Skip Navigation

Measuring and Interpreting Associations between Antibiotic Use and Penicillin Resistance in Streptococcus pneumoniae

  1. Marc Lipsitch
  1. Department of Epidemiology, Harvard School of Public Health, Boston
  1. Reprints or correspondence: Dr. Marc Lipsitch, Dept. of Epidemiology, Harvard School of Public Health, 677 Huntington Ave., Boston, MA 02115 (mlipsitc{at}hsph.harvard.edu).

  1. Presented in part: 2d International Symposium on Pneumococci and Pneumococcal Diseases, Sun City, South Africa, 19–23 March 2000 (abstract O74).

 
Next Section

Abstract

Studies of the relationship between antibiotic use and resistance in pneumococci have produced conflicting results, reflecting differences in study design, setting, and measures of association used. Mathematical models of pneumococcal transmission dynamics provide a framework for interpreting and reconciling these studies. The model predicts, and the review of published studies confirms, that treatment often has little effect in increasing an individual's absolute risk of carrying/being infected by penicillin-resistant Streptococcus pneumoniae (PRSP). However, treatment substantially increases a patient's risk of carriage of/infection by PRSP relative to that of penicillin-susceptible S. pneumoniae (PSSP). The appropriate measure of association depends on the question of interest. Antibiotic use can substantially increase the prevalence of risk in the community as a whole, even when there is a small or nonexistent effect of treatment on the absolute risk that a treated individual will carry a resistant organism. Recommendations for the design and analysis of future studies of antibiotic treatment and pneumococcal resistance are proposed.

Resistance of Streptococcus pneumoniae to penicillin and other antimicrobial agents is an increasing problem [1, 2]. Many studies have examined the relationship between antibiotic treatment of a patient (usually with a β-lactam or a cephalosporin) and that patient's risk of carrying, or being infected by, penicillin-resistant S. pneumoniae (PRSP). The results of these studies have varied considerably. This variation reflects differences in study design, time scale, and setting, as well as the measures of association used. This paper proposes a framework for interpreting these studies to draw conclusions about the individual and community-level effects of antimicrobial treatment on penicillin resistance in pneumococci.

The design and interpretation of epidemiological studies depend on the public health questions of interest. Studies of the association between use of antimicrobial agents and resistance in the pneumococcus can provide information on 3 important issues. The first issue is the individual effects of treatment on resistance. Does treatment with an antibiotic make the treated patient more likely to be colonized with, or infected by, a resistant organism? How do antibiotics or doses differ in this respect? The second issue is the community-wide effect of treatment on resistance. How does the use of antibiotics by individual patients affect the probability that other members of the community will be colonized with or infected by resistant organisms? How do changes in antibiotic use in the community as a whole affect the level of resistance in the pneumococcal population? The third issue is the informativeness of antibiotic treatment history for the prediction of resistance. If a patient presents with a likely or confirmed pneumococcal infection, what (if any) information does the patient's recent history of antibiotic use provide to the physician who is choosing an antibiotic for treatment of the patient's current infection?

The first part of this article describes how studies can be designed and interpreted to address each of these questions. The second part of this article describes a simple mathematical model that can be used as a framework for the analysis of the effects of antimicrobial treatment on pneumococcal colonization in observational and randomized studies of individual patients. The model is used to predict the way in which the magnitude of the association between resistance and previous antibiotic treatment will depend on the measure used and on the time interval between antibiotic exposure and measurement of pneumococcal carriage or disease. The third part of this article reviews, within the framework for analysis described, published studies of the association between antibiotic use and penicillin resistance in pneumococci. For the sake of brevity and clarity, the abbreviations PSSP (to denote penicillin-susceptible S. pneumoniae) and PRSP are used throughout the article, although many studies also distinguish between intermediately resistant and highly resistant pneumococci.

Previous SectionNext Section

Measuring the Relationship Between Antibiotic Use and Penicillin Resistance in Pneumococci

The association between antibiotic use and penicillin resistance in pneumococci has been measured by use of diverse study designs. Nearly all of these studies compare carriage (or infection) rates among treated versus untreated patients; treatment status may be measured before and after an intervention (e.g., in a randomized comparison of antibiotic regimens), or it may be defined as a history of antibiotic use during a defined time period before sampling (e.g., in observational studies). Such data fit into the categories shown in table 1.

Table 1
View larger version:
Table 1

Groups of subjects in a study of the association between antibiotic use and colonization with resistant pneumococci.

These data are usually analyzed in 1 of 2 ways. Some studies test for an association between exposure to the antibiotic(s) and carriage of PRSP, as opposed to noncarriage of PRSP. This association can be measured as an OR, which can be called the “simple OR” (ORS) because it compares carriage of PRSP with noncarriage of PRSP. ORS is defined as ORS = [RT(NU − RU)]/[RU(NT − RT)] = [RT(XU + SU)]/[RU]. Here, X is the proportion of subjects who are noncarriers of S. pneumoniae, S is the proportion of subjects carrying PSSP, and R is the proportion of subjects carrying PRSP. Subscripts U and T refer to untreated and treated subjects, respectively. The total numbers of treated and untreated hosts are denoted by NT and NU, respectively (see table 1).

In other studies, PRSP carriers (or infected persons) are compared with members of a different reference group: PSSP carriers (or people infected with PSSP). This association can be measured by a different OR, which can be called the “conditional OR” (ORC) because it refers to the odds of having a resistant pneumococcus, conditional on carriage of (or infection with) some pneumococcus. ORC is defined as ORC = RTSU/RUST. For any antibiotic that is effective in reducing total pneumococcal carriage, we will have XT > XU; as a result, we expect that, in all cases, ORC should be greater than ORS (i.e., ORC > ORS).

These measures provide different information about the effect of treatment on resistance. ORS measures the direct effect of treatment on a person's risk of colonization or infection with resistant pneumococci. If ORS > 1, it indicates (subject to the usual caveats in epidemiologic studies) that treatment of a person makes that person more likely to be infected or colonized with resistant organisms.

