Impact of a restrictive antibiotic policy on the acquisition of extended-spectrum beta-lactamase-producing Enterobacteriaceae in an endemic region: a quasi-experimental, before-and-after, propensity-matched cohort study in a Caribbean intensive care unit

Background: High-level antibiotic consumption plays a critical role in the selection and spread of extended-spectrum beta-lactamase-producing Enterobacteriaceae (ESBL-E) in ICU. Implementation of a stewardship program including a restrictive antibiotic policy was evaluated with respect to the ESBL-E acquisition (carriage and infection). Methods: We implemented a two-years, quasi-experimental, before-and-after intervention study with all consecutive adult patients admitted for >48 h in the medical-surgical 26-bed ICU of Guadeloupe University Hospital (French West Indies). A conventional strategy period (CSP), including a broad-spectrum antibiotic as initial empirical treatment, followed by de-escalation (period before) was compared to a restrictive strategy period (RSP) limiting broad-spectrum antibiotics and shortening their duration. Antibiotic therapy was delayed and initiated only after microbiological identication, except for septic shock, severe acute respiratory distress syndrome and meningitis (period after). A multivariate Cox proportional hazard regression model adjusted on propensity score values was performed. The main outcome was the median time of being free of ESBL-E in the ICU. Secondary outcome included all-cause ICU mortality. Results: The study included 1,541 patients, for a total of 738 in the CSP and 803 in the RSP. During the RSP, less patients were treated with antibiotics (46.8% vs 57.9%; p<0.01), treatment duration was shorter (5 vs 6 days; p<0.01), administration of antibiotics targeting anaerobic pathogens signicantly decreased (65.3% vs 33.5%; p<0.01) compared to the CSP. The incidence of ICU-acquired ESBL-E was lower (12.1% vs 19%; p<0.01) during the RSP. The median time of being free of ESBL-E was 22 days (95% condence interval, 16-NA days) in the RSP and 18 days (95% CI, 16-21 days) in the CSP. After propensity score weighting and adjusted analysis, the median time of being free of ESBL-E was independently associated with the RSP (hazard ratio, 0.746 [95% CI, 0.575-0.968]; p=0.02, and hazard ratio 0.751 [95% CI 0.578-0.977]; p=0.03, respectively). All-cause ICU mortality were lower in the RSP than in the CSP (22.5% vs 28.6%; p<0.01). Conclusions: Implementation of a program including a restrictive antibiotic strategy is feasible and is associated with less ESBL-E acquisition in the ICU without any worsening of patient outcome.


Introduction
Antibiotic resistance is among the most important public health concerns worldwide (1). Recently, the World Health Organization published a global priority list of antibiotic-resistant bacteria in which extended-spectrum beta-lactamase-producing Enterobacteriaceae (ESBL-E) are included in the priority 1 group (2). In intensive care units (ICUs), ESBL-E have been increasingly reported for many years, which strengthens the requirement for e cient prevention strategies (3). Occurrence of ESBL-E in ICU may result either from the introduction of an exogenous strain through newly-hospitalized patients with a possible further dissemination through cross-contamination, or from the in-vivo selection of resistant isolates from pre-existing strains, mainly in the gut microbiota, through horizontal gene transfer (4). Among risk factors for ESBL-E acquisition, antibiotic exposure to broad-spectrum cephalosporins and beta-lactam/betalactamase inhibitor combinations has been identi ed as an independent risk factor for colonization or infection with ESBL-E pathogens (5).
Antibiotic therapy is heavily used in ICUs where it has been reported that more than 70% of patients are treated with at least one antibiotic (6). Consequently, antibiotic overuse and the resulting selection pressure makes the ICU an important determinant of the spread of ESBL-E in the hospital (7,8). Different stewardship policies have been developed and implemented in many settings, including ICUs, to improve antibiotic use and clinical outcomes, and to reduce the overall antibiotic selective pressure (7)(8)(9). Among those strategies that have been implemented to optimize antibiotic prescription in ICUs, some restrictive policies, such as delaying the initiation of antibiotics in selected patients or avoiding broad-spectrum antibiotic therapy, have successfully been proposed (10).
In the Caribbean region, the prevalence of multidrug-resistant bacteria is high, including ESBL-E (11,12), and this particular local ecology often leads clinicians to use broad-spectrum antibiotics empirically (13).
Although stewardship programs are urgently needed, no such restrictive strategy has been evaluated in ICUs. In order to overcome this issue, we implemented a stewardship program based on a restrictive antibiotic policy. The aim of this study was to evaluate the impact this strategy on ESBL-E acquisition in the ICU, compared to a conventional and unrestricted antibiotic policy.

