Association between Antimicrobial Resistance among Pneumococcal Isolates and Burden of Invasive Pneumococcal Disease in the Community

Abstract

Treatment of infections with drug-resistant strains of Streptococcus pneumoniae (pneumococcus) may fail; whether drug resistance is associated with an increase in the number of serious infections in the community is unknown. We evaluated the relationship between the proportion of antimicrobial-resistant S. pneumoniae isolates and the number of cases of invasive pneumococcal disease. Linear regression models included 1996 county-level data from 38 counties participating in the US Centers for Disease Control and Prevention's Active Bacterial Core Surveillance. Separate models evaluated hospitalized children aged <5 years, nonhospitalized children aged <5 years, adults aged 18–64 years, and adults aged >64 years. The proportion of isolates resistant to ⩾3 drug classes was associated with invasive disease in both hospitalized (P =.06) and nonhospitalized (P =.001) children. The proportion of multidrug-resistant pneumococcal isolates did not predict invasive cases among adults. The increasing prevalence of multidrug-resistant pneumococci among children may be leading to an increase in invasive disease

Streptococcus pneumoniae infections are among the leading causes of illness and death among young children, persons with underlying medical conditions, and elderly persons. Each year in the United States, pneumococcal disease is estimated to cause 3000 cases of meningitis, 50,000 cases of bacteremia, 125,000 hospitalizations for pneumonia, and 6,000,000 cases of otitis media [1–11]. In the United States, acute otitis media results in >24 million visits to pediatricians per year; 62% of children experience an episode of acute otitis media during their first year of life, and nearly one-half have had ⩾3 episodes before their third birthday [12, 13]

During the 1980s, antimicrobial-resistant strains of S. pneumoniae were uncommon and had been documented only in case reports [14–16]. Surveillance data from the US Centers for Disease Control and Prevention (CDC; Atlanta) for 1979–1987 verified that invasive S. pneumoniae had low levels of antimicrobial resistance [17]. However, in the 1990s, antimicrobial-resistant strains, including those with reduced susceptibility to multiple antimicrobial drugs, became increasingly prevalent in many parts of the country [18–22]. Treatment failure associated with antimicrobial-resistant pneumococci has been reported for patients with otitis media [23–26] and meningitis [27–33], although the clinical significance of antimicrobial resistance for patients with pneumococcal pneumonia is less clear [2, 34–36]

The prevalence of antimicrobial-resistant strains of S. pneumoniae and the incidence of invasive disease vary by geographic region [18, 37]. Whether drug resistance leads to more cases of invasive disease in a community, perhaps because treatment failure associated with such syndromes as otitis media, is unknown. We conducted an ecological analysis to identify county-level factors (e.g., the proportion of S. pneumoniae isolates that are resistant to antimicrobials) associated with invasive pneumococcal disease

Materials and Methods

Case identification Cases were identified through the CDC's Active Bacterial Core Surveillance (ABCs) of the Emerging Infections Program Network. ABCs is an active, laboratory-based surveillance system, which, in 1996, conducted surveillance for invasive pneumococcal disease in selected counties in California (1 county), Georgia (8 counties), Maryland (6 counties), Minnesota (7 counties), Oregon (3 counties), and Tennessee (5 counties), in addition to the entire state of Connecticut (8 counties). Surveillance staff regularly contacted all clinical laboratories serving residents of the areas under surveillance and semiannually audited laboratory records to identify missed cases. A case of invasive S. pneumoniae disease was defined as one in which S. pneumoniae was isolated from a normally sterile site (e.g., blood, CSF, synovial fluid, pericardial fluid, pleural fluid, or peritoneal fluid) in a surveillance-area resident. We analyzed cases of invasive pneumococcal disease that occurred from 1 January 1996 through 31 December 1996. For the study period, the total surveillance-area population was estimated to be 15.1 million persons, including 1.1 million children aged <5 years

Microbiologic methods Pneumococcal isolates were sent to reference laboratories for susceptibility testing by broth microdilution, according to methods of the National Committee for Clinical Laboratory Standards [38]. Isolates from Georgia were tested at the CDC; all others were tested at the University of Texas Health Sciences Center at San Antonio. The reference laboratories used susceptibility testing panels that included penicillin, amoxicillin, cefotaxime, cefuroxime, meropenem, erythromycin, clindamycin, chloramphenicol, vancomycin, rifampin, and quinupristin-dalfopristin

