Background

Klebsiella pneumoniae is a major nosocomial pathogen that causes pneumonia, urinary tract infections, and bacteremia [1]. The widespread use of broad-spectrum cephalosporins over the last several decades has resulted in extended-spectrum β-lactamase (ESBL)–producing strains of K. pneumoniae becoming endemic in hospitals worldwide [2, 3]. ESBLs hydrolyze β-lactam antibiotics such as penicillins and most cephalosporins, thereby conferring resistance to these drugs [4, 5]. Thus, ESBL-producing K. pneumoniae are usually resistant to at least one extended-spectrum cephalosporin [6]. Delays in effective treatment result in worse outcomes including increased length of hospital stays, hospitalization costs, and mortality compared to infections caused by non-ESBL producing strains [6]. The burden caused by these strains is particularly high in Korean hospitals, where approximately 53% of K. pneumoniae infections are caused by ESBL-producing strains [7].

To overcome this β-lactamase–mediated resistance and to enhance the efficacy of β-lactam antibiotics, β-lactam/β-lactamase inhibitor (BLBLI) combinations such as ampicillin/sulbactam and piperacillin/tazobactams have been introduced in clinical practice [8]. Although these drugs are effective in treating infections caused by ESBL-producing organisms [8, 9], carbapenems have largely remained the treatment of choice [2, 11], as studies have reported conflicting results regarding the relative efficacy of therapy with BLBLIs compared to that with carbapenems [12]. However, the emergence of carbapenemase-producing Enterobacteriaceae [13, 14] has increased the need to utilize carbapenem-sparing treatment options, such as BLBLIs, for ESBL-related infections [10, 11]. However, it is now a growing concern that the increased use of BLBLIs will likely lead to increased resistance to these drugs [15, 16]. One previous molecular study supports this concern, indicating that the use of BLBLIs selects for point mutants in TEM penicillinases resistant to inhibitors [17].

Although there is a generally well-established link between overall antibiotic use and resistance within hospitals [18, 19], information is still lacking on the relationship between BLBLI consumption and antibiotic resistance in K. pneumoniae. In this study, we examined the temporal association between the consumption of antibiotics commonly used in hospitals, particularly BLBLIs, third-generation cephalosporins, and fluoroquinolones, and antimicrobial resistance in K. pneumoniae. This study aimed to increase understanding of the relationship between the use of and resistance to BLBLIs to improve decision-making and policies on the use of BLBLIs as a carbapenem-sparing option.

Methods

Antibiotic prescription data

The data for antibiotic utilization (total number of antibiotics prescribed) per quarter in a designated hospital was obtained from the database of the Korean Health Insurance Review and Assessment Service (KHIRA). KHIRA covers 98% of the Korean population and includes prescription data for patients from all Korean medical institutions based on financial claims [20]. The antibiotics from the database were coded according to the World Health Organization (WHO)‘s Anatomical Therapeutic Chemical (ATC) Classification [21]. Since classes of antibiotics have similar underlying mechanisms against bacteria, this classification is widely used to measure the relationship between prevalence of bacterial isolates’ resistance and the exposure to different antibiotics [22, 23]. Thus, systemic antibiotic prescriptions for all BLBLIs (ATC code: J01CR), all third-generation cephalosporins (J01DD), and all fluoroquinolones (J01MA) from the hospital were included.

Antimicrobial resistance data

Microbial culture data were extracted from the existing database of the hospital and aggregated to generate quarterly estimates of antibiotic resistance [24]. The hospital has 2200 beds and has been designated as a Korean antimicrobial resistance monitoring focal point by the WHO since 1988 [25]. The specimens were collected from outpatient and inpatient departments. Identification and susceptibility testing were performed using an automated system (Vitek; bioMerieux) or disk diffusion tests in accordance with Clinical Laboratory Standards Institute (CLSI) criteria [24]. Piperacillin/tazobactam, ceftazidime, and levofloxacin were used for the susceptibility test, representing BLBLIs, third-generation cephalosporins, and fluoroquinolones, respectively. Duplicate isolates and intermediate susceptibility were not included in the resistance data [26–28]. The quarterly rate of piperacillin/tazobactam, ceftazidime, and levofloxacin resistance of K. pneumoniae isolates were obtained. The resistance rates were calculated as the number of resistant isolates divided by the number of tests in each quarter/year.

