Introduction

The development of antimicrobial resistance is a growing problem worldwide, and infections due to multi-resistant pathogens, both Gram-positive and Gram-negative, have steadily increased over the years [13]. Infections with these organisms have been associated with greater mortality [4, 5], prolonged hospitalisation [6] and increased costs [710].

Multiple strategies have been employed to control the emergence and spread of antibiotic resistance. These include colonisation surveillance [2], barrier methods [2, 1113], antibiotic stewardship [2, 14], decision support [2, 15], formulary restriction [2, 16, 17] and antibiotic cycling or rotation [1825].

Antibiotic cycling or rotation is based on the scheduled substitution of an antibiotic class with another one having a comparable spectrum of activity but not sharing the same mechanism of resistance. In the ICU setting, cycling has been associated with conflicting results. A limited number of studies has shown a decreased incidence of antibiotic-resistant Gram-negative infections, particularly ventilator-associated pneumonia (VAP), a decreased resistance pattern of Pseudomonas aeruginosa and lower infection-related mortality rates [1921, 24]. Other studies have not observed any advantage of this strategy [26]. To date, the synthesis of these results is that there is insufficient evidence to recommend antibiotic cycling as a standard measure to reduce antibiotic resistance [27, 28].

The purpose of this study was to determine the effectiveness of an antimicrobial rotation programme in reducing the incidence of VAP attributed to antibiotic-resistant Gram-negative bacteria.

Materials and methods

Setting

The study was performed at two different ICUs of two different hospitals in Lombardy, Italy.

Centre A is represented by the medical/surgical ICU of the Istituti Ospitalieri di Cremona, with 10 beds and admitting about 400 patients per year. The ICU is structured as follows: 1 two-bed bays, 2 three-bed bays and 2 single bedrooms. The Istituti Ospitalieri di Cremona is a 600-bed community hospital that serves a population of approximately 150,000 inhabitants and admits about 28,000 patients every year.

Centre B is represented by the 2° Servizio di Anestesia e Rianimazione of the Spedali Civili di Brescia, it has 10 beds and admits about 400 medical/surgical patients per year. The ICU is structured as follows: 2 four- and 2 one-bed bays. The Spedali Civili di Brescia is an 1,800-bed teaching facility that serves a population of approximately 1,000,000 inhabitants and admits about 80,000 patients every year.

Study design

We carried out a “before-after” study of 2 years (from 1 January 2006 through 31 December 2007). Inclusion criteria were age >18 years, ICU stay >48 h and mechanical ventilation (MV) lasting more than 48 h.

The primary outcome of the study was to determine whether the incidence of VAP attributed to antibiotic-resistant Gram-negative bacteria could be reduced using a scheduled change of antibiotic classes. The secondary outcome was to determine the effectiveness of an antimicrobial rotation programme in reducing the antimicrobial resistance spectrum of antibiotic-resistant Gram-negative bacteria. We chose VAP as the nosocomial parameter to evaluate because it was the most frequent nosocomial infection occurring in both ICUs.

The study was divided into two periods:

  • Period 1 (P1), from 1 January 2006 through 31 December 2006, prior to the introduction of the antibiotic rotation programme.

  • Period 2 (P2), from 1 January 2007 to 31 December 2007, after the introduction of the antibiotic rotation programme.

Antibiotic prescription

During P1, the choice of antibiotic prescription for the treatment of VAP was left to the physician’s discretion, both for molecule selection and duration of therapy. However, the antibiotics used most were piperacillin/tazobactam and levofloxacin.

During P2 we introduced a protocol based on a quarterly rotation of antibiotics (piperacillin/tazobactam, quinolones, carbapenems and cefepime/ceftazidime) for the treatment of VAP. The new programme was as follows:

  • 1st trimester piperacillin/tazobactam: 4.5 g intravenously (i.v.), every 6 h.

  • 2nd trimester fluoroquinolones: ciprofloxacin 400 mg i.v. b.i.d. or levofloxacin 500 mg b.i.d.

  • 3rd trimester carbapenems: imipenem 1 g i.v. g8h or meropenem 1 g q8h.

  • 4th trimester 3rd/4th generation cephalosporins: ceftazidime or cefepime 6 g/day i.v. continuous infusion.

Standard duration of therapy was 8 days. Unless in the case of resistance, the antibiotic of choice was maintained for the treatment’s duration without applying a de-escalation strategy when the results from microbiology were available. In cases of different physicians’ judgement and when Pseudomonas aeruginosa infection was highly suspected, antibiotic treatment could be prolonged to 14 days. Given the observational nature of this study, informed consent was not required.

