Exposure to Children as a Risk Factor for Bacteremic Pneumococcal Disease: Changes in the Post–Conjugate Vaccine Era

Background  The introduction of a pneumococcal conjugate vaccine has been associated with a shift in the serotypes responsible for bacteremic pneumococcal disease. We examined recent trends in serotypes responsible for disease and current risk factors among adults.

Methods  Data were obtained from 48 acute care hospitals in the 5-county region surrounding Philadelphia, Pennsylvania, from October 1, 2002, through September 30, 2008, on all hospitalized adult patients with community-acquired bacteremic pneumococcal disease. Isolates were serotyped and patient characteristics were compared with data from a household survey of the adult population in the region.

Results  During the study period, the annual rate of disease due to vaccine serotypes declined by 29% per year, but the rate of disease due to nonvaccine serotypes increased 13% per year, yielding an overall 7% increase in the annual rate of disease among adults. Advanced age was a risk factor for infection with nonvaccine serotypes compared with vaccine serotypes. Comparing all patients with the source population, African Americans were at increased risk of infection, and the presence of additional children in the home was associated with decreased risk of disease. Smoking, advanced age, and diabetes mellitus remained important risk factors in adults.

Conclusions  New serotypes are replacing the serotypes covered in the conjugate vaccine. While some risk factors for pneumococcal disease remain unchanged, the observation that exposure to children in the home is associated with lower risk of disease suggests that the changing epidemiology of pneumococcal disease may be altering the dominant modes of transmission in the community.

In February 2000, the Food and Drug Administration licensed a new conjugate pneumococcal vaccine (Prevnar; Wyeth Lederle Vaccines, Philadelphia, Pennsylvania), which was subsequently recommended by the American Academy of Pediatrics and Advisory Committee on Immunization Practices for use in children 2 years and younger as well as older, high-risk children.^1,2 The vaccine includes 7 capsular polysaccharide serotypes (4, 6B, 9V, 14, 18C, 19F, and 23F) conjugated to an immunogenic carrier protein.

The conjugate vaccine has been shown to dramatically reduce the risk of invasive pneumococcal disease in infants.³ In addition, the vaccine significantly reduces the carriage of pneumococcal serotypes covered by the vaccine.^4,5 Introduction of the conjugate vaccine has been associated with statistically significant declines in the rates of invasive disease in adults.^6,7 However, recent reports raise concern about the emergence of nonvaccine serotypes among children^8,9 and adults⁷ with invasive disease.

Serotype replacement is an important process to study not only because of its effect on the overall benefit of the vaccine but also because different pneumococcal serotypes possess different absolute and relative potentials for establishing the carrier state and causing invasive disease.¹⁰ Thus, it is plausible that, as serotype replacement occurs, risk factors for infection may change.

The aim of this study was to examine trends in serotypes responsible for bacteremic pneumococcal disease in adults and identify risk factors for disease in the post–conjugate vaccine era. We were particularly interested in the role of exposure to children as a risk factor in the post–conjugate vaccine era.

Data were collected as part of population-based surveillance for bacteremic pneumococcal disease within the 5-county region surrounding Philadelphia (Bucks, Chester, Delaware, Montgomery, and Philadelphia counties). Adult (age ≥18 years) population surveillance was initiated in October 2002. The surveillance network currently encompasses 48 of the 49 acute care hospitals that serve the 3.7 million residents of the 5 counties. The 1 nonparticipating hospital is a small hospital, closed to external studies and accounting for less than 2% of all cases in the region.

Subjects were initially identified through the microbiology laboratories at all hospitals. Hospital personnel were contacted by study personnel on a regular basis throughout the surveillance period to ensure complete capture of new cases. We confirmed the total number of eligible cases through contact with laboratory directors and annual review of their log books, as well as comparison with data from the City of Philadelphia Health Department under mandatory reporting of cases of pneumococcal bacteremia.¹¹

Eligible patients were hospitalized adults residing in the 5-county region, with at least 1 set of blood cultures positive for Streptococcus pneumoniae drawn within 48 hours of hospitalization and no hospitalization within 10 days before the episode of pneumococcal bacteremia. Laboratory personnel provided the study with the name of the treating physician, and physician permission was obtained to contact eligible patients and conduct a telephone interview.

