The authors have declared that no competing interests exist.
Conceived and designed the experiments: RK EK. Performed the experiments: RK EK. Analyzed the data: RK EK. Wrote the paper: RK EK BG.
Methicillin-resistant
One of the most notorious cases of antibiotic-resistant bacteria is methicillin-resistant
Over the past ten years community-associated strains of methicillin-resistant
MRSA resistance is mediated by the integration of a staphylococcal cassette chromosome mec (SCC
Given the presumed relatively lower cost of resistance in CA-MRSA, it has been speculated that CA-MRSA strains may eventually replace HA-MRSA strains in the hospital
Intuitively, both outcomes, coexistence and replacement, are possible. On the one hand, replacement, or competitive exclusion, is the standard outcome expected by ecological theory for two strains occupying the same ecological niche. Accordingly, explaining observed coexistence in other bacterial pathogens has proven challenging
While this question of coexistence is an interesting ecological problem, it is also an important question for public health as the outcome of the interaction between CA-MRSA and HA-MRSA may have epidemiological and clinical consequences: HA-MRSA has a much broader resistance spectrum than CA-MRSA
We considered three epidemiological models of increasing complexity and assessed how the interaction between hospital and community populations could lead to stable coexistence between HA- and CA-MRSA. The basic model assumes that all individuals, regardless of age, have similar hospital admission and discharge rates as well as antibiotic usage rates. In this case the only difference between the hospital and the community is the usage frequency of antibiotics, which may lead to selection favoring one strain in the community and the other in the hospital. Next we considered two extensions of this model. First we examined how heterogeneity between age-classes with respect to hospitalization rates and antibiotic usage impacted coexistence between strains in each setting. Second, we explicitly distinguished between treated and untreated patients, thereby capturing the prophylactic effect of treatment, which is likely to be much stronger in the hospital than in the community.
The epidemiological dynamics of CA- and HA-MRSA were described by set of ordinary differential equations. Based on estimates from the literature and an analysis of public-use data in the United States, we assume differing dynamics of colonization, infection, and antibiotic use between the hospital and the community and consider how differing implementations of the host population structure impacts the dynamics of each strain and examine the parameter ranges over which coexistence occurs. In each of these models, the possibility of coexistence between HA- and CA-MRSA for a given parameter-combination was determined by an invasibility analysis: First, the system is allowed to reach the equilibrium with one strain only (burn-in time: 5×104 days); then the other strain is introduced at low abundance; if the introduced strain increases in frequency after the introduction, we say that it can invade the equilibrium determined by the resident strain. If both strains (HA-MRSA and CA-MRSA) can invade the equilibrium of the other strain, this indicates coexistence.
In the basic model we assume that populations in the hospital and community are homogenous. The two populations are connected through admission into the hospital population (with rate
The age-structured model is derived from the basic model by sub-dividing each compartment into 18 different age classes (five-year bins for the ages from 0 to 85 and one bin for 85+). For instance, the compartment
Susceptible individuals (S), which are structured into multiple age classes, can be colonized by CA- or HA-MRSA in the community or in the hospital. Colonized individuals, which are also structured by age, clear the pathogen either by treatment or through natural immune clearance. Individuals move between the hospital and the community at the same rate regardless of colonization status.
The treatment-structured model (
The treatment-and-age-structured model is derived from the treatment-structured model in the same way the age-structured model is derived from the basic model.
Our models integrate realistic values for drug-usage frequencies, resistance profiles, age structure, age-dependent contact patterns, and hospitalization rates. Usage frequencies, age distribution, hospitalization rates, and the mean length of stay in the hospital were estimated from publicly available data by five-year age groups.
Data are shown for the 18 age classes used in the age-structured model.