By contrast, ORC is poorly suited to measuring the effect of treatment on an individual's absolute risk for colonization or infection with resistant pneumococci. If antibiotic treatment clears carriage of susceptible pneumococci but has no effect on resistant organisms, then this OR will be elevated, even though a person's risk of carrying resistant organisms is unchanged. Moreover, an antibiotic that clears carriage even of “resistant” organisms in some patients may reduce the person's risk of carriage of resistant organisms, yet ORC could still exceed 1 if the antibiotic kills susceptible bacteria more efficiently than it kills resistant bacteria.

ORC is well suited to address the other 2 issues described in the introduction to this article: the community-wide effects of treatment and the informativeness of treatment history for the choice of antibiotics. To decide what antibiotic to prescribe, a physician would need to answer the following question: given that this patient has an infection (which is definitely or probably pneumococcal) and given that this patient has recently been treated with a particular antibiotic, is the infectious microorganism more likely to be resistant to that antibiotic than the same infection would be in someone who has not been treated recently? This is exactly the question that ORC answers.

With regard to community-level effects, the reason for the importance of ORC is more subtle. One important way in which antibiotic use promotes antibiotic resistance at the community level is by clearing carriage of drug-susceptible organisms. In this way, it produces indirect effects on the population as a whole by reducing the population's exposure to drug-susceptible organisms; this effect is similar to the herd immunity effects of vaccination [3]. In the case of antibiotics, however, this population effect also promotes the spread of resistant organisms, because different pneumococcal populations compete to colonize hosts [4, 5]. Therefore, any action (such as treatment) that reduces the successful transmission of susceptible organisms will promote the transmission of resistant ones at the population level.

In terms of mathematical transmission models, treatment reduces the basic reproductive rate (R0) [3] of susceptible organisms. Specifically, treatment of individuals who carry susceptible organisms reduces their infectiousness, their duration of carriage, or both. This results in an increased prevalence of resistant organisms [6]. The ORC reflects this process because it measures the reduction in carriage of susceptible pneumococci in a treated person as well as the increased carriage of resistant organisms. Although discussion on this topic has been framed in terms of univariate measures (ORs), the principles apply equally well to multivariate measures, such as multiple logistic regressions.

Previous SectionNext Section

Treatment and Resistance: A Simple Model of the Dynamics of Pneumococcal Carriage

In this section, a mathematical model of the dynamics of pneumococcal carriage among the subjects of a particular intervention study is used to predict how these measures of the association between antibiotic use and resistance will change during the interval from the time when a patient is treated to the time when his or her pneumococcal carriage is sampled. The purpose is to complete the framework for analysis of existing studies and to suggest principles for the design of future studies.

Structure of the model. The dynamics of colonization with susceptible and resistant pneumococci can be captured in a simple mathematical model, as illustrated in figure 1. (See figure 2 in the study by Feikin et al. [7] for illustration of a similar model.) For reasons of simplicity, the model shown in figure 1 is used to describe an intervention study in which subjects are treated with a particular antibiotic regimen, and their carriage rates before and after treatment are compared by use of the aforementioned ORs. With suitable modifications, this model could be used for cross-sectional or other observational study designs.

Figure 1
View larger version:
Figure 1

Structure of a simple model for analysis of an antibiotic treatment trial in a population exposed to colonization by both resistant and susceptible pneumococci. Left, Before treatment, susceptible patients (X) are colonized by penicillin-susceptible and penicillin-resistant pneumococci at rates of λS and λR, respectively. Colonization is lost at rates of μS and μR, respectively. These processes are described by equation [1] in the text. Right, Treatment clears carriage of susceptible bacteria with probability pSX, clears carriage of resistant bacteria with probability pRX, and transfers patients from primarily carrying susceptible bacteria to primarily carrying resistant bacteria with probability pSR. These processes are described by equation [2] in the text. At the end of treatment, patients are again exposed to the processes of colonization and loss at their pretreatment rates. X, noncarriers of Streptococcus pneumoniae; S, carriers of penicillin-susceptible pneumococci; R, carriers of penicillin-resistant pneumococci; arrows, transitions between categories. See table 2 for a list of parameter values for this illustrative model.

The model shown in figure 1 presents the colonization-loss dynamics among a group of study subjects that is relatively small in number compared with the size of the whole community. This model is in contrast to most transmission-dynamic models of antibiotic treatment [6, 8], which consider the dynamics in the whole community.

Consider a closed population before the trial starts. This population contains individuals who have not yet received treatment. These individuals are classified into 3 categories that correspond to the titles of columns 2–4 in table 1. The arrows in figure 1 represent transitions between categories. Patients who are not carrying pneumococci (X) are colonized by susceptible and resistant pneumococci at rates of λS and λR, which changes their classification to the S and R categories, respectively. Once colonized, these patients clear colonization with susceptible and resistant pneumococci at rates of μS and μR, respectively, in the absence of treatment. The equations for this system (for both treated and untreated hosts) are: Formula 1

If the source population for the study has reached a steady state, it will have proportions RU = λRμS/(λSμR +λRμS+μRμS) colonized with resistant pneumococci and SU = λSμR/(λSμR + λRμS + μRμS) colonized with susceptible pneumococci. These proportions are obtained by solving equation [1] for equilibrium.

Treatment may have several effects on the carriage status of patients. When patients with carriage of susceptible bacteria are treated, they will be cleared of carriage and will be reclassified as belonging to the “uncolonized” X category of patients, with a probability of pSX. Some patients in the S category may already carry a small subpopulation of resistant organisms, so we allow for the possibility that treatment will result in the patient category changing from S to R with a probability of pSR. Finally, some antibiotics will have partial efficacy in clearing the carriage of organisms that are classified as “resistant,” so we allow for the possibility that treatment will result in a change in the patient category changing from R to X with a probability of pRX.

