Nature Medicine9, 424 - 430 (2003)
Published online: 10 March 2003; | doi:10.1038/nm839
Geographic diversity and temporal trends of antimicrobial resistance in Streptococcus pneumoniae in the United States
Althea W. McCormick1, Cynthia G. Whitney2, Monica M. Farley3, Ruth Lynfield4, Lee H. Harrison5, Nancy M. Bennett6, William Schaffner7, Arthur Reingold8, James Hadler9, Paul Cieslak10, Matthew H. Samore11
& Marc Lipsitch1
1 Department of Epidemiology, Harvard School of Public Health, Boston, Massachusetts, USA
2 Active Bacterial Core surveillance and Emerging Infections Program Network, Centers for Disease Control and Prevention, Atlanta, Georgia, USA
3 Emory Department of Medicine, Emory University, Atlanta, Georgia, USA
4 Minnesota Emerging Infections Program, Minnesota Department of Health, Minneapolis, Minnesota, USA
5 Department of International Health, Johns Hopkins University Bloomberg School of Public Health, Baltimore, Maryland, USA
6 Monroe County Health Department, Rochester, New York, USA
7 Vanderbilt University School of Medicine, Vanderbilt Medical Center, Nashville, Tennessee, USA
8 School of Public Health, University of California, Berkeley, California, USA
9 Connecticut Emerging Infections Program, Connecticut Department of Public Health, Hartford, Connecticut, USA
10 Oregon Emerging Infections Program, Department of Human Services, Health Services, Office of Disease Prevention and Epidemiology, Portland, Oregon, USA
11 Department of Medicine, University of Utah School of Medicine, Salt Lake City, Utah, USA
Resistance of Streptococcus pneumoniae to antibiotics is increasing throughout the United States, with substantial variation among geographic regions. We show that patterns of geographic variation are best explained by the intensity of selection for resistance, which is reflected by differences between the proportions of resistance within individual serotypes, rather than by differences between the frequencies of particular serotypes. Using a mathematical transmission model, we analyzed temporal trends in the proportions of singly and dually resistant organisms and found that pneumococcal strains resistant to both penicillin and erythromycin are increasing faster than strains singly resistant to either. Using the model, we predict that by 1 July 2004, in the absence of a vaccine, 41% of pneumococci at the Centers for Disease Control and Prevention (CDC)'s Active Bacterial Core surveillance (ABCs) sites, taken together, will be dually resistant, with 5% resistant to penicillin only and 5% to erythromycin only.
Streptococcus pneumoniae is a leading cause of bacteremia, sinusitis, otitis media, bacterial meningitis and bacterial pneumonia worldwide1,
2. An increasing prevalence of S. pneumoniae strains resistant to important anti-pneumococcal drugs, including -lactams and macrolides, has been observed in both developing and developed countries3. In the United States, the proportion of penicillin-resistant S. pneumoniae strains increased from 21% of tested isolates in 1995 to 24% in 1998 (ref. 4), and the proportion of macrolide-resistant strains increased from 10.6% in 1995 to 20.4% in 1999 (refs. 5, 6). Use of antimicrobial agents is highly correlated with the increase in resistance of S. pneumoniae to -lactams and macrolides7,
8,
9,
10. Differential use of antimicrobial agents may be responsible for some of the observed regional and seasonal differences in the proportions of resistance11.
The proportions of resistance to various classes of antimicrobial agents vary substantially both among countries3,
4,
12,
13,
14 and among regions within countries5,
13,
14,
15,
16. The ABCs, a collaboration between the CDC and several health departments and universities participating in the Emerging Infections Program (EIP) network, collects population-based data through active surveillance of invasive pneumococcal infections at 8 sites around the US, covering approximately 21 million people. All of the collected isolates are tested for antimicrobial susceptibility and, since 1998, serotyped17. In the US in 1998, the proportion of penicillin resistance varied among the ABCs sites from 14.7% in New York State to 35.1% in Tennessee. Similar variation exists in the proportions of isolates resistant to erythromycin and in the proportions dually resistant to penicillin and erythromycin (a macrolide). Pneumococci are classified into approximately 90 serotypes, defined by their capsular polysaccharides. Worldwide, resistance to penicillin, erythromycin and, to some extent, other antibiotic classes is largely confined to a limited number of serotypes, most notably serotypes 6A, 6B, 9V, 14, 19A, 19F and 23F4,
13.
