Fluoroquinolone consumption and Escherichia coli resistance in Japan: an ecological study

The frequency of antimicrobial resistance (AMR) has steadily increased worldwide, induced by inappropriate use of antibiotics in a variety of settings [1]. Concurrent development of new antibiotics, however, has not necessarily been accelerated, rendering the issue of resistance a global concern. The World Health Organization endorsed the Global Action Plan on Antimicrobial Resistance in May 2015, encouraging all member countries to formulate national action plans within 2 years [1]. In response, Japan issued its National Action Plan on Antimicrobial Resistance in April 2016, explicitly specifying a number of priority goals to be achieved over the next 5 years to promote AMR measures [2]. Accordingly, various surveillance initiatives and resistance countermeasures are presently underway in the country.

In its Action Plan [2], Japan aims to lower the fraction of fluoroquinolone resistance in Escherichia coli from 45% in 2014 to less than 25% by 2020. Moreover, it aims to reduce daily fluoroquinolone use by 50% by 2020. E. coli belongs to the family Enterobacteriaceae, typically found in the lower gastrointestinal tract as microbial flora, and can cause urinary tract infection and sepsis if introduced into the bloodstream. β-lactam or quinolone antibiotics are frequently the first-choice treatment for E. coli infection [3]. Quinolones are highly bioavailable, permitting oral administration and good tissue distribution, and thus may be one of the most frequently used antibiotics at both outpatient and inpatient settings.

Increased use of antibiotics has been recognized as a major factor driving E. coli resistance over the last decades [4]. Fluoroquinolone resistance is easily established by point mutations in the DNA gyrase and via plasmid-mediated transfer [5], and various studies have indicated that the use of fluoroquinolone is most likely closely associated with the increase in fluoroquinolone resistance in E. coli [6‐17]. Moreover, a few published studies reported that lowering the use of fluoroquinolone decreased the resistance proportion [18‐20].

Published studies on the relationship between fluoroquinolone consumption and resistance in E. coli, however, have been mostly limited to specific healthcare facility settings, and findings on a national scale are yet to be reported. Moreover, the causal relationship must be carefully explored while adjusting for confounders. The purpose of the present study is to assess the ecological correlations between fluoroquinolone consumption and levofloxacin resistance in E. coli in Japan, using datasets from all 47 prefectures. Identifying the ecological association over the geographic space, one can further examine the causal link and underlying mechanisms that could potentially constitute future countermeasures.

Table 1 shows the summary statistics of the variables examined that were considered potentially associated with the geographic heterogeneity of levofloxacin-resistant E. coli proportions. In 2015 and 2016, median values (and IQR) of the total isolates of E. coli by prefecture were 4017 (IQR: 2141-6333) and 3969 (IQR: 2744-7133), respectively. Out of those samples of E. coli isolates, the overall proportion of levofloxacin-resistant E. coli (hereafter, referred to as the “resistance proportion”) was 37.4% in 2015 and 38.3% in 2016. Median consumption of fluoroquinolones was 2.8 DID in both 2015 and 2016. Of the total prescriptions, the proportion of orally administered fluoroquinolones was 98.6% in 2015 and 98.5% in 2016, respectively. Figure 1 shows the geographic distributions of fluoroquinolone consumption and levofloxacin resistance. Among the 47 prefectures in Japan, the proportion of levofloxacin resistance in E. coli was overall lower in the eastern region than in the west, with the lowest proportions estimated in Yamagata and Aomori (27.1 and 26.2%, in 2015 and 2016, respectively; Fig. 1a and b). The highest proportion was seen in the western Saga prefecture, and estimated at 51.1 and 55.0% in 2015 and 2016, respectively. Similarly, fluoroquinolone consumption was lower in eastern prefectures than in the west on a whole, and the lowest values were observed for Yamagata and Fukui. The highest value of fluoroquinolone consumption was observed in Tokushima, a western prefecture (Fig. 1c and d).

Table 2 summarizes the results of univariate correlation by year. The following variables were consistently and positively correlated with the proportion of levofloxacin resistance: Fluoroquinolone DID, the number of physicians, the number of nurses, and the number of hospitals and clinics. In contrast, the proportion of levofloxacin resistance was not significantly correlated with the number of nursing homes, the number of medical facilities registered in JANIS, the proportion of registered facilities with 500 or more beds, the average length of hospital stay, or the proportion of elderly individuals in the population. Figure 2 shows the crude univariate correlation between consumption of fluoroquinolone and the proportion of levofloxacin resistance. The linear regression coefficient was estimated at 0.50 and 0.52 for 2015 and 2016, respectively.

Table 3 shows the results from multivariate analyses. In advance of conducting multivariate linear regression, we identified a strong positive cross-correlation between fluoroquinolone consumption and the number of physicians (regression coefficient = 0.008 for both 2015 and 2016 for each one-physician increase per 100,000 individuals; p < 0.001). We therefore used the product of these two variables as an explanatory variable to address the possible interaction. When including all variables that were significantly correlated in the univariate analysis in the final multivariate model, no significant variables remained in the final model for either 2015 or 2016 data, and fluoroquinolone consumption was not significant (p = 0.09) with a large effect at 19.3 per DID increase in 2015, and not statistically significant (p = 0.16) with a large effect at 17.1 per DID increase in 2016. Employing the stepwise method, only fluoroquinolone consumption remained in the final model, with 5.0 and 5.5 per DID increase in 2015 and 2016, respectively (p < 0.01 for both years). The coefficient of determination was estimated at 0.25 for 2015 and 0.27 for 2016, with F-values of 14.8 and 17.0, respectively (p < 0.01 for both years), reflecting an overall success in identifying at least some factors involved in levofloxacin resistance.