Time (t) is measured from the moment t = 0, when each person is treated. Immediately after treatment, the values of the variables are reset as follows: Formula 2

For simplification, we make the assumption that no further changes occur during treatment; thus, for treatment of duration d, we set (for all t ⩽ d): ST(t) = ST(0), RT(t) = RT(0), XT(t) = XT(0). After treatment is completed, patients are recolonized at rates (λS and λR) equal to the rates at which they acquired colonization before treatment, and they also begin to lose colonization at pretreatment rates; therefore, for t > d, equation [1] takes effect for treated hosts, with starting conditions given by equation [2].

The values of the parameters for rates of colonization (λ) and loss (μ) will depend on the population studied, and the probabilities of the various events upon treatment (p) will additionally depend on the antibiotic being used and the susceptibility profile of the strains in the population. For example, if one were considering both a population that had many strains with MICs just slightly greater than the resistance breakpoint and an antibiotic that reached very high effective concentrations, pRX might be relatively high, whereas for an antibiotic that never reached in vivo concentrations that approached the MIC of resistant strains, pRX would be nearly 0. Table 2 shows the parameters used for the purpose of illustration in figure 2.

Table 2
View larger version:
Table 2

Parameter values for the illustrative model shown in figure 1.

Figure 2
View larger version:
Figure 2

Time course of a hypothetical study of the effect of an antibiotic treatment on carriage. Top, Proportions of the subjects in the trial who had carriage of resistant (thick line) or susceptible (thin line) pneumococci. Bottom, The 2 measures of association between treatment and resistance—simple OR (thick line) and conditional OR (thin line)—at various times after treatment. Parameters are the same as those shown in table 2.

Predictions of the model. Figure 2 shows the modeled dynamics of colonization in the subjects of a trial before and after treatment. Shown are the proportions of the population colonized with susceptible or resistant bacteria at a given interval after treatment (figure 2, top) and the values of the 2 aforementioned ORs as they change over time (figure 2, bottom).

In this example, 50% of subjects had carriage of susceptible bacteria before treatment and 20% had carriage of resistant bacteria. Treatment began at day 0 and lasted for 10 days. Treatment resulted in a 20% reduction in carriage of resistant bacteria (from 20% to 16%) and a 90% reduction in carriage of susceptible bacteria (from 50% to 5%). In many settings (including our hypothetical one), λS is greater than λR. This reflects the fact that most of the pneumococci to which people are exposed are susceptible to penicillin.

If we return to our hypothetical example and look at how the ORs will change over time, ORC jumps to 8 for patients whose carriage is measured just after treatment, ORC = [(16%)(50%)]/[(20%)(5%)] = 8, and it then declines rapidly from that peak as treated patients are recolonized. By contrast, ORS declines to slightly less than 1 immediately after treatment, ORS = [(16%)(80%)]/[(20%)(84%)] = 0.76. The decline reflects the reduction in carriage of resistant organisms that is caused by treatment. As time passes, ORS creeps upward, as the temporary glut of uncolonized patients created by treatment experiences recolonization. ORS reaches a peak value of 1.18 approximately 100 days after treatment. As time goes on, the population returns to its pretreatment equilibrium, which returns both ORs to 1.

These results are typical of those that are predicted by the model for reasonable values of the parameters. The model makes the following general predictions:

  1. ORC will usually reach its highest level immediately after treatment and will decline as the interval between treatment and sampling gets longer.

  2. ORS may be >1, 1, or <1 just after treatment. If pSR is not too large, ORS is expected to increase over time, reaching its maximum at ∼10–100 days after treatment and then declining back toward 1. ORS > 1 reflects an absolute increase in the carriage of resistant bacteria as a result of treatment. This will occur if more people move from category S to category R as a result of the outgrowth of a resistant subpopulation, in comparison with the number of those who move from category R to category X as a result of treatment-induced clearance of resistant carriage. If the opposite is true, we will observe ORS < 1.

  3. ORS is likely to be elevated when a study is performed in a population that has had a sudden increase in exposure to resistant organisms, such as in an outbreak of a resistant strain. The model assumed that the subjects' exposure to resistant and susceptible organisms remained constant over time. In the context of an outbreak of resistant organisms (or any sudden increase in exposure to colonization of resistant organisms), patients who are treated may be more likely to become colonized with resistant organisms of the outbreak strain than they would have been before they were treated. As a result, treatment during an outbreak may be an important risk factor for carriage of resistant organisms, even when measured by ORS, although it would not be if the same study were done in the same population at a different time (i.e., if it were not done during an outbreak).

Study design and analysis. The aforementioned considerations have several implications for study design. First, studies will virtually always have a greater power to detect an effect of treatment on resistance if they measure ORC than if they measure ORS, simply because the size of the effect is larger for ORC. Second, in any given trial, the ORC will be largest (and, therefore, easiest to detect) if a short time span between treatment and sampling is used; therefore, a greater effect is expected if antibiotic use during the previous month is tested as a risk factor for PRSP disease (vs. PSSP disease), rather than if use within the previous 2 months is considered. Third, ORS may be close to 1 at any given time, and the optimal sampling time for ORS will be variable.

If one is interested in determining the extent to which treatment clears carriage of resistant organisms, then patients should be sampled as soon as possible after treatment. Carriage of resistant organisms after treatment may peak at any time from weeks to a few months after treatment, as patients who have had their susceptible organisms cleared by treatment are recolonized (some of them, by resistant organisms; figure 2). These conclusions may change if the outgrowth of preexisting resistant subpopulations is a major factor, as it appears to be in some studies of other antibiotic and resistance associations in pneumococci (see the Discussion section) [7, 9, 10].

Although investigators may differ in their primary interest when undertaking any given study, many studies have the potential to provide additional information on several questions in addition to their primary end points. For most observational studies, data of the sort given in table 1, together with relevant univariate and multivariate analyses of both ORC and ORS, will provide maximal information and will allow the investigators to separate out the mechanistic effects of treatment on resistance; these effects include reduction of PSSP carriage, possible reduction of PRSP carriage, outgrowth of resistant subpopulations in treated hosts, and in the longer term, possible increased susceptibility to PSSP carriage.