Isolates were defined as susceptible, intermediate, or resistant to the agents tested, according to the definitions of the National Committee for Clinical Laboratory Standards [39]. We defined an isolate as nonsusceptible if it was either intermediate or resistant to the tested agent. In analyses of resistance to multiple drug classes, we grouped the penicillins, cephalosporins, and meropenem into 1 drug class; isolates that were not susceptible to any of these agents (penicillin, amoxicillin, cefotaxime, cefuroxime, or meropenem) were considered nonsusceptible to the β-lactam class; other agents were considered to be their own drug class

County-level data We obtained county-level data from a variety of sources (table 1). For each county, the number of cases of invasive disease and the proportion of isolates that were resistant to a variety of agents came from ABCs data. We examined several definitions of resistance, including the proportion of S. pneumoniae isolates that were nonsusceptible or resistant to penicillin, cefotaxime, trimethoprim-sulfamethoxazole, or erythromycin, and the proportion that were nonsusceptible to 2–3 different classes of antimicrobial agents. Because there were too few isolates to calculate meaningful proportions of nonsusceptible or resistant S. pneumoniae for each age group per county, we calculated the proportion of isolates that were resistant to antimicrobials for all available isolates per county, regardless of age category, for all 4 models

Table 1

County-level data sources, year data were obtained, and the number of counties for which data were available for each county characteristic for illness caused by pneumococci

We were able to access county-level estimates of previously identified individual risk factors for invasive pneumococcal disease (black race, presence of AIDS, and day care center attendance) [40–42] and previously identified risk factors for antimicrobial-resistant pneumococcal disease (recent antibiotic use) [43]. In addition, we used several sources to obtain county-level estimates of potential factors contributing to invasive pneumococcal disease (table 1)

Analytic methods We constructed multiple linear regression models to examine the relationship between proportion of antimicrobial-resistant S. pneumoniae isolates and the number of cases of invasive pneumococcal disease, controlling for other factors that could affect the number of cases of invasive pneumococcal disease in a county. We centered all variables around zero by subtracting the mean of the variable from each observation and then standardized them to a common scale by dividing the differences by the variable's standard deviation. We calculated correlation coefficients for all variables with the number of invasive cases per county included as the dependent variable. We also evaluated the association of each variable with the number of cases of invasive disease by use of bivariate linear regression models that controlled for county population. County-level estimates of potential risk factors for invasive S. pneumoniae disease with r >.60, county-level estimates of previously identified individual risk factors for invasive S. pneumoniae disease, and total population per county were evaluated as candidate variables for the multivariable linear regression models. We constructed a model for the analysis of the proportion of S. pneumoniae isolates that were nonsusceptible or resistant to either penicillin, cefotaxime, trimethoprim-sulfamethoxazole, or erythromycin, and we constructed a separate model to evaluate the proportion of isolates that were nonsusceptible to 2–3 different classes of antimicrobial agents

We performed multiple linear regression with stepwise selection using the regression procedure in the SAS software system (SAS Institute). We assessed colinearity diagnostics on all multivariable models and excluded highly correlated variables. We assessed interaction terms between the remaining independent variables one at a time, and, if identified, we included interaction terms in the multivariable models

We constructed separate models by using the number of invasive pneumococcal infections identified in the following categories per county as the dependent variable: hospitalized children aged <5 years, nonhospitalized children aged <5 years, adults aged 18–64 years, and persons aged >64 years. We selected these age groups because of the strong association between age and the risk of invasive pneumococcal disease and because previously identified and potential risk factors for invasive pneumococcal disease are often age specific (e.g., day care attendance and specific underlying diseases). Compared with nonhospitalized children, children hospitalized with invasive pneumococcal disease may be more likely to have underlying illnesses and to have more-severe disease presentation; therefore, we used separate models for hospitalized and nonhospitalized children. To control for the impact of population on the number of cases of invasive disease, we included the total age-group–specific population per county as an independent variable in all models

To confirm findings from the aforementioned analyses by a different approach, we assessed the correlation between disease incidence (number of cases per 100,000 population) in a county in 1996 and the proportion of pneumococcal isolates that were resistant to ⩾3 drug classes for the same year. We examined this relationship for the 4 aforementioned categories (hospitalized children aged <5 years, nonhospitalized children aged <5 years, adults aged 18–64 years, and adults aged >64 years). We calculated Pearson's correlation coefficients, and we considered P ⩽.05 to be statistically significant

