Continued Impact of Pneumococcal Conjugate Vaccine on Carriage in Young Children

OBJECTIVES: The goals were to assess serial changes in Streptococcus pneumoniae serotypes and antibiotic resistance in young children and to evaluate whether risk factors for carriage have been altered by heptavalent pneumococcal conjugate vaccine (PCV7). METHODS: Nasopharyngeal specimens and questionnaire/medical record data were obtained from children 3 months to <7 years of age in primary care practices in 16 Massachusetts communities during the winter seasons of 2000–2001 and 2003–2004 and in 8 communities in 2006–2007. Antimicrobial susceptibility testing and serotyping were performed with S pneumoniae isolates. RESULTS: We collected 678, 988, and 972 specimens during the sampling periods in 2000–2001, 2003–2004, and 2006–2007, respectively. Carriage of non-PCV7 serotypes increased from 15% to 19% and 29% (P < .001), with vaccine serotypes decreasing to 3% of carried serotypes in 2006–2007. The relative contribution of several non-PCV7 serotypes, including 19A, 35B, and 23A, increased across sampling periods. By 2007, commonly carried serotypes included 19A (16%), 6A (12%), 15B/C (11%), 35B (9%), and 11A (8%), and high-prevalence serotypes seemed to have greater proportions of penicillin nonsusceptibility. In multivariate models, common predictors of pneumococcal carriage, such as child care attendance, upper respiratory tract infection, and the presence of young siblings, persisted. CONCLUSIONS: The virtual disappearance of vaccine serotypes in S pneumoniae carriage has occurred in young children, with rapid replacement with penicillin-nonsusceptible nonvaccine serotypes, particularly 19A and 35B. Except for the age group at highest risk, previous predictors of carriage, such as child care attendance and the presence of young siblings, have not been changed by the vaccine.

The heptavalent pneumococcal conjugate vaccine (PCV7) has reduced substantially the rates of invasive pneumococcal disease (IPD) in children. [1][2][3][4][5] It also has produced modest decreases in rates of common childhood noninvasive diseases such as otitis media. 6 Through herd immunity, it has reduced rates of IPD in adults and immunocompromised hosts, 1,2,7 including disease among HIV-positive individuals. 8 The ability of this vaccine to maintain reductions in IPD rates has been questioned, however, because nonvaccine serotypes have increased in the absence of vaccine serotypes. [9][10][11] Evidence is mounting that nonvaccine serotypes are responsible for increasing proportions of IPD, 1,[12][13][14][15][16] which raises the question of whether reductions in IPD rates may be attenuated without use of a vaccine targeting broader serotypes. Numerous reports have highlighted the increase in nonvaccine serotype 19A involvement in both local disease and IPD [14][15][16][17] and provided evidence for recombination events between vaccine serotypes and nonvaccine serotypes that can confer an antibiotic-resistant profile to a previously susceptible serotype. [18][19][20][21] Changes in the serotype distribution of IPD are inextricably connected to the serotype distribution of Streptococcus pneumoniae colonization among children. [9][10][11]22 Because colonization rates are highest in young children, children serve as a natural reservoir for invasive strains and represent the environment in which continued evolution of S pneumoniae (including resistance and invasiveness factors) occurs. Therefore, it is both critical and informative to understand the changing population biology of this pathogen under the combined selective pressures of universal immunization and frequent antibiotic use in this age group. [23][24][25] We sampled the nasopharynx of young children in 16 Massachusetts communities in 2001 26 and 2004, 10 during pediatric office visits, and we observed substantial serotype replacement by nonvaccine serotypes and increases in penicillin-nonsusceptible S pneumoniae (PNSP) strains among nonvaccine serotypes in 2004, compared with 2001. 10,22 We showed that the increase in PNSP was driven by both clonal expansion of resistant strains in 15A and 35B serotypes and increasing strain diversity in 19A serotypes. 22 We now report continuing analysis of S pneumoniae carriage in 8 of the Massachusetts communities, to evaluate ongoing changes in serotype replacement and penicillin resistance. We include an assessment of the impact of recent changes in cutoff points for penicillin nonsusceptibility established by the Clinical and Laboratory Standards Institute (CLSI) and adopted by the Food and Drug Administration in March 2008. 27,28 These elevated breakpoints are intended to better reflect clinical relevance but may obscure incremental increases in resistance.
In addition, because colonization often is a precursor to IPD, we were interested in assessing whether known predictors of colonization have changed in association with the introduction of PCV7. In particular, we wanted to assess whether the strength of association with previously known predictors, such as group child care and young siblings, was reduced after PCV7 immunization.

