Herd immunity and pneumococcal conjugate vaccine: A quantitative model

Vaccine 25 (2007) 5390–5398

Herd immunity and pneumococcal conjugate vaccine: A quantitative model Michael Haber a,∗ , Albert Barskey b , Wendy Baughman c , Lawrence Barker d , Cynthia G. Whitney e , Kate M. Shaw d , Walter Orenstein f , David S. Stephens c,f a

f

Department of Biostatistics, Rollins School of Public Health, Emory University, Atlanta, GA 30322, USA b Department of Epidemiology, Rollins School of Public Health, Emory University, Atlanta, GA, USA c Georgia Emerging Infections Program and Research Service, VAMC (Atlanta), GA, USA d National Immunization Program, Centers for Disease Control and Prevention, Atlanta, GA, USA e National Center for Infectious Diseases, Centers for Disease Control and Prevention, Atlanta, GA, USA Division of Infectious Diseases, Department of Medicine, Emory University School of Medicine, Atlanta, GA, USA Received 7 January 2007; received in revised form 25 April 2007; accepted 27 April 2007 Available online 22 May 2007

Abstract Invasive pneumococcal disease in older children and adults declined markedly after introduction in 2000 of the pneumococcal conjugate vaccine for young children. An empirical quantitative model was developed to estimate the herd (indirect) effects on the incidence of invasive disease among persons ≥5 years of age induced by vaccination of young children with 1, 2, or ≥3 doses of the pneumococcal conjugate vaccine, Prevnar® (PCV7), containing serotypes 4, 6B, 9V, 14, 18C, 19F and 23F. From 1994 to 2003, cases of invasive pneumococcal disease were prospectively identified in Georgia Health District-3 (eight metropolitan Atlanta counties) by Active Bacterial Core surveillance (ABCs). From 2000 to 2003, vaccine coverage levels of PCV7 for children aged 19–35 months in Fulton and DeKalb counties (of Atlanta) were estimated from the National Immunization Survey (NIS). Based on incidence data and the estimated average number of doses received by 15 months of age, a Poisson regression model was fit, describing the trend in invasive pneumococcal disease in groups not targeted for vaccination (i.e., adults and older children) before and after the introduction of PCV7. Highly significant declines in all the serotypes contained in PCV7 in all unvaccinated populations (5–19, 20–39, 40–64, and >64 years) from 2000 to 2003 were found under the model. No significant change in incidence was seen from 1994 to 1999, indicating rates were stable prior to vaccine introduction. Among unvaccinated persons 5+ years of age, the modeled incidence of disease caused by PCV7 serotypes as a group dropped 38.4%, 62.0%, and 76.6% for 1, 2, and 3 doses, respectively, received on average by the population of children by the time they are 15 months of age. Incidence of serotypes 14 and 23F had consistent significant declines in all unvaccinated age groups. In contrast, the herd immunity effects on vaccine-related serotype 6A incidence were inconsistent. Increasing trends of non-vaccine serotypes, in particular 19A, were noted in most unvaccinated age groups, but these increases were substantially smaller than the concurrent decreases among the vaccine serotypes. Also, the model estimated PCV7 to have a greater (p = 0.014) indirect impact on the incidence of invasive pneumococcal disease caused by all vaccine serotypes among African-Americans of all ages than for whites. Thus, conjugate vaccines may be able to induce herd effects even in situations where vaccine coverage is far from complete or with schedules using fewer than 3 or 4 doses. Because the model was based on incidence rates and PCV7 coverage in Atlanta, our findings should be validated in other geographic areas. © 2007 Elsevier Ltd. All rights reserved. Keywords: Pneumococci; Conjugate vaccine; Herd immunity

1. Introduction

∗

Corresponding author. Tel.: +1 404 727 7698; fax: +1 404 727 1370. E-mail address: [email protected] (M. Haber).

0264-410X/$ – see front matter © 2007 Elsevier Ltd. All rights reserved. doi:10.1016/j.vaccine.2007.04.088

Streptococcus pneumoniae is a major cause of human disease ranging from mild syndromes, like otitis media, to severe invasive diseases, like meningitis. Invasive pneumococcal