Results

Pneumococcal disease During the study period, 37 counties provided complete surveillance data for children <5 years of age, and 38 counties provided complete surveillance data for adults 18–64 and >64 years of age. ABCs surveillance staff identified a total of 3863 cases of invasive pneumococcal disease in the 38 participating counties. Of those isolates, 3152 (82%) were available for antimicrobial susceptibility testing. The county-level incidence of invasive pneumococcal disease varied by age category and hospitalization status (table 2)

Table 2

Total number of invasive pneumococcal infections and county-level incidence rates of invasive pneumococcal infection, by age group

The proportion of isolates that were nonsusceptible or resistant to antimicrobials varied between counties. Ranges for the proportions of isolates with antimicrobial nonsusceptibility were as follows: trimethoprim-sulfamethoxazole, 0–0.50; cefotaxime, 0–0.35; penicillin, 0–0.50; and 2 or 3 different antimicrobial classes, 0–0.50 and 0–0.33, respectively. Ranges for antimicrobial resistance were are follows: trimethoprim-sulfamethoxazole, 0–0.46; erythromycin, 0–0.41; cefotaxime, 0–0.2; and penicillin, 0–0.33

Linear regression models We excluded 3 variables (total number of day care facilities per county, total number of physicians per county, and antibiotic sales per county) from the multivariable analysis because of colinearity with total population. Among children, the proportion of isolates that were resistant to ⩾3 antimicrobial drug classes in a county predicted both the number of hospitalized and the number of nonhospitalized children with invasive pneumococcal disease. For hospitalized children, the number of children aged 0–4 years in the county (P <.001), number of pediatric blood culture bottles sold, proportion of low-income households, and number of persons of black race with low income were also predictors of invasive disease (table 3). After the number of children aged 0–4 years (parameter estimate, 0.50), the number of blood culture bottles sold was the strongest predictor of the number of cases. For nonhospitalized children, population 0–4 years of age (P =.001) and black population were independent predictors of invasive disease (table 3). We found a similar association between the proportion of isolates that were resistant to erythromycin and the number of hospitalized and nonhospitalized persons with invasive pneumococcal disease. We did not find an association between resistance and the number of cases when other definitions for resistance were used

Table 3

Factors associated with the number of cases of invasive pneumococcal disease in a county for hospitalized children and children treated in the outpatient setting, by multivariable linear regression

In additional analyses stratified by race, we found that the association between resistance to ⩾3 antimicrobial drug classes and the number of cases in a county was stronger for white children than for black children. For hospitalized patients, the parameter estimate for white children was 0.35 (P <.001), compared with 0.06 (P =.08) for black children. For nonhospitalized patients, the parameter estimates for white and black children were 0.51 (P <.001) and 0.14 (P =.07), respectively

We did not find a significant association between drug-resistant pneumococci and the number of cases of invasive disease for either adult model, regardless of the definition of drug resistance used (nonsusceptibility or resistance to penicillin, cefotaxime, trimethoprim-sulfamethoxazole, or erythromycin, or multidrug resistance). However, among adults 18–64 years of age, the total number of AIDS cases, black population, the total number of adult blood culture bottles sold, and the proportion of low-income households were independent predictors of invasive pneumococcal disease (table 4). The total number of AIDS cases was the strongest predictor of the number of invasive pneumococcal cases. In this age group, the number of people 18–64 years of age was not a significant predictor of the number of cases when controlling for number of AIDS cases and total black population. Among adults >64 years of age, only the number of people aged >64 years (P <.001) was an independent predictor of invasive pneumococcal disease (table 4)

Table 4

Factors associated with the number of cases of invasive pneumococcal disease in a county for adults aged 18–64 and >64 years, by multivariable linear regression

Correlation coefficients The analysis assessing correlation of disease incidence and the proportion of resistant isolates in a county substantiated the findings from the multivariable linear regression models. For hospitalized and nonhospitalized children <5 years of age, we found a significant correlation between the proportion of isolates resistant to ⩾3 antimicrobial classes and the incidence of invasive pneumococcal disease per county (figure 1A, 1B). For adults >64 years of age, the correlation was not as strong and was not significant (figure 1C). For adults 18–64 years old, there was no correlation (figure 1D)

