Carbapenem-resistant Pseudomonas aeruginosa and carbapenem use in Japan: an ecological study.

Introduction

Human-induced antimicrobial resistance (AMR), which can develop through the inappropriate use of antimicrobial agents, has been recognized as one of the biggest threats to public health.[1] The World Health Organization (WHO) adopted the Global Action Plan on Antimicrobial Resistance in May 2015,[1] which encouraged each member country to design and issue national action plans within 2 years. Japan released its National Action Plan on Antimicrobial Resistance in April 2016, and one of its aims was to lower the proportion of carbapenem-resistant Pseudomonas aeruginosa isolates from 17% in 2014 to less than 10% by 2020,[2] because the bacterium is known to be one of the most common causes of opportunistic infections and death among immunocompromised hosts.[3-6]

Pseudomonas aeruginosa is a gram-negative rod-shaped bacterium that favours moist environments.[7] The bacterium has low virulence and does not usually elicit clinically severe disease in healthy people.[7,8] However, P. aeruginosa is commonly associated with nosocomial infections and is responsible for high mortality among immunocompromised hosts.[8] For severe infections, antipseudomonal drugs (e.g. antipseudomonal carbapenems such as imipenem and meropenem), fluoroquinolones (e.g. ciprofloxacin and levofloxacin) and aminoglycosides (e.g. amikacin) are used for antimicrobial therapy.[8] Resistance to antipseudomonal drugs is increasing and multidrug-resistant P. aeruginosa exhibiting resistance to three or more drug classes has emerged and spread across different regions worldwide.[3] This multi-drug resistance poses the serious threat of losing drug options to treat opportunistic infections in hospital settings. Resistance mechanisms of P. aeruginosa include selection of bacteria with drug-inactivating enzymes, efflux pumps and target-site or outer membrane modifications.[4,5] Epidemiological studies have identified several risk factors for carbapenem-resistant P. aeruginosa infection including use of medical devices, such as artificial respirators and central intravenous catheters, and admission to an intensive care unit with carbapenem use.[6,9-11]

To prevent and monitor the spread of carbapenem-resistant P. aeruginosa, the relationship between the prevalence of resistant strains and the use of carbapenems, fluoroquinolones and cephalosporin has been explored in various settings.[12-19] In some patient-based studies, no correlation between carbapenem use and the prevalence of carbapenem-resistant P. aeruginosa strains was reported.[20-24] However, the published reports have been mostly limited to epidemiological studies conducted at a single healthcare facility or a single geographical location.[20-24] Thus, a nationwide evaluation of this relationship is called for. Moreover, potential links between imipenem/meropenem use and the prevalence of carbapenem-resistant P. aeruginosa strains should be explored while appropriately adjusting for confounders.

In Japan, both nationwide AMR surveillance and a national database of health insurance claims and specific health check-ups publicly share their statistical data. This offers an excellent opportunity to explore the relationship between carbapenem use and the prevalence of resistant P. aeruginosa strains. The purpose of this study was to assess ecological associations between carbapenem use and the prevalence of imipenem/meropenem resistance in P. aeruginosa using datasets from all 47 prefectures in Japan.

Materials and methods

Study design

This study was designed to explore the ecological relationship between the prevalence of carbapenem-resistant P. aeruginosa isolates and carbapenem use across different regions in Japan. The study aimed to exploit geographical heterogeneities in the observed datasets and was conducted in a cross-sectional manner. The AMR datasets from 2015–2016 were obtained and the spatial datasets over these years were analysed.

This study analysed publicly available data. Microbial testing results were collected as part of the national surveillance activities without individual identifications and the submission of testing results alone did not require informed consent. The datasets used in this study were de-identified and fully anonymized in advance, and the analysis of publicly available data without identifying information did not require ethical approval. All of the datasets were already available in published articles.[25,26]

Epidemiological data collection

Data on P. aeruginosa resistance to imipenem/meropenem and epidemiological data were obtained from the clinical laboratory division of the Japan Nosocomial Infections Surveillance (JANIS) system.[25] This system includes microbiological testing results of all registered medical facilities. The corresponding data encompass all testing results among both inpatient and outpatient facilities regardless of the characteristics such as infection status, colonization or carrier status. The drug sensitivity testing results were extracted for imipenem or meropenem whenever P. aeruginosa was detected from patient laboratory samples. Publicly shared data do not classify results by body samples (i.e. providing summed results of all samples including those obtained from urine, blood and sputum); and from an identical individual, the samples are sometimes taken on multiple occasions during the course of treatment. Based on the sensitivity testing data for each prefecture, prevalence was calculated as the number of resistant isolates detected per 100 000 population. The proportion of resistant P. aeruginosa isolates out of the total number of infections was calculated.

