Results
Study selection turned out to be challenging. This was partly due to the characteristics of the intervention studied which preclude blinding and limit options for even partial randomization. The search process retrieved a sole cluster randomised trial and for this reason we decided to include less robust study types and perform a sensitivity analysis in case of conflicting results.
Fifteen relevant studies formed our dataset including seven prospective trials and eight simple before-and-after studies (Table 1). Nine studies evaluated the effects of antibiotic cycling versus a control group[13][14][15][16][17][18][19][20][21]. Three papers compared antimicrobial cycling with antibiotic mixing[22][23][24], that is administering the scheduled antimicrobial agents on a successive patient basis. The last three assessed the resistance potential of each of the alternating on-cycle antibiotics, that is the variations in risk of antibiotic resistant infection and/or colonization during cycles of different predominant antibiotic use[25][26][27].
Fixed durations of each cycle ranged from one week to eight months. The rotating agents were piperacillin-tazobactam with cefepime in two cases[13][25], fluoroquinolones with beta lactams in three cases[18][26][27]. The rest rotated the aforementioned agents with carbapenems and aminoglycosides in varying combinations. In some protocols de-escalation to suitable narrow-spectrum agents was permitted but in others it was not, with six teams proceeding to de-escalation in view of bacterial susceptibility results[16][17][19][23][24][27], five teams avoiding de-escalation to increase the on-cycle antimicrobial use[14][15][18][21][26] and four teams not clarifying their practices enough for their readers to be able to ascertain specifically what they did[13][20][22][25]. Four studies provided bacterial typing data to assist in the evaluation of cross-transmission dynamics[14][18][25][27]. Furthermore, methodologies differed as to whether surveillance cultures or cultures from clinically presumed infections, unit-wide or patient-specific, were recorded as indicators of resistance incidence.
Among those studies which compared an experimental with a control cohort there were seven simple before-and-after and two prospective trials. Seven of these provided data with regard to antimicrobial protocols in the control group[14][15][16][18][19][20][21] and two did not set out their standard practice[13][17]. Oddly, many studies fail to state any explicit goal of their chosen intervention, but the available information suggests that the institution of an antimicrobial rotation policy aimed to increase heterogeneity of antimicrobial administration in the intervention group by utilising more antimicrobial classes of similar spectrum in a scheduled fashion. The results, however, appear rather conflicting.
In particular, if one takes into account bacterial susceptibilities to the rotated agents which are apparently a more straightforward indicator of the policy’s effectiveness four studies did not achieve any measurable success and five reported variable improvement (Table 1). The most noteworthy study in the group reporting negative findings is probably the trial conducted by Toltzis et al. Its main distinctive feature is the use of a contemporary control group, and its use of bacterial typing data facilitates interpretation of the available findings. The researchers observed no benefits even when only clonally discordant isolates were taken into account[14].
The group reporting positive findings encompassed two studies which observed an increase in P. aeruginosa susceptibility to one and two of the rotated agents respectively[17][18] and two studies which reported improvements in Extended-spectrum Beta Lactamase (ESBL) incidence (p<0.05)[20][21]. One of the latter used a rather small sample while none of the aforementioned seemingly successful studies utilized bacterial typing. Thus, the possibility that the observed findings could be a result of horizontal transfer of bacterial clones due to breaks in infection control cannot be excluded as in the study conducted by Toltzis et al.
Nijssen et al reported lower colonization rates for ciprofloxacin-resistant isolates in the intervention group but no changes for cephalosporin-resistant isolates[18]. Highly homogeneous prescription of fluoroquinolones in the control arm, a radical reduction in ciprofloxacin administration in the intervention arm along with the main mechanism of fluoroquinolone resistance which incurs spontaneous chromosomal mutations favoured by increased selective pressures could perhaps explain the observed results, but no firm interpretation is possible.
Frequency of cycling did not appear to be associated with the possibility of positive or inconclusive outcomes as it varied widely in both groups. Furthermore, the fact that universal lack of randomization and blinding would potentially predispose to some degree of selection and information bias in favour of more positive outcomes, and while no specific biases were evident, this inevitable contextual bias should be taken into account.
Three studies assessed antimicrobial rotation compared to administering the agents on a successive patient basis to maximise antibiotic heterogeneity, a practice known as antibiotic mixing. Two of those, including one using the robust cluster-randomised cross-over design, observed no significant differences[23][24]. Jayashree et al reported lower resistance rates in both cycling and mixing periods compared to a three-month baseline period. The latter, however, was too short to be informative[24]. The third reported higher cefepime susceptibility rates for P. aeruginosa during cycling (p=0.01) but no further improvements[22]. De-escalation as well as combination therapy were permitted in two instances[23][24], and their allowability was not clarified in the third[22]. None of the teams used typing data to assess cross-transmission dynamics.
As for the remaining studies, Ginn et al cycled piperacillin-tazobactam with cefepime and found that cefepime showed as a more important driver for the onset of bacterial resistance with the proportion of admissions complicated by resistant infections during cefepime cycles being more than twice as high compared to piperacillin-tazobactam cycles (p<0.001)[25]. Van Loon et al cycled levofloxacin with cefpirome and piperacillin-tazobactam. concluding that levofloxacin use was associated with higher levofloxacin-resistance rates, but cefpirome was seemingly not prone to the selection of cefpirome-resistant strains[26]. Tsukayama et al rotated fluoroquinolones with piperacillin-tazobactam but did not find any significant correlations between the on-cycle antibiotic class and the probability of resistance onset. However, the authors report high use of off-cycle antibiotics which could potentially act as a confounding factor[27].
Finally, all but two studies provided some data regarding the on- and off-cycle antimicrobial consumption during the experimental period, while seven studies measured variable side effects as indicators of the policy’s potential collateral damage including morbidity and/or mortality rates reported by six studies[15][16][19][22][23][24]. None of these recorded worrying trends in intervention groups.