Parameter | Explanation | Default Value |
|
Admission rate to hospital | 0.00032 d−1 |
|
Discharge rate from hospital | 0.25 d−1 |
|
Treatment rate in community | 0.0015 |
|
Treatment rate in hospital | 0.2 |
|
Probability that a treated HA-MRSA colonized individual receives effective drugs in the hospital | 0.1 |
|
Probability that a treated CA-MRSA colonized individual receives effective drugs in the hospital | 0.9 |
|
Probability that a treated HA-MRSA colonized individual receives effective drugs in the community | 0.03 |
|
Probability that a treated CA-MRSA colonized individual receives effective drugs in the community | 0.7 |
|
Probability that patient clears infection given appropriate treatment | 0.5 |
|
Base-line clearance rate | 1/300 d−1 |
|
Basic reproductive ration of CA-MRSA in community | 1.4 |
|
Single-admission basic reproductive number of HA-MRSA in hospital | Variable (0.5–1.5) |
|
Reduction of transmissibility of HA-MRSA ( |
Variable (0.1–0.5) |
|
Transmission rate of strain x in setting y | Determined by R0 and s |
Average transmission and discharge rate in the US (see
Average number of antibiotic prescriptions in the US (data from the data from the National Ambulatory Medical Care Survey (NAMCS) and the National Hospital Ambulatory Medical Care Survey (NHAMCS)); see
Polk et al
The basic reproductive ratio of CA-MRSA in community was determined under the assumption that this strain pays no cost of resistance; i.e. it has the same transmission rate as methicillin-sensitive
The single-admission basic reproductive number of HA-MRSA in the hospital corresponds to the number of secondary infections caused by a single colonized individual admitted to a hospital containing only susceptible individuals and is given by the dominant eigenvalue of the next-generation matrix B V−1. The matrix B is given by Bij =
The number of hospitalizations and average length of stay for each age group was estimated from the Nationwide Inpatient Sample (NIS), Healthcare Cost and Utilization Project, Agency for Healthcare Research and Quality for the year 2008. The NIS contains data on ∼8 million records of hospital stays annually from about 1,000 hospitals, approximating a 20% stratified sample of US community hospitals, and includes all nonfederal, short-term, general, and specialty hospitals, such as obstetrics-gynecology, ear-nose-throat, orthopedic, and pediatric institutions. The NIS includes public hospitals and academic medical centers but excludes long- and short-term acute rehabilitation facilities, psychiatric hospitals, and alcoholism and chemical dependency treatment facilities. Hospitalization rates were calculated as the average number of hospitalizations per-person per-day by age group. The numbers of individuals for each age group were obtained from the US Census bureau's annual estimates of the resident population by five-year age groups (
Antibiotic usage in the community was estimated based on data from the National Ambulatory Medical Care Survey (NAMCS) and the National Hospital Ambulatory Medical Care Survey (NHAMCS) for 2008. NAMCS is an annual national survey of visits to non-federally employed office-based physicians who are primarily engaged in direct patient care, and NHAMCS is designed to collect data on the utilization and provision of ambulatory care services in hospital emergency and outpatient departments and in ambulatory surgery centers. Weighted patient level data was used to estimate the annual number of prescriptions for antibiotics that were written for each age group. The usage rate was calculated as the average number of prescriptions written per person per day per age group. Antibiotic usage in the hospital was estimated based on the data from
To calculate the approximate effectiveness of community antibiotic usage on CA- and HA-MRSA, we calculated the number of prescriptions for each antibiotic class and, based on assumptions about the effectiveness of each antibiotic against CA- and HA-MRSA, we estimated the percentage of drug usage that was effective against each pathogen (Supplementary
In order to explore the possibility of coexistence between HA-MRSA and CA-MRSA, we consider a series of epidemiological models of increasing complexity. The simplest, basic model ignores all population structure beyond the distinction between hospital and community. We then extend this basic model by incorporating age- and treatment-heterogeneities in accordance with published data (see
For the basic model, which ignores both age- and treatment-structure, we find that the interaction between the hospital and the community can in principle generate coexistence between HA- and CA-MRSA. However, we observed this outcome only for a relatively narrow band of fitness-costs for HA-MRSA (
The blue area indicates the parameter-combinations for which HA-MRSA and CA-MRSA coexist. The dark grey region indicates the parameter-combinations in which HA-MRSA cannot be invaded by CA-MRSA. The light-grey region indicates parameter-combinations in which CA-MRSA cannot be invaded by HA-MRSA. The range between the two red lines corresponds to fitness costs for which selection in the hospital and community act in opposite directions (i.e. CA-MRSA is fitter in the community and HA-MRSA is fitter in the hospital). The x-axis corresponds to the fitness disadvantage of HA-MRSA compared to CA-MRSA in the absence of effective therapy. The y-axis corresponds to the average number of secondary infections caused by a single colonized individual admitted to a hospital containing only susceptible individuals (single-admission reproduction number
We included age-dependent transmission rates for the community by assuming that transmission rates are proportional to the rate of physical contact
The red area indicates co-existence in the age-structured model. Axes and parameter values are the same as in
An additional source of heterogeneity is treatment itself. We take this heterogeneity into account by explicitly tracking the treatment status of patients and assuming that individuals receiving a given antibiotic cannot be colonized with strains that are susceptible to this drug. Including treatment heterogeneity in this way leads to an additional, substantial extension of the parameter range over which coexistence occurs (
The red area indicates coexistence in the treatment- and age-structured model. Axes and parameter values are the same as in
HA-MRSA is most likely adapted to the hospital environment in other ways than by its broad antibiotic resistance spectrum (e.g. tolerance to disinfectants, smaller requirements of invasibility, etc.)