Data on carbapenem use were extracted from the National Database of Health Insurance Claims and Specific Health Check-ups of Japan (NDB), representing the total amount of drug used in each prefecture from 2015 to 2016.[26] These datasets have been publicly shared over the internet by the AMR Clinical Reference Centre of the National Centre for Global Health and Medicine. Carbapenem use data were standardized in accordance with the Anatomical Therapeutic Chemical classification, using defined daily dose as a measurement unit as recommended by the WHO Collaborating Centre for Drug Statistics and Methodology.[27] The population-weighted consumption of oral and parenteral antimicrobial drugs was expressed as the defined daily dose per 1000 inhabitants per day (DID).

As potential confounders of spatial variations in carbapenem resistance, the study also collected the following information in prefectures from 2015 to 2016: (i) the number of physicians per 100 000 individuals;[28] (ii) the number of nurses per 100 000 individuals;[29] (iii) the number of hospitals and clinics per 100 000 individuals;[30] (iv) the number of nursing homes per 100 000 individuals;[31] (v) the number of registered medical facilities in JANIS and the number of data-submitting facilities with more than 500 beds;[25] (vi) the proportion of elderly individuals in the population;[32] and (vii) the mean length of a hospital stay.[33]

Statistical analyses

The statistical analyses were conducted in three steps. First, the study investigated the descriptive characteristics of variables, including their distributions and summary statistics. Because some variables yielded variance-to-mean ratios greatly exceeding 1, medians and interquartile ranges were consistently used to summarize the distribution of all variables. The data were log-transformed for some variables with skewed distributions so that the data were approximately normally distributed. Secondly, defining the outcome as the proportion of P. aeruginosa isolates resistant to carbapenems, the univariate relationships between this proportion and explanatory variables including use of imipenem and meropenem were explored. Univariate linear regression was used for testing ecological correlations. Thirdly, multiple linear regression analysis was used to adjust for the effects of confounding variables. To appropriately select variables to be included in the final model, multicollinearity between variables was evaluated. Subsequently, variables showing strong multicollinearity were not included in multivariate regression models. All statistical data were analysed using JMP Pro version 14.0 (SAS Institute Inc., Cary, NC, USA). A P-value < 0.05 was considered statistically significant.

Results

Table 1 summarizes the descriptive features of the data. The median prevalence of imipenem- and meropenem-resistant P. aeruginosa was 24.4 and 17.4 per 100 000 individuals in 2015, respectively; and 24.1 and 17.8 per 100 000 individuals in 2016, respectively. The median proportion of imipenem- and meropenem-resistant P. aeruginosa isolates was 18.7% and 12.8% in 2015, respectively; and 17.2% and 11.9% in 2016, respectively. The median consumption of carbapenems was 0.11 DID in both 2015 and 2016.

Table 1.

Descriptive characteristics of variables potentially related to imipenem/meropenem-resistant Pseudomonas aeruginosa in Japan in 2015 and 2016.