There are at least two possible mechanistic explanations for the observed geographic differences in the proportions of resistance. Different geographic sites may have differential selection pressures for resistance (mechanism 1). If such differential selection were operative, it could have two possible effects on the pneumococcal population: areas of higher selection pressure might have greater proportions of isolates from serotypes that tend to be resistant (mechanism 1a); or areas of higher selection pressure might have a greater proportion of resistance within each serotype although they have serotype distributions similar to those in areas of lower selection pressure (mechanism 1b). An alternative mechanism not involving differential selection pressures for resistance is that certain resistant clones of S. pneumoniae might be widely disseminated in some regions but not in others (mechanism 2). Worldwide, a small number of multi-resistant clones have shown remarkable success, not only in competition with drug-sensitive clones, but in competition with other singly and multi-resistant pneumococci18, suggesting that factors besides antibiotic resistance may be partly responsible for the success of these clones.
We sought to distinguish between mechanisms 1 and 2. If mechanism 1b were the cause of the geographic variation, then we would expect that the proportions of resistance should show consistent (highly correlated) patterns across different serotypes; areas with high levels of resistance in one serotype should also have high levels of resistance in most other serotypes10. By contrast, if mechanism 2 were operative, such consistency across serotypes would not be expected. Each of the defined multi-resistant clones would be restricted to one or a small number of serotypes18. Therefore, differential success of one or a few multi-resistant clones would result in elevated proportions of resistance in one or a few serotypes in those areas where the clones spread successfully, but not in geographical correlations in the resistant proportions across a large number of serotypes.
We also tested whether geographic variation in the resistant proportions reflects differences in serotype composition (mechanism 1a), or variation in the resistant proportions within individual serotypes (mechanism 1b). First, to test the importance of geographic differences in serotype distribution, we statistically eliminated these differences by calculating standardized resistant proportions for each site, using the population's average serotype distribution but each site's own serotype-specific resistant proportions. If mechanism 1a alone were operative, this standardization would remove site-to-site variation in the resistant proportions. By contrast, if mechanism 1b alone were operative, the standardized resistant proportions for each site should be the same as the crude resistant proportions for that site, preserving site-to-site variation.
Second, we statistically removed site-to-site variation in the resistant proportions within each serotype and calculated standardized resistant proportions for each site, using the population's average resistant proportions within a serotype, but each site's own serotype distribution. This standardization should have the opposite effect from the previous one, removing differences caused by mechanism 1b but preserving differences caused by mechanism 1a.
Pneumococcal resistance to both -lactams and macrolides is increasing in the US4. To understand the reasons for this increase and to predict future trends, we developed a mathematical model of the transmission of susceptible, singly and dually resistant pneumococci. Under assumptions described in the Methods, the model predicts the shape of the curve describing changes in the prevalence of each of these pneumococcal types over time. We have fitted this model to subsets of ABCs data from 1996 to 1999 and used these fitted curves to predict trends in resistance from 2000 to 2004.
Geographic variation is consistent across serotypes To determine whether geographic differences in the proportions of resistance were replicated across different serotypes, we first calculated the proportions resistant to penicillin, erythromycin or both for each of the eight most common serotypes in counties of eight sites present in the 1998−1999 data set (Table 1). Using these values, the Spearman correlation coefficient for each pair of serotypes was calculated to form a correlation matrix. We found a significantly positive sum of the correlations observed between serotypes by site in the proportions resistant to erythromycin (P < 0.001), penicillin (P < 0.01) and both (P < 0.001). The significant association of site with resistant proportions in different serotypes supports the hypothesis that differences in selection pressure account for geographic variation in the proportions of resistance.
