while penicillin and beta-lactamase inhibitor combinations decreased by 14% with a significant level change of -0.96 (95%CI -1.51 to -0.42). The largest reduction (-71%) and a level change of -0.4 DDD/100BD (95%CI -0.7 to -0.1, p<0.01) was observed in sultamicillin. However, decreases were strongest in the pre-intervention phase.

Trimethoprim/sulfamethoxazole prescriptions dropped immediately after the intervention by -0.4 DDD/100BD (95%CI -0.7 to -0.1, p<0.01). Prescriptions increased steadily, though not significantly, when routine Pneumocystis jirovecii PCR was introduced in 2016 (Fig 6).

Vancomycin consumption decreased pre-intervention and converted into a steady state following the intervention which was associated with a level change of +0.4 DDD/100BD (95%CI 0.1 to 0.7, p<0.01) and trend change of +0.04 DDD/100BD per month (95%CI 0.02 to 0.05, p<0.001). Additionally, we observed a level change of prescriptions of oral clarithromycin of +0.8 DDD/100BD (95%CI 0.4 to 1.1, p<0.001).

Furthermore, we ran the ITSA with a subset of the data that excluded wards receiving regular AMS rounds, accounting for 51% of antibiotic consumption documented in the main dataset (S1 Table). Most of the changes observed in the main dataset persisted, e.g. the increase in cefotaxime consumption and a decrease in the prescription rates of fluoroquinolones, ceftriaxone, and sultamicillin. Changes in the consumption of clarithromycin and trimethoprim/sulfamethoxazole were not significant.

Table 3 shows annual rates of resistance to “focus antibiotics” in isolates of K. pneumoniae, E. coli, S. aureus, and P. aeruginosa.

More comprehensive tables including further antibiotic agents as well as E. faecium, E. faecalis, and S. epidermidis isolates can be found in S2 Table, and ITSA results of resistance rates can be found in S3 Table. Rates of resistance to the majority of tested antibiotics, including broad-spectrum penicillins, second and third generation cephalosporins, fluoroquinolones and trimethoprim/sulfamethoxazole decreased significantly in all analyzed pathogens, except in enterococci, where no significant trends were observed. In most cases, resistance rates were already decreasing in the pre-intervention phase and continued to decrease with a more moderate trend after the intervention. Negative trend changes were observed for ciprofloxacin resistance in E. coli, P. aeruginosa and S. aureus as well as cefotaxime resistance in E. coli, but these changes were not significant.

Fluoroquinolone-resistance rates among Enterobacterales, staphylococci, and P. aeruginosa decreased strongly, with relative reductions of ciprofloxacin-resistance ranging between 12% in P. aeruginosa (-1.8% per year, 95%CI -2.6 to -0.9, p<0.01) to 16% in S. aureus (-2.9% per year; 95%CI -5.2 to -0.6, p<0.05).

Resistance to second- and third-generation cephalosporins, tested with cefuroxime and cefotaxime, decreased significantly in Enterobacterales. For instance, rates of cefuroxime-resistant K. pneumoniae decreased by 1.8% per year (95%CI -3.0 to -0.6, p<0.05) from 40.5% (366 of 902 cases) in 2012 to 20.1% (200 of 996 cases) in 2020. Similarly, we observed reduced rates of ceftazidime resistance in P. aeruginosa (-1.4% per year; 95%CI -2.4 to -0.5; p<0.05).

We observed reduced rates of resistance to broad-spectrum penicillins among Enterobacterales, as well as staphylococci and pseudomonas. Ampicillin/sulbactam-resistant E. coli decreased significantly by 2.4% per year (95%CI -3.2 to -1.5; p<0.001) accounting for a relative reduction of almost 25% over the study period. A comparable trend of -1.9 (95%CI -2.9 to -0.9, p<0.01) was observed for ampicillin/sulbactam-resistant K. pneumoniae. In staphylococci, significantly decreased rates of oxacillin-resistance are indicative of reduced resistance rates to other broad-spectrum penicillins. Interestingly, rates of penicillin G resistance in S. aureus, which is expected in penicillinase-producing strains, showed a significant decreasing trend of -2.3% per year (95%CI -3.5 to -1.1; p<0.01), accounting for a relative reduction of 15% from 2012 to 2020. Rates of piperacillin/tazobactam resistance decreased significantly, resulting in a relative reduction of almost 20% in K. pneumoniae (-2% per year; 95%CI -2.9 to -1.2; p<0.01), and 17% in P. aeruginosa. (-1.6% per year; 95%CI -2.4 to -0.8; p<0.01).

