Antibiotics involved in the occurrence of antibiotic-resistant bacteria: a nationwide multilevel study suggests differences within antibiotic classes

Abstract

Objectives

To identify the antibiotics potentially the most involved in the occurrence of antibiotic-resistant bacteria from an ecological perspective in French healthcare facilities (HCFs).

Methods

This study was based on data from the French antimicrobial surveillance network (ATB-RAISIN, 2007–09). Antibiotics were expressed in defined daily doses per 1000 patient-days. Antibiotic-resistant bacteria were considered as count data adjusted for patient-days. These were third-generation cephalosporin (3GC)- and ciprofloxacin-resistant Escherichia coli, cefotaxime-resistant Enterobacter cloacae, methicillin-resistant Staphylococcus aureus and ceftazidime-, imipenem- and ciprofloxacin-resistant Pseudomonas aeruginosa. Three-level negative binomial regression models were built to take into account the hierarchical structure of data: level 1, repeated measures each year (count outcome, time, antibiotics); level 2, HCFs (type and size); and level 3, regions (geographical area).

Results

A total of 701 HCFs from 20 French regions and up to 1339 HCF-years were analysed. The use of ceftriaxone, but not of cefotaxime, was positively correlated with incidence rates of 3GC- and ciprofloxacin-resistant E. coli. In contrast, both 3GCs were positively correlated with the incidence rate of cefotaxime-resistant E. cloacae. Higher levels of use of ciprofloxacin and/or ofloxacin, but not of levofloxacin, were associated with higher incidence rates of 3GC- and ciprofloxacin-resistant E. coli, cefotaxime-resistant E. cloacae, methicillin-resistant S. aureus and ceftazidime- and ciprofloxacin-resistant P. aeruginosa.

Conclusions

Our study suggests differences within antibiotic classes in promoting antibiotic resistance. We identified ceftriaxone, ciprofloxacin and ofloxacin as priority targets in public health strategies designed to reduce antibiotic use and antibiotic-resistant bacteria in French HCFs.

Introduction

The acquisition of antibiotic resistance in bacteria limits therapy options in clinical practice, and can lead to a therapeutic dead end. These antibiotic-resistant bacteria, which are sometimes multiresistant or even panresistant, have become a major public health concern worldwide. This situation is worrisome since the development of new antibacterial drugs is stagnating.¹ Antibiotic use and poor compliance with hygiene measures are currently recognized as the main determinants of emergence and spread of antibacterial resistance.^2,3 In addition, the relationship between antibiotic use and antibiotic resistance is complex. This relationship has been widely explored at different levels using various epidemiological approaches,^4–8 as well as in in vitro studies.^9–11 In France, the ATB-RAISIN network (RAISIN stands for Network for early warning, investigation and surveillance of nosocomial infections) is responsible for surveillance of antibiotic consumption and antibiotic resistance in healthcare facilities (HCFs). This surveillance network was created under the aegis of the five coordinating centres for nosocomial infection control (Figure 1, geographical areas of responsibility) and the French Institute for Public Health Surveillance (InVS). Thus, data from the ATB-RAISIN network enabled us to examine the issue of the relationship between antibiotic use and antibiotic resistance from an ecological perspective across a national network of HCFs. Few studies have explored this issue at a nationwide level. Published data often include a sample of hospitals^12,13 that is not necessarily representative of the whole country. Moreover, the underlying issue, which remains a challenge, is that of the existence of differences between antibiotics from the same class in their ability to promote the development of antibiotic resistance. We aimed here to identify antibiotics that are potentially those most involved in the occurrence of antibiotic-resistant bacteria in French HCFs.

Methods

Design, setting and period of study

This multicentre ecological study was based on data from the ATB-RAISIN national network obtained between 2007 and 2009. The geographical distribution of participating HCFs is presented according to regions and geographical areas of France in Figure 1. Our study was restricted to HCFs that participated in the two types of surveillance (antibiotic consumption and antibiotic resistance). Therefore, it was limited to four out of five geographical areas (south-east not included) and 20 out of 26 French regions. Moreover, psychiatric hospitals were excluded because of their low levels of antibiotic consumption and low incidence of antibiotic-resistant bacteria.