Table 1. Percentage of resistant isolates from serotypes 4, 14, 6A, 6B, 9V, 19A, 19F and 23F, at 8 sites.
Source of geographic variation in resistant proportion When the resistant proportions in each site were standardized to reflect the population's average serotype composition, standardized proportions remained almost identical to crude proportions (Table 2, second column of each section). This pattern was consistent when applied to penicillin resistance, erythromycin resistance or dual resistance. This indicates that geographic differences in serotype composition contribute little to geographic differences in the resistant proportion. On the other hand, standardized proportions reflecting the population's average resistant proportions within each serotype showed almost no geographic variation; in other words, removing geographic differences in resistant proportions within each serotype removed the geographic differences in resistance (Table 2, third column of each section). Again, patterns were similar among penicillin, erythromycin and dual resistance. These patterns were quantified by the regression of crude and standardized resistant proportions (see Methods). Regression slopes for the first standardization were near one, whereas the slopes for the second standardization were near zero and rejected the null hypothesis of no effect (Table 2).
Table 2. Crude and standardized proportions resistant to penicillin, erythromycin and both.
These findings indicate that differential selection pressures in various geographic areas have resulted in geographic variation in the proportions of resistance, and that this variation is due primarily to differences in the resistant proportions in each serotype. Thus, the findings reject the hypothesis that differential success of a few clones explains differences in resistance, and they also reject the hypothesis that differential selection pressure results in significant shifts in pneumococcal populations toward a greater representation of the highly resistant serotypes.
Dynamics of resistance: model predictions To characterize temporal trends in resistance, we fit ABCs resistance data to a multinomial logistic regression based on a model of transmission of four classes of pneumococci: susceptible to penicillin and erythromycin, resistant to penicillin only, resistant to erythromycin only and dually resistant to penicillin and erythromycin (Fig. 1).
Figure 1. Model of pneumococcal transmission (described in Methods).
X, uncolonized host; S, strains susceptible to both penicillin and erythromycin; P, strains resistant to penicillin only; E, strains resistant to erythromycin only; D, strains resistant to both penicillin and erythromycin; , transmission rate constant; u, clearance rate; P, effective rate of treatment with -lactam; E, effective rate of treatment with macrolide.
Dynamic patterns Overall penicillin resistance increased from 21.7% in 1996 to 26.6% in 1999, and erythromycin resistance increased from 10.8% to 20.2%. We hypothesized that strains resistant to both classes of antibiotics would increase in frequency faster than strains resistant to either class alone. To determine the relative rates of increase of dually and singly resistant strains, we fit the data to a multinomial logistic regression versus time (Fig. 2), based on the transmission model described in the Methods. The proportion of strains resistant to penicillin only decreased between 1996 and 1999, with a multinomial coefficient of -0.068/year (95% confidence -0.11 to -0.022), whereas the proportion of strains resistant to erythromycin only increased very slowly, with a coefficient of 0.13/year (95% confidence 0.041 to 0.22). Only dually resistant strains increased substantially between 1996 and 1999, with a multinomial coefficient of 0.25/year (95% confidence 0.2 to 0.29). These results suggest that under current usage patterns of antimicrobial agents, strains resistant to both penicillin and erythromycin have a substantial fitness advantage over singly resistant strains.
Figure 2. Actual and predicted proportions of resistance.
a−e, Points represent averages for time periods from 1 January to 31 December of the next year; curves represent fitted multinomial logistic regressions. Shown are proportions for all counties and all serotypes (a); counties in Georgia (b); counties in Georgia and only serotypes present in the heptavalent conjugate vaccine (c); counties in Minnesota (d); and counties in Minnesota and only serotypes in the heptavalent conjugate vaccine (e). The symbols represent the actual proportion of strains resistant to both erythromycin and penicillin (), penicillin only () and erythromycin only (). The lines represent the predicted proportion of strains resistant to both erythromycin and penicillin (solid), penicillin only (dashed) and erythromycin only (dotted).