M. Haber et al. / Vaccine 25 (2007) 5390–5398

disease (IPD) includes bacteremia, bacteremic pneumonia and empyema, meningitis and arthritis, and is most prevalent among children, the elderly, and those who are immunocompromised [1]. For children less than 5 years of age in the United States, S. pneumoniae caused approximately 17,000 cases of IPD annually before the introduction of the pneumococcal conjugate vaccine in 2000 [2]. Minority groups have shared a disproportionate burden of invasive pneumococcal disease. African-American children have had an incidence of 400/100,000, Alaskan Natives an incidence of 625/100,000, and Native Americans an incidence of 2400/100,000 [2]. In developing countries the incidence also is high; 500 cases per 100,000 children is estimated for children <12 months in the Gambia [3]. While the highest incidence rates for pneumococcal disease occur among young children, a significant burden of disease and mortality occurs in adults, particularly the elderly. The pneumococcus is an asymptomatic colonizer of the human nasopharynx or upper respiratory tract. In practically all cases of IPD, the pneumococcus causing disease is present in the upper respiratory tract or recently acquired from an asymptomatic carrier [4]. Ninety disease-associated serotypes exist for S. pneumoniae. However, most disease, especially in children in the United States, is caused by relatively few serotypes (4, 14, 6B, 9V, 18C, 19F and 23F). Vaccination of young children with a pneumococcal conjugate vaccine, Prevnar® or PCV7, markedly decreases the risk of pneumococcal disease and antibiotic-resistant pneumococcal disease in older children and adults by decreasing exposure and acquisition of the pneumococcus (i.e. herd immunity) [5,6]. Fig. 1 presents the yearly incidence rates of IPD associated with PCV7 serotypes and with non-vaccine serotypes among persons aged 5 years and above in the Atlanta metropolitan area in the years 1994–2003. The figure also shows the mean number of doses of PCV7 Atlanta children had received by 15 months of age. As PCV7 coverage increased, there was a significant decrease in the incidence of IPD associated with the seven serotypes included

Fig. 1. PCV7 coverage by 15 months of age and incidence rates of invasive pneumococcal disease among persons aged 5 years and above in the Atlanta area, 1994–2003.

5391

in the vaccine, with little or no decrease in IPD caused by non-PCV7 serotypes. The decline in disease related to the vaccine serotypes in unvaccinated individuals represented herd immunity effects of PCV7 vaccination [5,6]. Development of valid models is important to estimation of vaccine effectiveness, cost benefit, and the best methods of introducing PCV7 into new communities or populations. In both developing and in industrialized countries, decisions about the introduction of PCV7 may be significantly influenced by the indirect herd immunity effect on the unimmunized populations [7,8]. Further, the average number of doses each child should receive to induce optimal herd immunity substantially influences cost-effectiveness calculations. We developed a Poisson regression model to estimate the impact of receipt of a specified number of doses of PCV7 by 15 months of age on the incidence of IPD among unvaccinated children and adults ≥5 years of age. The model was derived from data generated in 9 years of active, population-based surveillance for invasive pneumococcal disease in metropolitan Atlanta as well as vaccination coverage rates among children in Atlanta.

2. Methods 2.1. Pneumococcal disease case surveillance Since 1994, cases of invasive pneumococcal disease have been prospectively identified in Georgia Health District 3 (HD-3) (comprised of the following eight metropolitan Atlanta counties: Clayton, Cobb, DeKalb, Douglas, Fulton, Gwinnett, Newton, and Rockdale) by active laboratory-based surveillance and defined as the isolation of S. pneumoniae from normally sterile sites. This region had a 2000 census population of 3.1 million, which was 62.4% white, 34.5% African-American, and 3.1% other races. Surveillance, collection and validation methods have been described elsewhere [9–11]. Cases were identified and pneumococcal isolates were collected from all laboratories and hospitals in HD-3. Demographic data about the patients from whom isolates were obtained were also collected. To ascertain the sensitivity of the surveillance system in identifying cases, laboratory audits were conducted at a minimum of every 6 months to ensure accurate reporting. Isolates were confirmed as S. pneumoniae based on susceptibility to ethylhydrocupreine (optochin) and bile solubility [12]. Serotyping of isolates was also conducted using the quellung reaction with type-specific anti-serum [12]. Of the 7143 cases identified from 1994 to 2003, 84.4% had isolates available for further testing [5]. Incidence rates were determined by year, age group (<2-year-olds, 2–4-year-olds, 5–19-year-olds, 20–39-yearolds, 40–64-year-olds, and 65+ year-olds) and serotype (all serotypes, PCV7 serotypes (4, 14, 6B, 9V, 18C, 19F, 23F), serotypes 6A and 19A, and all non-PCV7 serotypes). Incidence rates were also determined by race and county.