Data collection

The antimicrobial surveillance network retrospectively collected three types of data from each HCF each year. First, HCF characteristics were collected: type, size (number of beds), geographical location (region and geographical area) and activity (number of patient-days in full hospitalization). Second, data on antibiotic use (dispensing data) were retrieved from hospital pharmacy records, as described elsewhere.¹⁴ Use of antibiotics was expressed in defined daily doses (DDDs) per 1000 patient-days, in accordance with the 2009 version of the Anatomical Therapeutic Chemical-DDD classification from the WHO.¹⁵ The following antibiotics were studied: aminoglycosides (J01G; amikacin, gentamicin), carbapenems (J01DH), cephalosporins (J01DB + J01DC + J01DD + J01DE; cefotaxime, ceftriaxone, ceftazidime and cefepime), glycopeptides (J01XA), macrolides, lincosamides and streptogramins (MLS; J01F), penicillin (J01C; group M penicillins, aminopenicillins ± β-lactamase inhibitors, amoxicillin, amoxicillin/clavulanate, piperacillin/tazobactam), fluoroquinolones (J01MA; ciprofloxacin, levofloxacin and ofloxacin), imidazoles (J01XD + P01AB). Third, antibiotic-resistant bacteria tracked were third-generation cephalosporin (3GC) and ciprofloxacin-resistant Escherichia coli, cefotaxime-resistant Enterobacter cloacae, methicillin-resistant Staphylococcus aureus (MRSA) and ceftazidime-, imipenem- and ciprofloxacin-resistant Pseudomonas aeruginosa. Data on antibiotic-resistant bacteria referred to the number of non-duplicate clinical isolates that were non-susceptible to a specific antibiotic. These data were expressed in terms of proportions within bacterial species (%) and incidence rates (per 1000 patient-days). All clinical isolates were from inpatients and surveillance cultures were not included. Antimicrobial susceptibility was interpreted according to the guidelines of the Antibiogram Committee of the French Society for Microbiology.¹⁶

Data analysis

For each of the antibiotic-resistant bacteria, the analysis was restricted to HCFs that tested at least 10 non-duplicate clinical isolates for antimicrobial susceptibility every year of participation. A multilevel approach (three levels) with negative binomial regression models was used to take into account the hierarchical structure of data: repeated measures each year (first level) nested within HCFs (second level), which were nested within regions (third level). The count outcome variable was defined at the first level: number of occurrences of antibiotic-resistant bacteria adjusted for patient-days (i.e. the log of the number of patient-days was treated as an offset parameter in models). Explanatory variables were defined at the first level (time and antibiotic use), at the second level (type and size of HCF) and at the third level (geographical area). Such an approach is more appropriate in case of possible correlation of data than traditional regression techniques, which, by ignoring data clustering, may lead to biased statistical inferences.¹⁷ We decided to use negative binomial regression rather than Poisson regression because the count outcome variable exhibited overdispersion. Moreover, negative binomial regression is more suitable for ‘contagious processes’ (non-independent occurrences of antibiotic-resistant bacteria due to patient-to-patient transmission).¹⁸ First, a univariable analysis was performed using an ordinary negative binomial regression in order to evaluate the crude correlation between the outcome variable and each explanatory variable. This analysis enabled us to identify the most parsimonious fitting variable form (continuous or categorical) for time and HCF size. Second, multivariable analysis was carried out by building multilevel negative binomial models with random intercepts, i.e. in which the intercept of the model is allowed to vary randomly across HCFs and regions. Additionally, random intercept has a fixed part and a random part. The random part corresponds to residuals at higher levels (i.e. HCFs and regions). These residuals are assumed to be normally distributed, with mean zero. Their variances are HCF-level variance and region-level variance. Multilevel modelling was performed according to a three-step process. In practice, three consecutive multilevel models were developed. Model A (unconditional model), which included only the intercept (i.e. no explanatory variables), was assessed to determine the initial distribution of the variance of the outcome variable between the three levels. This model enabled us to determine the initial HCF-level variance and region-level variance. These variances correspond, respectively, to the heterogeneity of HCFs within regions and the heterogeneity of regions in terms of incidence rates of antibiotic-resistant bacteria. Thus, a region-level variance that is significantly different from zero indicates a significant variability in incidence rates of antibiotic-resistant bacteria between regions. In model B, we introduced first- and second-level variables. Finally, the region-level variable was included in model C (final model). This model provided adjusted incidence rate ratios (aIRRs) that were obtained by exponentiating the parameter estimates (coefficients of variables). The latter were estimated using penalized quasi-likelihood (PQL) with the second-order Taylor expansion procedure.¹⁷ The aIRRs measured the magnitude of correlations between antibiotic use and incidence rates of antibiotic-resistant bacteria. For the multivariable analysis, only explanatory variables with a P value of <0.20 in the univariable analysis were considered, except time, which was maintained in the final model in all cases. Each final model was analysed thoroughly with respect to satisfying statistical assumptions, including normal distribution of HCF- and region-level residuals that was graphically checked. A P value of <0.05 was considered to be statistically significant. The software package Stata, version 10.0 (Stata Corp., College Station, TX, USA), was used for the univariable analysis. Multilevel modelling was performed using MLwiN software version 2.02 developed by the Centre for Multilevel Modelling (www.bristol.ac.uk/cmm/).