Data Collection
Nasopharyngeal specimens were collected from children 3 months to Ͻ7 years of age in pediatric practices in 8 Massachusetts communities during the winter respiratory virus season, from October 2006 to April 2007. These 8 communities were a subset of 16 geographically distinct areas chosen on the basis of evidence that few children crossed community boundaries for pediatric care. 29 Similarly designed and conducted specimen collections were performed in these 16 communities during the winter respiratory virus seasons of 2001 (March to May) 26 and 2003-2004 (November to April), 10 as part of a cluster-randomized trial of a community-level intervention to promote judicious antibiotic-prescribing for children. 29 Eight practices participated in 2007. These 8 practices were among the largest of the 23 participating practices in 2004 and the 31 practices in 2001.
Parents of participating children completed a brief survey (which was used previously) on possible predictors of pneumococcal carriage, with questions on group child care, young siblings, recent antibiotic use, and concurrent illness. 26 Parents of participating children also consented to subsequent medical record review for information on vaccines received, recent antibiotic use, and symptoms and diagnoses at the time of the visit. Parental consent was obtained by study staff members, and nasopharyngeal swabs were obtained by trained nurses during well-child or sick visits. All study procedures were approved by the Harvard Pilgrim Health Care institutional review board.

Microbiologic Processing
Nasopharyngeal samples were processed within 24 hours. All pneumococcal isolates underwent antibiotic susceptibility testing and serotyping (using the Quellung reaction) as described previously. 10,26 CLSI susceptibility breakpoints were used to classify organisms as susceptible, intermediate, or resistant to the following antibiotics: penicillin, amoxicillin, ceftriaxone, erythromycin, clindamycin, trimethoprimsulfamethoxazole, levofloxacin, and vancomycin. 27,28