Study Design and Setting
We conducted a retrospective, observational, quasi-experimental, before-and-after intervention study from 1 January 2014 to 31 December 2015 in a 26-bed ICU admitting medical and surgical patients at Guadeloupe University Hospital (French West Indies). The ethics committee of the French Society of Intensive Care Medicine (CE SRLF 18-44) approved the study and granted a waiver for informed consent because that both treatment methods were classi ed as standard care. This manuscript follows the STROBE statement for reporting of cohort studies.
Rectal swabs (ESBL-E screening) were performed at ICU admission and once-weekly until discharge, as well as upon admission to the next unit. In that latter case, a positivity in ESBL-E carriage was attributed to the ICU. Contact isolation precautions were applied for each patient until the rst swab results were obtained. Alcohol-based handrub was routinely used for hand hygiene. None of these procedures was modi ed during the study period. All patients admitted to the ICU >48 h during the study period were included in the analysis and followed up until hospital discharge or death. Patients for whom ESBL-E carriage was unknown on ICU admission were not included.
The outcome of interest was the median time of being free of ESBL-E in the ICU, de ned by the time to acquire an ESBL-E in a competing event of death during the follow-up. Secondary outcomes were the incidence of ICU-acquired ESBL-E, duration of antibiotic therapy, antibiotic-free days until ICU discharge, all-cause hospital and ICU mortality, ICU and hospital length of stay, ICU-acquired infections and bacteremia with ESBL-E, and relapse or recurrence of sepsis. A subgroup of ICU patients receiving antibiotic therapy was also analyzed to investigate outcomes in those directly exposed by the restrictive antibiotic stewardship strategy.

Procedures
The 2-year study period was split into two 1-year periods, which differed by the antibiotic policy employed. During the rst year, the "conventional strategy period" (CSP), antibiotic therapy was prescribed at the physician's discretion based on national and international guidelines. This strategy included the use of a broad-spectrum antibiotic as initial empirical treatment in the case of sepsis or suspected infection, followed by de-escalation after 48 to 72 h, based on microbiological data. The main regimens were combination therapies with a cephalosporin and aminoglycoside for community-acquired infections, and carbapenem or piperacillin/tazobactam combined with amikacin for hospital-acquired infections. Dosage, timing and duration followed French guidelines (14).
As part of a stewardship program, a new set of guidelines with a very restrictive antibiotic protocol was established by the ICU team, approved by a multidisciplinary team and implemented on 1 January 2015. The "restrictive strategy period" (RSP) was based on seven principles. 1) For suspected infection, microbiological samples were taken immediately, and antibiotic therapy was initiated only after microbiological identi cation, except for septic shock, severe acute respiratory distress syndrome (ARDS) and meningitis. 2) For non-documented septic shock and severe ARDS, an empiric combination therapy including a cephalosporin and an aminoglycoside was immediately started after microbiological sampling according to the ICU protocol. Combined therapy included either second or third-cephalosporin (cefuroxime, cefotaxime or ceftriaxone) for community-acquired septic shock, or cefoxitin for hospitalacquired septic shock (owing the resistance to the previously-listed cephalosporins and the high rate of susceptibility to cefoxitin of the ESBL-E) or an anti Pseudomonas aeruginosa cephalosporin (ceftazidime or cefepime) for late (>5 days) ventilation-acquired pneumonia (VAP). The second antibiotic was amikacin, unless a Gram-positive pathogen was highly suspected. Due to the very low prevalence of methicillin-resistant Staphylococcus aureus (MRSA) in our hospital, the rst-line anti-staphylococcal treatment was cefazolin. 3) No use of piperacillin/tazobactam and carbapenems for empirical treatment, only for a documented infection without an alternative option. 4) Limited coverage on P. aeruginosa, unless clearly indicated. 5) Limited coverage on subdiaphragmatic anaerobes, unless clearly indicated. 6) Monotherapy as a de nitive treatment. 7) Other characteristics of antibiotic treatment were short duration, high doses, and de-escalation as soon as possible to the narrowest alternative (15), with a focus on penicillin, rst-and second-generation cephalosporins, according to the attending physician and following ICU protocols (see detailed protocol in Additional le 1).