5392

M. Haber et al. / Vaccine 25 (2007) 5390–5398

Population data were based on the U.S. census reports. Since the goal was to model herd immunity effects, the incidence rates for the <2-year-old and 2–4-year-old age groups were not included in the present study because many of the children in these age groups were vaccinated, and therefore, the decline in incidence would be reflective of both direct and indirect effects of the vaccine. 2.2. PCV7 vaccine coverage surveillance The National Immunization Survey (NIS) is a random digit dialed survey used to estimate vaccine coverage rates for non-institutionalized children living in the U.S. and aged 19–35 months at the time of the survey [13,14]. The NIS collects information from each of the 50 states and 28 selected urban areas. Fulton and DeKalb counties, included in Atlanta HD-3, are one of the 28 urban areas. In the NIS telephone interview, parents are asked to give the child’s vaccination history by reading the vaccine type and date from the immunization card. In the event that the immunization card is not available, respondents are asked to recall vaccination history from memory. Permission to contact the child’s vaccine providers is requested. If verbal consent is obtained, vaccine providers are contacted by mail and asked to record the child’s immunization history and return the form [13]. Demographic and socioeconomic information are also collected from the family. The pneumococcal conjugate vaccine (PCV7) was licensed in February 2000 and administered in Atlanta beginning in early to mid 2000. We used the data from Fulton and DeKalb counties to assess PCV7 introduction in HD-3. The PCV7 coverage data from the NIS used for the analysis were the percentages of children aged 19–35 months residing in Fulton and DeKalb counties who received 0, 1, 2, 3 or 4 PCV7 doses based on the 2001–2004 surveys. For this analysis, PCV7 coverage throughout HD-3 was assumed to be the same as what was found in Fulton and DeKalb counties. Because of shortages of the vaccine, few children received all four recommended doses. Thus, children who received >3 doses were grouped together. Coverage data were also available in race subcategories (white and African-American). The NIS determines immunization status at the time the family of a 19–35-month-old child is interviewed, regardless of when immunizations were received. Most vaccinations are delivered in the first 2 years of life (in the 2004 NIS, 92% of doses administered to 19–35-month-old children from Fulton and Dekalb counties were delivered prior to 16 months of age). Using the data from the NIS survey conducted in a given year, we determined the PCV7 coverage among these 19–35-month-old children in the year prior to the survey. These data provided an estimate of the coverage among 15month-old children in the previous year, since the median age of 19–35-month-old children in the year prior to the survey would be 15 months. For example, in the 2002 survey the proportions of 19–35-month-old children who received

Table 1 Estimated number of pneumococcal conjugate vaccine (PCV7) doses received by children with a median age of 15 months, Fulton and Dekalb counties, 2000–2003a Year

2000 2001 2002 2003

Doses 0

1

2

≥3

Mean number of dosesb

76.9% 36.8 15.1 15.1

9.8% 11.8 12.0 12.4

7.7% 13.3 18.9 22.4

5.6% 38.1 54.0 50.1

0.42 1.53 2.12 2.08

a Information derived from the National Immunization Surveys of 19–35months-old children in 2001–2004. Coverage at 15 months inferred assuming the vast majority of doses are delivered before this age. b In calculating the mean, children who received 3 or more doses were considered as having received 3 doses.

0, 1, 2 and ≥3 doses of PCV7 no later than 1 year prior to the date of the survey were 0.37, 0.12, 0.13 and 0.38, respectively. These proportions were used as the vaccine coverage data among 15-month-old children in 2001 (see Table 1). Finally, we determined the mean number of doses given to a 15-month-old child each year from the proportions of children who received 0, 1, 2 or ≥3 doses. In the above example, the mean number of doses of PCV7 was 0 × 0.37 + 1 × 0.12 + 2 × 0.13 + 3 × 0.38 = 1.53. 2.3. Statistical analysis Poisson regression models were applied to the data described in Sections 2.1 and 2.2 to investigate the herd immunity relationship between the incidence rates of invasive pneumococcal disease in unvaccinated age groups and PCV7 coverage among 15-month-old children as measured by the mean number of doses per child. Separate models were derived for each of the four unvaccinated age groups (5–19, 20–39, 40–64 and 65+ years). For each of these age groups, the models related the natural logarithm of the yearly incidence to the mean number of doses: ln(c/N) = α + βM, where c and N are the number of cases and the population size, respectively, for that age group, and M is the mean number of doses per 15-month-old child. SAS software [15] was used to estimate the coefficients α and β and their standard errors. For example, for the 65+ years age group, the model’s estimated coefficients were α = −7.2737, and β = −0.2369. Therefore, if every 15-month-old child was given two doses (M = 2), the natural logarithm of the expected incidence rate among persons 65+ years of age was: ln(c/N) = −7.2737 − 0.2369 × 2 = −7.7475. Exponentiating, we got e−7.7475 = 0.0004318, or 43.18 cases per 100,000 people. The herd immunity effectiveness of vaccination with a mean coverage of x doses per 15-month-old child was calculated as one minus the ratio of the expected incidence rate using x doses and the expected incidence rate using 0 doses. In the example mentioned above, the expected incidence

M. Haber et al. / Vaccine 25 (2007) 5390–5398

rates for 0 and 2 doses were 69.35 and 43.18 per 100,000, respectively. Therefore, the corresponding herd immunity effectiveness was 1 − (43.18/69.35) = 0.377. In other words, vaccination of young children with 2 doses of PCV7 was expected to prevent 37.7% of IPD cases in the 65+ years age group. 2.4. Per-dose herd immunity effectiveness To obtain a standardized measure of the herd immunity effect that can be compared across serotypes or age groups, we defined the per-dose herd immunity effectiveness of vaccination with PCV7 as the relative decrease in the incidence of IPD in unvaccinated persons resulting from an increase of one dose in the average coverage of 15-month-old children: per-dose herd immunity effectiveness =1−