Statistical analyses

We used regression analysis to determine the trends of annual antibiotic prescriptions and antimicrobial resistance during the study period. P values less than 0.05 with R-squared values greater than 0.3 were considered statistically significant [29]. Furthermore, to identify the temporal relationship between the quarterly number of antibiotics prescriptions and the quarterly isolation rate of antibiotic susceptibility, a cross-correlation function test was performed. This cross-correlation function test is widely used to identify the time lags of one time series (antibiotic prescribing) with the possible predictors of another time series (antibiotic resistance) [30]. The Box-Jenkins method was applied to fit the time series data to seasonal autoregressive moving average models [31, 32]. Stationary was evaluated using the augmented Dickey-Fuller test to determine whether differencing was required to rule out spurious correlations. The Akaike information criterion test and the portmanteau test of the residuals were used to identify the best model fit. The residuals from each time series model were used to evaluate the temporal relationship between antibiotics prescription and isolation rate of K. pneumoniae with antibiotic resistance [33]. All statistical analyses were performed using R version 3.2.4 (R Foundation for Statistical Computing, Vienna, Austria).

Results

The average annual number of antibiotics prescriptions during the study period was 69,100 (range 27,371–84,128) for all BLBLIs; 53,300 (range 22,789–70,514) for all third-generation cephalosporins; and 28,536 (range 12,324–35,368) for all fluoroquinolones (Table 1). There was no significant trend in prescribing rates over the study period.

Table 1 Average quarterly number of antibiotic prescriptions in a tertiary hospital from 2012 to 2016
Full size table

The yearly numbers of K. pneumoniae isolates from 2012 to 2016 for this analysis were 1524, 1609, 1585, and 1754. The mean rates of K. pneumoniae resistance to piperacillin/tazobactam, ceftazidime, and levofloxacin were 41% (range 22–49), 34% (range 14–44), and 30% (range 23–44), respectively (Table 2). There was a stable trend in resistance rates against ceftazidime and levofloxacin; however, the resistance rate against piperacillin/tazobactam significantly increased from 34% (range: 22–46) in 2012 to 46% (range: 43–49) in 2016; (p < 0.01).

Table 2 Quarterly antimicrobial resistance in K. pneumoniae in a tertiary hospital from 2012 to 2016
Full size table

In a bivariate analysis, the lagged (two quarters) quarterly number of all BLBLI prescriptions significantly correlated with K. pneumoniae resistance to piperacillin/tazobactam (β = 0.66; p < 0.01), ceftazidime (β = 0.54; p = 0.02), and levofloxacin (β = − 0.60; p = 0.01). Similarly, the lagged quarterly number of all third-generation cephalosporin prescriptions significantly correlated with K. pneumoniae resistance to ceftazidime (β = 0.64; p < 0.01; two quarters) and levofloxacin (β = 0.50; p = 0.03; one quarter). The lagged quarterly number of all quinolone prescriptions correlated with K. pneumoniae resistance to ceftazidime (β = 0.14; p < 0.01; two quarters), although there was no significant association with levofloxacin resistance (β = 0.23; p = 0.35) (Table 3).

Table 3 Cross-correlation coefficients between quarterly antibiotics prescriptions and quarterly isolation rates of antibiotic-resistant K pneumoniae from 2012 to 2016
Full size table

Discussion

The emergence of ESBL-producing K. pneumoniae strains and the increasing rates of carbapenem resistance have reduced therapeutic options for infections caused by these organisms [2, 6, 10]. BLBLIs such as piperacillin/tazobactam have been suggested as alternatives to carbapenems [10, 11, 12] and may be comparable in efficacy to carbapenems against ESBL-producing K. pneumoniae infections [34]. However, although the evidence for their use as an alternative to carbapenems has been conflicting [12], their use has been increasing. This has led to increasing concern about resistance, which undermines the efficacy of these drugs [12].