Results

HCF characteristics, antibiotic use and resistance

Seven hundred and one HCFs (and up to 1339 HCF-years) were included in this study between 2007 and 2009 (Figure 1). They represented 25% of the HCFs eligible for the French antimicrobial surveillance network and about 40% of hospital beds in France. Additionally, 40% of these HCFs participated in every year of the surveillance. In two-thirds of the cases, the HCFs had 200 beds or fewer, and only 3.0% of them had more than 1000 beds. HCF types were distributed as follows: university hospitals (4.3%), general hospitals (33.2%), cancer hospitals (1.4%), private hospitals (31.5%), local hospitals (8.3%) and rehabilitation and long-term care facilities (21.3%). Data on antibiotic use are provided in Table 1. The lowest levels of antibiotic use were observed in western France. Penicillins, especially amoxicillin and amoxicillin/clavulanate, represented more than half of antibiotic consumption in French HCFs. Between 2007 and 2009, the use of carbapenems (+33.3%), ceftriaxone (+27.8%), levofloxacin (+31.5%) and imidazoles (+27.5%) greatly increased among HCFs that participated in every year of the surveillance. Conversely, consumption of ceftazidime (−3.1%) and ciprofloxacin (−3.5%) decreased slightly during the same period. Table 2 provides the antibiotic resistance data. The proportions and the incidence rates of 3GC-resistant E. coli and cefotaxime-resistant E. cloacae greatly increased from 2007 to 2009. In contrast, the proportions and the incidence rates of MRSA and ceftazidime-resistant P. aeruginosa greatly decreased over that period. As with antibiotic use, the incidence rates of antibiotic-resistant bacteria were lower in western France than in the rest of the country. All explanatory variables showed a P value of <0.20 in the univariable analysis and were considered for the multivariable analysis.

Multilevel modelling results

The unconditional models (not shown) exhibited a region-level variance that was significantly different from zero for 3GC- and ciprofloxacin-resistant E. coli, cefotaxime-resistant E. cloacae and imipenem- and ciprofloxacin-resistant P. aeruginosa. In the final models (Tables 3 and 4), we found significantly lower incidence rates of 3GC- and ciprofloxacin-resistant E. coli, cefotaxime-resistant E. cloacae and ceftazidime-resistant P. aeruginosa in western France than in the other geographical areas. Similarly, significantly lower incidence rates of MRSA and imipenem-resistant P. aeruginosa were recorded in western France than in the north and south-west of the country. Additionally, our results revealed significant variations of the incidence rates of 3GC- and ciprofloxacin-resistant E. coli, MRSA and ceftazidime-resistant P. aeruginosa over the surveillance period. These final models, which included specific antibiotic molecules, indicated a positive correlation between the use of amoxicillin/clavulanate and the incidence rates of ciprofloxacin-resistant E. coli and ciprofloxacin-resistant P. aeruginosa, whereas the use of amoxicillin was positively correlated with the incidence rate of cefotaxime-resistant E. cloacae. The variable aminopenicillins ± β-lactamase inhibitors, which replaced amoxicillin and amoxicillin/clavulanate in the model with MRSA, was also positively correlated with the incidence rate of MRSA. A higher level of ofloxacin consumption was associated with higher incidence rates of all antibiotic-resistant bacteria, except imipenem-resistant P. aeruginosa. In contrast, the use of levofloxacin was not correlated with the incidence rates of antibiotic-resistant bacteria. Consumption of ciprofloxacin was positively correlated with the incidence rates of ciprofloxacin-resistant E. coli and ciprofloxacin-resistant P. aeruginosa, but also with those of MRSA and ceftazidime-resistant P. aeruginosa. The use of ceftriaxone, but not cefotaxime, was positively correlated with the incidence rates of 3GC- and ciprofloxacin-resistant E. coli. In contrast, both 3GCs were positively correlated with the incidence rate of cefotaxime-resistant E. cloacae. The use of ceftazidime was positively correlated with the incidence rates of imipenem- and ciprofloxacin-resistant P. aeruginosa, but not with that of ceftazidime-resistant P. aeruginosa. Gentamicin and carbapenem use was positively correlated with the incidence rates of ceftazidime- and imipenem-resistant P. aeruginosa. Furthermore, the consumption of imidazoles was positively correlated with the incidence rates of 3GC-resistant E. coli and MRSA. Finally, MLS and glycopeptide use was positively correlated with MRSA incidence rate.