Definitions

For definitions of infection, systemic inflammatory response syndrome, sepsis, severe sepsis and septic shock, we referred to the 2001 SCCM/ESICM/ACCP/ATS/SIS International Sepsis Definitions Conference [29].

For the definition of VAP we referred to the guidelines of the 2005 American Thoracic Society-Infectious Disease Society of America [30]. VAP was defined as any lower respiratory tract infection that developed after 2 days of MV. Clinical suspicion of VAP was defined as a new, progressive, or persistent (>24 h) infiltrate on the chest radiograph, with two or more of the following criteria:

  1. 1.

    Fever >38.3°C or hypothermia <36°C

  2. 2.

    Purulent endotracheal aspirate

  3. 3.

    Leukocytes count >10,000/mm3 or <4,000/mm3

Every patient suspected of having pneumonia underwent, within 24 h, lower respiratory tract microbiological sampling, always before initiating empirical antimicrobial treatment. To establish a microbiological diagnosis we performed, whenever feasible, a broncho-alveolar lavage (BAL); as an alternative method, when bronchoscopy was not immediately available, we used a blind sampling technique (blind mini-BAL) whose diagnostic accuracy has been widely established [31, 32]. A case of VAP was defined as microbiologically confirmed when bacteria were isolated in significant quantities from BAL samples (≥104 CFU/ml). We defined “early VAP” as those occurring during the first 5 days of MV and “late VAP” as those occurring after 5 days of MV.

For definitions of infections, we referred to the International Sepsis Forum Consensus Conference on Definitions of Infections in the Intensive Care Unit [33]. The organism associated with infection was defined by the isolation of the germ in a biological material in the presence of signs and symptoms of infection.

Micro-organism susceptibility

Antibiotic-resistant Gram-negative bacteria were defined as germs being resistant to at least one class of antibiotics typically used in the treatment of Gram-negative bacterial infections. We defined as multi-drug resistant (MDR) a germ that was resistant to at least three antibiotic classes. We analysed data regarding resistant Pseudomonas aeruginosa, Stenotrophomonas maltophilia, Acinetobacter baumanii and extended spectrum β-lactamase (ESBL) Enterobacteriaceae. The antibiotics we considered for resistance were: aminoglycosides (gentamycin, tobramycin, amikacin), third/fourth generation cephalosporins (ceftazidime/cefepime), extended-spectrum penicillins (piperacillin/tazobactam), fluoroquinolones (ciprofloxacin, levofloxacin) and carbapenems (imipenem/meropenem).

Compliance with the rotation protocol

For each patient we evaluated if the prescribed therapy was in accordance or not with the rotation protocol. We defined three categories:

  1. 1.

    Exact antibiotic class prescribed

  2. 2.

    Acceptable deviation, e.g., allergy, drug interaction, antibiotic-resistant germ

  3. 3.

    Major deviation

We considered categories 1 and 2 as adhering to the rotation protocol.

Antibiotic utilisation

Although analysis of antibiotic consumption was not an objective of this study; we reported the frequency of use of commonly prescribed antibiotic classes for the treatment of VAP before and after the introduction of rotation, expressed as the number of antibiotic cycles per 1,000 patient days.

According to the Anatomical, Therapeutic, Chemical (ATC) classification system with Defined Daily Doses (DDDs), the amount in grams for an antimicrobial agent was converted into a number of DDDs [34]. The consumption density for each antibiotic class was defined as the division of the number of DDDs by the number of patient-days.

Infection control measures

All the standard infection control measures were maintained active and were not modified during the two periods. All patients were placed, whenever possible, in a 30–45° (semi-recumbent) position. Health care workers were invited to wash their hands with chlorhexidine-based soap or rub hands with alcohol-based solutions before and after patient contact. In the case of colonisation or infection caused by a multi-resistant germ, either Gram-positive or Gram-negative, standard contact precautions, including hand hygiene and use of gloves, gowns, masks and caps were applied during all medical and nursing procedures. Those patients were isolated in single rooms or cohorted, depending on bed availability [35, 36]. No specific interventions to modify infection control measures were performed in the two units over the study period.

Data collection and validation

Data were collected by ICU medical staff members: ER and LC at Centre A, and ER and SD at Centre B. Data were registered in a dedicated database. For each patient we recorded: age, sex, admission diagnosis, Glasgow Coma Scale (GCS), Simplified Acute Physiology Score (SAPS) 2 and 3, isolated germs, susceptibility pattern, type, dose and length of antibiotic therapy, days of MV, days of ICU stay and crude ICU mortality. Microbiological results were obtained from the Microbiology Laboratory computerised system of each hospital. All copy-isolates were excluded; only the first positive culture of each germ for each patient was included. Data on antibiotic use were obtained from the Hospital Pharmacy.