The red area indicates coexistence in the age-structured model.
The above analysis was based on the ability of one strain to invade the equilibrium defined by the presence of the other strain. This method indicates where the two strains can coexist at equilibrium and therefore allows one to assess the main ecological forces underlying coexistence and competitive exclusion. However, it has three disadvantages: First, the equilibrium might be attained only very slowly: for instance two strains might coexist for a transient period which can extend over decades even though the equilibrium analysis indicates that one strain should exclude the other. Second, even if the two strains coexist one of them might attain only very low levels (i.e. even though the two strains can coexist in theory, almost all infections are caused by one single strain). Third, the pairwise-invasibility approach only allows an analysis of the competitive interaction of two strains, whereas, in reality, several
In order to address these issues, we considered a more pragmatic definition of coexistence: We initiated the population either with HA-MRSA as the only resident strain or with two resident strains (HA-MRSA and MSSA) and ran the simulation for 30 years. Then we add the new strain (CA-MRSA) and examined how the frequencies of each changed over time. Specifically, we tested after 10, 20, 50, 100, and 200 years, which strains still exist in substantial frequency (using a threshold of 5%). Note that we focused here only on the invasion of HA-MRSA/MSSA equilibrium by CA-MRSA rather than the opposite, since the former describes the current epidemic development (whereas the latter is merely of theoretical interest).
We first considered the interaction between CA-MRSA and HA-MRSA (in analogy to the above analyses). We find that the two strains can coexist during a long transient phase (10–50 years) for a broad range of conditions, which do not support coexistence at equilibrium (see
Colors indicate which strains have frequencies >5% among the colonized patients in the hospital (HA-MRSA) and the community (CA-MRSA): Blue indicates coexistence (i.e. both strains >5%), dark grey indicates HA-MRSA only, and light grey CA-MRSA only. The dashed orange line delimits the parameter region in which HA-MRSA can invade MSSA (criterion for invasion: frequency of MRSA >5%, 50 years after its introduction; see
When we consider the interaction between all three strains by including MSSA as one of the initial resident strains, we find that the parameter range in which all three strains can coexist shrinks successively with increasing time (see
Colors (see legend) indicate which strains have frequencies >5% among the colonized patients in the hospital (HA-MRSA) and the community (MSSA/CA-MRSA).
We examined how differences in age-structured patterns of antibiotic use and hospitalization rates can promote coexistence of CA- and HA-MRSA. Overall, our results show that hospital and community-associated strains of MRSA can coexist if the broader resistance spectrum of the hospital-associated strains is balanced by intermediate fitness-disadvantages in the absence of treatment. For such intermediate fitness costs, the hospital-associated strains have higher fitness in the hospital, where treatment rates are high, whereas community-associated strains have a higher fitness in the community were treatment rates are low. Despite opposite directions of selection, both strains are present in both environments if there is coexistence at all (see
Including heterogeneity in the form of realistic age- and treatment-structures into the model significantly increases the range of parameters over which coexistence can occur, making it a likely outcome. Furthermore, the fitness cost of HA-MRSA in the absence of treatment is presumably weaker in the hospital than in the community because of factors such as easier invasion due to open wounds, catheters, etc., as well as increased use of antiseptics to which the hospital strain might be better adapted. Taking this possibility into account leads to an additional, substantial increase in the range over which coexistence is likely. Thus, coexistence between HA-MRSA and CA-MRSA is a likely outcome due to the combined effect of hospital-community interactions, age-structure, treatment-structure, and possibly setting dependent fitness costs in the absence of treatment.
Coexistence is mainly dependent upon the cost of HA-MRSA being neither too high nor too low. It should be noted, however, that the upper bound for resistance costs is, in this context, more informative than the lower bound. For costs of HA-MRSA below the lower bound, we would expect that CA-MRSA could not invade the HA-MRSA equilibrium. However, such an invasion is exactly what occurred during the 1990s. Thus, we know that fitness-costs of HA-MRSA are high enough to allow the invasion of CA-MRSA. The crucial question is whether they are low enough for this invasion to stop before CA-MRSA has completely replaced HA-MRSA.