Variable	Median (interquartile range)
	2015	2016
Prevalence of imipenem resistance, per 100 000	24.4 (19.2–30.7)	24.1 (20.0–33.0)
Prevalence of meropenem resistance, per 100 000	17.4 (14.0–21.0)	17.8 (14.1–22.5)
Proportion of imipenem resistance, %	18.7 (16.8–21.9)	17.2 (16.1–19.8)
Proportion of meropenem resistance, %	12.8 (11.4–15.1)	11.9 (10.3–14.0)
Carbapenem consumption, DID × 10−1	1.1 (0.9–1.3)	1.1 (0.9–1.3)
Number of physicians per 100 000 individuals	242.4 (217.0–279.8)	242.4 (217.0–279.8)
Number of nurses per 100 000 individuals	965.5 (817.0–1104.2)	965.5 (817.0–1104.2)
Number of hospitals per 100 000 individuals	7.1 (5.7–10.0)	7.2 (5.7–10.0)
Number of clinics per 100 000 individuals	81.1 (72.8–90.2)	81.3 (90.7–72.7)
Number of nursing homes per 100 000 individuals	7.3 (6.2–8.3)	6.9 (5.8–8.1)
Number of registered medical facilities	21.0 (15.0–37.0)	26.0 (17.0–40.0)
Number of facilities with over 500 beds	4 (3–7)	4 (2–8)
Proportion of elderly individuals in the population, %	28.7 (26.8–30.1)	28.7 (26.8–30.1)
Mean length of hospital stay, days	32.0 (29.2–34.1)	31.8 (28.8–33.8)

DID, defined daily dose per 1000 inhabitants per day.

Figure 1 shows the geographical distributions of the proportion of imipenem- and meropenem-resistant P. aeruginosa isolates and carbapenem use. There were tendencies toward higher proportions of resistant strains and higher use of carbapenems in the western as compared with the eastern prefectures. In 2015, the highest proportions of imipenem and meropenem resistance were 26.1% and 19.7% in Oita prefecture, located on the westernmost large island, Kyushu. In 2016, the highest proportions of imipenem and meropenem resistance were 27.5% in Miyazaki prefecture and 18.3% in Ehime prefecture, respectively. Again, both prefectures were located in the west of Japan. In 2015 and 2016, the highest use of carbapenem was 0.16 DID in Oita prefecture and Kagawa prefecture.

Figure 1.

Geographical distribution of carbapenem-resistant Pseudomonas aeruginosa isolates and carbapenem use in Japan in 2015 and 2016. The proportion of P. aeruginosa isolates (per 100 000 individuals) resistant to imipenem (a) and meropenem (b) among total infections in 2015 and 2016. Carbapenem use (c) in 2015 and 2016 expressed as the defined daily dose per 1000 inhabitants per day (DID).

Table 2 shows the results of univariate analyses of associations between the proportion of imipenem- or meropenem-resistant P. aeruginosa isolates and other factors. Carbapenem consumption was positively correlated with the proportion of imipenem- and meropenem-resistant P. aeruginosa isolates in both years examined (P < 0.05 for all comparisons). The following variables were positively correlated with the proportion of imipenem-resistant isolates in 2015: number of nurses (P = 0.04), number of hospitals (P = 0.02) and mean length of hospital stay (P = 0.04). The number of physicians was not correlated with the proportion of imipenem- and meropenem resistant isolates both in 2015 and 2016.

Table 2.

Univariate correlations between the proportion of imipenem or meropenem-resistant Pseudomonas aeruginosa strains and selected variables in Japan in 2015 and 2016.

Variable	Proportion resistant in 2015				Proportion resistant in 2016
	Imipenem		Meropenem		Imipenem		Meropenem
	Coefficient	Statistical significance	Coefficient	Statistical significance	Coefficient	Statistical significance	Coefficient	Statistical significance
Carbapenem consumption, DID[a]	0.53	P < 0.01	0.46	P < 0.01	0.35	P = 0.01	0.30	P = 0.04
Number of physicians per 100 000 individuals	0.18	NS	0.13	NS	0.07	NS	0.12	NS
Number of nurses per 100 000 individuals	0.30	P = 0.04	0.17	NS	0.16	NS	0.08	NS
Number of hospitals per 100 000 individuals[a]	0.35	P = 0.02	0.27	NS	0.21	NS	0.17	NS
Number of clinics per 100 000 individuals	0.22	NS	0.20	NS	0.09	NS	0.15	NS
Number of nursing homes per 100 000 individuals	0.26	NS	0.15	NS	<0.01	NS	–0.04	NS
Number of medical facilities included in surveillance[a]	–0.10	NS	–0.03	NS	0.10	NS	0.15	NS
Number of facilities with > 500 beds included in surveillance[a]	–0.09	NS	–0.02	NS	–0.05	NS	0.05	NS
Mean length of hospital stay, days	0.31	P = 0.04	0.23	NS	0.11	NS	0.05	NS
Proportion of elderly individuals in the population, %	0.25	NS	0.16	NS	<0.01	NS	–0.06	NS

aIndicates log-transformed data.