Discussion

The present study was carried out at a time of changes in the epidemiology of antibiotic-resistant bacteria. First, the proportion of MRSA within S. aureus isolates is stabilizing or decreasing in most European countries, particularly in France.¹⁹ In contrast, the dramatic increase in extended-spectrum β-lactamase (ESBL)-producing Enterobacteriaceae^20,21 is continuing to such an extent that ESBL-producing E. coli could shortly supplant MRSA at the first rank of multidrug-resistant bacteria in French hospitals. Owing to the study design (ecological-level investigation and ‘contemporaneous relationships’, i.e. no time-lagged correlation), the results were interpreted in light of the biological plausibility of correlations and the epidemiological data available.

Antibiotic-resistant Enterobacteriaceae

Our multilevel models indicated a significant increase of the incidence rates of antibiotic-resistant E. coli over the study period. Hence, for the time variable, aIRRs were 1.269 and 1.043 for 3GC- and ciprofloxacin-resistant E. coli respectively. In other words, the incidence rates of these antibiotic-resistant bacteria increased by 26.9% and 4.3% per year respectively. The resistance to extended-spectrum cephalosporins (ESCs) in E. coli is mainly conferred by the production of ESBLs.²² In addition, ESBL-producing E. coli strains are frequently resistant to multiple antibiotic classes, including fluoroquinolones.^23,24 Thus, the significant increase in antibiotic-resistant E. coli presumably reflects the worldwide spread of ESBL-producing E. coli in community and hospital settings.^20,21 Interestingly, the use of ceftriaxone, but not cefotaxime, was positively correlated with incidence rates of 3GC- and ciprofloxacin-resistant E. coli, even though ESBL producers have a similar resistance level to both compounds.²⁵ Such a difference could be explained by the pharmacokinetic properties of these ESCs.²⁶ Indeed, ceftriaxone has high biliary elimination (up to 45%) whereas the elimination route of cefotaxime is mainly renal. Consequently, ceftriaxone exerts a higher selective pressure on gastrointestinal flora and promotes the emergence of resistant pathogens.²⁷ A substantial proportion of ESBL-producing E. coli strains that cause healthcare-associated infections is imported from the community by inpatients with intestinal carriage on admission.²⁸ These infections often occur after antibiotic exposure, especially to oxyimino-cephalosporins and fluoroquinolones.²² In addition, the positive correlation between the use of imidazoles (for example metronidazole) and the incidence rate of 3GC-resistant E. coli corroborates the findings of previous studies.^29,30 In contrast to what is observed with antibiotic-resistant E. coli, the incidence rate of cefotaxime-resistant E. cloacae was correlated with both ceftriaxone and cefotaxime use. Such a difference between the two pathogens is difficult to explain. Nevertheless, it should be noted that, unlike antibiotic-resistant E. coli, which is partly of community origin,²⁸ cefotaxime-resistant E. cloacae is mainly responsible for hospital-acquired infections.³¹ Therefore, we speculate that this difference in their epidemiology may explain differences in their correlations with hospital use of ceftriaxone and cefotaxime. In the multilevel model for cefotaxime-resistant E. cloacae, there was no significant difference between the aIRRs of cefotaxime and ceftriaxone use, even though that of cefotaxime use tended to be higher. Surprisingly, the use of carbapenems (antibiotics of choice for the treatment of severe infections with ESC-resistant Enterobacteriaceae) was not significantly correlated with incidence rates of 3GC-resistant E. coli and cefotaxime-resistant E. cloacae. The existence of other determinants of carbapenem prescriptions may explain this result (e.g. empirical therapy in patients who are at high risk of death and infections with other multidrug-resistant Gram-negative pathogens such as P. aeruginosa and Acinetobacter spp., especially in intensive care units).³² Regarding ciprofloxacin-resistant E. coli, positive correlations were found with the use of ciprofloxacin and ofloxacin, but not with levofloxacin. These results are consistent with those of an in vitro study¹⁰ that assessed the influence of different growth conditions on the development of resistant mutants among E. coli strains exposed to quinolones. A higher number of mutants were recovered under anaerobic conditions (as present in the human gut) than under aerobic conditions with ciprofloxacin and ofloxacin, but not with levofloxacin. Additionally, a higher concentration of ciprofloxacin was required under an anaerobic atmosphere to prevent the emergence of resistant mutants. In contrast, the mutant prevention concentration of levofloxacin was lower under anaerobic conditions.¹⁰