Data Analysis
To assess whether analyses should be restricted to the 8 communities that participated in all 3 sampling periods or whether data from all 16 communities could be used in the evaluation of trends, we performed 2 analyses. First, we assessed whether the populations of serotypes were significantly distinct among communities within the same sampling period. We did this by using Monte Carlo hypothesis testing, 30 repeatedly reassigning serotypes randomly to communities within a sampling period, and assessing where the observed serotype distribution lay among the many random assessments. Specific permutations of serotypes were compared by using a classification index described elsewhere. 31 Second, we used a second Monte Carlo test to assess whether communities retained a unique mixture of serotypes over time. The intracommunity distribution of serotypes between sampling periods was compared by using the classification index. This was subsequently compared with observed distributions when random reassignments of communities as pairs were made. The first assessment showed no evidence that communities differed in profiles from the grouped communities. The second assessment detected no evidence that single communities were stable over time. Together, these tests support the use of all communities available in each of the 3 sampling periods (see Appendix). In addition, previous work with prescribing data from several large health insurers 29 showed identical antibiotic usage rates (1.5 antibiotics per person-year) among children 2 to 4 years of age when communities included in 2007 were compared with those not included.
We calculated the proportions of children carrying any S pneumoniae isolate and non-PCV7 serotypes during the 3 sampling periods. In contrast to the 2007 sample, previous sampling periods included infants Ͻ3 months of age. Therefore, we excluded all infants Ͻ3 months of age, for comparability across periods. Age groups assessed were 3 to Ͻ6 months, 6 to Ͻ24 months, 24 to Ͻ36 months, and 36 to Ͻ84 months. 10,26 We also calculated the proportions of children carrying S pneumoniae isolates that were not susceptible to antibiotics. Previous susceptibility breakpoints for penicillin (minimal inhibitory concentration [MIC] of Յ0.06 mg/L) 28 were compared with new 2008 breakpoints (MIC of Յ2.0 mg/L) established by the CLSI. 27 Two definitions of multidrug resistance were used, that is, resistance to the ␤-lactam class plus 1 other antibiotic class (multidrug-resistant group 1) and resistance to Ն3 antibiotic classes (multidrug-resistant group 2). 10 Penicillins, cephalosporins, and carbapenems were considered a single class.
We further evaluated whether the prevalence of serotype-specific carriage in 2007 was associated with the proportion of serotype-specific penicillin nonsusceptibility. To do this, we created a scatterplot and fit a logistic regression curve, reporting the odds ratio (OR) for the association between antibiotic resistance and serotype prevalence.
Changes in specific serotypes were calculated on the basis of their respective proportions in previously reported sampling periods. 10 Fisher's exact tests were used to assess differences among the sampling periods. All analyses were performed with SAS 9.1 (SAS Institute, Cary, NC).
We calculated the proportions of children with characteristics (determined on the basis of survey or chart review) associated with pneumococcal carriage in previous sampling periods, 26 including age, group child care, young siblings, recent antibiotic use, and acute respiratory tract infection (RTI). Acute RTI was defined as confirmation of symptoms on questionnaires or any medical record diagnosis of any upper or lower RTI, including otitis media, sinusitis, bronchitis, pharyngitis, cough illness, and pneumonia. These characteristics were entered into generalized linear mixed models predicting carriage of S pneumoniae, PNSP, and non-PCV7 serotypes. From models restricted to individual sampling periods, we calculated not only the ORs of various covariates for the specified outcomes but also the probabilities of carriage associated with certain combinations of covariates. In a second set of models predicting the same outcomes, we included data from all sampling periods and evaluated sampling year as a covariate. In these models, based upon effect modification sampling year was assessed for all significant variables. All models accounted for clustering of data within communities by using general linear mixed models. 32  (Table 3). Age, RTI, and other potential confounders were adjusted for in the final model (described below).

Changes in Predictors of Carriage
Models assessing predictors of S pneumoniae carriage showed that group child care was a strong predictor of S pneumoniae, PNSP, and non-PCV7 carriage in 2007, as in previous studies. However, age and RTI had different effects across the sampling periods, as shown by the significance of interaction terms between year and these covariates (Table 5). When these effects were assessed in stratified models for each sampling period, older children seemed relatively less likely to carry S pneumoniae, compared with younger children, after the introduction of PCV7 (Table 5). This effect was attributable to the increase in carriage among the youngest age group (Ͻ6 months). RTI remained predictive of S pneumoniae carriage in all sampling periods, but this association was somewhat decreased in later time periods because of a lack of effect on PNSP carriage.

FIGURE 1
Proportions of children carrying non-PCV7 pneumococcal serotypes within each sampling period, according to age. P values of Յ.05 based on 2-tailed Fisher's exact tests are indicated.

FIGURE 2
Distribution of non-PCV7 pneumococcal serotypes, as proportions of total serotypes, according to sampling period. P values of Յ.1 based on 2-tailed Fisher's exact tests evaluating differences in serotype-specific proportional carriage are indicated.
In addition to the relative increase in risk (ORs) indicated in Table 5, we provided calculated estimates of carriage prevalence for children with specific risk factors, on the basis of model-based predicted probabilities of non-PCV7 carriage (