Data Collection
Clinical and laboratory ndings were collected from the patient's medical records. In the CSP, diagnosis of infection and sepsis was based on the clinical judgment of the attending physician. During the RSP, the diagnosis required the identi cation of a pathogen and/or a source of infection. Septic shock was de ned by an infection associated with the need of vasoactive drugs.
ESBL-E carriage on ICU admission was de ned by a positive rectal swab or positive culture without evidence of clinical infection. ICU-acquired ESBL-E was de ned by a positive swab or positive culture 48 h or more after a negative culture at admission. ESBL-E infection was de ned as a positive culture with evidence of clinical infection and detected by chromID ESBL® (bioMérieux, Marcy l'Etoile, France), a ready-to-use chromogenic selective medium for ESBL-producing Enterobacteriaceae. Antibiotic susceptibility was tested by the disk diffusion method on Mueller-Hinton agar (Bio-Rad, Hercules, CA, USA), and production of ESBL was con rmed by the double-disk synergy test according to the guidelines of the European Committee on Antimicrobial Susceptibility Testing (16). Due to the very low incidence in our hospital, Carbapenemase-producing Enterobacteriaceae were not analyzed in this study. VAP was diagnosed using standard criteria (17). Diagnosis of VAP required microbiological con rmation by quantitative culture of a common pathogens. Other ICU-acquired infections were diagnosed using standard criteria, with microbiological documentation for all cases.

Statistical Analysis
Data are reported as frequencies and proportions for categorical variables and mean, median, standard deviation, and 1st and 3rd quartiles for continuous variables. Patients hospitalized in the CSP were compared with those hospitalized in the RSP using the chi-square or Fisher's tests for categorical variables and Wilcoxon-Mann-Whitney tests for continuous variables. To assess the relationship between the strategy period and the median time of being free of ESBL-E, we calculated the Kaplan-Meier curve and used a univariate Cox proportional hazard regression model to estimate the hazard ratio (HR) and its 95% con dence interval. The follow-up time used for survival analyses corresponds to the time between ICU admission and ESBL-E acquisition or ICU discharge from intensive care (or death) if no ESBL-E acquisition occurring during the ICU stay.
To balance confounding factors, a propensity score to receive the restrictive strategy was calculated using a logistic regression model including either clinically relevant or statistically signi cant covariates (age, Simpli ed Acute Physiology Score [SAPS] II, chronic renal failure, diabetes, immunosuppression, antibiotherapy prior to admission, sepsis [at admission or occurring during ICU stay], other multidrug bacteria carriage at admission, and length of ventilation >48 hours) ( Figure 1, Additional le 1). Fifty-two patients were excluded due to missing values on relevant variables. A comparison between excluded and included patients showed a signi cantly superior number of included patients with less than 48 hours of ventilation (52% vs 17.3%; p<0.001) and sepsis at admission (33.2% vs 5.8%; p<0.001). The area under the ROC curve estimating the predictive score ability was 0.612 (95% CI, 0.609-0.614) ( Figure 2, Additional le 1). In the weighted dataset, all absolute standardized differences were inferior to 5%, thus re ecting the good comparability of both groups. This score was used to calculate the inverse probability of treatment weights, assigning patients receiving a restrictive strategy a weight of 1÷ (propensity score) and those receiving a conventional strategy a weight of 1÷ (1 -propensity score), with the use of stabilized weights to reduce variability. Balance among covariates was assessed in the weighted dataset using absolute standardized differences and all results were inferior to 5%. These weights were then used to estimate the relationship between the strategy used and the median time of being free of ESBL-E in a univariate Cox proportional hazard regression weighted model. Kaplan-Meier curves of both groups from the weighted dataset were also estimated. A sensitivity analysis was performed by estimating the impact of the intervention using a multivariate Cox proportional hazard regression model adjusted on the propensity score values. All tests were conducted at a two-sided alpha risk of 5%. Analyses were performed using the 4.0.3 version of R.