Dynamics and predictions We extrapolated the multinomial logistic regression to predict the proportion of penicillin-only, erythromycin-only and dually resistant strains in the years after 1999 (Fig. 2a). The predictions show strains resistant to penicillin only decreasing from 13.7% to 5.3%, strains resistant to erythromycin only increasing from 2.4% to 4.6% and dually resistant strains increasing from 8.6% to 40.6%, from 1 July 1996 to 1 July 2004. These predictions assume that selective pressures (such as use of antimicrobial agents and vaccination status) remain constant from 1999 onward.
Increasing use of the heptavalent pneumococcal conjugate vaccine (PCV), approved in the US in February 2000, will violate our assumption of constant selection pressures. Because the vaccine is expected to reduce the incidence of infection with the serotypes included in the vaccine, and because those serotypes are among the types with the highest proportions of resistance to both -lactams and macrolides, it is likely that the vaccine will reduce the frequency of resistant pneumococcal infections more than that of susceptible ones19. Although vaccine use is expected to reduce the proportion of all infections caused by vaccine-type pneumococci, it should not change the proportion of resistance within serotypes (except perhaps indirectly by reducing antibiotic use).
To test the model prospectively, we require predictions that will hold even if vaccine uptake changes over time. Predictions of trends in resistance for vaccine serotypes alone should not be affected by changes in serotype composition resulting from increased use of the vaccine. Such predictions were possible only for Georgia and Minnesota, the two sites for which serotype information was available for the full period from 1 January 1996 to 31 December 1999 (Fig. 2b−e). For both sites, the proportions of strains resistant to penicillin only and erythromycin only decrease between 1996 and 2004 in the vaccine serotypes and across all serotypes, while the proportions of dually resistant strains increases.
Geographic variation in temporal trends In the static analysis it was evident that there were geographic differences in the proportions resistant to each class of antimicrobials. We sought to determine whether those sites with high proportions of resistance in 1999 were already high in 1996, or whether resistance increased faster in these sites from 1996 to 1999, or both. The proportions of resistance and the rates of increase in resistance (Table 3) show that most regions with high proportions of resistance in 1999 were already high by 1996, but were not those with the highest rates of increase in resistance during the period, especially for single resistance.
Table 3. Changes in resistance over time, by site.
Discussion Resistance in S. pneumoniae has been increasing worldwide, with some areas developing resistance faster than others. Our analyses show that variation in the recent (1998−1999) proportions of strains resistant to -lactams and macrolides, between geographic sites in the US, is best explained by geographic variation in selection pressure for resistance, rather than by clonal dynamics. The effect of these selective differences has been primarily to change the proportions of resistance within individual serotypes, rather than to shift the balance in 'high-resistance' areas in favor of serotypes in which high proportions of isolates are resistant.
These inferences have been made without regionally and temporally specific data on the use of antimicrobial agents. Such data would be valuable in determining the extent to which the geographic differences in the rate of increase of resistance in the US are correlated with antimicrobial use. Nevertheless, our testable predictions, which distinguish between mechanisms for geographic differences in resistance, provide evidence that site-to-site differences in the proportions of resistant strains are largely attributable to regional differences in the selective pressures favoring those strains. Besides antimicrobial agents, other factors that may be causing these differences include day care attendance, prevalence of immunodeficiencies, and other risk factors for resistance20.
We expected that strains of pneumococci resistant to more than one widely used drug would be increasing at a faster rate than singly resistant strains, because they would be unaffected by treatment with a wider variety of antibiotics. This prediction was confirmed in the case of penicillin and erythromycin resistance; dually resistant strains are increasing much faster than strains resistant to either drug alone. Resistance to erythromycin is increasing among both penicillin-susceptible and penicillin-resistant strains of pneumococci. However, the rate of this increase is much faster among penicillin-resistant pneumococci. Resistance to penicillin is increasing only among erythromycin-resistant strains and is decreasing among erythromycin-susceptible strains.
