Impact of Antibiotic Exposure Patterns on Selection of Community-Associated Methicillin-Resistant Staphylococcus aureus in Hospital Settings^,

Abstract

Community-associated methicillin-resistant S. aureus (CA-MRSA) is increasingly common in hospitals, with potentially serious consequences. The aim of this study was to assess the impact of antibiotic prescription patterns on the selection of CA-MRSA within hospitals, in a context of competition with other circulating staphylococcal strains, including methicillin-sensitive (MSSA) and hospital-associated methicillin-resistant (HA-MRSA) strains. We developed a computerized agent-based model of S. aureus transmission in a hospital ward in which CA-MRSA, MSSA, and HA-MRSA strains may cocirculate. We investigated a wide range of antibiotic prescription patterns in both intensive care units (ICUs) and general wards, and we studied how differences in antibiotic exposure may explain observed variations in the success of CA-MRSA invasion in the hospitals of several European countries and of the United States. Model predictions underlined the influence of antibiotic prescription patterns on CA-MRSA spread in hospitals, especially in the ICU, where the endemic prevalence of CA-MRSA carriage can range from 3% to 20%, depending on the simulated prescription pattern. Large antibiotic exposure with drugs effective against MSSA but not MRSA was found to promote invasion by CA-MRSA. We also found that, should CA-MRSA acquire fluoroquinolone resistance, a major increase in CA-MRSA prevalence could ensue in hospitals worldwide. Controlling the spread of highly community-prevalent CA-MRSA within hospitals is a challenge. This study demonstrates that antibiotic exposure strategies could participate in this control. This is all the more important in wards such as ICUs, which may play the role of incubators, promoting CA-MRSA selection in hospitals.

INTRODUCTION

Methicillin-resistant Staphylococcus aureus (MRSA) is one of the leading causes of nosocomial infections. Community-associated MRSA (CA-MRSA), which acquire methicillin resistance outside the hospital, are increasingly found in hospitals worldwide. In the United States, CA-MRSA now represents more than 30% of all nosocomial MRSA strains (7). This spread occurs in an environment where a complex, and poorly understood, balance exists between methicillin-sensitive S. aureus (MSSA) and hospital-associated MRSA (HA-MRSA).

The introduction and dissemination of CA-MRSA into hospitals are current threats, especially at a time when the implementation of effective policies for control of cross-transmission in hospitals has led to the reduction of HA-MRSA infections in many developed countries (21). Indeed, while CA-MRSA strains are currently susceptible to more antibiotics than HA-MRSA, they are more transmissible and may acquire additional resistances under the high antibiotic selective pressure found in hospitals (24). In this context, it is important to gain a better understanding of the factors that may promote the selection of CA-MRSA in hospital settings.

Simulation models have long been used to analyze pathogen dissemination in hospital settings and control strategies (3, 22). However, the complex functioning of a hospital ward must often be overly simplified in these approaches, for example, ignoring spatial detail or structure of contact networks. Such limitations may be overcome using stochastic agent-based simulation, as illustrated in the study of pandemic influenza or bioterrorist attacks (10, 12).

Agent-based models are well suited to modeling complex phenomena. In this bottom-up approach, agents (i.e., individuals) are described as entities in interaction with others according to explicit rules. The dynamics of the system as a whole is studied from the individual level toward the population level (5, 17, 25).

Here, an agent-based, spatially explicit model of pathogen transmission in a hypothetical intensive care unit or general medicine ward was described, where agents represented patients and staff. The model includes detailed data on antibiotic exposure, as well as data on antibiotic susceptibility of all staphylococcal strains.

First, we use model simulations to investigate the impact of a wide range of hypothetical antibiotic prescription patterns on CA-MRSA prevalence in two different ward types: an intensive care unit (ICU) and a general ward (GW). Then, we interpret our findings in the context of several European countries and the United States.

MATERIALS AND METHODS

Model structure.

We developed and used an agent-based, stochastic, discrete time, spatially explicit computerized model to simulate the spread of bacterial strains in a hypothetical 20-bed hospital ward among patients and health care workers (HCWs) (35, 36).