In many intervention studies and in some longitudinal ones, a more informative option is available: stratification of the antibiotic effect according to the carriage status of the patient before treatment (as shown in table 3) [7, 11]. When data are presented in this form, it becomes natural to describe the effect of treatment in each of the category transitions shown in figure 1: clearance of PSSP carriage, which changes the patient category from S to X; clearance of PRSP carriage, which changes the patient category from R to X; and clearing PSSP carriage but permitting the outgrowth of a PRSP strain (which may have been there originally in large numbers, or which may be acquired during or just after treatment), which changes the patient category from S to R. From a statistical perspective, such an analysis has the benefit of controlling for some (although not all) of the nonindependence of observations of the same patient before and after treatment, by conditioning on colonization status before treatment.

Table 3
View larger version:
Table 3

Stratification of patients in a treatment study, according to pneumococcal carriage status before treatment.

Previous SectionNext Section

Interpretation of Existing Studies

Table 4 presents the results of 14 published studies of the association between antibiotic use and infection with penicillin-resistant pneumococci; table 5 shows the results of 17 such studies of carriage (one of which includes some invasive disease isolates). A total of 1033 English-language abstracts were obtained by searching MEDLINE records through the end of 1999 for studies of humans by use of the key words “Streptococcus pneumoniae” and “penicillin,” along with the terms “resistantrd or “resistance.” From these abstracts, we identified 45 possible articles that appeared as if they might meet the criteria of being either (1) prospective treatment studies with data on carriage of PSSP and PRSP before and after treatment, or (2) observational studies in which treatment with an antimicrobial agent or with classes of antimicrobial agents, performed during a defined time period, was examined as a risk factor for carriage of or infection with penicillin-resistant pneumococci.

Table 4
View larger version:
Table 4

Studies of the association between antibiotic use and penicillin-nonsusceptible pneumococcal infection.

Table 5
View larger version:
Table 5

Studies of the association between antibiotic use and penicillin-nonsusceptible pneumococcal carriage.

Studies were included only if 1 or both of the aforementioned ORs (or corresponding RRs) were calculated, or if they could be calculated from the data given. Studies of HIV-infected patients or patients with sickle-cell anemia were excluded. Of the 45 studies, 26 were included; of the remaining studies, 1 was unobtainable and 18 did not provide the relevant data. Five additional references were identified through the bibliographies of these and other sources.

The methodologies used in these studies were described in various ways, and they have been classified as case-control studies, cross-sectional (or in 1 case, longitudinal) prevalence studies, or before-after (intervention) studies. For the studies that reported relevant ORs, the ORs have been reproduced in tables 4 and 5. For studies that analyzed the data in other ways (or reported only 1 of the 2 ORs), other ORs were calculated whenever possible; calculations that are not reported in the original article are shown in parentheses. Statistical significance (which is indicated in the footnotes of the tables) reflects calculations reported in each original article, when available. In assessing the significance of ORs calculated for the present analysis and not present in the original article, the χ2 test or Fisher's exact test was used (EpiInfo 6; CDC).

In the studies of pneumococcal disease, only ORC could be calculated. Nine of the 14 studies, including most of those that had larger sample sizes, reported a statistically significantly greater risk of PRSP disease than of PSSP disease among patients who had received antibiotics or some class of antibiotics. This is in accord with my prediction that ORC generally should be large and relatively easy to detect. Therefore, these studies collectively provide considerable evidence to indicate that, when a patient has a pneumococcal infection, it is more likely to be penicillin resistant if that patient was recently treated with an antibiotic or with a β-lactam in particular.

The studies of carriage were more variable in their design and included 4 intervention (before-after) studies and 13 observational studies. In nearly all of the studies for which ORC was available, ORC was significantly greater than 1. In 5 studies, ORC was not significantly greater than 1 at any point in time [11, 27, 31, 33, 34]. Possible explanations include limited power [11, 27, 31] and, in 1 study, the use of cefaclor, which has poor activity against pneumococci in the nasopharynx [11]. In the 2 cases where ORC was calculated for prior antibiotic use over different time periods, it was largest for the shortest time period and then declined, as predicted [35, 36].

The model predicted that ORS would rarely be much bigger than 1, except in the context of an outbreak of a resistant organism. All 3 studies conducted during an outbreak of microorganisms concluded that the treated subjects were at significantly higher absolute risk of carriage of resistant organisms than were untreated subjects (ORS > 1) [37–39]. In most of the remaining cases, ORS was not significantly greater than 1. Indeed, in 3 studies—all of which were intervention trials in which patients were sampled at 7–14 days after treatment with amoxicillin-clavulanate—treated patients were less likely to carry resistant organisms than were untreated ones (ORS < 1), and this association reached statistical significance. This suggests that amoxicillin-clavulanate may be particularly effective for clearance of nasopharyngeal carriage of “resistant” (including intermediately resistant) S. pneumoniae, and it is also consistent with the prediction that ORS may be depressed below 1 just after treatment.

There were just 2 studies in which ORS was significantly greater than 1, apart from those performed in the context of resistant outbreaks. In one of these studies, only low doses of antibiotics or treatment of long duration were implicated, and the frequency of colonization was extremely low, probably as a result of the use of oropharyngeal sampling [40]. The other study also involved a small number of children (n = 48), although several samples were obtained from each child [41].

The following conclusions may be drawn from the existing studies:

  1. Recent antibiotic treatment makes it more likely that a patient who is carrying a pneumococcus has a resistant (rather than a susceptible) organism (ORC > 1). This suggests that antibiotic (in particular, β-lactam or cephalosporin) treatment is exerting an important effect at the community level by reducing carriage of susceptible organisms, thereby indirectly promoting the transmission of resistant ones [6, 42].

  2. In the setting of an outbreak of a resistant pneumococcus, antibiotic treatment puts a patient at a higher risk for colonization by the resistant strain (ORS > 1). This presumably occurs because clearance of the patient's own flora makes new colonization more likely, and the patient will have substantial exposure to the resistant strain.