We used a simple model of transmission of susceptible, singly resistant and dually resistant pneumococci21 to predict the shape of the curves describing the changes in frequencies of each of these types over time. We fit ABCs data to a multinomial logistic regression, based on the model. We then extrapolated the proportions of resistance of the different classes out to 2004, using the model, for the pneumococcal population as a whole and for serotypes included in the heptavalent conjugate vaccine. We predict that by 1 July 2004, 41% of pneumococci, in all ABCs sites taken together, will be dually resistant, with 5% resistant to penicillin only and 5% to erythromycin only. For Georgia and Minnesota, two sites for which long-term serotype data were available, we considered resistance only in those serotypes included in the vaccine. By 1 July 2004, we predict that in Georgia, 65% of the vaccine-type pneumococci will be dually resistant (Georgia overall, 55% dually resistant), and that in Minnesota 67% of the strains included in the vaccine will be dually resistant (Minnesota overall, 53% dually resistant). We did not consider the possibility of penicillin- or erythromycin-resistant strains converting to dual resistance because acquisition of resistance by a strain is rare relative to transmission of dually resistant strains22.
The method for generating these predictions has the important strength of being based on a mechanistic (transmission-dynamic) model. However, it also has several limitations. Ideally, one would like to estimate parameters of the underlying transmission model (such as use of antimicrobial agents, duration of colonization with pneumococci and differences in transmissibility between susceptible and resistant strains) independently, and use these estimates to predict changes in the proportions of resistant pneumococci. This was not possible because geographically specific antimicrobial use data were not available, nor were independent estimates of the effects of various types of resistance on pneumococcal fitness (transmissibility)23. Instead, the functional form (multinomial logistic regression with time as a co-variate) was specified by the transmission model, and the parameters for that function were obtained by curve fitting. The following assumptions are required to obtain the predicted functional form. (i) Selective pressure for each type of resistance remains constant over time; selective pressure is determined by use of antimicrobial agents and the fitness cost of resistance. (ii) Different pneumococcal strains are interchangeable; in other words, the only factor influencing a strain's expected prevalence in a given year is its resistance pattern. (iii) The model is designed for carried pneumococci, whereas our data set (as with nearly all large pneumococcal data sets) includes only pneumococci causing invasive disease. Strictly, therefore, the model should only be predictive of disease isolates if the disease strains are randomly sampled from the carriage population. Although these assumptions, like any modeling assumptions, are only approximations of reality, the limited data that are available are consistent with assumption (i). Data from the CDC's National (Hospital) Ambulatory Medical Care Survey show that -lactam and macrolide use remained nearly stable in the US from 1996 to 199924,
25,
26. Data collected over the coming years will tell how strongly departures from the remaining assumptions influence the predictive validity of the model.
An improved understanding of the factors responsible for geographic variations in resistance patterns will help in the design and evaluation of measures to slow the increase in resistance to clinically important antibiotics. Predictions of trends in resistance serve several purposes. They may be used in planning, to estimate the lifespan of existing drugs. They may serve as a baseline for evaluating interventions, such as vaccination or changes in antimicrobial use, that are expected to reduce the rate of increase of resistance and (optimistically) to reduce the absolute proportions of resistance in the pneumococcal population. Predictions that can be tested with prospective data serve as a means of testing the adequacy of our understanding (as formalized in a model) of the reasons for the current rate of increase of resistance. Further progress will require integration of independent estimates of model parameters, especially regionally specific antibiotic use, into the predictions.
Our findings indicate that geographic differences in selection pressures between sites in the United States are responsible for significant differences in the proportions of resistance to two major classes of antibiotics, and that rapid increases in the proportions of resistance (and especially multiple resistance) are expected to continue. We hope that the conjugate vaccine for children will reverse this trend and that better vaccines for adults will soon be available. These findings underline the urgency of limiting the unnecessary use of antimicrobial agents and of finding new agents that will be effective against multi-resistant strains.