In this model, each patient and health care worker (HCW) was represented as an ‘′agent'' with a specific internal state and a geographical situation. Patients were characterized by spatial location, length of stay in the hospital, colonization status, and exposure to antibiotics; HCWs were characterized by their daily schedule and colonization status. Every day, the model simulated the actions of each agent, such as patient visits by HCWs and patient admission or discharge. Patients could require low or high levels of care, the latter leading to more frequent contacts with HCWs. HCW schedules reflected those of nurses or those of physicians. Contacts between HCWs and patients and patient admission or discharge were simulated. Figure S1 in the supplemental material provides a schematic representation of the modeled ward.

Two types of hospital wards were simulated: a GW and an ICU, with different rates of contacts and antibiotic usage. The main parameters of the model are presented in Tables 1 and 2. The simulation platform is described in more detail and available for download at http://sites.google.com/site/nososim/.

Table 1.

Main model parameters

Model parameter	Value	Source(s)
No. of beds	20	Assumed
Occupancy rate (%)	90	Assumed
Length of stay (gamma distribution with mean) of patient with:
Low level of care	5 (shape, 10; scale, 0.5)	29, 33
High level of care	14 (shape, 28; scale, 0.5)
Portion of patient with low level of care	0.9	33
Portion of patient with high level of care	0.1	33
Fixed length of nurse visit (min) for patient with:		29
Low level of care	20
High level of care	100
Fixed length of physician visit (min) for patient with:		16
Low level of care	25
High level of care	25
Prevalence (%) of patients colonized at hospital admission with:
MSSA	18	1
CA-MRSA	1	13
HA-MRSA	5	14
Length (days) of patient's colonization in absence of antibiotic exposure	100 (mean of exponential distribution)	4
Length (days) of HCW's superficial colonization	1 (mean of exponential distribution)	37
Length of treatment (days)	8	32
Antibiotic resistance (%) to:		Assumed
Group A
MSSA	100
CA-MRSA	100
HA-MRSA	100
Group B
MSSA	0
CA-MRSA	100
HA-MRSA	100
Group C
MSSA	0
CA-MRSA	0
HA-MRSA	100
Group D
MSSA	0
CA-MRSA	0
HA-MRSA	0

Table 2.

Specific parameters for the two types of hospital wards

Parameter	GW	ICU	Source(s)
Patient-to-HCW ratio			16, 19
Physicians	1:6	1:6
Nurse (day shift)	1:4	1:2
Nurse (night shift)	1:6	1:2
No. of patient visits with:			15, 16
Physicians	2	2
Nurse (day shift)	3	3
Nurse (night shift)	2	3
Transmission (min of contact−1) of:			Calibrated values based on reference 18
MSSA	0.003	0.0072
CA-MRSA	0.0028–0.003	0.0054–0.0072
HA-MRSA	0.0028	0.0054
Prevalence of antibiotic exposure (%)	27	60	31

Staphylococcus aureus colonization. (i) Circulating strains.

All three categories of S. aureus strains were considered in the simulations. The differences between strains were modeled as differences in susceptibility to antibiotics and transmissibility. The prevalence of carriage on hospital admission for MSSA, HA-MRSA, and CA-MRSA was fixed at, respectively, 18%, 5%, and 1% (1, 13, 14).

(ii) Colonization transmission.

Transmission of S. aureus to patients occurred only by contact with transiently colonized HCWs. The probability of S. aureus transmission from a colonized HCW to a patient or from a colonized patient to an HCW was calculated as the product of the transmission rate per minute and the duration of contact. We denoted the transmission rate per minute as p_S for MSSA, p_CA for CA-MRSA, and p_HA for HA-MRSA. We calibrated p_S and p_HA to reproduce an observed prevalence for HA-MRSA of 17% (reported range, 9 to 23% [18, 20] and twice as high as for MSSA [11]). The values of p_S and p_HA were calibrated independently for GWs and ICUs. We hypothesized that p_S = p_CA ≥ p_HA, as CA-MRSA appears to be more closely related to MSSA and the transmissibility of HA-MRSA is reduced in comparison with MSSA and CA-MRSA (8).