The width of the coexistence range depends strongly on how effectively MRSA can transmit in the hospital. In our simulations we quantified this transmissibility as the average number of secondary cases caused by the admission of one patient to the hospital containing only susceptible patients (RA). If this value is considerably smaller than one (i.e. hospitals cannot maintain the spread of MRSA on their own), then the coexistence range becomes very narrow. This is because coexistence relies on opposite directions of selection in the hospital and community environment. If however, one of these environments does contribute only very weakly to transmission, this balancing effect cannot take place. The only published estimate for RA we are aware of found values of 0.68 (0.47–0.95) and 0.93 (0.71–1.21) for two Dutch hospitals; one implies a broad and one a narrow coexistence range (The same study also reported an RA value of 0.16, which however corresponded to an animal derived strain)
Even though our model realistically includes several levels of population structure, our analysis might still underestimate the range over which CA-MRSA and HA-MRSA can coexist. First, other types of heterogeneity might promote coexistence in a similar way as the ones discussed here. Examples include spatial heterogeneities like rural vs. urban areas, small vs. large hospitals (which would impose different levels of stochastic effects and thereby affect strain abundances
Even though our model can explain the coexistence between HA-MRSA and CA-MRSA, we did not find any parameter combination that supports coexistence at equilibrium between more than two strains (HA-MRSA, CA-MRSA and MSSA). This suggests that the system as described by our model corresponds to only two ecological niches. This implies that the maintenance of the diversity within HA-MRSA and CA-MRSA has to be explained by mechanisms not included in our model (such as the geographical and temporal variation mentioned above). Moreover,
Our model also describes a static situation in which the properties of the strains and the age structure do not change over time. However, both demographic change in the human population and evolutionary change of the MRSA strains are likely to occur and their impact on coexistence between competing strains is an interesting question for future studies. Demographic change will most likely increase the proportion of old people in the US and most western countries. In the context of our model this means that selection will tend to favor hospital adapted strains, as the hospitalization rates are considerably higher for the old age classes. However, the direction of evolutionary change depends very strongly on the physiological constraints underlying antibiotic resistance. For instance, if CA-MRSA can increase its resistance spectrum while maintaining a high transmissibility, it could eventually out-compete HA-MRSA. If on the other hand a higher fitness cost is the inevitable consequence of a broad resistance spectrum, then such a replacement is unlikely to occur. Such evolutionary changes may be particularly important given the very long transient phases during which CA- and HA-MRSA can coexist. These long transient phases provide the opportunity for evolutionary adaptation of the inferior strain (by way of compensatory mutations or extension of the resistance spectrum), which could allow it to persist, even though coexistence is not expected on the basis of the current pathogen fitness.
The classical ecological paradigm of niche overlap states that two species can coexist if their resource usage differs sufficiently
More fundamentally, the transmission route of MRSA in hospitals might change. The current view is that MRSA in hospitals is mainly transmitted indirectly through short-term contaminated health-care workers
Equations summarizing the Treatment-Structured Model.
(DOCX)
The treatment-structured model is derived from the basic model by subdividing each compartment according to treatment status. Specifically, we distinguish between 4 treatment classes: 1) Untreated; 2) treated with a drug that is effective against neither CA-MRSA nor HA-MRSA; 3) treated with a drug that is effective against CA-MRSA but not HA-MRSA; and 4) treated with a drug that is effective against both CA-MRSA and HA-MRSA. Upon treatment initiation with an effective drug, the infection is either cleared immediately or alternatively remains colonized (this is an approximation to the real dynamics in which the patient would clear after a given amount of time). Finally, treated individuals stop treatment at a rate that is the inverse duration of antibiotic use.
(DOCX)
Invasion of HA-MRSA into the MSSA equilibrium: Simulations are initiated with MSSA only and run until they reach equilibrium. Then HA-MRSA is introduced (at a frequency of 0.1%). Color indicates frequency of HA-MRSA and MSSA at a given time point (for each strain we compute its frequency among S. aureus infections in the hospital and the community and use the larger of the two values): HA-MRSA alone (light blue = HA-MRSA>5% and MSSA<5%); Coexistence (dark blue = HA-MRSA>5% and MSSA>5%); MSSA alone (red = HA-MRSA<5% and MSSA>5%).
(DOCX)
Example runs for the treatment- and age-structured model with
(DOCX)
Summary of the percentage of drug usage that was effective against CA- and HA-MRSA.
(DOCX)