expected incidence with x + 1 doses expected incidence with x doses

Under the Poisson regression model, the preventive effectiveness due to an increase of 1 dose is the same when the average number of doses increases from 0 to 1, from 1 to 2, or from 2 to 3. The per-dose herd immunity effectiveness defined above is the relative reduction in incidence due to an increase of a single dose. For an increase of y doses, the overall herd immunity effectiveness, i.e., the overall relative reduction in incidence, can be calculated as: overall herd immunity effectiveness = 1 − (1 − per-dose herd immunity effectiveness)y . For example, if the per-dose herd immunity effectiveness is 0.8, i.e. each additional dose decreases the estimated incidence by 20%, then increasing the number of doses by 2 is expected to decrease the incidence by 1 − 0.82 = 0.36, or 36%. The per-dose herd immunity effectiveness is statistically significant if the estimate of the parameter β in the Poisson regression model is statistically different from zero. If β = 0 then the incidence does not depend on the number of vaccine doses, i.e., vaccination of young children does not affect the incidence rates of older persons. 2.5. Assessing the fit of the model To examine the goodness of fit of the models to the data, the above equation was rearranged: ln(c) = α + βM + ln(N). This enabled us to estimate the expected number of cases in a given year by substituting the actual value of M in the above equation, and comparing this expected frequency to the observed one. In the above example, the mean number of doses in 2002 was M = 2.12, and the population (N) in the 65+ year age group in 2002 was 240,116. Therefore, the natural log of the expected number of cases was: ln(c) = −7.2737 − 0.2369 × 2.12 + ln(240116) = 4.6129 and the expected number of cases in 2002 among 65+ year-old persons was e4.6129 = 100.78 cases. The observed number of cases for that age group in 2002 was 95,

5393

so in this example the fit of the model was very good. To further assess the model’s fit, the value ((observed cases − expected cases)2 /expected cases) was calculated for each year and age group. Using the numbers from the above example, we got (95 − 100.78)2 /100.78 = 0.331. When the model fits the data then the distribution of each of the ratios (observed − expected)2 /expected is approximately ␹squared with one degree of freedom. Hence, if the model fits well then about 95% of these ratios are expected to be less than 4.0 [16]. We used these Poisson regression models to investigate the association of the decline of incidence rates of IPD caused by individual serotypes and by serotype groups with the mean dose coverage. In addition, we measured the perdose herd immunity effectiveness by age group and serotype as described earlier. We also used these models to explore the relationship between incidence of IPD resulting from all serotypes combined and race by fitting separate models to the observed incidence rates in blacks and whites.

3. Results 3.1. PCV7 coverage data The proportion of 15-month-old children who received 0, 1, 2 or ≥3 doses of PCV7 in Georgia HD-3 for 2000–2003 and the calculated mean number of PCV7 doses per child appear in Table 1. The proportions of children receiving at least 1 dose were 23%, 63%, 85% and 85% for the years 2000, 2001, 2002 and 2003, respectively. 3.2. Model fit Table 2 shows the observed and expected frequencies of IPD (all serotypes combined) for each age group for the prePCV7 years 1994–1999 (combined) and the PCV7-use years 2000, 2001, 2003 and 2004. Since only one of the 20 values in Table 2 was greater than 4 when we assessed the values of (observed cases − expected cases)2 /expected cases for each combination of age group by year, we concluded that the expected cases based on the Poisson regression model fit well with the observed cases. 3.3. Estimation of incidence For each of the four unvaccinated age groups (5–19years old, 20–39-years old, 40–64-years old, 65+ years old), Fig. 2a–d shows the incidence estimated by our model as the average number of doses of PCV7 received by the time children are 15 months old varied from 0 to 3. The curves in Fig. 2a–d correspond to the estimated incidence for all serotypes, for the PCV7 serotypes, and for all other serotypes, respectively. For all serotypes, the model estimated a significant decrease in incidence for each age group (p = 0.046 in 5–19-year-olds, and p < 0.001 in the other three age groups)

5394

M. Haber et al. / Vaccine 25 (2007) 5390–5398

Table 2 Expected and observed cases for all serotypes combined Year

Age groups (years) 5–19

1994–1999 2000 2001 2002 2003 *

20–39

40–64

65+

Obs*

Exp*

D*

Obs

Exp

D

Obs

Exp

D

Obs

Exp

D

31 16 24 21 16

24 26 21 18 19

2.17 3.86 0.48 0.34 0.43

158 135 104 76 81

152 145 98 79 80

0.26 0.67 0.34 0.12 0.01

197 207 205 158 180

198 213 184 172 180

0.01 0.14 2.34 1.12 0.00

153 126 136 95 97

145 144 114 101 104

0.48 2.24 4.33 0.33 0.44

Obs = observed number of cases, Exp = expected number of cases, D = (Obs − Exp)2 /Exp.

when the average number of doses received by 15-montholds increased. Similarly, the incidence of PCV7 serotypes was estimated to significantly decline in each of the four unvaccinated age groups (p = 0.0025 in 5–19-year-olds, and p < 0.001 in 20–39, 40–64 and 65+ year-olds). For age group 40–64-years (Fig. 2c), the incidence of disease caused by all other (non-PCV7) serotypes was estimated to increase as the average number of doses of vaccine given to 15-month-old children increased, but the estimated increases in this group did not achieve statistical significance (p = 0.53). Interestingly, for age groups 5–19 (Fig. 2a), 20–39 (Fig. 2b), and 65+ (Fig. 2d), the model estimated a lower overall incidence of other (non-PCV7) serotypes with an increasing average number of doses, and this decrease in incidence was significant for the 20–39-year-old age group (p = 0.004) and non-significant for the 5–19 and 65+ age groups (p = 0.80 and 0.28, respectively). For all the age groups the changes in the non-PCV7 serotypes associated with an increase in the mean number of doses were small in absolute terms in contrast to the marked reductions seen for all serotypes combined and for the PCV7 serotypes.