(iii) Colonization clearance and immunity.

We assumed that the mean time to decolonization with S. aureus was 100 days in the absence of antibiotic exposure in patients and that colonization of HCWs was always transient (1 day).

In patients, clearance was followed by a temporary immunity period of 4 days, during which recolonization by the same strain was not allowed (2). Transient hand carriage by HCWs was not associated with a period of immunity.

(iv) Simultaneous carriage and competition.

Simultaneous carriage of strains was allowed. The probability of acquisition of another strain was, however, reduced by 50% in already-colonized individuals to reflect competition between strains (6).

Antibiotic exposure.

Patients could be exposed to one or several antibiotics during their stay. Antibiotic exposure of a colonized patient cleared carriage of sensitive strains but had no impact on resistant strains. The mean duration of antibiotic exposure was 8 days (32) for all antibiotic exposures.

(i) Antibiotic class categorizations.

We reviewed the systemic antimicrobials prescribed in hospitals, as found in group J01 of the Anatomical Therapeutic Chemical (ATC) classification system (39). The susceptibility of MSSA, CA-MRSA, and HA-MRSA isolates to each of these antimicrobials was assessed.

Then, we split the ATC classes into 4 subgroups, according to their activity on each of the three S. aureus strains: group A (e.g., J01CA [ampicillin]), to which all three strains of S. aureus were resistant; group B (e.g., J01CF [methicillin]), to which MSSA isolates were sensitive and MRSA isolates were resistant; group C (e.g., J01FF [clindamycin]), to which MSSA and CA-MRSA isolates were sensitive but HA-MRSA isolates were resistant; group D (e.g., J01XA [vancomycin]), to which all three strains were sensitive. Table S1 in the supplemental material provides the susceptibility data for all ATC classes as well as the classification into 4 groups of these classes, taking into account uncertainties due to 5 ATC classes for which variations in susceptibility have been reported.

Baseline scenario.

We used as a baseline scenario the best-case scenario for antibiotic efficacy, where CA-MRSA and HA-MRSA are susceptible to antibiotics for which variations in susceptibility have been reported.

Antibiotic prescription patterns.

In order to study the impact of different patterns of antibiotic use on CA-MRSA dissemination, we systematically investigated different hypothetical usage frequencies for the 4 groups (A, B, C, and D), which ranged from 5% to 80% of the overall antibiotic consumption, which was fixed. The explored values were 5%, 10%, 15%, 20%, 30%, 40%, 50%, 60%, 70%, and 80%, leading to 252 hypothetical antibiotic prescription patterns (e.g., A = 50%, B = 15%, C = 20%, and D = 15%). Figure S2 in the supplemental material provides all 252 investigated antibiotic prescription patterns.

Antibiotic exposure frequency.

We assumed 27% of patients were exposed daily to antibiotics in the GW, based on data from France (31). As ward-specific data have shown 2 to 3 times more antibiotic exposure in ICUs than in GWs (31), we assumed a total of 60% of patients were exposed daily to antibiotics in the ICU.

Applications: baseline scenario.

We used data from the ESAC study on drug consumption in European hospitals (9), as well as data from U.S. hospitals (28), in order to locate the antibiotic prescription practices in several countries (United States, Denmark, Finland, France, Poland, and Greece) among the 252 investigated patterns. When available, ward-specific data were used (28, 34).

As an example, Table 3 provides the percentages represented by groups A, B, C, and D among the total antibiotic consumption within GWs and ICUs among French hospitals, as well as the corresponding proportion of patients exposed daily to antibiotics from these 4 groups. Both the baseline scenario and the range due to uncertainties in the classification are provided.

Table 3.