Trimethoprim/sulfamethoxazole resistance showed an overall significant decreasing trend in Enterobacterales and staphylococci (e.g. -1.7% per year [95%CI -2.0 to -1.4, p<0.001] in E. coli, and -2.9% per year [95%CI -4.0 to -1.7, p<0.001] in S. epidermidis), yet in all of them an increase could be observed in 2020.

Vancomycin-resistance in staphylococci stayed at an extremely low level (0–0.2%).

S4 Table shows annual resistance rates of E. coli, K. pneumoniae, P. aeruginosa, and S. aureus in Germany between 2012 and 2019 [26]. Methicillin-resistant S. aureus (MRSA) rates decreased by 1.2% per year (95%CI -1.3 to -1.0, p<0.001), while decreases in other pathogens were very moderate. In the case of third-generation cephalosporin-resistant E. coli, rates actually increased (0.38%; 95%CI 0.18 to 0.58, p<0.001). A graphic comparison of resistance rates to fluoroquinolones and third-generation cephalosporins in Enterobacterales is given in Fig 7.

We observed a large reduction of CDI cases. In 2014, when the ISGs were introduced, culture positive cases had declined by 47% when compared with the previous year (279 vs. 523 cases). Over the whole study period annual CDIs were reduced by 65%, from 501 in 2012 to 174 in 2020 (Table 4).

Since this study is based on routinely documented data without a control group, we can only assess associations and not causations. To analyze antibiotic consumption, we used the pharmacy records of the dispensed antibiotics. Dispensing data holds several limitations. Most importantly, it does not include information on when and whether the dispensed antibiotics were actually used and whether the prescriptions were indicated. Complete surveillance for antibiotic consumption and resistance was only available from the year 2012, when most AMS interventions started. It was therefore not possible to analyze changes due to previous AMS interventions using ITSA.

The fact that antibiotic consumption and resistance rates were already strongly decreasing before the implementation of ISGs makes analyzing their impact with ITSA challenging. We regard other AMS interventions as the strongest confounder. Aside from these AMS interventions, there are many other potential confounders affecting antibiotic consumption, such as new standard operating procedures (SOP) and other department-specific recommendations, as well as stockouts or changed antibiotic prices. Additionally, the current COVID-19 pandemic likely impacted prescriptions in 2020. We addressed such factors wherever possible: by adding a second intervention in the case of fluoroquinolones and trimethoprim/sulfamethoxazole, removing data points of ampicillin/sulbactam affected by a stockout, and analyzing consumption data again after excluding wards receiving AMS rounds.

Resistance rates and CDI numbers were only available as annual aggregates. Hence, a clearly separated microanalysis of resistance rates stratified by those wards receiving regular AMS rounds and those not (especially “heavv users” such as hematologic oncology wards) using aggregated data, is not feasible. The small number of data points and the strong pre-intervention trend make ITSA unreliable for this data. It is therefore not possible to draw clear conclusions regarding the effect of the ISGs on resistance rates and CDIs. Aside of the antimicrobial treatment, there are several confounders that might have impacted resistance rates in our hospital. For example, infection prevention and control interventions may have reduced hospital-acquired infections and thus contributed to lower incidences of CDIs and antibiotic resistance. In addition, given the various factors contributing to the rise and spread of AMR, it is expected that rates of resistance will be largely influenced by external factors, including treatments from other providers. To address this, we compared the resistance trends in our hospital with national trends.