The present study examined the ecological correlation between fluoroquinolone consumption and levofloxacin-resistant E. coli in Japan, adjusting for potential confounding variables. Comparing fluoroquinolone consumption and levofloxacin resistance by prefecture across Japan, we identified relatively higher values in the western region and lower values in the eastern region. To explain the underlying mechanisms of such geographic heterogeneities, we conducted univariate analyses, and identified not only fluoroquinolone consumption but also the numbers of physicians, nurses, hospitals, and clinics per 100,000 individuals as positively correlated with the proportion of levofloxacin-resistant E. coli. Multivariate analyses validated the positive correlation between fluoroquinolone consumption and levofloxacin resistance.

The present study’s largest contribution is its demonstration of the link between fluoroquinolone consumption and levofloxacin resistance using nationwide surveillance data, while appropriately addressing potential confounding variables. Our study rested on a digitalized laboratory-based surveillance system, which covers 20% of healthcare facilities and 80% of medical facilities with 500 or more beds; to our knowledge, no published study has explored this relationship using a comparably large national-scale dataset. Antibiotic consumption has already been shown to be positively and strongly correlated with prescription [30], and thus, our study endorses published conclusions [6‐17] that the prescription of fluoroquinolones induces the emergence (and perhaps maintenance) of resistant E. coli. Various factors may contribute to the development of levofloxacin resistance in E. coli, but our study did not identify any significant healthcare services-associated variables that were correlated with levofloxacin resistance. A high density of physicians could lead to competition and reduce hospital visits by individuals feeling unwell, in turn promoting inappropriate (and increased) prescription of fluoroquinolones. This notion was in line with the univariate analysis, but the number of physicians was removed in the final multivariate model, indicating that the information that was contained in our dataset (i.e. only the variations among the 47 prefectures) was not sufficient to establish a relationship between the number of physicians per 100,000 individuals and levofloxacin resistance. Only fluoroquinolone consumption was left in the final multivariate model using the stepwise method, as this variable is a strong predictor of geographic heterogeneity of levofloxacin-resistant E. coli.

An important implication of the correlation between fluoroquinolone consumption and levofloxacin resistance in E. coli is that the restricted use of fluoroquinolones could potentially contribute to lowering the resistance proportion, if the resistant E. coli is evolutionarily unfit. The recovery of sensitivity at the population level depends on the fitness cost (i.e. transmission and self-maintenance capacity) of resistant E. coli. The cross-sectional design has ignored two important aspects of the emergence: (i) The induction period, i.e. the time from prescription to an increase in resistant E. coli, and (ii) the length of survival of resistant E. coli as part of the human gut microflora or elsewhere [31]. In addition to these known factors, it is worth noting that cross-resistance from the use of other antibiotics is also indicated [11, 32, 33]. The population benefit of restricted prescription requires additional epidemiological evidence and analyses. In particular, the present study was not able to account for disease burden that is associated with resistant E. coli, including urinary tract infections and respiratory tract infections, because they are not categorized as notifiable diseases in Japan. The recovery of sensitivity to fluoroquinolones should ideally be attributed to the reduction in those disease burden estimates via an appropriate epidemiological study design.

In Japan, medical expenditure has been shown to exhibit geographically heterogeneous patterns, mostly indicating higher expenditure in the west and lower in the east [34]. We have shown that fluoroquinolone consumption and resistance are no exceptions: Despite the management of healthcare services under the same system, such geographic heterogeneities are evidently observed, and no clear mechanisms were identified to explain the differences. Our study demonstrated that geographic heterogeneity of levofloxacin resistance was most likely regulated by that of fluoroquinolone consumption, and the mechanisms underlying the increase in resistance should be determined in the context of antimicrobial stewardship [35‐40]. A top-down approach to restricting inappropriate prescription of antibiotics may yield evidence of the causal link suggested by our analyses.

Several limitations must be noted. First, the present study rests on an ecological study design that is known to be prone to unobserved confounding. What has been identified does not immediately lead to a causal link at the individual level. Second, the JANIS data that we mined were not accompanied by information on the type of clinical specimen or sensitivity to other antimicrobial agents. More detailed insights into levofloxacin resistance could be gained by exploring a body part from which the samples were obtained, and also examining drug sensitivity of the same bacterium to other antimicrobial agents. Third, antimicrobial consumption was measured in the present study by defined daily dose, but this approach overlooks individual heterogeneities in the prescribed dose (e.g. a lower dose in children). Published studies at the individual level can overcome this problem. Fourth, we used the number of physicians as one of variables, but the number was not stratified by hospital and primary care physicians. For the entire Japan, we have an access to the total count, i.e., there are 101,884 physicians working for clinics out of the total of 296,845 (34.3%), but we did not have an access to the corresponding numbers by prefecture. Fifth, we used consumption and resistant frequency data, but the fluoroquinolone use at outpatient and inpatient pharmaceutical sales were not separately recorded, and moreover, microbial samples were also not stratified by inpatient and outpatient services. The distinction has an important implication for considering future restriction of antimicrobial use. Sixth, we examined only 2015 and 2016 data, because of limited access to antimicrobial consumption data. The longer the data series are collected, the more plausible that we will have a chance to elucidate the causal link, especially in the case fluoroquinolone use may be restricted from some future point in time.

In conclusion, we showed an ecological correlation between fluoroquinolone consumption and levofloxacin resistance in Japan, analyzing prefectural datasets from two recent years. The geospatial pattern of fluoroquinolone use correlated with that of resistance. The relationship between the two must be elucidated using additional studies with different epidemiological designs, so that any possible countermeasures, including alternative prescription, can be considered in the future.