Distribution of prescribed antibiotics and daily frequency of antibiotic exposure in French hospitals among the study's four antibiotic groups^a

Study group	Baseline classification (uncertainty range)
	% of total antibiotic consumption		Prevalence (%) of patients exposed to antibiotics
	GW	ICU	GW	ICU
A	21	11	5.7	6.6
B	44 (44–72)	63 (63–82)	11.9 (11.9–19.4)	37.8 (37.8–49.2)
C	18 (0–30)	17 (0–24)	4.8 (0–8.1)	10.2 (0–14.4)
D	17 (5–17)	9 (6–9)	4.6 (1.3–4.6)	5.4 (3.6–5.4)
Total	100	100	27	60

^aBased on a classification of detailed drug consumption data from the ESAC and the French Coordination Centers to fight Nosocomial Infections (C-CLIN) study (9, 34) in ICUs and GWs. The data represent the baseline values and the ranges due to uncertainties in the classification into groups B, C, and D.

Table 4 provides patterns in antibiotic use for all considered countries under the baseline scenario assumption.

Table 4.

Antibiotic prescription patterns in U.S., French, Danish, Finnish, Polish, and Greek hospitals

Group	% of total antibiotic consumption by drug groupa in:
	United States		France		Denmark	Finland	Poland	Greece
	GW	ICU	GW	ICU
A	15	11	21	11	45	13	19	14
B	57	58	44	63	29	32	46	49
C	18	16	18	17	14	15	13	19
D	10	15	17	9	12	40	22	18
Total	100	100	100	100	100	100	100	100

^aThe data represent the baseline classifications for all prescribed antibiotics into four antibiotic groups, A, B, C, and D. Ward-specific data were available only for U.S. and French hospitals (9, 28, 34).

Model simulations.

For each possible scenario, we introduced a single initial CA-MRSA-colonized patient within the ward and we simulated the resulting 3-strain dynamics for 30 days. Over these 30 days, following this first admitted colonized patient, 1% of admitted patients were assumed to be colonized with CA-MRSA (13). At the end of the simulated period we calculated the prevalence of CA-MRSA among all S. aureus patient carriers. This outcome was determined as the average over 1,000 simulation replicates required to hold stochastic components of the model constant at their average values.

Investigation of antibiotic prescription patterns.

In order to assess the impact of antibiotic prescription patterns on CA-MRSA dissemination in the hospital, we performed a sensitivity analysis based on simulations using our 252 prescription patterns. We computed Kendall partial rank correlation coefficients (PRCC) between CA-MRSA endemic prevalence and the exposure to antibiotic groups A, B, C, and D (expressed as a fraction of total antibiotic exposure in the ward).

Sensitivity analyses.

We assessed the sensitivity of our predictions in terms of CA-MRSA colonization prevalence on several model parameters related to S. aureus colonization and transmission. In this analysis, in the pre-CA-MRSA era, we investigated prevalences of HA-MRSA that ranged between 9 and 23%, as well as prevalences of MSSA between 20 and 40%. We also investigated transmission rates per minute for CA-MRSA (p_CA), which ranged from p_S (transmission rate for MSSA) to p_HA (transmission rate for HA-MRSA).

In another analysis, we studied the consequences of future changes in CA-MRSA susceptibility to fluoroquinolone antibiotics. Although CA-MRSA strains are currently mostly sensitive to fluoroquinolones, previous experience has demonstrated that resistance to this class may be selected due to antibiotic selection pressure. Indeed, recent data from the United States suggests that some CA-MRSA isolates already exhibit a significant decrease in sensitivity to fluoroquinolones (24). In order to perform this sensitivity analysis, we determined which of the 252 investigated prescription patterns best described antibiotic prescription practices in the six countries we studied, assuming that CA-MRSA had become resistant to fluoroquinolones.

RESULTS

Model calibration.

Using a least square criterion, we calibrated the transmission rates of MSSA and of HA-MRSA in order to best reproduce observed carriage prevalence. Transmission rates (per minute of contact) for the ICU were 0.0072 for p_S^ICU and 0.0054 for p_HA ^ICU, and rates for GWs were p_S^GW of 0.003 and p_HA^GW of 0.0028 per minute of contact.

In the following sections, the probability of CA-MRSA transmission (p_CA) was therefore equal to p_S^ICU in the ICU and to p_S^GW in the GW.

Impact of antibiotic prescription patterns.

Figure 1 depicts the range of predicted endemic prevalence of CA-MRSA among all S. aureus carriers under different antibiotic prescription patterns in the ICU and in the GW.

Fig. 1.