Methods Data. We analyzed invasive pneumococcal isolates from the ABCs27 collection, collected by active population-based surveillance at sites in the United States between 1996 and 1999. For each analysis, we included only isolates from counties under surveillance for the entire time frame considered. Intermediate and highly resistant strains (penicillin minimum inhibitory concentration 0.12 g/ml; erythromycin minimum inhibitory concentration 0.5 g/ml)28 were grouped as resistant and considered together.
Statics. To determine whether sites with high proportions of resistance were consistently high across all serotypes, we calculated Spearman rank correlation coefficients for the resistant proportions by site, for each pair of serotypes from the eight most common serotypes in the collection, producing a symmetric 8 8 matrix of correlation coefficients C. Each row and column corresponds to a serotype, and each entry cij represents the correlation by site between the resistant proportions of serotypes i and j. Nonrandom overall correlation was assessed by permutation tests using the sum of C as a test statistic. In this permutation test, the null distribution of the sum of C was approximated by computing test statistics for 1,000 matrices, obtained by permuting the resistant proportions of each serotype randomly by site. This analysis included 7,406 isolates from 1998 and 1999, the years for which all sites had serotype data.
For the same 2-year interval, we analyzed by standardization the effects of serotype composition and resistant proportions within serotypes on geographic differences in the proportions of resistance. Crude resistant proportions were calculated for each geographic site, and 2 different methods were used to calculate the standardized values for each of these sites, using the following formula:
where Ri is the standardized resistant proportion for site i. For the standardization to remove the effects of serotype composition, pj is the population's proportion of isolates that are serotype j, and rij is the resistant proportion among isolates of serotype j at site i. For the standardization to remove the site-specific differences in the resistant proportion of each serotype, pj is the population's resistant proportion among isolates of serotype j, and rij is the proportion of isolates from site i that are serotype j.
We used inverse-variance weighted linear regression to examine the relationship between the standardized and crude resistant proportions for each geographic area. The slope of this regression should be 1 if the standardizing factor has no effect on geographic differences in the proportions of resistance, and 0 if the standardizing factor is completely responsible for the differences. Intermediate regression slopes indicate that some, but not all, of the inter-site variations in resistance are explained by the standardizing factor.
Dynamics. To analyze changes in resistance over time, we generalized a previously published mathematical model of competition between resistant and susceptible strains of S. pneumoniae in a closed host population21. The state variables in the model represent the frequencies of 5 types of hosts: uncolonized (X) or colonized with a strain that is susceptible to both penicillin and erythromycin (S), penicillin-resistant (P), erythromycin-resistant (E) or resistant to both penicillin and erythromycin (D). Parameter subscripts follow the same convention. The model assumes that each individual can only carry one strain at a given time. The following differential equations describe the rates, with respect to time, at which individuals move between states:
i represents the transmission rate constant and ui the clearance rate. This model also considers that an individual can be treated with a -lactam at an 'effective rate' (incidence of treatment with each antibiotic probability that such treatment clears colonization with pneumococci susceptible to that antibiotic) of P, or a macrolide at an effective rate of E
Assuming that X stays approximately constant over time, this model predicts that the changes in the proportion of S (the reference group), P, E and D can be described by a multinomial logistic regression, with the 4 strains as the outcomes and time as a co-variate. We calculated the rates of increase of P, E or D compared with S, using 12,989 isolates, for counties with data from 1 January 1996 to 31 December 1999. Rates of increase described in the text represent the coefficient of time in this regression for each strain relative to S; because of the properties of multinomial regression, the sign of this coefficient need not be the same as the sign of the change in frequency of the corresponding strain. Assuming that the rate of increase will remain constant, the regression predicted the future proportion of resistance for each group and determined whether dually resistant strains were more successful than singly resistant strains. A multinomial logistic regression stratified by site was performed to determine whether sites with the greatest prevalence of resistance in 1996 had rates of increase in resistance that were faster or slower than those sites that had smaller proportions of resistance in 1996.