DISCUSSION
Some replacement in pneumococcal carriage with nonvaccine serotypes has been reported since the release of PCV7. [9][10][11]22 This was associated with initial sharp decreases in rates of IPD attributable to vaccine serotypes, [1][2][3][4][5]7,8 followed by increases in rates of IPD attributable to nonvaccine strains. 1,[12][13][14][15][16][17] Some of these effects on colonization and IPD rates were demonstrated in Massachusetts. 10,22,34,35 Through the continued sampling of young children in Massachusetts communities, we now show that nonvaccine serotypes have virtually completely replaced colonizing vaccine serotypes within the 7 years since the release of PCV7. Replacement was correlated temporally with vaccine penetration and was rapid and relatively complete, compared with expectations based on incomplete effects on colonization during vaccine trials. [36][37][38] These results suggest that the risk of pneumococcal disease, although clearly lower than in the prevaccine era, may continue  to change in the absence of immunization against noncovered serotypes.
As vaccine serotypes disappear, nonvaccine serotypes have been increasing rapidly in all age groups. Common nonvaccine serotypes found in 2004 have become even more prevalent, with the proportions of infrequent serotypes decreasing. We report significant increases in 19A, 35B, 23A, and 7F prevalence, as well as trends toward increased prevalence of 23B and 16F. Although 19A has already been shown to be responsible for an increasing proportion of IPD, it remains to be seen whether these other serotypes will have similar invasive potential.
We suggest that the selective advantage of penicillin nonsusceptibility may explain in part the increase in specific serotypes. High rates of antibiotic use for common conditions such as acute otitis media may provide pressure in favor of resistant strains among colonized children. [39][40][41] Previous genetic typing work suggested that clonal expansion of serotypes 19A, 35B, and 15A has favored strains that are resistant to penicillin, such as sequence type 199 among 19A serotypes, sequence type 558 among serogroup 35B isolates, and sequence type 63 among 15A serotypes. 22,42 The revised national breakpoints for penicillin susceptibility are intended to reflect more closely the likelihood of clinical failure in response to antibiotic therapy. However, attention to the expansion of clones associated with resistance genes and incremental increases in MICs for common antibiotics can warn of impending loss of therapeutic efficacy before detection with these higher laboratory thresholds. Continued evaluation of the changing epidemiological features of pneumococcal serotypes, strain types, and phenotypic resistance is important during a time when the population dynamics of S pneumoniae are changing rapidly and invasive potential remains to be seen.
Despite the nearly complete replacement of vaccine serotypes by nonvaccine serotypes, we found that predictors of carriage remained quite similar between 2001 and 2007, with the exception of age and RTI. This change in age as a risk factor came as a surprise and seemed to be associated with higher colonization rates in young infants. Because we had no previous hypothesis in this direction, additional work will be required to determine whether this apparent trend is real. In later sampling periods, we found that RTI was less strongly associated with S pneumoniae carriage, a finding that seemed to be mediated by a loss of  Three generalized linear mixed models were calculated by using data limited to each sampling period. Participants missing covariate values were excluded from the models. Because of rounding, some estimates with P Ͼ .05 show 95% CIs that include 1.0. a P values were derived from a separate multivariate model in which data from all 3 sampling periods were included. The P values evaluating the interactions between year and each covariate are provided. b ORs with P Յ .05. c PNSP is based upon the nonsusceptibility breakpoint used before 2008 (MIC of Ͼ0.06 mg/L). effect on PNSP carriage. Whether this is related to the changing serotype representation of PNSP isolates is not known. Child care persisted as the strongest predictor of pneumococcal carriage, with young siblings and RTI continuing to confer added independent risk. Together, these risk factors conferred a 2.5-to 3-fold increased risk of carriage in all age groups examined.
The persistence of pneumococcal carriage in 30% of young children through the post-PCV7 era and the continued influence of these common characteristics on carriage suggest that reductions in IPD rates will be influenced by the capacity of specific serotypes to produce IPD when carried, rather than by a vaccine-induced reduction in overall carriage rates. Knowledge of the dynamic changes in carried serotypes will be invaluable for understanding invasive potential. For example, we observed minimal evidence of serotype 3 colonization (Ͻ1%) in this large multicommunity study of young children.
Recent reports of increased rates of serotype 3 IPD in some areas of the United States might be better under-stood in the context of whether increases in serotype 3 carriage prevalence occurred previously. 43,44 There are several limitations to this study. We sampled serially during winter respiratory virus seasons to assess changes in the S pneumoniae reservoir during peak times of annual carriage and transmission risk. We cannot know whether changes in carriage or the distributions of S pneumoniae serotypes are similar throughout the year or whether we adjusted completely for differences in respiratory illness rates between sampling periods. In addition, sampling in physician offices may bias our data toward serotypes that are cultured more easily during respiratory illnesses. The frequency of well-child visits and our adjustment for respiratory illness should have mitigated these effects, and the inclusion of children with respiratory illnesses allows assessment of pneumococcal carriage related to a common illness state in this age group. Finally, we did not examine the possibility that children carried multiple strains simultaneously. If this was common, this might have altered per-ceptions of antibiotic resistance and the relative proportions of nonvaccine serotypes.