Range of predicted endemic prevalence of CA-MRSA among carried staphylococcal strains under 252 hypothetical different antibiotic prescription patterns. Each box-plot displays the minimum, first quartile, median, third quartile and the maximum predicted prevalence in the ward.

Antibiotic prescription patterns had an important impact on CA-MRSA colonization dynamics, especially in the ICU. The computed prevalence varied by a factor of 3.25 in the GW, ranging from 4% to 13%, and by a factor of 6.7 in the ICU, ranging from 3% to 20%.

The sensitivity analysis (see Table S2 in the supplemental material) showed that changes in groups B and C had the greatest influence on the prevalence of CA-MRSA (in the ICU, PRCC = 0.81 and PRCC = −0.59, versus PRCC = 0.41 and PRCC = −0.26 for groups A and D, respectively). Table S2 also provides computed PRCCs for the predicted prevalence rates of MSSA and HA-MRSA.

Figure 2, which depicts the prevalence of CA-MRSA among carried staphylococcal strains in the ICU depending on exposure to antibiotics from groups A, B, C, and D, highlights the importance of group B antibiotic exposure for CA-MRSA dissemination.

Fig. 2.

Predicted endemic CA-MRSA prevalence among carried staphylococcal strains in the ICU setting (depicted on a color scale, from white [low prevalence] to red [high prevalence]), depending on relative exposure to antibiotics from groups A, B, and C (a) and B, C, and D (b). (a) Exposure to group A was fixed at 10%, 30%, 50%, and 80%, respectively; exposure to groups B and C varied from 5% to 80% of the overall antibiotic consumption. (b) Exposure to group D was fixed at 10%, 30%, 50%, and 80%, respectively, exposure to groups B and C vary from 5% to 80% of the overall antibiotic consumption.

Simulations in a GW setting give similar results (see Fig. S3 in the supplemental material).

Applications. (i) Baseline scenario.

Figure 3 locates the six antibiotic prescription patterns most similar to observed prescription levels in the United States, Denmark, Finland, France, Poland, and Greece among the range of our predictions for both the GW and the ICU settings.

Fig. 3.

Application of French-like, Greek-like, Danish-like, Finnish-like, Polish-like, and U.S.-like patterns of antibiotic use in hospitals to our predictions in both GW and ICU settings. Levels of antibiotic use were based on a classification of detailed consumption data from the ESAC (9), C-CLIN study (34) and from the NNIS system report (28) (Table 4). Other model parameters (e.g., prevalence) were not country specific.

High use of antibiotics from group B (e.g., β-lactamase-resistant penicillins) and low use of antibiotics from groups C (e.g., fluoroquinolones), as in the “Polish-like” prescription pattern, or D (e.g., glycopeptides), as in the “French-like” or the “U.S.-like” antibiotic prescription patterns, tended to promote dissemination of community strains in hospitals.

On the other hand, high use of antibiotics from group D, as in the “Finland-like” antibiotic prescription pattern, prevented major CA-MRSA diffusion.

Sensitivity analyses.

Figure 4 depicts a tornado diagram of the effects of transmission and colonization parameter values on the predicted prevalence of CA-MRSA in the “French-like” scenario as well as on the variance of predicted prevalences among the 252 investigated scenarios. The predicted prevalence decreased, as well as its variance, when a lower transmissibility of CA-MRSA, lower initial MSSA prevalence, or higher initial HA-MRSA prevalence was assumed. They increased when a higher initial MSSA prevalence or lower initial HA-MRSA prevalence was assumed.

Fig. 4.

Tornado diagram of the effects of model parameters on predicted endemic CA-MRSA prevalence among carried staphylococcal strains in the ICU in the French-like scenario (a) and variance of this predicted prevalence among the 252 investigated scenarios (b). Blue bars are projections associated with the lower parameter values; red bars show projections associated with the higher parameter values.