3.4. Per-dose herd immunity effectiveness of PCV7 Fig. 3 shows the per-dose herd immunity effectiveness for all unvaccinated age groups combined (i.e. for all persons of age 5 or older) for all serotypes combined, each of the individual PCV7 serotypes, serotypes 6A and 19A, and for all other serotypes combined. For example, for serotype 4, each additional dose was estimated to decrease the overall incidence by 0.342 (34.2%). The model estimates the decrease in incidence due to an increase of 2 doses as 1 − (1 − 0.342)2 = 0.567 (or 56.7%). For an increase of 3 doses the estimated decrease in incidence was 1 − (1 − 0.342)3 = 0.715 (or 71.5%). In the figure, a star above a bar indicates a statistically significant effectiveness (p < 0.05). The herd immunity effectiveness was significant for all serotypes combined and for each of the PCV7 serotypes. For serotype 6A and for all other serotypes (not including PCV7 serotypes, 6A and 19A) combined, there was a positive but non-significant herd immunity effectiveness. On the other hand, for serotype 19A the incidence was estimated to increase with each additional dose (p = 0.046).

Fig. 2. Estimated herd immunity effect of vaccinating children <5 years old on incidence rates of IPD by mean number of doses and age group: (a) 5–19-years old, (b) 20–39-years old, (c) 40–64-years old and (d) 65+ years old.

M. Haber et al. / Vaccine 25 (2007) 5390–5398

5395

4. Discussion

Fig. 3. Estimated per-dose herd immunity effectiveness by serotype for ages 5 years and above. Vertical lines represent the standard errors of the estimates. Stars indicate a statistically significant effect (p < 0.05).

Finally, we used our model to compare the decline in the incidence rates of IPD from all serotypes (combined) between whites and African-Americans. We fit separate Poisson regression models to the incidence rates in whites and African-Americans using the same population-wide coverage data as the explanatory variable in both models. When the average number of doses of PCV7 given to 15-month-old children regardless of race is increased, our model estimated a decrease in the incidence among whites of all ages and among African-Americans of all ages (Fig. 4). The decreases were highly significant (p < 0.001) in both whites and AfricanAmericans. When the Poisson regression model was used to study the interaction between race and the average number of doses, the incidence in African-Americans was estimated to decline significantly faster (p = 0.014) than the incidence in whites. The decline in incidence among African-American was more pronounced than that in whites in both absolute and relative terms.

Fig. 4. Estimated incidence rates of IPD for Whites and African-Americans (all serotypes).

The introduction of the pneumococcal conjugate vaccine in the United States has had a dramatic impact on the incidence of pneumococcal disease. The decline in disease has greatly exceeded what was expected based on the coverage in the population (e.g., children <5 years) [17–19] because of herd immunity. Following the success of large clinical trials showing 86–94% efficacy in infants and young children [20,21], the FDA approved PCV7, and the US Advisory Committee on Immunization Practices (ACIP) recommended in February of 2000 that all infants between the ages of 6 and 8 weeks (2 months) should routinely receive the first dose of PCV7. Two more doses were recommended 6–8 weeks apart (i.e. at 4 and 6 months of age), and a booster dose was recommended between the ages of 12 and 15 months. However, vaccine shortages between 2001 and 2004 significantly influenced the introduction of the vaccine. In February 2004, the ACIP requested that all vaccine providers suspend the use of a fourth dose (i.e. the booster) to conserve supplies of PCV7 [22]. In the following month, the ACIP requested providers to only administer the first 2 doses to healthy children, to ensure more children can at least begin the recommended immunization series [23]. Children at high risk of IPD were recommended to receive all 4 doses. As vaccine production levels began to rise, the ACIP reinstated its 3rd and 4th dose recommendations in July and September 2004, respectively [24,25]. The objectives of this study were to fit a model that could describe herd immunity induced by PCV7 among populations not targeted for vaccination and to estimate the herd immunity benefits induced with each successive dose in the schedule of up to 4 doses. Our results indicate that use of PCV7 among young children significantly decreases the incidence of IPD with vaccine serotypes in older children and all age groups of adults. The greater the number of doses delivered to children, the greater the estimated reductions. A cost-effectiveness analysis in 2006, based on the observed costs and benefits in the first 5 years of PCV7 use in the U.S., suggested that inclusion of herd immunity effects in the calculations markedly reduced the cost per case prevented and per life saved [26]. A cost-effectiveness study in England and Wales came to the conclusion that full vaccination would not be economical if only the direct effects of the vaccine are considered [8]. The marked indirect impact of vaccination of young children on invasive pneumococcal disease among unvaccinated older populations, again verified by our analysis, suggests that analyses that do not include herd immunity effects may substantially underestimate the cost-effectiveness of PCV7 both on health care costs and on societal benefits. For all unvaccinated age groups, IPD caused by all serotypes combined and by all PCV7 serotypes combined was estimated to significantly decrease as the average number of PCV7 doses given increased. While this was not surprising, based on the substantial herd immunity effects of PCV7