Statistical analysis

Continuous variables with normal distribution were expressed as mean ± standard deviation (SD); those with a non-normal distribution (Shapiro–Wilk W test) were expressed as median values and interquartile range (25th–75th percentile). To compare the categorical variables, we used the χ2 test or Fisher’s exact test when necessary. To compare continuous variables the t test or the Kruskal–Wallis test was used, as appropriate. Relative risks (RRs) and their 95% confidence intervals (CI) were calculated using standard methods.

The SAPS2 was used during 2006 and SAPS3 during 2007. These two parameters have been compared by calculating the predicted in-hospital mortality associated with each SAPS value.

The incidence rate of VAP was expressed as the number of cases per 1,000 patient days and per 1,000 days of MV.

The analysis of rate differences between periods was performed utilising the segmented regression analysis, a powerful method for estimating intervention effect in interrupted time series studies [37]. In this study we have an intervention determining two segments, namely: P1 and P2. Each segment is divided into periods of 1 month. Two parameters defined each segment: level and trend. The level is the value of the series at the beginning of a given time interval and the trend is the slope of a given segment. A change in level constitutes an abrupt intervention effect; a change in trend is defined by an increase or decrease in the slope of the segment after the intervention compared with the segment preceding the intervention itself, i.e., represents a gradual change in infection rate during the segment. We fitted the following model: Y = β 0  + β 1 x time + β 2 x interention1 + β 3 x timeafterintervention1 +  e where β 0 represents the intercept, β 1 the baseline time trend, β 2 the level change after the introduction of the rotation protocol, β 3 the trend change after the introduction of the rotation protocol and e the error term. Ordinary least squares regression analysis assumes that error terms associated with each observation are uncorrelated. As time is a predictor in segmented regression analysis, error terms of consecutive observations could be correlated (first order autocorrelation): that is why we applied the Durbin–Watson statistics (values close to 2.00 indicate no serious autocorrelation) and visually inspect a plot of residuals against time.

All the statistical tests were 2-tailed and were considered significant when P <0 .05. The statistical analysis was conducted using STATA software (Stata Statistical Software, release 8.0, 2003; StataCorp, College Station, TX, USA).

Results

General population

From 1 January 2006 to 31 December 2007 a total of 1,649 patients were admitted, 827 and 822 during P1 and P2 respectively. Mean age was 58.99 ± 20.11 and 61.24 ± 18.76 years in P1 and P2 respectively (p = 0.05). There were a total of 10,553 patient days; mean length of ICU stay (median; 25th–75th percentile) was 6.88 (3; 1–9) days in P1 and 5.91 (2; 1–7) days in P2 (p = 0.006). The total number of days on mechanical ventilation were 6,839; the mean duration of mechanical ventilation (median; 25th–75th percentile) was 5.57 (2; 1–7) days in P1 and 4.43 (1; 1–5) days in P2 (p = 0.023).

Expected in-hospital mortality at 28 days calculated on the basis of the SAPS (SAPS2 during 2006 and SAPS3 during 2007) was 32.75% in P1 and 44.11% in P2 (p < 0.0001). Median GCS (25th–75th percentile) was 11 (7–15) in P1 and 11 (7–15) in P2 (p > 0.99).

Patients with VAP

We enrolled 112 patients with VAP; their baseline characteristics are summarised in Table 1. Epidemiological features were similar in both periods. In-hospital mortality predicted by SAPS was 34.08% in P1 and 45.13% in P2 (p = 0.03), and the crude ICU mortality was similar in both periods (Table 2). Among the pathologies on admission, we observed a reduction in patients with trauma (44.1% in P1 vs 29.5% in P2; p = 0.12), and an increase in patients with infection (7.4% in P1 vs 25% in P2; p = 0.004).

Table 1 Baseline characteristics of the study cohort
Table 2 Main clinical outcomes of the study cohort

There were a total of 79 VAP episodes in P1 and 44 in P2 (Table 2); the proportion of late onset VAP was similar in the two periods: 57% in P1 and 56.8% in P2 (p = 0.88). The mean incidence rate of VAP was 13.94 episodes per 1,000 patient days (20.96 episodes per 1,000 days of MV) in P1 and 9.05 episodes per 1,000 patient days (14.97 episodes per 1,000 days of MV) in P2 (Fig. 1). Segmented regression analysis showed no significant difference between periods.