Figure 5 locates the six antibiotic prescription patterns most similar to observed patterns in the United States, Denmark, Finland, France, Poland, and Greece among the range of our predictions for both the GW and the ICU settings, assuming that CA-MRSA has become resistant to fluoroquinolones. Our results show that in countries with high levels of fluoroquinolone use, such as the United States, France, or Greece, such an evolution in CA-MRSA susceptibility could have a major impact on CA-MRSA selection in hospitals, in particular in the ICU, where the antibiotic pressure is higher (see Fig. S5 in the supplemental material).

Fig. 5.

Effects of the acquisition of resistance to fluoroquinolones in CA-MRSA on their spread in hospitals where antibiotic use patterns are similar to those observed in France, Greece, Denmark, Finland, Poland, and the United States in both GW and ICU settings.

DISCUSSION

In this study, we examined the impact of antibiotic prescription patterns on the spread of CA-MRSA and other staphylococcal strains among hospitalized patients, using an agent-based model.

We showed that, even at a constant antibiotic consumption level, the selection of prescribed antibiotic classes may have a major impact on the dynamics of nosocomial spread of microorganisms. This implies that investigating antibiotic usage strategies may help provide tools for control. It may also help explain differences in the hospital epidemiology of CA-MRSA across different countries.

CA-MRSA epidemiology in the general community.

Outpatients may play a major role in the community-to-hospital spread of CA-MRSA. In this study, we assumed an incoming flow of CA-MRSA-colonized patients into health care settings but did not include any country-specific observed data on the community prevalence of CA-MRSA.

This means that the application of our results using data from 6 countries cannot and should not be interpreted as actual predictions for these countries. Rather, we use country-specific data in order to illustrate the variability among current patterns of hospital antibiotic use worldwide and to provide actual examples on how such levels may influence CA-MRSA dissemination in health care settings.

This is underlined by the fact that the “French-like” and the “U.S.-like” prescription patterns have similar impacts on CA-MRSA dissemination in the ICU (Fig. 3), while observed data show that CA-MRSA is currently more frequent in American hospitals than in French hospitals (7, 27).

Model hypotheses.

In this study, we used an agent-based model for modeling the spread of S. aureus in a hospital setting. This approach allowed for increased realism in the description of individual behavior in a small hospital environment. The simulations demonstrated that invasion of CA-MRSA in hospital wards could be facilitated depending on typical antibiotic use. However, our analysis disregards several points which may also be at play in the epidemiology of CA-MRSA, such as community dynamics, strain-specific characteristics, differences in health care system organization, etc.

Despite the applicability of agent-based models for modeling complex biological phenomena, this approach presents several limitations. Agent-based modeling requires detailed and reliable data for model building or validation, which is not always easily available. What is more, the increase in behavioral detail provided by agent-based models leads to much greater computational intensity and makes carrying out extensive sensitivity analyses difficult.

Here, we developed a virtual hospital ward where all S. aureus transmissions between patients occurred via direct contacts with HCWs. In the absence of detailed information on transmission of S. aureus in hospital wards, we assumed equal HCW-to-patient and patient-to-HCW transmissibility rates, and we ignored direct HCW-to-HCW transmissions as well as environmental contamination.

Simulations were performed for two different types of hospital wards: a GW and an ICU, where overall antibiotic exposure was higher and patient-HCW contacts more frequent. Intensive care units have been noted to play an important role in the selection and spread of antibiotic-resistant bacteria within hospitals. Selective pressure of antimicrobials and presence of patients with severe illness combined in a relatively small and crowded area promote MRSA spread within ICUs, increasing the risk of MRSA infections (21).

Our results suggest that the chosen antibiotic prescription strategy may have a larger impact on CA-MRSA selection in the ICU than in GWs. However, this conclusion is important at the hospital level, as ICU patients are frequently transferred between hospitals and wards, thereby increasing the risk for intra- and interhospital dissemination of resistant strains.

Finally, hospitals are often indicated as a source of emergent resistant strains, but this spread is not unidirectional, as illustrated by outbreaks of community-acquired MRSA in health care facilities. In this study we ignored the community dynamics of S. aureus strains and assumed a fixed rate of colonized patients on hospital admission. In future studies, it would be interesting to describe the spread of resistant pathogens from the community to hospital settings, as well as the interaction between transmission dynamics within a hospital and the surrounding community.

Antibiotic efficacy.