5396

M. Haber et al. / Vaccine 25 (2007) 5390–5398

already observed [5,6,18], the magnitude of these effects with only a single dose were remarkable. Although optimal effectiveness may not be achieved unless 2 or more doses are administered in the first 6 months of life, a recent casecontrol study reported that significant individual protection was provided for approximately 6 months after a single dose of PCV7 during early infancy [27]. Under the Poisson regression model, the greatest reductions due to herd immunity in absolute terms come from the first dose compared to zero doses. Subsequent doses lead to similar relative reductions from the preceding dose, but because the overall incidence is lower, the absolute differences are less. We used Poisson regression models to describe the association between the incidence of IPD and coverage of PCV7 because these kinds of models are most often used to describe and analyze trends in incidence data. The Poisson regression models assume that the relative reduction in incidence rates due to an increase of 1 dose in coverage is the same when we move from 0 to 1, 1 to 2 or 2 to 3 doses. The good fit of the model to the data indicates that the data do not contradict this assumption. The model however does not discriminate between the cumulative effect of herd immunity due to a more vaccinated population and the effect of the number of doses per child on herd immunity, since these two phenomena occur concomitantly. One dose given to 1500 children may have a greater effect on herd immunity than 3 doses given to 500 children. So how the vaccine is given (1 dose to many or multiple does to fewer) could make a difference in the actual herd immunity effects. The rapid declines in disease due to herd immunity with the introduction of the first doses of the vaccine support this concept. Nasopharyngeal colonization with S. pneumoniae is an important prerequisite for invasive pneumococcal disease [4]. Although transmission dynamics of S. pneumoniae are complex and serotype-dependent, repeated periods of exposure and contact with a carrier are a significant risk factor for colonization with pneumococci. Children, and especially young children, have significantly higher pneumococcal carriage rates than adults; and adults living with children have higher carriage rates than adults not living with children [28]. Although our data did not measure rates of pneumococcal carriage, studies using randomized trials show that pneumococcal conjugate vaccines reduce the rates of carriage of vaccine-containing serotypes of S. pneumoniae in those who received the vaccine [29,30]. The herd immunity of the bacterial conjugate vaccines is believed to result from an overall reduction in vaccine-serotype carriage [31]. Combined direct and indirect effects of a vaccine can be measured by observational, epidemiologic studies or by randomized trials [32]. This study estimated indirect effects using a model based on active surveillance and vaccine survey data. The study assumed changes in incidence among the unvaccinated population were due to herd immunity, which decreased vaccine serotypes, or to serotype-replacement, which increased non-vaccine serotypes. Other explanations for changes in incidence have been addressed [5]. The effects

of the 23-valent pneumococcal polysaccharide vaccine, for example, which is usually given to older individuals, are potential confounders. However, no significant changes in 23-valent polysaccharide vaccine coverage between 1999 and 2002 were noted [33]. We believe that the decrease in IPD incidence can be attributed to the increase in the coverage of PCV7. As we can see from Fig. 1, the incidence of IPD changed little between 1994 and 2000, i.e., prior to the introduction of PCV7. This further strengthens the conclusion that the decline in incidence between 2000 and 2003 did not just result from a natural decline in pneumococcal disease rates over time. The incidence rates associated with non-vaccine serotype 19A consistently increased in all age groups after introduction of the vaccine. This increase in incidence (negative effectiveness) for a non-PCV7 serotype is a concern. Our data support the hypothesis that serotype 19A has increasingly caused invasive disease as the incidence of disease caused by the vaccine-serotypes has become less common. Increases in non-invasive disease, such as acute otitis media, caused by non-vaccine serotypes have also been observed in those who received PCV7 [20,34]. Such increases have not yet had a significant impact on the overall decline in the incidence of pneumococcal disease since the baseline incidence rates of the vaccine serotypes were so much higher than the nonvaccine serotypes and because the decreases in incidence in the vaccine serotypes have been much greater than the increases in 19A. However, serotype replacement will need to be closely monitored. Although analyses of direct effects [5,34] have demonstrated that the PCV7 serotype 6B antigen has been reducing the incidence of disease caused by serotype 6A, none of the changes in the incidence due to herd immunity for serotype 6A were significant. Minority groups share a disproportionate burden of IPD [2]. Our model estimates a reduction in the incidence of IPD among persons too old to have received PCV7 for both the white and African-American populations. However, the indirect effects of PCV7 were significantly greater in African-Americans, consistent with the observed reduction in racial disparities since PCV7 licensure [35]. The benefits of the pneumococcal conjugate vaccine may be greatest in populations with high burdens of pneumococcal disease. The reasons for these differences remain unclear. Crowding and high pneumococcal transmission rates, and thus greater impact of the vaccine on transmission, are possible explanations. In the U.S., 4 doses of PCV7 are recommended over the first 18 months of life. This was the schedule used in the pre-licensure clinical trials in the U.S. Our model estimates a substantial herd immunity impact when the average child in a community has received fewer than the recommended 4 doses. The achievement of marked decreases in pneumococcal disease incidence nationwide, despite vaccine shortages and delays in the introduction of the vaccine, is consistent with our findings that significant herd immunity can be attained when children receive fewer than 3 doses.