CONCLUSIONS
We report the virtual disappearance of vaccine serotypes among young children in multiple Massachusetts communities. Under the abrupt and powerful selective pressure of immunization against select pneumococcal serotypes, it was plausible that overall pneumococcal colonization would decrease substantially. Instead, replacement with nonvaccine serotypes has enabled S pneumoniae prevalence to remain at ϳ30%, with persisting risk factors for pneumococcal carriage, such as child care attendance. Furthermore, increases were seen among serotypes with elevated penicillin MICs, such as 19A and 35B. This work and previous work continue to suggest that serotype-specific increases may be favored by penicillin resistance. 22 Given antibiotic selection pressures and the demonstrated ability of S pneumoniae to acquire genetic factors that promote both resistance and invasion, we should be prepared for continuing changes in the incidence of local disease and IPD, as well as the responsible serotypes. Understanding the population biology underlying the "moving target" of pneumococcal colonization should help us understand the long-term impact of pneumococcal vaccines, and potentially other vaccines, that target only a subset of strains within a species.

APPENDIX: CUSTOM MONTE CARLO TESTS TO ASSESS 8 VERSUS 16 COMMUNITIES
The question we addressed with these tests was whether we should use all of the communities we had at each time point or whether we should use the 8 communities observed at all 3 time points (or use something more complicated). We operationalized this by assessing whether the assortment of communities from which we sampled could be treated as a population or whether they were sufficiently distinct that each represented a unique subpopulation. We divided the question into 2 tests, namely, whether there were significant differences between communities at each time point and whether there was consistency within communities across time points. Stable communities, all distinct, might result in apparently mixed populations at each time point.
The tests both depend on a classification index. 31 This is constructed as where ij is the frequency of serotype i in population j and is the average frequency of across the 2 populations. When the 2 populations have identical proportions of each serotype, the index is 0; as the populations become more dissimilar, the index increases toward 1.
In the first test, we calculate the pairwise classification index between each of the pairs of communities within each season. Then we take their average; this is the observed statistic. We perform a Monte Carlo process, in which we permute the cases by reassigning them randomly to communities. This is a random reordering of the community membership. Effectively, each individual can be assigned to any community with a probability equal to the proportion of individuals actually observed in that community. We then calculate the average classification in- In the second test, we perform a similar Monte Carlo test but, in this case, we calculate the classification index between consecutive time points for each community. The average of these is the observed statistic in this case. The Monte Carlo process for this step involves not reassigning individuals randomly but pairing communities randomly across time. In other words, community A at time 1 may be paired with any of the communities at time 2. The average classification index is calculated. This process is repeated. The Monte Carlo P value (constructed as described above) assesses the null hypothesis that there is no continuity within a community, consistent with mixing across a broader population. The There was no evidence against the null hypothesis of no community stability.