Received 7 October 2002; Accepted 24 January 2003; Published online: 10 March 2003.
REFERENCES
Hausdorff, W.P., Bryant, J., Paradiso, P.R. & Siber, G.R. Which pneumococcal serogroups cause the most invasive disease: implications for conjugate vaccine formulation and use, part I. Clin. Infect. Dis.30, 100–121 (2000). | Article | PubMed | ISI | ChemPort |
Prevention of pneumococcal disease: recommendations of the Advisory Committee on Immunization Practices (ACIP). MMWR Recomm. Rep.46, 1–24 (1997).
Baquero, F. Pneumococcal resistance to -lactam antibiotics: a global geographic overview. Microb. Drug Resist.1, 115–20 (1995). | PubMed | ISI | ChemPort |
Whitney, C.G. et al. Increasing prevalence of multidrug-resistant Streptococcus pneumoniae in the United States. N. Engl. J. Med.343, 1917–1924 (2000). | Article | PubMed | ISI | ChemPort |
Geographic variation in penicillin resistance in Streptococcus pneumoniae—selected sites, United States, 1997. MMWR Morb. Mortal. Wkly Rep.48, 656–661 (1999).
Garcia-Rey, C., Aguilar, L., Baquero, F., Casal, J. & Dal-Re, R. Importance of local variations in antibiotic consumption and geographical differences of erythromycin and penicillin resistance in Streptococcus pneumoniae. J. Clin. Microbiol.40, 159–164 (2002). | Article | PubMed | ISI | ChemPort |
Diekema, D.J., Brueggemann, A.B. & Doern, G.V. Antimicrobial-drug use and changes in resistance in Streptococcus pneumoniae. Emerg. Infect. Dis.6, 552–556 (2000). | PubMed | ISI | ChemPort |
Doern, G. Antimicrobial use and the emergence of antimicrobial resistance with Streptococcus pneumoniae in the United States. Clin. Infect. Dis.33 (suppl. 3), S187–S192 (2001). | Article | PubMed | ISI | ChemPort |
Jones, M. Relationship between antibiotic resistance in Streptococcus pneumoniae and that in Haemophilus influenzae: evidence for common selective pressure. Antimicrob. Agents Chemother.46, 3106–3107 (2002). | Article | PubMed | ISI | ChemPort |
McEllistrem, M. et al. Distribution of penicillin-nonsusceptible pneumococcal clones in the Baltimore metropolitan area and variables associated with drug resistance. Clin. Infect. Dis.34, 704–707 (2002). | Article | PubMed | ISI |
Scott, J.A. et al. Serotype distribution and prevalence of resistance to benzylpenicillin in three representative populations of Streptococcus pneumoniae isolates from the coast of Kenya. Clin. Infect. Dis.27, 1442–1450 (1998). | PubMed | ISI | ChemPort |
Fenoll, A., Jado, I., Vicioso, D., Perez, A. & Casal, J. Evolution of Streptococcus pneumoniae serotypes and antibiotic resistance in Spain: update (1990 to 1996). J. Clin. Microbiol.36, 3447–3454 (1998). | PubMed | ISI | ChemPort |
Pihlajamaki, M. et al. Macrolide-resistant Streptococcus pneumoniae and use of antimicrobial agents. Clin. Infect. Dis.33, 483–488 (2001). | Article | PubMed | ISI | ChemPort |
Arason, V.A. et al. Do antimicrobials increase the carriage rate of penicillin resistant pneumococci in children? Cross sectional prevalence study. BMJ313, 387–391 (1996). | PubMed | ChemPort |
de Neeling, A.J., Overbeek, B.P., Horrevorts, A.M., Ligtvoet, E.E. & Goettsch, W.G. Antibiotic use and resistance of Streptococcus pneumoniae in the Netherlands during the period 1994-1999. J. Antimicrob. Chemother.48, 441–444 (2001). | Article | PubMed | ISI | ChemPort |
Schuchat, A. et al. Active Bacterial Core surveillance of the Emerging Infections Program Network. Emerg. Infect. Dis.7, 92–99 (2001). | PubMed | ISI | ChemPort |