To improve policies around the potential replacement of carbapenems with BLBLIs, it is important to examine the relationship between BLBLI use and antimicrobial resistance using a seasonal analysis, which previous studies used to demonstrate the linkage between seasonality and antibiotic use [22, 35].

The greatest positive temporal association was found between all BLBLI utilization and piperacillin/tazobactam resistance in K. pneumoniae. However, no significant correlation was found for the use of the third-generation cephalosporins and fluoroquinolone. This was presumably due to the selection pressure of antibiotic use on the prevalence of resistant isolates. This finding is consistent with results of a previous study that identified a significant correlation between piperacillin/tazobactam use and subsequent piperacillin/tazobactam resistance in Enterobacteriaceae [36].

We also found that ceftazidime resistance in K. pneumoniae temporally correlated with prescriptions for all BLBLIs, all third-generation cephalosporins, and all fluoroquinolones. Consistent with our findings, a previous report indicated that the presence of ceftazidime-resistant Klebsiella spp. is significantly positively associated with cephalosporin and fluoroquinolone use at the hospital level [37]. Regarding the positive association between fluoroquinolone use and the prevalence of ceftazidime resistance to K. pneumoniae, this may be explained by the transferrable plasmid (qnr gene) among ESBL-producing strains which leads to quinolone resistance [38]. Thus, these correlations are likely due to the ability of ESBLs to confer resistance to multiple classes of antibiotics [39].

However, we did not find an association between levofloxacin resistance in K. pneumoniae and the consumption of quinolones; this result was in contrast to some previous reports which found that increasing fluoroquinolone use was associated with rising incidence of quinolone resistance in Enterobacteriaceae [40, 41]. Additionally, we found that levofloxacin resistance in K. pneumoniae was negatively associated with all BLBLI utilization and positively associated with third-generation cephalosporin utilization. These results may be due to the low level of fluctuation from the quarterly number of fluoroquinolones used, and the low magnitude of utilization likely affected the resistance emergence rates [42].

This is the first study to describe the temporal relationship between BLBLI use and the resistance to them in K. pneumoniae in a tertiary level hospital. The strong correlations reported in our study suggest that replacing carbapenems with BLBLIs is not an ideal long-term solution, as resistance is likely to spread rapidly. A previous study in Korea showed that broad-spectrum penicillins including BLBLIs were more frequently prescribed than narrow-spectrum penicillins [43]. Another study demonstrated that this is presumably due to the physician’s desire to avoid missing an unlikely or unidentified resistant pathogen [44]. Thus, BLBLIs should be suggested to be used with caution as a carbapenem-sparing option. Further research is needed to evaluate the impact of antimicrobial stewardship programs to balance the use and resistance across drug classes.

Although the results from our study were significant, there are some limitations. First, the study was ecological in nature and may not represent biological associations at the individual level. However, ecological studies are often used to suggest the relationship between use of an antibiotic and resistance to that antibiotic on population level [35, 45]. Second, we used the total number of antibiotics prescriptions as a measure of antibiotic consumption instead of using defined daily doses (DDDs) in this study. However, a previous study indicated that evaluating the total number of antibiotics prescriptions is a useful method to measure the strength of the relationship between antibiotic consumption and antibiotic resistance [41], and DDDs are typically highly correlated with the total number of prescriptions. Third, other factors such as the inconsistency of the patient population, transferred patients having resistant organisms from different hospitals, possible epidemics, and infection control measures during the study the period may have affected the level of resistance in our findings.

Conclusions

In this study, we identified a significant temporal correlation between BLBLI use and piperacillin/tazobactam resistance in K. pneumoniae. The increasing trend in the rates of resistance to piperacillin/tazobactam in K. pneumoniae observed in this study raises concerns regarding the appropriate use of BLBLIs as carbapenem-sparing antibiotics. Our results suggest that new interventions aimed at reducing the use of BLBLIs may help retain their efficacy as a carbapenem-sparing alternative.