Fig. 1
figure 1

Incidence of overall VAP episodes (dashed line) and VAP episodes caused by Pseudomonas aeruginosa (solid line) per 1,000 days of mechanical ventilation

Microbiology

Ventilator-associated pneumonia

There were 79 microbiologically documented cases of VAP (100%) in P1 and 41 (91%) in P2. A total of 149 micro-organisms were isolated: 93 in P1 and 47 in P2 (Table 3). Poly-microbial VAP was diagnosed in 14 cases (17.7%) during P1 and in 5 (10.6%) in P2 (p = 0.32). There were 42 episodes of VAP caused by antibiotic resistant Gram-negative bacteria (45.2%) in P1 and 16 (34%) in P2 (p = 0.21).

Table 3 Micro-organisms associated with VAP

The number of VAP caused by Pseudomonas aeruginosa passed from 8.35 per 1,000 days of MV in P1 to 2.33 per 1,000 days of MV in P2 (p = 0.02). We used segmented regression to analyse VAP episodes caused by Pseudomonas aeruginosa per 1,000 days of MV (Fig. 1). We observed:

  1. 1.

    A significant baseline trend (β1: 0.64; IC95 0.005–1.28; p = 0.048)

  2. 2.

    A significant rate reduction after the introduction of the rotation programme (β2: −7.54; IC95 from −13,81 to −1.27; p = 0.02)

  3. 3.

    A significant trend change after the introduction of the rotation programme (β3:−1.07; IC95 from −1.98 to −0.17; p = 0.02)

Both Durbin–Watson statistics and visual inspection of the plotting of residuals against time showed an absence of autocorrelation.

Microbial flora

Routine surveillance cultures were not performed. A total of 420 micro-organisms associated with infection were isolated through clinical cultures during the study (data not shown). The proportion of isolates for any single species was similar during the two study periods.

Bacterial resistance

Percentages of resistance to antimicrobial agents of potentially resistant Gram-negative bacteria responsible for VAP are summarised in Table 4. We observed a reduction in the percentage of Pseudomonas aeruginosa resistant to cefepime (p = 0.05) and to aminoglycosides (p = 0.2). For Klebsiella spp., Enterobacter spp., Serratia spp. and E. coli, we observed a reduction in resistance toward cefazolin (p = 0.004).

Table 4 Number (%) of antibiotic-resistant micro-organisms among the principal Gram-negative bacteria responsible for VAP

Antibiotic utilisation and compliance with rotation protocol

During the study there was a significant reduction in aminoglycoside use for the treatment of VAP: 7.73 vs 2.88 antibiotic courses per 1,000 patient days in P1 and P2 respectively (p = 0.014). No differences were observed for the other classes. Concerning the overall antibiotic use (Table 5), we observed a significant increase in the use of extended-spectrum penicillins, carbapenems and metronidazole. Conversely, there was a significant reduction in the use of third/fourth generation cephalosporins, fluoroquinolones, glycopeptides, aminoglycosides and macrolides. Quarterly consumption density of the most frequently prescribed antimicrobial groups in the two ICUs during 2007 is summarised in Fig. 2.

Table 5 Consumption density of the most frequently prescribed antimicrobial groups in the two ICUs during the study period
Fig. 2
figure 2

Consumption density per quarter of the most frequently prescribed antimicrobial groups in the two ICUs during 2007

Compliance with rotation protocol was high and similar in both ICUs. Adherence to the rotation protocol was 88% in centre A and 83% in centre B.

Discussion

This study analyses the effectiveness of an antibiotic rotation programme to reduce the incidence of VAP caused by antibiotic-resistant Gram-negative bacteria in the ICU. We observed a reduction in the VAP rate caused by Pseudomonas aeruginosa per 1,000 days of MV, although there was not an overall reduction in VAP episodes. There was also an augmented susceptibility of Pseudomonas aeruginosa toward cefepime and a reduced frequency of aminoglycosides use, an antibiotic class that is not considered nowadays to be first/second line treatment for VAP. Antibiotic use significantly changed during the study: we observed a significant increase in the use of extended-spectrum penicillins and carbapenems and a significant reduction in the use of third/fourth generation cephalosporins, fluoroquinolones, glycopeptides and aminoglycosides.