Trained telephone interviewers completed a telephone interview with each subject (or a proxy in the case of impaired subjects or children) covering demographic and clinical areas as previously reported.¹² In particular, subjects reported the number and age of children living in the household before the onset of illness, as well as additional information on race/ethnicity and underlying chronic illnesses. Because we were unable to validate self-reported vaccine histories, we did not include vaccine history in these analyses. The interview instrument is provided in an Appendix (available from the authors on request). Twelve percent of all adults identified with bacteremic pneumococcal disease died before contact for this study. In these cases, interviews were attempted with family members; if they were not available, we were granted a waiver of authorization to abstract data from the medical record. Forty-one percent of these patients had their medical records reviewed, and in 33% we completed an interview with a family member or caregiver.

We obtained a waiver of authorization to abstract information from the hospital record for eligible subjects who could not be contacted by telephone (35% of all eligible subjects including deceased patients as noted in the preceding section). Trained medical record abstractors collected information parallel to the telephone interview, including demographic variables and medical comorbidities.

For the analysis of overall population trends in disease rates, we used adult population estimates from the US Census population between-census estimates for 2002, 2003, 2004, 2005, 2006, and 2007. These county-level population estimates are based on the 2000 decennial census, with annual population adjustments based on sampling and boundary adjustments (http://www.census.gov/popest/archives/2000s/vintage_2007/). Because our analyses focused on October 1 through September 30 analysis periods, we used the estimated population denominator of the year at the start of the observation period for each annual incidence rate calculation. For the analysis of population risk groups for the period from October 1, 2005, through September 30, 2008, we estimated population denominator counts on the basis of a 5-county random-digit-dial survey of the surveillance region. The Philadelphia Health Management Corp conducts a biannual weighted probabilistic survey of the population in the 5-county region, the Southeastern Pennsylvania Household Health Survey (SPHHS).¹³ We used population estimates from the 2006 SPHHS for these analyses.

The 2006 survey included 10 100 adults 18 years or older. Each individual in the data set (or his or her proxy) was contacted by telephone interview and data were collected that included information on basic demographic and household factors, along with long-term health information, such as whether participants had ever been diagnosed with diabetes mellitus or asthma or were currently smoking.

All pneumococcal isolates were transported to the central laboratory at the Hospital of the University of Pennsylvania, Philadelphia. Serotyping of all pneumococcal isolates was performed according to standard methods using the quellung reaction.^14,15 The reagents for serotyping constituted 14 pool serum samples, 22 type-specific serum samples, and 62 factor-specific serum samples (Statens Serum Institut, World Health Organization Collaborating Centre for Reference and Research on Pneumococci, Copenhagen, Denmark). A total of 5.5% of all pneumococcal isolates could not be serotyped (3.7% were not viable after shipping and 1.7% were untypeable). To account for untyped isolates, rates of disease by vaccine and nonvaccine serotypes were proportionally inflated in each year to match the total rate of disease in that year.

We calculated overall and serotype-specific rates of disease in each year (October 1 through September 30) from October 1, 2002, through September 30, 2008. We analyzed secular trends by means of Poisson regression using annual count data and fitting linear terms for each surveillance year. To adjust for overdispersion in the Poisson regression model, Pearson scale adjustment was applied. We explored higher-order terms for year and also attempted to fit cubic splines, but the simple linear model was the best fit for the data. We categorized all isolates with serotype results into vaccine serotypes (1 of the 7 included in the conjugate vaccine) vs nonvaccine serotypes (all remaining serotypes), and calculated annual rates of disease attributed to each of these 2 subgroups.

We conducted 2 comparative analyses for all cases identified from October 1, 2005, through September 30, 2008. First, we compared patient characteristics between cases of pneumococcal bacteremia due to vaccine serotypes vs nonvaccine serotypes. Second, we examined potential population risk factors for bacteremic pneumococcal disease including all pneumococcal serotypes. For the first set of analyses, we used simple logistic regression for comparisons, fitting 1-variable models for each tested characteristic. We planned to fit multivariable models only if multiple factors were significant on initial bivariable analysis. For the second set of analyses, we calculated demographic and clinical subgroup-specific annual rates of disease, with 95% confidence intervals (CIs), using weighted population estimates from the SPHHS survey. The SPHHS uses probabilistic geographic sampling across the 5 counties. The inverse of the sampling probabilities provided sample weights for creating population projections and accounted for the complex sample design in the SPHHS. Multivariable Poisson regression incorporating the projection weights was then used to estimate adjusted relative rates and their respective 95% CIs, in which population appears as an offset in the model. To address the issue of missing data in the model, we completed a second multivariable regression using multiple imputation. We used imputation by chained equations in Stata^16,17 (StataCorp, College Station, Texas) to impute missing values of the covariates in the regression models. Imputation by chained equations is a multivariable approach that uses the conditional distribution of each covariate, given other predictor variables, to cycle between filling the missing values for each covariate. We implemented the default approach, which repeats the imputation process 5 times, to create 5 data sets with complete data. We then conducted Poisson regression on each imputed data set and combined the results by using Proc MI in SAS (SAS Institute Inc, Cary, North Carolina). As a separate analysis, we reran the multivariable model excluding patients with documented human immunodeficiency virus (HIV) infection to assess the degree to which HIV infection might confound the relationship between children in the home and disease risk.