DID, defined daily dose per 1000 inhabitants per day; NS, no significant association (P ≥ 0.05).

Figure 2 shows the univariate correlation between carbapenem use and the prevalence of carbapenem-resistant P. aeruginosa strains in 2015 and 2016. The proportion of resistant strains was strongly correlated with carbapenem consumption and the correlation coefficient did not greatly differ between the 2 years examined.

Figure 2.

Univariate relationships between carbapenem consumption and carbapenem resistance in Pseudomonas aeruginosa in Japan in 2015 and 2016. Correlation between carbapenem consumption and the proportion of P. aeruginosa strains resistant to imipenem (2015, a; 2016, b) and meropenem (2015, c; 2016, d). Each dot represents the observed value in a single prefecture. Each panel contains 47 dots that reflect observations from all 47 prefectures in Japan. The black lines show the results of univariate linear regression analysis.

Table 3 shows the results of multivariate analyses of the proportion of imipenem- and meropenem-resistant P. aeruginosa strains. Due to strong collinearity between the numbers of nurses and hospitals, the latter variable was removed from the model. Thus, three variables (carbapenem consumption, number of nurses and mean length of hospital stay) were explored. In the final multivariate model, only carbapenem consumption remained significantly associated with the proportion of resistant strains (P < 0.05 for both 2015 and 2016).

Table 3.

Multiple regression analysis of the proportion of imipenem or meropenem-resistant Pseudomonas aeruginosa strains in Japan in 2015 and 2016.

Variable	Proportion resistant in 2015								Proportion resistant in 2016
	Imipenem				Meropenem				Imipenem				Meropenem
	PRC	SE	t	Statistical significance	PRC	SE	t	Statistical significance	PRC	SE	t	Statistical significance	PRC	SE	t	Statistical significance
Intercept constant	41.9	8.4	5.00	P < 0.01	32.6	6.6	4.95	P < 0.01	35.8	9.3	3.86	P < 0.01	27.2	7.3	3.72	P < 0.01
Carbapenem consumption, DID[a]	9.4	2.7	3.49	P < 0.01	7.1	2.1	3.37	P < 0.01	6.9	3.0	2.31	P = 0.03	5.2	2.4	2.18	P = 0.03
Number of nurses per 100 000 individuals	>–0.1[a]	<0.1	–0.20	NS	>–0.1[a]	<0.1	–0.28	NS	>–0.1[a]	<0.1	–0.36	NS	>–0.1[a]	<0.1	–0.21	NS
Mean length of hospital stay, days	>–0.1[a]	0.1	–0.23	NS	–0.1	0.1	0.89	NS	>–0.1[a]	0.1	–0.2	NS	–0.1	0.1	–0.83	NS

aIndicates log-transformed data.

Summary statistics for assessing the fit of multivariate models: 2015, imipenem: R2 = 0.29, F = 5.88, P < 0.01; 2015, meropenem: R2 = 0.25, F = 4.67, P = 0.01; 2016, imipenem: R2 = 0.14, F = 2.26, P = 0.10; 2016, meropenem: R2 = 0.13, F = 2.06, P = 0.12.

PRC, partial regression coefficient; SE, standard error; DID, defined daily dose per 1000 inhabitants per day; NS, no significant association (P ≥ 0.05).

Discussion

This study examined the ecological correlation between the proportion of carbapenem-resistant P. aeruginosa isolates and carbapenem consumption by examining the spatial distribution of these factors in 47 prefectures in Japan in 2015 and 2016. Univariate analysis showed that carbapenem consumption, numbers of nurses, numbers of hospitals and mean length of hospital stay were positively correlated with the proportion of carbapenem-resistant P. aeruginosa isolates. In a multivariate analysis, only carbapenem consumption remained significantly associated with the proportion of carbapenem-resistant P. aeruginosa isolates. Mapping the geographical distribution of resistant strains, western Japan was found to have higher proportions of resistant strains in 2015 and 2015 than the rest of Japan. Importantly, this geographical variation over different prefectures was consistent with that of carbapenem consumption.