M. Haber et al. / Vaccine 25 (2007) 5390–5398

This is also consistent with case-control data that have documented high levels of direct protection after only 2 doses [27]. While our model describes the effect of herd immunity on incidence rates based on an average number of doses received by young children, it says nothing about how those doses are distributed. For example, the model could exaggerate the benefits of a given number of average doses received if populations at highest risk for transmitting S. pneumoniae received more than the average number of doses and low risk populations received fewer than the average. Our method has several limitations: First, this is an ecological study, hence we do not claim that we found a causal association between the increase in vaccination and the decrease in incidence. Second, because we had only three observations (one for each post-PCV7 year) in each age group, we only used the average number of doses as the predictor, rather than the separate frequencies of children who received 0, 1, 2 and 3 doses. In addition, we did not account for the sampling variance of this average. Third, our predictor variable is vaccine coverage among 15-month-old children. Using coverage among all children 0–4 years of age may have provided a fuller picture of how coverage was driving indirect effects. Fourth, we assumed that the coverage among 7–23-month-old children represents the coverage among 15month-old children. Fifth, the coverage data represent Fulton and Dekalb counties, while the incidence data is based on the HD3 area, which includes six additional counties. Finally, our study is based on a rather small number of cases from a single population. For these reasons we plan to conduct a more extensive study that will include data from several additional U.S. geographic areas. In summary, we presented a quantitative model to describe and estimate the reductions in invasive pneumococcal diseases with each successive dose of PCV7 among unvaccinated populations. We concluded that herd immunity effects may substantially increase the benefits and cost-effectiveness of introducing conjugate pneumococcal vaccines. The estimated magnitude of the herd immunity effects of the pneumococcal conjugate vaccine are quite remarkable. Among unvaccinated persons over 5 years of age, the incidence of pneumococcal disease caused by PCV7 serotypes as a group are estimated to drop 38.4%, 62.0% and 76.6% for 1, 2 and 3 doses, respectively, received on average by children by the time they are 15-month-old. Since the model was derived from data from a single population, it will need to be validated with other data sets. However, the model may be a useful tool for the introduction of PCV7 into new regions and populations.

Acknowledgements We wish to thank Monica Farley, Bernard Beall, Chris Van Beneden and Tami Skoff for their help in conducting this study. We also thank two reviewers for their helpful comments.

5397

References [1] Robinson KA, Baughman WS, Rothrock G. Epidemiology of invasive Streptococcus pneumoniae infections in the United States, 1995–1998—opportunities for prevention in the conjugate vaccine era. JAMA 2001;285:1729–35. [2] Preventing pneumococcal disease among infants and young children: recommendations of the Advisory Committee on Immunization Practices (ACIP). MMWR 2000;49(RR-9):1–35. [3] Obaro SK. Prospects for pneumococcal vaccination in African children. Acta Trop 2000;75:141–53. [4] Bogaert D, de Groot R, Hermans PWM. Streptococcus pnueumoniae colonisation: the key to pneumococcal disease. Lancet Infect Dis 2004;4:144–54. [5] Stephens DS, Zughaier SM, Whitney CG, Baughman WS, Barker L, Gay K, et al. Incidence of macrolide resistance in Streptococcus pneumoniae after introduction of the pneumococcal conjugate vaccine: population-based assessment. Lancet 2005;365:855–63. [6] Whitney CG, Farley MM, Hadler J, Harrison LH, Bennett NM, Lynfield R, et al. Decline in invasive pneumococcal disease after the introduction of protein–polysaccharide conjugate vaccine. N Engl J Med 2003;348:1737–46. [7] Lieu TA, Ray GT, Black SB. Projected cost-effectiveness of pneumococcal conjugate vaccination of healthy infants and young children. JAMA 2000;283:1460–8. [8] Melegaro A, Edmunds WJ. Cost-effectiveness analysis of pneumococcal conjugate vaccination in England and Wales. Vaccine 2004;22:4203–14. [9] Hofmann J, Cetron MS, Farley MM, Baughman WS, Facklam RR, Elliott JA, et al. The prevalence of drug-resistant Streptococcus pneumoniae in Atlanta. N Engl J Med 1995;333:481–6. [10] Gay K, Baughman WS, Miller Y, Jackson D, Whitney CG, Schuchat A, et al. The emergence of Streptococcus pneumoniae resistant to macrolide antimicrobial agents: a 6-year population-based assessment. J Infect Dis 2000;182:1417–24. [11] Hyde TB, Gay K, Stephens DS, Vugia DJ, Pass M, Johnson S, et al. Macrolide resistance among invasive Streptococcus pneumoniae isolates. JAMA 2001;286:1857–62. [12] Facklam RR, Washington JAI. Streptococcus and related catalasenegative gram-positive cocci. In: Balows A, Hausler WJJ, Herrmann KL, Isenberg HD, Shadomy HJ, editors. Manual of clinical microbiology. Washington, DC: American Society for Microbiology; 1991. p. 238–57. [13] Smith PJ, Hoaglin DC, Battaglia MP. Statistical methodology of the National Immunization Survey, 1994–2002. Vital and Health Statistics 2: National Center for Health Statistics, 2005. [14] Smith PJ, Battaglia MP, Huggins VJ, Hoaglin DC, Roden AS, Khare M, et al. Overview of the sampling design and statistical methods used in the National Immunization Survey. Am J Prev Med 2001;20:17–24. [15] SAS Institute, Cary, NC. [16] Agresti A. Categorical data analysis. 2nd ed. New York: Wiley–Interscience; 2002. [17] Black SB, Shinefield HR, Hansen JR, Elvin L, Laufer D, Malinoski F. Postlicensure evaluation of the effectiveness of seven valent pneumococcal conjugate vaccine. Pediatr Infect Dis J 2001;20: 1105–7. [18] Black S, Shinefield H, Baxter R, Austrian R, Bracken L, Hansen J, et al. Postlicensure surveillance for pneumococcal invasive disease after use of heptavalent pneumococcal conjugate vaccine in Northern California Kaiser Permanente. Pediatr Infect Dis J 2004;23:485–9. [19] Lexau CA, Lynfield R, Danila R, Pilishvili T, Facklam R, Farley MM, et al. Changing epidemiology of invasive pneumococcal disease among older adults in the era of pediatric pneumococcal conjugate vaccine. JAMA 2005;294:2043–51. [20] Black SB, Shinefield HR, Fireman B, Lewis E, Ray P, Hansen JR, et al. Efficacy, safety and immunogenicity of heptavalent pneumococcal conjugate vaccine in children. Pediatr Infect Dis J 2000;19:187–95.