Patient Characteristics and Outcomes
We included 1,541 patients in the study (CSP: 738; RSP; 803) ( Figure 1). Baseline characteristics are summarized in Table 1. RSP patients had a lower SAPS II, less diagnosis of sepsis on admission. Less patients in the RSP presented with sepsis or septic shock (40.7% vs 51.5%; p<0.01), but there was no statistical difference in the proportion of patients receiving vasoactive drugs for septic shock during the two periods (20.3% vs 24.3%; p=0.06). Similar ICU mortality was observed in patients admitted for a length of stay <48 hours as safety criteria. No difference in a speci c sepsis category was reported at ICU admission. However, more pulmonary infections and less intra-abdominal infections were diagnosed during ICU stay in the CSP group (Table 1, Additional le 1). During the RSP, the number of patients treated with antibiotics was signi cantly lower than during the CSP (46.8% vs 57.9%, respectively; p<0.01). Median duration of antibiotic treatment was shorter by one day in the RSP (5 vs 6 days; p<0.01) ( Table 2). Third-generation cephalosporins (ceftriaxone, cefotaxime) were the most commonly-administered antibiotics in both periods. More patients were treated with amoxicillin and cefuroxime in the RSP, and less patients received amoxicillin-clavulanate. Moreover, the administration of piperacillin/tazobactam and carbapenem was signi cantly reduced in the RSP (4.5% vs 39.8%, p<0.01 and 3.5% vs 12.5%, p<0.01, respectively). Conversely there were more patients treated with ceftazidime, cefepime and cefoxitin in the RSP compared to the CSP. Importantly, the use of antibiotics targeting anaerobes pathogens (amoxicillin/clavulanic acid, piperacillin/tazobactam, carbapenem, metronidazole, cefoxitin and clindamycin) decreased in the RSP (65.3% vs 33.5%; p<0.01) and a large number of patients did not receive any antibiotic treatment during ICU stay.  Figure 2 and Table 4).

Secondary Outcomes
All-cause ICU mortality were lower in the RSP than in the CSP (22.5% vs 28.6%; p<0.01), including in the subgroup of patients receiving antibiotic therapy. Similar ICU mortality was observed in patients admitted for a length of stay <48 hours as safety criteria (Table 3).

Discussion
This study is the rst to describe the effect of a very restrictive antibiotic stewardship program in ICU, with a particular emphasis on the impact on ESBL-E acquisition. An important nding was a reduction in ESBL-E acquisition associated to the antibiotic stewardship program with no excess in mortality rate, in the context of high prevalence in an endemic region, such as the French West Indies (18). This nding is consistent with previous studies showing a signi cant reduction of the incidence of antibiotic-resistant bacteria, including in ICU setting, following the implementation of antibiotic stewardship programs (19,20). The lower acquisition rate in our ICU was associated with a signi cant reduction of the use of broadspectrum antibiotics, mainly piperacillin/tazobactam and carbapenem and, more generally, antibiotics targeting anaerobic microbiota. Several authors have reported the role of antibiotic therapies with activity against anaerobic microbiota, as well as the beta-lactamase inhibitor, in the acquisition of multidrugresistant Gram-negative bacteria, including ESBL-E (5,21). Furthermore, preserving the microbiota against antibiotics is one of the key strategies against ESBL-E acquisition (22).
The reluctance of ICU physicians to rationalize antibiotic use is often a major limitation in stewardship programs. We were able to overcome this unwillingness in our ICU, resulting in a signi cant reduction in global antibiotic consumption, including broad-spectrum antibiotics, a shorter duration of therapy, and an increased use of narrow-spectrum molecules. Interestingly, we observed a dramatic decrease of the consumption of carbapenems in the RSP, even though the prevalence of ESBL-E was high. The precise cause of the lower incidence of ESBL-E acquisition in the RSP is di cult to determine, but the strong decrease in the use of piperacillin/tazobactam may have played a role, as well as the choice of alternative antibiotics, such as cefoxitin. In the RSP, antibiotic duration was reduced by one day, although the baseline duration was already short. This is consistent with the current trend of a shorter treatment duration for infections such as VAP (23), intra-abdominal infections (24,25), or bacteremia (26). Discontinuation of antibiotic therapy in the case of negative microbiological cultures could also explain the short treatment duration, especially in VAP (27,28). Antibiotic duration is frequently longer than recommended in clinical practice (29), in spite of no better outcome, and the likely promotion of bacterial or fungal superinfections (27). Indeed, Daneman and colleagues showed that up to two-thirds of ICU patients with bacteremia did not receive a short course as recommended (30).
Lower antibiotic consumption during the RSP was also the result of the use of a lower threshold for the initiation of antibiotic therapy (7). In the case of hemodynamically-stable patients with suspected infection, antibiotic therapy was initiated only after clinical evidence of infection and microbiological documentation. In some patients, an alternative diagnosis was identi ed and antibiotic therapy was not initiated. This "conservative" strategy has been evaluated by Hranjec and colleagues in critically-ill surgical patients and resulted in a higher appropriate initial antibiotic treatment, reduced duration of antibiotic treatment, and a lower mortality rate (10). Despite the contradiction of delaying the initiation of antimicrobial treatment with pre-existing dogma and the promoted de-escalation strategy, several arguments support this approach (31). First, a high proportion of patients with fever received antimicrobial therapy in the ICU, despite non-infectious disease (32). Similarly, overdiagnosis of VAP was reported in 68% of patients, resulting in an overuse of antimicrobial therapy (33). Second, the linear association between the timing of antibiotic therapy and mortality has been challenged (34), suggesting no bene t in an immediate start of antimicrobial drugs in less severely-ill patients. Third, this position has been recently advocated by many experts (35). Of note, the Infectious Diseases Society of America decided not to endorse the 2018 Surviving Sepsis Campaign experts highlighted its recommendation of prompt antibiotic therapy (36), which they regarded as "an oversimpli ed approach" that should be reconsidered and potentially delayed in less severely-ill patients to avoid "treating some infections inadequately and others excessively" (37).
Our study has some limitations. First, it is a single-center study. Nevertheless, our results regarding the implementation of the stewardship program and its feasibility could be extrapolated to other ICUs.
Second, outcomes related to the restrictive strategy cannot be addressed accurately as patients in the RSP were less severe, based on SAPS II at admission. However, SAPS II were included in the propensity score analysis. Furthermore, the standardized mortality ratio (ratio of actual ICU-mortality to SAPS IIpredicted mortality) was similar (0.93 and 0.95, respectively) in both periods, suggesting no excess in the RSP mortality rate. Third, infection may have been overdiagnosed during the CSP, where the diagnosis was based on the physician's judgment. As previously mentioned, the diagnosis of infection may be di cult in the ICU and many non-infectious diseases may present with fever or in ammatory syndrome, especially when it comes to VAP (33). Conversely, during the RSP, a microbiological con rmation and/or a severity signs of sepsis were required to de ne infection. In the data collection process, those situations were called "infection", probably leading to some falsely reported infections during the CSP. As a suspicion of infection was often the trigger for initiating an empirical antibiotic therapy during the CSP, it is likely that the overdiagnosis of infection has played a role in the broader use of antibiotics, compared to the RSP in which a microbiological con rmation of infection and/or a severity sign were required to initiate empirical antibiotics. Neverless, the proportion of patients treated with vasoactive drugs for septic shock was similar, suggesting no difference in the incidence of the most severe patients between groups.
Fourth, the rate of initial appropriate antibiotic therapy in septic shock was not reported. The use of cefoxitin as empirical treatment for acquired non-pulmonary septic shock was based on our local epidemiology, i.e., a high prevalence of cefoxitin susceptibility to ESBL-E, low incidence of AmpChyperproducing Enterobacteriaceae, and a very rare incidence of carbapenemase-producing Enterobacteriaceae and MRSA. Although administration of cefoxitin was appropriate when referring to antimicrobial susceptibility testing, few data support its clinical use in severe infections, other than urinary tract infection (38). To address this issue, we used a combination therapy of cefoxitin and amikacin for empirical treatment in acquired septic shock.