MRSA

Our data also confirmed the decline of MRSA.¹⁹ Indeed, the multilevel model for MRSA showed a significant annual decrease in the incidence rate of MRSA of 8.3%. This decrease could be attributed to control efforts aimed at interrupting the spread of this multidrug-resistant pathogen within and between French HCFs (i.e. infection control measures such as implementation of barrier precautions and use of alcohol-based hand rub). In this study, we confirmed the role of fluoroquinolone use as risk factor for MRSA. Nevertheless, a previous patient-level study⁴ indicated a possible differential effect of fluoroquinolones on MRSA emergence, showing a stronger association with levofloxacin than with ciprofloxacin. Such a result was not found in our hospital-level study, in which levofloxacin use was not correlated with MRSA incidence. This discrepancy in results could be due to a lower use of levofloxacin compared with ciprofloxacin (1.5-fold higher) and ofloxacin (2-fold higher), and a possible threshold effect. We confirmed that the use of MLS, aminopenicillins ± β-lactamase inhibitors and imidazoles is a risk factor for MRSA, as found in previous studies.^4,33,34 As almost all MRSA strains in France remain susceptible to glycopeptides, the correlation between glycopeptide consumption and MRSA incidence reflects the use of glycopeptides for treating MRSA infections.

Antibiotic-resistant P. aeruginosa

The use of ceftazidime was not significantly correlated with ceftazidime-resistant P. aeruginosa incidence. This lack of correlation was reported in previous studies.^35–37 These studies showed that the ability to induce resistance to itself in P. aeruginosa was lower for ceftazidime than for imipenem and ciprofloxacin, whose use was distinctly associated with imipenem- and ciprofloxacin-resistant P. aeruginosa respectively. These results were also observed in our study. Overexpression of the chromosomal AmpC β-lactamase is the most common mechanism of resistance to β-lactams (including ceftazidime) in P. aeruginosa.³⁸ Such ceftazidime-resistant mutants often accumulate determinants of resistance to multiple classes of antibiotics,³⁹ including carbapenems, aminoglycosides and fluoroquinolones. Additionally, resistance to carbapenems resulting from loss of the porin protein OprD requires the presence of AmpC β-lactamase.⁴⁰ Furthermore, upregulated efflux also contributes to the development of multiple resistances to antipseudomonal agents and is mediated by the three-component active efflux systems that belong to the resistance–nodulation–division family: MexAB–OprM, MexCD–OprJ, MexEF–OprN and MexXY–OprM.^38,39 Thus, significant correlations in our study between antibiotic use and antibiotic-resistant P. aeruginosa incidence generally corroborate the results of previous studies and/or are biologically plausible.^35–37 Nevertheless, unlike some studies,^8,9,12,13,41 which found that levofloxacin was more likely than ciprofloxacin to result in ciprofloxacin-resistant P. aeruginosa, the present study did not lead to the same finding. As previously mentioned with regard to MRSA, such a discrepancy may be due to a lower use of levofloxacin, and a possible threshold effect. A positive correlation such as that between ceftazidime use and ciprofloxacin-resistant P. aeruginosa incidence is partly due to the use of this antibiotic to treat ciprofloxacin-resistant P. aeruginosa infections.

Geographical variations

It is noted that the region-level variance in incidence rates of 3GC- and ciprofloxacin-resistant E. coli, and imipenem- and ciprofloxacin-resistant P. aeruginosa remained significantly different from zero in model B (adjusted for time, antibiotic use, type and size of HCF). This result (not shown) suggests that factors other than use of antibiotics may explain the regional heterogeneity of these incidence rates, such as region-level factors that remain to be identified. Furthermore, these geographical variations were highlighted in the final models, with a lower incidence rate of antibiotic-resistant bacteria in western France.