  3. Outside the setting of a resistant outbreak, there is little evidence that treatment makes a patient more susceptible to carrying a resistant strain than an untreated individual patient would be. Indeed, in some circumstances, at least in the short term, treatment can clear a person of colonization with even a “resistant strain”; this has been shown in trials only of amoxicillin-clavulanate [33, 34, 43]. Therefore, the possibility that treatment puts a person at greater risk for resistant carriage (or infection), although perhaps suggested by 2 aforementioned studies [40, 41], should not be accepted without further study.

Previous SectionNext Section

Discussion

Growing concern about antibiotic resistance in general and about penicillin resistance in pneumococci in particular, has led to a growing number of studies searching for modifiable risk factors for antibiotic-resistant infection or colonization. A practical goal of these studies is to identify changes in prescribing practices or choices of antibiotics that will maximize the “life expectancy” of existing antibiotics while maintaining appropriate treatment for individual patients. A related goal is to document the risks of inappropriate treatment, both to the treated patient and to others in the community, to motivate more judicious antibiotic use [44, 45].

Accomplishing these goals requires careful attention to the study design and interpretation. Although the reported results of differently designed studies that examine the effect of antibiotic treatment on resistance have been quite heterogeneous, the conclusions are in fact fairly consistent if the relevant characteristics of the studies are isolated. Treatment repeatedly has been shown to be a risk factor for infection or colonization with penicillin-resistant (as opposed to penicillin-susceptible) pneumococci, but it has rarely been shown to increase a patient's absolute risk for penicillin-resistant colonization or infection. The largest exceptions to the latter generalization occur during outbreaks of a resistant organism, as predicted by our model. Antibiotics differ in their effects on nasopharyngeal carriage of both susceptible and resistant organisms; for example, amoxicillin-clavulanate repeatedly has been shown to reduce the carriage of resistant and susceptible organisms, whereas other drugs have shown little effect on carriage of either. The importance of selecting the correct comparison group for studies of use of antimicrobial agents and resistance recently has been noted for other bacterial species [46, 47].

In many circumstances, treatment will increase only slightly (or may even decrease) carriage of resistant organisms in the treated patient. Furthermore, even when there is an effect on the patient, it will be relatively short-lived (figure 2); this prediction was confirmed in a recent study of resistance to trimethoprim-sulfamethoxazole (TMP-SMX) in Malawi [7]. If treatment produces, at most, a small and short-lived increase in resistance in the patient, how does it create selective pressure for resistance in the community? The answer, alluded to above, is that although treatment may reduce or only slightly increase carriage of resistant organisms, it typically has a considerably larger effect in reducing carriage of susceptible organisms. When 2 populations of organisms compete for a resource (in this case, hosts to colonize), any circumstance that reduces the fitness of one population will increase that of its competitor. Although hosts can carry both susceptible and resistant organisms simultaneously, there is evidence from several sources that carriage of one strain can inhibit acquisition of other strains, resulting in competition for hosts [4, 5].

In terms of mathematical models for the spread of antibiotic resistance in a population, resistant organisms will increase in frequency, roughly speaking, when their basic reproductive number (R0) is greater than that of susceptible ones. R0 for each type is given by the transmission rate constant for that type (β) multiplied by the average duration of carriage of that type. If treatment is more effective in clearing susceptible organisms than in clearing resistant ones, then an increase in rates of treatment in the community at large will result in a greater decline in R0 for susceptible organisms than in R0 for resistant ones. As a result, if treatment becomes sufficiently frequent, it will result in an increase in the frequency of resistant organisms.

Although the principles described in the present report should apply to any antibiotic or class and its corresponding resistance in pneumococcus, there may be important differences. At least 3 intervention studies have examined the effect of other drugs: azithromycin in 2 cases [9, 10] and TMP-SMX in a third [7]. In these cases, carriage of organisms resistant to the drug used for treatment increased significantly after treatment, compared with before (ORS > 1). It will be important to understand why this phenomenon seems to be more frequent in these other drug classes (albeit in a limited number of studies) than for resistance to β-lactams and penicillin.

Part of the explanation—at least for TMP-SMX—may be that outgrowth of resistant subpopulations is relatively frequent compared to killing of “resistant” organisms; this is consistent with the results of Feikin et al. [7], who measured these processes separately. Furthermore, it may be relatively easy for some β-lactams or cephalosporins to eradicate an organism that has been defined as nonsusceptible [48, 49] because susceptibilities (MICs) to these drugs form a relatively smooth continuum, in contrast to macrolides, for example, in which there are fairly sharp breaks in MICs between susceptible and resistant isolates [50].

Randomized studies of antibiotic use and resistance have 2 important advantages over most observational studies. First, they allow investigators to separate the different effects of treatment: (1) killing susceptible organisms; (2) in some cases, allowing for the outgrowth of resistant subpopulations; and (3) possibly increasing susceptibility to future colonization with a resistant strain. Such knowledge is important as the scientific community attempts to understand the impact of treatment at the community level. Second, randomized intervention studies avoid many of the problems of confounding and reverse causality that can cloud the interpretation of observational studies. For example, in an observational study, confounders such as age and day care attendance might produce a spurious association between use of antimicrobial agents and carriage of resistant organisms. A randomized intervention study would reduce this confounding. Alternatively, multivariate methods can be used to control for such confounding in observational studies.

These studies raise a practical question: how can an antibiotic regimen be chosen that will not only be effective in treating the patient's infection but will also: (1) minimize the probability that the treated patient will subsequently develop resistant infection, and (2) minimize the selective pressure for resistant pneumococci in the community as a whole? ORS is an appropriate measure of the direct effect of treatment on a patient's own risk of resistant infection, so drug regimens that produce the lowest value for this measure, perhaps by killing relatively resistant organisms, may be preferable for the patient.