Conclusions
A restrictive strategy delaying the initiation of antibiotic treatment in less severely-ill patients and using the narrowest spectrum has the potential to avoid broad-spectrum antibiotics, particularly those targeting intestinal anaerobe microbiota. This strategy resulted in a dramatic decrease in antibiotic consumption and was independently associated with a reduction of ESBL-E acquisition in the ICU.
Authors' contributions: CLT, MV and GT conceived and designed the study. They had full access to all data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. CLT, MV and RR collected the data. CLT, PB, EO, MC and GT analysed the data. CLT, MV and GT drafted the paper. CLT and MV contributed equally to this work. All authors interpreted the data and critically revised the manuscript for important intellectual content and gave approval for the nal version to be published. All authors agree to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.
The manuscript's guarantors (CLT, MV and GT) a rm that the manuscript is an honest, accurate, and transparent account of the study being reported; that no important aspects of the study have been omitted; and that any discrepancies from the study as planned have been explained.
Funding: No funding was received for this work.
Availability of data and materials: After publication, the data will be made available upon reasonable request to the corresponding author. A proposal with a detailed description of study objectives and statistical analysis plan will be needed for evaluation of the reasonability of requests. Additional materials might also be required during the process of evaluation. De-identi ed participant data will be provided after approval from the corresponding author and the Guadeloupe University Hospital, French West Indies.

Declarations
Ethics approval and consent to participate: The ethics committee of the French Society of Intensive Care Medicine (CE SRLF 18-44) approved the study and granted a waiver for informed consent because that both treatment methods were classi ed as standard care. AdditionalFile.pdf