Total grams for each fluoroquinolone used during each year were electronically extracted from individual patient billing records and aggregated to reflect hospitalwide usage. The total number of patient days (PD) for the corresponding time period at each hospital was determined from the sum of individual patient lengths of stay. These data were used to express normalized antimicrobial drug use in defined daily doses per 1,000 patient days (DDD/1,000PD) as recommended by the World Health Organization (WHO) (http://www.whocc.no/atcddd/). The DDDs used were levofloxacin 500 mg, moxifloxacin 400 mg, gatifloxacin 400 mg, and ciprofloxacin 1,000 mg. (Because the MMIT database does not indicate the proportion of intravenous versus oral ciprofloxacin used, we used the oral DDD for ciprofloxacin of 1,000 mg for all ciprofloxacin use).

The results of this longitudinal, multicenter study of fluoroquinolone use and bacterial resistance suggest a complex relationship between fluoroquinolone use in hospitals and percentage of methicillin-resistant S. aureus and fluoroquinolone-resistant P. aeruginosa when viewed from an ecologic level. Fluoroquinolone use was linked to greater percent resistance in study years by univariate linear regression. This observation is consistent with the results of other studies linking fluoroquinolone use to resistance in these pathogens (4–11), although the pathways for selection of resistance differ between the 2 pathogens. Fluoroquinolone resistance in P. aeruginosa is believed to arise largely from the selection of organisms with point mutations in the topoisomerase enzymes that are targets for the fluoroquinolones (13). This hypothesis is supported by studies demonstrating the emergence of resistance during therapy with fluoroquinolones (4). For methicillin-resistant S. aureus, de novo emergence of resistance as seen with P. aeruginosa is not a common event (14); rather, patients are generally believed to acquire methicillin-resistant strains of S. aureus from the environment (e.g., through cross-transmission); antimicrobial drug use may increase the likelihood of colonization or amplify the resistant population after colonization (15). Fluoroquinolones are not active against most methicillin-resistant isolates (16), providing a selective pressure for MRSA. While any antimicrobial agent that is not active against MRSA should increase a patient's risk for infection, fluoroquinolones may be particularly likely to do so, since they have appear to have unique effects on the expression of MRSA resistance determinants (17) and fibronectin-binding proteins (18).

Using a model incorporating the previous year's percent resistance as well as fluoroquinolone use and time over the entire study period, we did not find an additional effect of total fluoroquinolone use on percent resistance. Many factors beyond the volume of use of an antimicrobial agent affect the emergence and spread of antimicrobial resistance to that drug in the hospital setting. Cross-transmission between patients, acquisition of organisms from the hospital environment, and the use of different antimicrobial agents with linked resistance to the agent under study are all factors that also affect the number of resistant isolates in a given hospital (19–21). These factors are difficult to control for, since quantitative measures of infection control are lacking, and examining the effects of multiple antimicrobial agents complicates analysis. We also did not account for antimicrobial drug use in the community, which we have previously reported to be associated with hospital resistance rates (22). Because of the large number of variables that may influence resistance in the hospital setting, we would expect hospital fluoroquinolone use to have at best a modest effect on resistance. The results of our model suggest that the ecologic effect size of fluoroquinolone use is such that substantial changes in fluoroquinolone use may be required to affect percent resistance. Thus, a larger sample size, which incorporates more hospitals over a greater period, may be required to demonstrate significance, if such effects exist.

Because percent antimicrobial resistance in a given period is highly correlated to previous percent resistance (i.e., the observations are not independent), we incorporated the previous year's percent resistance into our longitudinal model. Failing to account for this autocorrelation may result in spurious associations (23). However, our use of a 1-year lag period was imposed by the nature of our data (which was aggregated on a yearly basis) rather than biologic considerations and may not be the most accurate method of modeling this relationship. We did not time-lag the effect of fluoroquinolone use. Monnet et al. used time-series analysis with autoregressive integrative moving average (ARIMA) functions to model the changes in percent MRSA in a hospital in Scotland during a 3-year period (24). Their transfer function model for predicting percent MRSA used the previous month's percent MRSA; the effect of fluoroquinolone use had a 4-month lag in the effect of changes in fluoroquinolone use on percent MRSA. No study has specifically examined lag effects with fluoroquinolone use and fluoroquinolone-resistant P. aeruginosa, although Lopez-Lozano et al. used a lag period of 3 and 5 months in percent resistance and a 1-month lag in ceftazidime use to model the relationship between ceftazidime use and ceftazidime-resistant gram-negative bacilli (25). Thus, use of yearly antibiograms may not be adequate to properly model the relationships between percent resistance in a given year, previous resistance, and fluoroquinolone use. Future studies should attempt to attain more detailed data to allow for more flexibility in modeling the relationship between antimicrobial use and resistance.