From the perspective of the community, the choice is more complicated for 2 reasons. The first reason was given above: even a drug that kills resistant organisms can nonetheless exert selection in favor of resistance, if it is more effective against susceptible ones. Future work in mathematical modeling should address the relative merits (at a population level) of, for example, a drug that is highly effective against susceptible organisms (producing marked selection in favor of resistance) but that also clears carriage of resistant ones (reducing the selective benefits of resistance), in comparison with a drug that is less effective against carriage of susceptible organisms but that also rarely kills resistant ones. Second, at the community level, one effect of using a drug that can kill “resistant” organisms may be to select for still higher levels of resistance, which places a microorganism beyond the reach of even the more potent antibiotic. In vitro and animal studies [51, 52] suggest that more potent treatment regimens can select for ever-higher MICs, whereas less potent ones will provide little selective advantage for MICs higher than those required to resist drug action.

Previous SectionNext Section

Acknowledgments

I thank Matthew Samore and Stephanie Schrag for helpful discussions of these issues, and Scott Dowell, Susan Huang, and 2 reviewers for helpful comments on the manuscript.

  • Received May 19, 2000.
  • Revision received August 9, 2000.
Previous Section
 

References

  1. ↵
    1. Fenoll A,
    2. Jado I,
    3. Vicioso D,
    4. Perez A,
    5. Casal J
    . Evolution of Streptococcus pneumoniae serotypes and antibiotic resistance in Spain: update (1990 to 1996). J Clin Microbiol 1998;36:3447-54.
    FREE Full Text
  2. ↵
    1. Doern GV,
    2. Brueggemann AB,
    3. Huynh H,
    4. Wingert E
    . Antimicrobial resistance with Streptococcus pneumoniae in the United States, 1997–98. Emerg Infect Dis 1999;5:757-65.
    MedlineWeb of Science
  3. ↵
    1. Anderson RM,
    2. May RM
    . Infectious diseases of humans: dynamics and control. Oxford: Oxford University Press; 1991.
  4. ↵
    1. Lipsitch M,
    2. Dykes JK,
    3. Johnson SE,
    4. et al
    . Competition among Streptococcus pneumoniae for intranasal colonization in a mouse model. Vaccine 2000;18:2895-901.
    CrossRefMedlineWeb of Science
  5. ↵
    1. Lipsitch M
    . Bacterial vaccines and serotype replacement: lessons from Haemophilus influenzae and prospects for Streptococcus pneumoniae. Emerg Infect Dis 1999;5:336-45.
    MedlineWeb of Science
  6. ↵
    1. Austin DJ,
    2. Kristinsson KG,
    3. Anderson RM
    . The relationship between the volume of antimicrobial consumption in human communities and the frequency of resistance. Proc Natl Acad Sci USA 1999;96:1152-6.
    Abstract/FREE Full Text
  7. ↵
    1. Feikin DR,
    2. Dowell SF,
    3. Nwanyanwu OC,
    4. et al
    . Increased carriage of trimethoprim/sulfamethoxazole-resistant Streptococcus pneumoniae in Malawian children after treatment for malaria with sulfadoxine/pyrimethamine. J Infect Dis 2000;181:1501-5.
    Abstract/FREE Full Text
  8. ↵
    1. Bonhoeffer S,
    2. Lipsitch M,
    3. Levin BR
    . Evaluating treatment protocols to prevent antibiotic resistance. Proc Natl Acad Sci USA 1997;94:12106-11.
    Abstract/FREE Full Text
  9. ↵
    1. Leach AJ,
    2. Shelby-James TM,
    3. Mayo M,
    4. et al
    . A prospective study of the impact of community-based azithromycin treatment of trachoma on carriage and resistance of Streptococcus pneumoniae. Clin Infect Dis 1997;24:356-62.
    Abstract/FREE Full Text
  10. ↵
    1. Morita JY,
    2. Kahn E,
    3. Thompson T,
    4. et al
    . Impact of azithromycin on oropharyngeal carriage of group A Streptococcus and nasopharyngeal carriage of macrolide-resistant Streptococcus pneumoniae. Pediatr Infect Dis J 2000;19:41-6.
    CrossRefMedlineWeb of Science
  11. ↵
    1. Dagan R,
    2. Leibovitz E,
    3. Greenberg D,
    4. Yagupsky P,
    5. Fliss DM,
    6. Leiberman A
    . Dynamics of pneumococcal nasopharyngeal colonization during the first days of antibiotic treatment in pediatric patients. Pediatr Infect Dis J 1998;17:880-5.
    CrossRefMedlineWeb of Science
    1. Bedos JP,
    2. Chevret S,
    3. Chastang C,
    4. Geslin P,
    5. Regnier B
    . Epidemiological features of and risk factors for infection by Streptococcus pneumoniae strains with diminished susceptibility to penicillin: findings of a French survey. Clin Infect Dis 1996;22:63-72.
    Abstract/FREE Full Text
    1. Nava JM,
    2. Bella F,
    3. Garau J,
    4. et al
    . Predictive factors for invasive disease due to penicillin-resistant Streptococcus pneumoniae: a population-based study. Clin Infect Dis 1994;19:884-90.
    Abstract/FREE Full Text
    1. Deeks SL,
    2. Palacio R,
    3. Ruvinsky R,
    4. et al
    . Risk factors and course of illness among children with invasive penicillin-resistant Streptococcus pneumoniae: the Streptococcus pneumoniae Working Group. Pediatrics 1999;103:409-13.
    Abstract/FREE Full Text
    1. Kronenberger CB,
    2. Hoffman RE,
    3. Lezotte DC,
    4. Marine WM