Limitations

This study has certain limitations. First, our approach was ecological, with inherent limitations of this type of study such as ecological fallacy and lack of patient-level data. Thus, the significant correlations found should be interpreted with caution, in particular in case of P values, which are not small enough to convincingly rule out chance in view of the problem of multiple comparisons. These correlations are suggestive and do not necessarily imply the existence of a cause–effect relationship. Additionally, lack of information about the resistance mechanisms involved (no molecular investigation) and possible co-resistances and cross-resistances make the interpretation of our results difficult. Therefore, further research is needed to confirm some of these findings. Second, we could not distinguish reliably between community- and hospital-associated cases of antibiotic-resistant bacteria from inpatients. In fact, data were not available to determine the origin of these cases on a time-based criterion (e.g. by defining cases acquired in HCFs as those occurring more than 2 days after patient admission as opposed imported cases). However, as these cases concerned hospitalized patients only (full hospitalization), the impact of this bias was probably limited. In addition, for a given HCF, some imported cases are from the community and others are from another HCF as a result of inter-hospital patient transfers. These patient transfers more often occur between HCFs in the same region than between those in different regions. Furthermore, the network of HCFs located in the same region covers a catchment population with a level of antibiotic use and a level of antibiotic-resistant bacterial carriage in the community that may differ from those of other regions. Multilevel modelling, which is the strong point of this study, allowed us to take into account the clustering effects at region level that may be due to the aforementioned confounding factors. Third, we had no data regarding the policy of antibiotic prescribing and compliance with infection control measures, which may differ between HCFs and vary over time (e.g. following interventions from antimicrobial stewardship committees⁴² and promotion of hand hygiene). These interventions could have influenced antibiotic use and cross-transmission, and therefore the occurrence of antibiotic-resistant bacteria. Fourth, even though DDD measurements are useful for benchmarking, they present biases mentioned elsewhere,² including inadequate measurement of antibiotic use in children. Additionally, information about types of hospital wards were not available for all HCFs (e.g. presence of intensive care units and paediatrics). Therefore, we could not take into account this potential confounding factor in the analysis.

Conclusions

Despite these limitations, our study suggests differences between antibiotics from the same class in their ability to promote the development of antibiotic resistance in bacteria. Current public health strategies designed to reduce antibiotic use and antibiotic resistance rely on the improvement of antibiotic prescribing practice. The present study, if the findings are confirmed, emphasizes the importance of taking into account the ecological adverse effects of each antibiotic in the clinical decision to choose a particular molecule rather than another within a given antibiotic class. For example, cefotaxime is to be preferred to ceftriaxone, whenever possible, in treating E. coli infections, especially in the current context of ESBL-producing Enterobacteriaceae increase. We also identify antibiotics (ceftriaxone, ciprofloxacin, ofloxacin) whose use could be limited in priority to achieve better control of antibiotic-resistant bacteria in French HCFs and the preservation of the effectiveness of available antibiotics.

ATB-RAISIN network steering committee

S. Alfandari, O. Ali-Brandemeyer, P. Angora, X. Bertrand, S. Boussat, A. Carbonne, B. Coignard, C. Dumartin, M. Giard, P. Jarno, L. Lacavé, F. L'Hériteau, A. Machut, S. Maugat, F. Nguyen, M. Péfau, E. Remy, A.-M. Rogues, K. Saby, A. Savey, B. Schlemmer, S. Touratier and S. Vaux.

Funding

This work was supported by a grant from the French Ministry of Health (East Delegation for Clinical Research): APJ, 2012. The sponsor has no role in the study. The ATB-RAISIN network is partly funded by the French Institute for Public Health Surveillance (InVS).

Transparency declarations

None to declare.

Author contributions

H.G.-H. and X.B. wrote the study project. C.D. and M.P. were responsible for collection and validation of data within the framework of the ATB-RAISIN network. H.G.-H., F.L., D.H., A.-M.R. and X.B. participated in the analysis and the interpretation of the data. H.G.-H. and D.H. prepared the first and revised draft of the manuscript. All authors contributed to the final version of the manuscript.

Acknowledgements

We thank Dr Stephan Harbarth (University of Geneva Hospitals, Geneva, Switzerland) for critical reading of the manuscript, and Frances Sheppard (Clinical Investigation Center (Inserm CBT 506), Besançon, France) for her editorial assistance. We also thank the French hospitals that contributed to data collection within the framework of the ATB-RAISIN network.

References

Author notes