McGee, L. et al. Nomenclature of major antimicrobial-resistant clones of Streptococcus pneumoniae defined by the Pneumococcal Molecular Epidemiology Network. J. Clin. Microbiol.39, 2565–2571 (2001). | Article | PubMed | ISI | ChemPort |
Black, S. et al. Efficacy, safety and immunogenicity of heptavalent pneumococcal conjugate vaccine in children. Northern California Kaiser Permanente Vaccine Study Center Group. Pediatr. Infect. Dis. J.19, 187–195 (2000). | Article | PubMed | ISI | ChemPort |
Schrag, S.J., Beall, B. & Dowell, S.F. Limiting the spread of resistant pneumococci: biological and epidemiologic evidence for the effectiveness of alternative interventions. Clin. Microbiol. Rev.13, 588–601 (2000). | Article | PubMed | ISI | ChemPort |
Lipsitch, M. The rise and fall of antimicrobial resistance. Trends Microbiol.9, 438–444 (2001). | Article | PubMed | ISI | ChemPort |
Lipsitch, M. Interpreting results from trials of pneumococcal conjugate vaccines: a statistical test for detecting vaccine-induced increases in carriage of nonvaccine serotypes. Am. J. Epidemiol.154, 85–92 (2001). | Article | PubMed | ISI | ChemPort |
National Ambulatory Medical Care Survey: 1993-1999 Summary. (National Center for Health Statistics, Centers for Disease Control and Prevention Hyattsville, Maryland, 2001).
National Hospital Ambulatory Medical Care Survey: 1993-1999 Outpatient Department Summary. (National Center for Health Statistics, Centers for Disease Control and Prevention, Hyattsville, Maryland, 2001).
National Hospital Ambulatory Medical Care Survey: 1993-1999 Emergency Department Summary (National Center for Health Statistics, Centers for Disease Control and Prevention, Hyattsville, Maryland, 2001).
Robinson, K.A. et al. Epidemiology of invasive Streptococcus pneumoniae infections in the United States, 1995-1998: opportunities for prevention in the conjugate vaccine era. JAMA285, 1729–1735 (2001). | Article | PubMed | ISI | ChemPort |
Methods for dilution of antimicrobial susceptibility tests for bacteria that grow aerobically. Approved Standard NCCLS Document M7-A5 (National Committee for Clinical Laboratory Standards, Wayne, Pennsylvania, 2000).
Acknowledgments This work was supported by a National Institutes of Health grant AI48935 to M.L. and by an Ellison Foundation New Scholar Award in Global Infectious Diseases to M.L. We thank S. McCoy and E. Zell for compiling antimicrobial use data and J. Robins for helpful suggestions.
Competing interests statement:
The authors declare that they have no competing financial interests.
-lactams and macrolides, has been observed in both developing and developed countries
, transmission rate constant; u, clearance rate; 
P, effective rate of treatment with 
), penicillin only (
) and erythromycin only (
). The lines represent the predicted proportion of strains resistant to both erythromycin and penicillin (solid), penicillin only (dashed) and erythromycin only (dotted).
0.12 
g/ml; erythromycin minimum inhibitory concentration 
8 matrix of correlation coefficients C. Each row and column corresponds to a serotype, and each entry cij represents the correlation by site between the resistant proportions of serotypes i and j. Nonrandom overall correlation was assessed by permutation tests using the sum of C as a test statistic. In this permutation test, the null distribution of the sum of C was approximated by computing test statistics for 1,000 matrices, obtained by permuting the resistant proportions of each serotype randomly by site. This analysis included 7,406 isolates from 1998 and 1999, the years for which all sites had serotype data.