All analyses were completed with Stata version 10 and SAS version 9.1, with 2-sided tests of hypothesis and P < .05 as the criterion for statistical significance.

From October 1, 2002, through September 30, 2008, we identified a total of 2418 adult cases of pneumococcal bacteremia from the acute care hospital network within the 5-county Philadelphia region. The Figure plots the annual rate of adult cases for the study period, starting at 12.8 cases per 100 000 in 2002-2003 and ending at 15.4 cases per 100 000 in 2007-2008. The annual rate of disease demonstrated a significant positive linear trend (P = .007) with an estimated relative increase of 7% per year. In 2002-2003, annual rates of disease were 5.8, 12.9, 30.0, and 63.2 cases per 100 000 for adults aged 18 to 49, 50 to 64, 65 to 79, and 80 or more years, respectively. In 2007-2008, annual rates of disease were 7.9, 18.9, 26.8, and 53.6 cases per 100 000 for these age ranges, respectively.

Overall, 12.6% of adult cases were due to 1 of the 7 serotypes contained within the conjugate vaccine. The Figure plots the annual rate of disease due to vaccine and nonvaccine serotypes during the study period. The rate of disease due to vaccine serotypes significantly declined during the observation period (rate ratio [RR], 0.71 per year; 95% CI, 0.67- 0.75). In contrast, we observed a significant linear increase in nonvaccine serotypes (RR, 1.13 per year; 95% CI, 1.08-1.19). Among specific nonvaccine serotypes, the greatest relative increase was seen in serotype 19A, from 0.8 cases per 100 000 in 2002-2003 to 3.4 cases per 100 000 in 2007-2008.

Detailed clinical and demographic information was collected on all cases from October 1, 2005, through September 30, 2008 (Table 1). In all, 42.7% of patients were 65 years or older. Fifty percent of patients were male, 62.7% were white and non-Hispanic, and 48.4% lived in the city of Philadelphia. Current smoking was documented for 66.9% of patients, HIV infection for 12.6%, and asthma for 15.3%.

During this 3-year period, 6.4% of cases were due to 1 of the 7 serotypes contained in the conjugate vaccine. We compared vaccine-serotype cases with nonvaccine-serotype cases and found that patients with vaccine-serotype cases were younger but otherwise did not differ demographically or clinically (Table 2).

Table 3 summarizes subgroup-specific annual rates of bacteremic pneumococcal disease. Adults aged 18 to 49 years experienced 8.3 cases per 100 000 compared with 59.4 cases per 100 000 among adults 80 years or older. Annual rate of disease was higher in urban Philadelphia County (19.1 cases per 100 000) than in outlying suburban and rural counties (13.0 cases per 100 000). Annual rate of disease was higher in African Americans (26.4 cases per 100 000) than in whites (13.7 cases per 100 000). Rate of disease also declined with increasing number of children in the home. Adults living in homes with no children younger than 7 years experienced 21.5 cases per 100 000 compared with 3.3 cases per 100 000 among adults living with 2 or more children in the home. Rates of disease were higher in smokers and patients with diabetes mellitus.

In multivariable models adjusting for all of the factors examined in Table 3, increasing numbers of children in the home remained a protective factor (Table 4). For example, compared with having no children younger than 7 years in the home, adults with 2 or more children in the home had a 76% relative reduction in the risk of bacteremic pneumococcal disease, even after adjusting for age (RR, 0.24; 95% CI, 0.18-0.33). African American race, current smoking, and diabetes mellitus remained significant risk factors for adults in the adjusted models.