Although a cross-sectional survey of over 200 hospitals reporting potential links between antimicrobial consumption and the prevalence of resistant P. aeruginosa isolates has already been published,[34] this current study used national datasets both for carbapenem consumption (NDB) and for surveillance data (JANIS) that encompassed about 20% of all healthcare facilities. More than 50% of larger hospitals with more than 200 beds and more than 80% of those with more than 500 beds contribute data to JANIS. Using a larger sample size that better represented the population of Japan, this current study demonstrated a relationship between the prevalence of carbapenem-resistant P. aeruginosa isolates and carbapenem use based on national databases. This current study also endeavoured to identify and account for other possible explanatory factors for this association.

The link between the prevalence of carbapenem-resistant P. aeruginosa isolates and carbapenem use is not novel.[12-19] However, this current study shows that publicly accessible datasets can jointly be analysed and that such associations can be monitored in a longitudinal manner. For instance, the Japanese government has started to financially subsidize healthcare facilities in which antimicrobial stewardship teams (ASTs) are organized and actively engaged in regulating the use of antimicrobial agents. ASTs restrict the use of carbapenems. If the proportion of resistant strains decreases and carbapenem sensitivity recovers, that would strongly indicate a causal relationship between carbapenem use and the development of resistance. Thus, this current study should be used as the methodological basis for establishing such a causal link.

The potential causal link between the prevalence of carbapenem-resistant P. aeruginosa isolates and carbapenem use suggests that strong selection pressure at the population level, leading to de novo emergence of resistance, is the dominant mechanism of resistance as opposed to the spread via transmission. P. aeruginosa is an environmentally abundant bacterium and a common nosocomial infection.[8] Although transmission of resistant strains and also the spread of resistance via plasmids are not refuted by this current study, these data were based on prefectural units and the mode of spread was not separately captured from the dataset. Similarly, the use of medical devices including intravenous catheters[6,9-11] was not captured in this current study’s dataset, and if such datasets were collected at greater resolution (e.g. by hospital or by hospital ward), associations between these risk factors and P. aeruginosa resistance in Japan could be analysed.

Medical expenditure in Japan is geographically heterogeneous, with higher costs in the west compared with the east.[35] Both increased carbapenem consumption and a higher proportion of resistant P. aeruginosa strains in the west may have reflected the less restricted use of antibiotics in the west compared with the east.[36] While medical services are provided under identical rules and regulations, geographical heterogeneity apparently still arises although the underlying mechanisms have yet to be explored in detail.

This current study had several limitations. First, the study had an ecological design, which is prone to confounding. In addition, the results at the population level cannot be extended to a direct causal link at the individual level. Secondly, as acknowledged in the Methods section, the JANIS data did not include information on type of clinical specimens (i.e. including those obtained from urine, blood and sputum) or sensitivity to other antimicrobial agents. Additional insights into carbapenem resistance could be gained by exploring additional sampling details and microbiological testing results. Similarly, the study was not able to classify functional types of hospitals (e.g. acute or chronic care hospitals) among those registered with the surveillance system. Thirdly, antimicrobial consumption was measured by defined daily dose, but this method overlooks individual heterogeneity in the prescribed dose (e.g. lower doses in children). Fourthly, the study was unable to evaluate factors potentially associated with nosocomial transmission due to the analysis at the prefectural resolution. Fifthly, as test samples were possibly taken from an individual patient on multiple occasions during their course of treatment, the post-treatment samples were more likely to be resistant. Improvements in design and sampling schemes are called for.

In conclusion, despite these limitations, this current study has shown an ecological correlation between carbapenem consumption and the prevalence of carbapenem-resistant P. aeruginosa isolates in Japan by analysing prefectural datasets. The geospatial patterns of carbapenem use correlated with the prevalence of resistance in P. aeruginosa at the national scale. The use of two large-scale databases showed their potential usefulness to monitor and evaluate causal links and future declines of carbapenem-resistant P. aeruginosa in a longitudinal manner.