    . Invasive penicillin-resistant pneumococcal infections: a prevalence and historical cohort study. Emerg Infect Dis 1996;2:121-4.
    MedlineWeb of Science
    1. Block SL,
    2. Harrison CJ,
    3. Hedrick JA,
    4. et al
    . Penicillin-resistant Streptococcus pneumoniae in acute otitis media: risk factors, susceptibility patterns and antimicrobial management. Pediatr Infect Dis J 1995;14:751-9.
    MedlineWeb of Science
    1. Tan TQ,
    2. Mason EO Jr,
    3. Kaplan SL
    . Penicillin-resistant systemic pneumococcal infections in children: a retrospective case-control study. Pediatrics 1993;92:761-7.
    Abstract/FREE Full Text
    1. Clavo-Sanchez AJ,
    2. Giron-Gonzalez JA,
    3. Lopez-Prieto D,
    4. et al
    . Multivariate analysis of risk factors for infection due to penicillin-resistant and multidrug-resistant Streptococcus pneumoniae: a multicenter study. Clin Infect Dis 1997;24:1052-9.
    Abstract/FREE Full Text
    1. Reichler MR,
    2. Rakovsky J,
    3. Sobotova A,
    4. et al
    . Multiple antimicrobial resistance of pneumococci in children with otitis media, bacteremia, and meningitis in Slovakia. J Infect Dis 1995;171:1491-6.
    Abstract/FREE Full Text
    1. Pallares R,
    2. Gudiol F,
    3. Linares J,
    4. et al
    . Risk factors and response to antibiotic therapy in adults with bacteremic pneumonia caused by penicillin-resistant pneumococci. N Engl J Med 1987;317:18-22.
    MedlineWeb of Science
    1. Doone JL,
    2. Klespies SL,
    3. Sabella C
    . Risk factors for penicillin-resistant systemic pneumococcal infections in children. Clin Pediatr (Phila) 1997;36:187-91.
    Abstract/FREE Full Text
    1. Rauch AM,
    2. O'Ryan M,
    3. Van R,
    4. Pickering LK
    . Invasive disease due to multiply resistant Streptococcus pneumoniae in a Houston, Tex, day-care center. Am J Dis Child 1990;144:923-7.
    Abstract/FREE Full Text
    1. Moreno F,
    2. Crisp C,
    3. Jorgensen JH,
    4. Patterson JE
    . The clinical and molecular epidemiology of bacteremias at a university hospital caused by pneumococci not susceptible to penicillin. J Infect Dis 1995;172:427-32.
    Abstract/FREE Full Text
    1. Saah AJ,
    2. Mallonee JP,
    3. Tarpay M,
    4. Thornsberry CT,
    5. Roberts MA,
    6. Rhoades ER
    . Relative resistance to penicillin in the pneumococcus: a prevalence and case-control study. JAMA 1980;243:1924-7.
    CrossRefMedline
    1. Bradley JS,
    2. Kaplan SL,
    3. Tan TQ,
    4. et al
    . Pediatric pneumococcal bone and joint infections: the Pediatric Multicenter Pneumococcal Surveillance Study Group (PMPSSG). Pediatrics 1998;102:1376-82.
    Abstract/FREE Full Text
    1. Arason VA,
    2. Kristinsson KG,
    3. Sigurdsson JA,
    4. Stefansdottir G,
    5. Molstad S,
    6. Gudmundsson S
    . Do antimicrobials increase the carriage rate of penicillin resistant pneumococci in children? Cross sectional prevalence study. BMJ 1996;313:387-91.
    Abstract/FREE Full Text
  27. ↵
    1. Boken DJ,
    2. Chartrand SA,
    3. Moland ES,
    4. Goering RV
    . Colonization with penicillin-nonsusceptible Streptococcus pneumoniae in urban and rural child-care centers. Pediatr Infect Dis J 1996;15:667-72.
    CrossRefMedlineWeb of Science
    1. Ussery XT,
    2. Gessner BD,
    3. Lipman H,
    4. et al
    . Risk factors for nasopharyngeal carriage of resistant Streptococcus pneumoniae and detection of a multiply resistant clone among children living in the Yukon-Kuskokwim Delta region of Alaska. Pediatr Infect Dis J 1996;15:986-92.
    CrossRefMedlineWeb of Science
    1. Robins-Browne RM,
    2. Kharsany AB,
    3. Koornhof HJ
    . Antibiotic-resistant pneumococci in hospitalized children. J Hyg (Lond) 1984;93:9-16.
    Medline
    1. Melander E,
    2. Molstad S,
    3. Persson K,
    4. Hansson HB,
    5. Soderstrom M,
    6. Ekdahl K
    . Previous antibiotic consumption and other risk factors for carriage of penicillin-resistant Streptococcus pneumoniae in children. Eur J Clin Microbiol Infect Dis 1998;17:834-8.
    CrossRefMedlineWeb of Science
  31. ↵
    1. Mainous AG III,
    2. Evans ME,
    3. Hueston WJ,
    4. Titlow WB,
    5. McCown LJ
    . Patterns of antibiotic-resistant Streptococcus pneumoniae in children in a day-care setting. J Fam Pract 1998;46:142-6.
    MedlineWeb of Science
    1. Kellner JD,
    2. Ford-Jones EL
    . Streptococcus pneumoniae carriage in children attending 59 Canadian child care centers. Toronto Child Care Centre Study Group. Arch Pediatr Adolesc Med 1999;153:495-502.
    Abstract/FREE Full Text
  33. ↵
    1. Cohen R,
    2. Navel M,
    3. Grunberg J,
    4. et al
    . One dose ceftriaxone vs. ten days of amoxicillin/clavulanate therapy for acute otitis media: clinical efficacy and change in nasopharyngeal flora. Pediatr Infect Dis J 1999;18:403-9.
    CrossRefMedlineWeb of Science
  34. ↵
    1. Dabernat H,
    2. Geslin P,
    3. Megraud F,
    4. et al
    . Effects of cefixime or co-amoxiclav treatment on nasopharyngeal carriage of Streptococcus pneumoniae and Haemophilus influenzae in children with acute otitis media. J Antimicrob Chemother 1998;41:253-8.
    Abstract/FREE Full Text
  35. ↵