We conducted additional analyses to explore the relationship between the number of children in the home and risk of bacteremic pneumococcal disease. To better adjust for age in the analyses, we conducted stratified analyses using 10-year age increments (20-29 years, 30-39 years, etc.) In these stratified analyses, the presence of 2 or more children in the home was associated with a significantly reduced risk of disease for adults aged 30 to 39, 40 to 49, 50 to 59, and 80 years or older (RR, 0.2, 0.1, 0.5, and 0.2, respectively; all P < .05). For the age groups 60 to 69 and 70 to 79 years, the association between children in the home and risk of disease was not significant. With these finer age increments, compared with adults without children in the home, the adjusted risk of disease was 44% lower for adults with 1 child in the home (RR, 0.56; 95% CI, 0.45-0.71) and 78% lower for adults with 2 or more children in the home (0.22; 0.17-0.30). To address the role of missing data in the multivariable analysis, we completed multiple imputation to generate multiple data sets without missing data. In the combined multivariable adjusted result with all imputed data sets, compared with adults with no children in the home, adults with 2 or more children had a 78% reduced risk of bacteremic pneumococcal disease (RR, 0.22; 95% CI, 0.17-0.30). Finally, to exclude confounding effects of HIV infection, we also examined the association between children in the home and risk of disease in the subgroup of cases without documented HIV infection. In this analysis, compared with adults without children in the home, the adjusted incidence rates of disease were 47% lower for adults with 1 child in the home (RR, 0.53; 95% CI, 0.40-0.71) and 70% lower for adults with 2 or more children in the home (0.3; 0.21-0.41).

Active surveillance for bacteremic pneumococcal disease in adults since 2002 in the greater Philadelphia region demonstrated a small but significant increase in the annual rate of disease. This increase was caused by increased rates of disease due to nonvaccine serotypes, while disease due to vaccine serotypes continued to decline during this period. Patients with bacteremic pneumococcal disease due to vaccine serotypes were demographically and clinically similar to patients with disease due to nonvaccine serotypes, except that patients with nonvaccine serotypes were older. In terms of overall risk of bacteremic pneumococcal disease, certain well-established risk factors remained important, including African American race, advanced age, smoking, and diabetes mellitus. However, one striking new finding is that the presence of additional children in the home was associated with decreased risk of disease, even after adjusting for other demographic and clinical factors.

Risk factors for pneumococcal disease have been recognized for some time and used to inform vaccine policy. These factors include extremes of age, African American race, and several comorbidities including chronic heart and lung diseases, HIV infection, diabetes mellitus, and smoking history.^18-20 Recent studies have demonstrated that the increased risk of disease among African Americans compared with whites has narrowed since the introduction of conjugate vaccine,²¹ but our study demonstrates that the risk remains elevated. Our study was unable to explore the continuing role of other established risk factors (eg, HIV) because data on the population distribution of these factors were not available in the household survey we used to measure the distribution of these factors in the source population.

Previous studies from the pre–conjugate vaccine era demonstrated that the presence of children in the home, particularly children in daycare, was associated with increasing risk of pneumococcal disease.¹⁹ Children are well known to be a major reservoir for pneumococcal bacteria and thought to be a primary source for bacterial spread in the community. However, with the introduction of the conjugate vaccine, patterns of pneumococcal carriage in children have shifted dramatically. As a result, it is possible that children are no longer a primary source of infection for adults, and the absence of children in the home may be a marker for other factors that lead adults to have increased exposure to alternative sources of pneumococci. Adults without children in the household may have increased exposure to other populations that transmit pneumococcal disease. For example, they may spend more time outside the home in settings that are higher risk for transmission now that transmission from children has been reduced by the vaccine. Alternatively, adults with children in the home may have reduced prevalence of other comorbidities that are associated with disease risk, including chronic heart and lung diseases or other immunosuppressive disorders.

Our results suggest that replacement disease is disproportionately affecting older adults. A recent report from the Centers for Disease Control and Prevention's Active Bacterial Core Surveillance Network also noted that, since introduction of the conjugate vaccine, adults with comorbidities represent an increasing proportion of cases with invasive pneumococcal disease.²² Earlier reports indicated that invasive disease was declining in older adults as a result of pediatric vaccination campaigns, a form of herd immunity. However, serotype replacement may ultimately reverse these gains and also increase the prevalence of disease in higher-risk adults.