The efficacy of systemic antibiotics for eliminating carriage of S. aureus strains has been demonstrated in several trials, but the antibiotics were often used in combination with topical agents (23). In our study, we assumed that antibiotic exposure of a colonized patient led to complete clearance of carriage for sensitive strains but had no impact on resistant strains.

In order to investigate the impact of this hypothesis, we performed simulations assuming that 10 to 50% of colonizations with sensitive strains persisted following antibiotic exposure. The distribution of predicted CA-MRSA prevalence (see Fig. S4 in the supplemental material) was not significantly different from our baseline predictions (Wilcoxon test, P = 0.2 to 0.28).

Data on antibiotic use.

In order to apply our predictions, we determined which of the 252 investigated antibiotic prescription patterns best reflected observed levels in antimicrobial use in hospitals from the United States and several European countries, in terms of the distribution of prescribed antibiotic classes. However, recent studies showed wide variations in antimicrobial use in hospital care, both between countries and at the national level (38). This means that the country-level antibiotic use patterns we used are only a crude description of actual levels in the countries we considered. This is particularly true for large countries, such as the United States, as evidenced by significant differences in reported hospital antibiotic use between different American studies (28, 30).

Furthermore, although we performed simulations in both a GW and an ICU setting, we were not always able to use ward-type-specific antibiotic use data, because for several countries (Denmark, Finland, Poland, and Greece) we only had access to average hospital antibiotic use patterns.

Finally, the data obtained from the United States was incomplete. The report from the NNIS did not cover use of tetracyclines, macrolides, or aminoglycosides (28). Whether this was due to low overall consumption of these agents or to an oversight remains unclear.

Transmission rates.

To calibrate transmission rates, we assumed that MRSA strains represented 33% of all S. aureus isolates recovered in hospitals, which reflects the mean MRSA rate in European hospitals and is close to the French situation in the early 2000s (11).

Applying our model to countries where infection control programs are different and where the proportion of MRSA among all S. aureus isolates carried in hospitals is either much larger (e.g., Greece) or much smaller (e.g., Finland) than the 33% may lead to errors in the estimation of transmission rates. This in turn may lead to underestimated or overestimated MRSA prevalence rates in these countries.

However, while this means that our predictions cannot be used to predict accurately CA-MRSA carriage prevalence in a given country, we feel that it does not impede our capacity to assess qualitatively the impact of antibiotic prescription patterns in these countries on the spread of CA-MRSA in hospitals.

Epidemiological data suggest that CA-MRSA, like MSSA, is more transmissible than HA-MRSA (8). For this study, we assumed the transmissibility of CA-MRSA to be as high as that of MSSA, which is a worst-case scenario for CA-MRSA invasion in hospitals. In order to assess the impact of this hypothesis, we also performed simulations assuming that CA-MRSA transmissibility was only equal to that of HA-MRSA. Our main conclusions still held.

Future evolution of CA-MRSA.

Fluoroquinolones are among the most commonly prescribed classes of antibiotics in the hospital as well as in the community in some countries (38). Several studies suggest that fluoroquinolone exposure may predispose patients to infections with or carriage of HA-MRSA, eradicating most susceptible strains. What is more, recent data suggest that CA-MRSA strains may evolve to become fluoroquinolone resistant. We performed a sensitivity analysis to assess the potential consequences of this phenomenon. Our results suggest that such an evolution in CA-MRSA susceptibility may have a major impact on its selection within hospitals of some countries, including France and the United States (Fig. 5).

Conclusions.

Strains of community-acquired MRSA have been introduced into hospital settings worldwide and have become the most frequent cause of skin and soft tissue infections in the emergency departments in some areas of the United States (26). Although routine surveillance and isolation procedures have proved successful in controlling HA-MRSA in hospitals in some countries, they cannot be efficient in community settings. For this reason, the presence of a community reservoir from which resistant strains can repeatedly be introduced into health care settings to potentially cause secondary outbreaks is a growing challenge, and more research is needed to better define optimal control measures. Based on this study, these control measures may include selected antibiotic patterns of use and strategies that could minimize the risk of dissemination within the hospital.