    1. Tsolia M,
    2. Kouppari G,
    3. Zaphiropoulou A,
    4. et al
    . Prevalence and patterns of resistance of Streptococcus pneumoniae strains isolated from carriers attending day care centers in the area of Athens. Microb Drug Resist 1999;5:271-8.
    MedlineWeb of Science
  36. ↵
    1. Arnold KE,
    2. Leggiadro RJ,
    3. Breiman RF,
    4. et al
    . Risk factors for carriage of drug-resistant Streptococcus pneumoniae among children in Memphis, Tennessee. J Pediatr 1996;128:757-64.
    CrossRefMedlineWeb of Science
  37. ↵
    1. Melander E,
    2. Ekdahl K,
    3. Hansson HB,
    4. et al
    . Introduction and clonal spread of penicillin- and trimethoprim/sulfamethoxazole-resistant Streptococcus pneumoniae, serotype 9V, in southern Sweden. Microb Drug Resist 1998;4:71-8.
    MedlineWeb of Science
    1. Radetsky MS,
    2. Istre GR,
    3. Johansen TL,
    4. et al
    . Multiply resistant pneumococcus causing meningitis: its epidemiology within a day-care centre. Lancet 1981;2(8250):771-3.
    MedlineWeb of Science
  39. ↵
    1. Reichler MR,
    2. Allphin AA,
    3. Breiman RF,
    4. et al
    . The spread of multiply resistant Streptococcus pneumoniae at a day care center in Ohio. J Infect Dis 1992;166:1346-53.
    Abstract/FREE Full Text
  40. ↵
    1. Guillemot D,
    2. Carbon C,
    3. Balkau B,
    4. et al
    . Low dosage and long treatment duration of β-lactam: risk factors for carriage of penicillin-resistant Streptococcus pneumoniae. JAMA 1998;279:365-70.
    Abstract/FREE Full Text
  41. ↵
    1. Yagupsky P,
    2. Porat N,
    3. Fraser D,
    4. et al
    . Acquisition, carriage, and transmission of pneumococci with decreased antibiotic susceptibility in young children attending a day care facility in southern Israel. J Infect Dis 1998;177:1003-12.
    Abstract/FREE Full Text
  42. ↵
    1. Austin DJ,
    2. Kakehashi M,
    3. Anderson RM
    . The transmission dynamics of antibiotic-resistant bacteria: the relationship between resistance in commensal organisms and antibiotic consumption. Proc R Soc Lond B Biol Sci 1997;264:1629-38.
    Abstract/FREE Full Text
  43. ↵
    1. Cohen R,
    2. Bingen E,
    3. Varon E,
    4. et al
    . Change in nasopharyngeal carriage of Streptococcus pneumoniae resulting from antibiotic therapy for acute otitis media in children. Pediatr Infect Dis J 1997;16:555-60.
    CrossRefMedlineWeb of Science
  44. ↵
    Centers for Disease Control and Prevention. Defining the public health impact of drug-resistant Streptococcus pneumoniae: report of a working group. MMWR Morb Mortal Wkly Rep 1996;45(RR-1):1-14.
    Medline
  45. ↵
    1. Dowell SF,
    2. Schwartz B
    . Resistant pneumococci: protecting patients through judicious use of antibiotics. Am Fam Physician 1997;55:1647-54, 1657-8.
    MedlineWeb of Science
  46. ↵
    1. Carmeli Y,
    2. Samore MH,
    3. Huskins C
    . The association between antecedent vancomycin treatment and hospital-acquired vancomycin-resistant enterococci: a meta-analysis. Arch Intern Med 1999;159:2461-8.
    Abstract/FREE Full Text
  47. ↵
    1. Harris AD,
    2. Samore MH,
    3. Carmeli Y
    . The importance of control group selection in correctly identifying risk factors for antibiotic-resistant pathogens. Clin Infect Dis 2001;32:1055-61. (in this issue).
    Abstract/FREE Full Text
  48. ↵
    1. Gehanno P,
    2. Nguyen L,
    3. Barry B,
    4. et al
    . Eradication by ceftriaxone of Streptococcus pneumoniae isolates with increased resistance to penicillin in cases of acute otitis media [published erratum appears in Antimicrob Agents Chemother 1999; 43(6):1532]. Antimicrob Agents Chemother 1999;43:16-20.
    Abstract/FREE Full Text
  49. ↵
    1. Lister PD,
    2. Pong A,
    3. Chartrand SA,
    4. Sanders CC
    . Rationale behind high-dose amoxicillin therapy for acute otitis media due to penicillin-nonsusceptible pneumococci: support from in vitro pharmacodynamic studies. Antimicrob Agents Chemother 1997;41:1926-32.
    Abstract/FREE Full Text
  50. ↵
    1. Jacobs M
    . Increasing antibiotic resistance among otitis media pathogens and their susceptibility to oral agents based on pharmacodynamic parameters. Pediatr Infect Dis J 2000;19:S47-56.
    CrossRefMedlineWeb of Science
  51. ↵
    1. Negri MC,
    2. Morosini MI,
    3. Loza E,
    4. Baquero F
    . In vitro selective antibiotic concentrations of β-lactams for penicillin-resistant Streptococcus pneumoniae populations. Antimicrob Agents Chemother 1994;38:122-5.
    Abstract/FREE Full Text
  52. ↵
    1. Negri MC,
    2. Lipsitch M,
    3. Blazquez J,
    4. Levin BR,
    5. Baquero F
    . Concentration-dependent selection of small phenotypic differences in TEM-β-lactamase mediated antibiotic resistance. Antimicrob Agents Chemother 2000;44:2485-91.
    Abstract/FREE Full Text
| Table of Contents

This Article

  1. Clin Infect Dis. (2001) 32 (7): 1044-1054. doi: 10.1086/319604
  1. AbstractFree
  2. » Full Text (HTML)Free
  3. Full Text (PDF)Free

Classifications

Share

  1. Email this article

Navigate This Article

Current Issue

  1. March 1, 2012 54 (5)
  1. Clinical Infectious Diseases
  1. Alert me to new issues

Most