Our observations are consistent with other surveillance studies that have highlighted the rise in invasive disease due to nonvaccine serotypes. During a 5-year period, we observed a significant increase in the rates of disease due to nonvaccine serotypes, including a greater than 4-fold rise in the rate of disease due to serotype 19A. During this period, vaccine-related disease declined by 80%. Notably, our surveillance began 2 years after introduction of the conjugate vaccine, so it is highly likely that the full impact of the vaccine on disease due to vaccine serotypes has been even greater and the rate of disease in adults in 2008 is likely still lower than in the pre–conjugate vaccine era. For example, results from the Centers for Disease Control and Prevention's Active Bacterial Core Surveillance program reported the incidence of invasive pneumococcal disease in 1998 as 23, 46.4, and 98.5 cases per 100 000 for adults aged 50 to 64, 65 to 79, and 80 years or older, respectively,¹⁸ which are all higher than the age group–specific rates we observed in 2007-2008. Regardless, the total rate of disease in adults has risen during the period from 2002 through 2008. In addition, the observation that disease due to vaccine serotypes declined while disease due to nonvaccine serotypes increased suggests that these patterns are not the result of changes in testing of adults with suspected bacteremic disease.

From a health policy perspective, it is unclear whether the observed increases in invasive and noninvasive disease due to nonvaccine serotypes, especially 19A, are the result of the introduction of conjugate vaccine. In the United States, multiple reports have emphasized that serotype replacement was temporally related to the introduction of conjugate vaccine in 2000.^9,20,22,23 However, in other countries, serotype replacement was observed before the introduction of conjugate vaccine,²⁴ suggesting that other secular forces (eg, antimicrobial drug use patterns or socioeconomic conditions) may underline trends in pneumococcal serotype epidemiology.²⁵ The emergence of these serotypes emphasizes the importance of newer conjugate vaccine formulations, including a conjugate vaccine with 13 serotypes that includes 19A, which is currently under review.

Our study has some limitations, most notably that we were unable to compare the full range of potential risk factors for bacteremic pneumococcal disease because of limited data on the source population. Also, some clinical and demographic data were missing on many of the subjects for whom we were unable to complete telephone interviews and therefore used medical record abstractions. In addition, our measures of population rates of disease relied on hospital-based clinical testing, which would be sensitive to changes in testing practices and would fail to detect nonhospitalized cases of the disease.

In conclusion, during the past 6 years, there has been a significant increase in the rate of disease due to pneumococcal serotypes not included in the conjugate vaccine. At the moment, the reduction in disease due to vaccine serotypes still exceeds the level of increase due to nonvaccine serotypes. The changing epidemiology of pneumococcal disease may be leading to important shifts in risk factors for the disease. Continued surveillance will allow us to monitor ongoing patterns of disease so as to develop and target newer strategies for the prevention of pneumococcal disease.

Correspondence: Joshua P. Metlay, MD, PhD, Department of Medicine, University of Pennsylvania School of Medicine, 712 Blockley Hall, 423 Guardian Dr, Philadelphia, PA 19104 (jmetlay@mail.med.upenn.edu).

Accepted for Publication: October 2, 2009.

Author Contributions: Dr Metlay had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Study concept and design: Metlay, Lautenbach, and Edelstein. Acquisition of data: Metlay and Edelstein. Analysis and interpretation of data: Metlay, Lautenbach, Li, and Shults. Drafting of the manuscript: Metlay. Critical revision of the manuscript for important intellectual content: Metlay, Lautenbach, Li, Shults, and Edelstein. Statistical analysis: Li and Shults. Obtained funding: Metlay and Edelstein. Administrative, technical, and material support: Shults and Edelstein. Study supervision: Metlay and Edelstein.

Financial Disclosure: Dr Lautenbach has received research funding from Merck, Ortho-McNeil, Astra-Zeneca, and Cubist Pharmaceuticals.

Funding/Support: This project was supported by grants R01-AI46645 and K24-AI073957 (Dr Metlay) from the National Institute of Allergy and Infectious Diseases, National Institutes of Health.

Role of the Sponsor: The funding agency had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; or preparation, review, or approval of the manuscript.

Additional Contributions: Linda Crossette, MPH, coordinated the activities of this study, and the staff at the Clinical Microbiology Laboratory of the Hospital of the University of Pennsylvania provided the microbiology testing.