The results of this study raise a number of questions for further investigation. Why did the association between fluoroquinolone use and resistance become weaker in the later study years? Does this finding represent random fluctuation or an underlying trend? Mean fluoroquinolone use, after increasing through the first 3 study years, reached a plateau in the last 2 years (Figure 1). Perhaps associations with resistance are reflected most strongly when antimicrobial drug use is increasing, as was seen in the first 3 study years. The study by Zervos et al. (11) found a significant association between changes in fluoroquinolone use and changes in resistance in P. aeruginosa in 10 teaching hospitals from 1991 to 2000. During this period, fluoroquinolone use increased by a mean of 97% in the participating hospitals; in the present study, the mean increase during the study period was 23%. Also, the percentage of resistance among MRSA and P. aeruginosa was relatively high in all hospitals from the beginning of this study. Monnet et al., in their study of the impact of antimicrobial drug use on an outbreak of MRSA in Scotland, observed that "…antimicrobial drug use was a more important ecologic risk factor at the start of the outbreak than once MRSA had become endemic in the hospital" (24). After antimicrobial use "drives" resistance to a certain point, other effects such as cross-transmission may become the dominant mode of spread. From a scientific standpoint, this finding suggests that ecologic studies may be more likely to detect a significant effect when a particular form of resistance is in its "infancy" (current examples might be linezolid-resistant S. aureus or fluoroquinolone-resistant Streptococcus pneumoniae). From a clinical point of view, this finding suggests that measures to ensure proper antimicrobial drug use would have the most effect before resistance becomes widespread.

Also of interest is the effect of individual fluoroquinolones on resistance. In our univariate analyses, levofloxacin use had a much stronger association with resistance in both pathogens than did use of ciprofloxacin. Levofloxacin also showed the only significant relationship (with percent MRSA) among the longitudinal models. However, the univariate models did not control for prior resistance levels. Also, the population-averaged modeling approach results in a pooling of effects across all hospitals, resulting in some degree of effect attenuation. Indeed, the estimates from a GEE analysis can sometimes be smaller than those estimates produced from a corresponding mixed-effect model (12). This apparent effect attenuation can be viewed as a trade-off for the reduced vulnerability to model misspecification, as previously discussed. Relationships between individual fluoroquinolone use and resistance should be interpreted with caution, as it is possible that our study is biased (an unknown confounder that is causally associated with MRSA or resistant P. aeruginosa occurs more frequently in hospitals that use more levofloxacin). However, other ecologic studies have found similar results. At a single teaching institution in an 8-year period, Mohr et al. observed an association between increasing levofloxacin use and increasing percent resistance in P. aeruginosa; this effect was not observed for total fluoroquinolone use or use of other fluoroquinolones (26). An ecologic study across 174 hospitals across 6 years by Bhavnani et al. showed that increasing use (measured as drug expenditures) of levofloxacin and ofloxacin, but not ciprofloxacin, was associated with increased percent resistance to ciprofloxacin in P. aeruginosa (10). This ecologic effect might arise from a greater in vitro potential for levofloxacin to select for resistant mutants of P. aeruginosa, as Gilbert and colleagues have observed (27). For MRSA, a different relationship between fluoroquinolone use and resistance is relevant, since the emergence of methicillin resistance during therapy is not of concern. When viewed from the perspective of selecting for preexisting methicillin-resistant, fluoroquinolone-resistant Staphylococcus aureus from a population that contains both resistant and susceptible isolates, levofloxacin might be more likely to select for the preexisting resistant clones because of its greater activity against the susceptible organisms. In such a scenario, the proportion of resistant isolates (percent resistance) may increase, although the incidence of infections due to resistant bacteria may be relatively stable, as argued by Schwaber et al. (28). Thus, analysis of incidence rates of isolation of resistant organisms may tell a different story from analysis of percentage of resistant organisms. We did not determine incidence rates because some of our study hospitals did not include the number of organisms isolated on their antibiograms. Although percent resistance is most likely of concern to a clinician, the incidence rates determine the overall effect on impact of resistance from an ecologic perspective. Future studies should incorporate resistance rates alongside changes in percent resistance to give the complete picture of the effect of antimicrobial drug use on resistance. The differential effects of individual fluoroquinolones on antimicrobial drug resistance are an important area for future study, as hospitals manipulate their formularies with regard to use of individual fluoroquinolones, often for economic reasons.

This study has a number of limitations. The study hospitals do not represent a random sample of US hospitals; further studies are required to determine whether the results are broadly applicable. We were unable to control for differences between hospitals in their methods of antibiogram construction, including methods and reporting of duplicate isolates, which can affect reported resistance (29), as well as hospital culturing practices (such as MRSA screening). Standardization of methods to report antibiograms has been advocated by the Clinical and Laboratory Standards Institute (formerly NCCLS), but adherence is poor (30). We also were not able to control for differences in infection control measures between hospitals, which are a likely source of variability in the prevalence of resistant organisms. Finally, the associations between antimicrobial drug use and resistance found in ecologic studies such as this may not always coincide with those observed on an individual patient level (31). Case-control studies are more appropriate to quantify the risk associated with antimicrobial drug exposure in individual patients.

Our results suggest that the ecologic relationship between the hospital use of fluoroquinolones and antimicrobial resistance in P. aeruginosa and S. aureus is complex. Future studies to better define this relationship would be worthwhile, in that they would help hospitals determine where to best invest their resources to reduce the overall impact of resistance in their institutions, whether through enhanced infection control measures or more active antimicrobial stewardship, although both measures are likely important. Meanwhile, judicious use of fluoroquinolones is advocated to prevent the loss of this valuable therapeutic class.