Much attention has been focussed in the past on the scheduled rotation of antibiotic therapy as a potential means of limiting the spread of resistant pathogens, but to date it has not been extensively studied to allow definitive conclusions. Although methodological differences make previous studies difficult to compare, a proportion of them reported positive results in terms of decreased incidence of antibiotic-resistant Gram-negative infections, particularly VAP [19, 20, 24], infections caused by multi-resistant pathogens [1921, 24] and infection-attributable mortality [21]. Gruson et al. [19] demonstrated that antimicrobial cycling in a medical ICU, significantly reduced VAP incidence and increased antimicrobial susceptibility, particularly of Pseudomonas aeruginosa, Burkholderia cepacia and methicillin-resistant Staphylococcus aureus (MRSA); these results have also been confirmed in the long term [20]. In a study by Kollef et al. [24] a change in empirical therapy for Gram-negative infection from ceftazidime to ciprofloxacin resulted in a decline in antibiotic-resistant Gram-negative VAP and overall VAP rates without a significant change in BSI incidence or mortality. Raymond et al. [21] studied the impact of rotating empirical antimicrobial regimens among surgical patients during a 2-year period, demonstrating a significant reduction in multi-drug resistant bacterial infections (both Gram-positive and Gram-negative) and infection-related mortality. In a study by Bennet et al. the implementation of an antibiotic rotation protocol in a surgical ICU resulted in overall improvement in the antibiotic susceptibility profile of Gram-negative microorganisms [38]. Despite these positive results, these studies, as highlighted by Nijssen et al., have important design limitations that increase the risk of confounders such as sub-optimal study design and end point, missing data regarding determination of the germs, different acquisition routes or data regarding infection control practices, such as adherence to hand hygiene [39]. Two different reviews [28, 40] emphasised the poor quality of most clinical trials available, concluding that cycling should not be implemented as a routine means of optimising antibiotic prescription and reducing antibiotic resistance. Similar indications are reported by the 2007 Infectious Disease Society of America (IDSA) guidelines on antibiotic stewardship: there have not been enough data in favour of antibiotic cycling to date to recommend this practice [27]. Contrariwise, another review [41] concluded that, although further study is needed, cycling could be an option for controlling antibiotic-resistant Gram-negative infections.

During this study we did not observe any significant reduction in:

  1. 1.

    Overall VAP incidence

  2. 2.

    Mean duration of ICU stay

  3. 3.

    Mean duration of MV

  4. 4.

    ICU mortality

Moreover, no significant change in microbial flora isolated through clinical cultures was observed. Interestingly, the rate of VAP caused by Pseudomonas aeruginosa, which represents the most common pathogen, was significantly reduced from P1 (8.35 per 1,000 days of MV) to P2 (2.33 per 1,000 days of MV; p = 0.02). We may only suppose that this reduction could be related to the introduction of the rotation programme, since, during the study, all the principal infection control measures were similar and no new interventions had been introduced, although we cannot exclude a “surveillance effect.”

We observed a reduced Pseudomonas aeruginosa resistance rate toward cefepime (p = 0.05). This could be explained by a low utilisation of the drug in both centres; it is well known, in fact, the correlation between antibiotic use and development of bacterial resistance [4244]. The change in resistance to cefazolin by KES. and E. coli should be explained by the very infrequent use of this antibiotic in both ICUs during the study. We have no evidence of clonal spread during the first period, although we cannot exclude it. No molecular investigation was performed to identify clonal spreading of these germs. We reported a decreased utilisation of aminoglycosides in P2 (p = 0.014). This could be attributed to the exclusion of this drug from the rotation programme and also to an increase in mono-therapy compared with association therapy for the empiric treatment of VAP. Regarding the overall antibiotic use, we found a reorganisation of the antibiotic prescription behaviour in both centres. Third/fourth generation cephalosporins and fluoroquinolones were partly substituted by extended-spectrum and anti-pseudomonal penicillins or carbapenems. Aminoglycosides were rarely used as first-line empiric treatment or prophylaxis; macrolides use was limited to the treatment of community-acquired pneumonia; glycopeptides were mainly prescribed for the treatment of microbiologically documented infections caused by methicillin-resistant Gram-positive cocci.

The main limitation of this study is represented by the small study population, determined mainly by the low VAP incidence observed in both centres, compared with that published in other reports [19, 24], possibly related to the baseline infection control measures, which were already active at the beginning of the study. Second, as VAP was the only infection studied, we cannot estimate the effect of antibiotic rotation on other infections such as bloodstream infection, peritonitis or urinary tract infection. Third, we did not perform routine surveillance cultures in order to evaluate colonisation with antibiotic-resistant germs nor did we perform isolates genotyping in order to define cross-transmission, endogenous selection of resistance or resistance development. Fourth, we did not analyse the costs of antibiotic prescription. Finally, as suggested by McGowan et al. [45], the “before–after” design did not represent the ideal model for the study of cycling.

We were able to conclude that, despite global microbial flora not being affected by such a programme, antibiotic therapy rotation may reduce the incidence of VAP caused by antibiotic-resistant Gram-negative bacteria in the ICU, such as Pseudomonas aeruginosa. The application of this programme may also improve antibiotic susceptibility. However, further studies are needed to confirm our results.