5398

M. Haber et al. / Vaccine 25 (2007) 5390–5398

[21] O’Brien KL, Moulton LH, Reid R, Weatherholtz R, Oski J, Brown L, et al. Efficacy and safety of seven-valent conjugate pneumococcal vaccine in American Indian children: group randomised trial. Lancet 2003;362:355–61. [22] Limited supply of pneumococcal conjugate vaccine: suspension of recommendation for fourth dose. MMWR 2004;53:108–9. [23] Updated recommendations on the use of pneumococcal conjugate vaccine: suspension of recommendation for third and fourth dose. MMWR 2004;53:177–8. [24] Updated recommendations for use of pneumococcal conjugate vaccine: reinstatement of the third dose. MMWR 2004;53:589–90. [25] Pneumococcal conjugate vaccine shortage resolved. MMWR 2004;53:851–2. [26] Ray GT, Whitney CG, Fireman BH, Ciuryla V, Black SB. Costeffectiveness of pneumococcal conjugate vaccine—evidence from the first 5 years of use in the United States incorporating herd effects. Ped Inf Dis J 2006;25(6):494–501. [27] Whitney CG, Pilishvili T, Farley MM, Schaffner W, Craig AS, Lynfield R, et al. Effectiveness of seven-valent pneumococcal conjugate vaccine against invasive pneumococcal disease: a matched case-control study. Lancet 2006;368:1495–502. [28] Hendley JO, Sande MA, Stewart PM, Gwaltney JMJ. Spread of Streptococcus pneumoniae in families. I. Carriage rates and distribution of types. J Infect Dis 1975;132:55–61.

[29] Dagan R, Melamed R, Muallem M, Piglanski L, Greenberg D, Abramson O, et al. Reduction of nasopharyngeal carriage of pneumococci during the second year of life by a heptavalent conjugate pneumococcal vaccine. J Infect Dis 1996;174:1271–8. [30] Dagan R, Givon-Lavi N, Zamir O, Sikuler-Cohen M, Guy L, Junco J, et al. Reduction of nasopharyngeal carriage of Streptococcus pneumoniae after administration of a 9-valent pneumococcal conjugate vaccine to toddlers attending day care centers. J Infect Dis 2002;185: 927–36. [31] Anderson MR, May RM. Immunisation and herd immunity. Lancet 1990;335:641–5. [32] Halloran ME, Haber M, Longini IM, Struchiner CJ. Direct and indirect effects in vaccine efficacy and effectiveness. Am J Epidemiol 1991;133:323–31. [33] CDC. Influenza vaccination coverage among adults aged ≥50 years and pneumococcal vaccination coverage among adults aged ≥65 years, United States, 2002. MMWR 2003;52:987–92. [34] Eskola J, Kilpi T, Palmu A, Jokinen J, Haapakoski J, Herva E, et al. Efficacy of a pneumococcal conjugate vaccine against acute otitis media. N Engl J Med 2001;344:403–9. [35] Flannery B, Schrag S, Bennett NA, Lynfield R, Harrison LH, Reingold A, et al. Impact of childhood vaccination on racial disparities in invasive Streptococcus pneumoniae infections. JAMA 2004;291(18): 2197–203.