Figure 1

Regression graphs demonstrating the relationship between the incidence rate of invasive pneumococcal disease (cases per 100,000 population) and the proportion of isolates that are nonsusceptible to ⩾3 antimicrobial classes per county for hospitalized children aged <5 years (A), nonhospitalized children aged <5 years (B), adults aged >64 years (C), and adults aged 18–64 years (D) Data are from the 38 counties participating in the Active Bacterial Core Surveillance, Emerging Infections Program Network, 1996

Discussion

We found that, in counties with higher proportions of S. pneumoniae isolates that were nonsusceptible to ⩾3 classes of antimicrobial drugs, there were more cases of invasive pneumococcal disease (requiring and not requiring hospitalization) among children <5 years of age. This finding suggests that, in addition to causing antimicrobial treatment failure for invasive and noninvasive pneumococcal infections [27–33], the increasing prevalence of antimicrobial-resistant S. pneumoniae has led to more cases of invasive pneumococcal disease among children. We did not find this association among adults aged 18–64 years or adults aged >64 years during 1996

The association between multidrug-resistant S. pneumoniae and invasive pneumococcal disease is biologically plausible. Noninvasive S. pneumoniae infections that fail to respond to initial antimicrobial therapy could progress to invasive disease. Thus, geographic regions that have a greater prevalence of multidrug-resistant S. pneumoniae would be expected to have more treatment failures for noninvasive pneumococcal infections and, subsequently, more cases of invasive pneumococcal disease. Because recent ear infection has been identified as a risk factor for invasive pneumococcal disease [42, 44, 45], and because otitis media is more common among children than it is among adults, the impact of an increased prevalence of antimicrobial-resistant S. pneumoniae on invasive pneumococcal disease may be more demonstrable among children than adults. Because certain pneumococcal serotypes are associated with antibiotic resistance [18], serotype patterns of circulating pneumococci could also play a role in disease burden. There are no data, however, to suggest that resistant strains are more likely to cause invasive disease than are susceptible strains. We did find a trend toward an increased number of cases in counties that had higher proportions of antimicrobial-resistant pneumococcal isolates for both adult categories; a large study may have identified such a relationship in those age groups

Among children, we identified several other factors that were predictive of a greater amount of invasive pneumococcal disease. For hospitalized children, counties with larger low-income black populations and higher blood culture bottle sales had more cases of invasive disease. For nonhospitalized children, counties with larger black populations, regardless of income level, had more cases of invasive disease. In our adult models, risk factors for invasive pneumococcal disease varied by age group. For adults aged 18–64 years, total number of AIDS cases, total black population, total number of adult blood culture bottles sold, and percentage of low-income households were associated with more cases of invasive disease. For adults aged >64 years, only the number of people aged >64 years in the county was associated with the amount of invasive disease

Our study corroborates the finding of other investigators that black race and low income level are associated with a higher incidence of invasive pneumococcal disease [1, 40, 41, 46–49]. Elevated rates among black persons have been attributed to socioeconomic factors, including limited access to health care and increased household crowding [47–49], and biologic factors, such as the increased prevalence of underlying diseases among black persons that predispose them to invasive disease (e.g., sickle-cell disease and diabetes) [40]. Low income may represent limited access to health care or increased household crowding

We found that, among hospitalized children or adults aged 18–64 years, counties with higher sales of blood culture bottles had more cases of invasive pneumococcal disease. This finding may be related to geographic differences in medical care–provider blood culturing practices [50]; for example, providers in some regions may be more likely to empirically treat patients without obtaining blood samples for culture

The only county-level factor unique to adults aged 18–64 years was the total number of AIDS cases. High rates of invasive pneumococcal disease have been documented among patients with AIDS [51, 52]. In San Francisco, 55% of the cases of invasive pneumococcal disease among adults aged 18–64 years can be attributed to HIV infection; the risk of invasive pneumococcal disease is 46 times higher in persons with AIDS than it is in persons without known HIV infection [41]

Among adults >64 years of age, only total population was predictive of invasive pneumococcal disease. Compared with nonhospitalized children, elderly individuals had a narrow range of incidence rates between counties; this suggests that host factors may play a more important role in this age group than do environmental, behavioral, or medical care issues. Given the high prevalence of underlying illnesses among elderly persons and the waning immune system function associated with advanced age, for the elderly population, socioeconomic factors and physician blood culturing practices may not be as influential in determining the amount of invasive pneumococcal disease in an area

Limitations of this study include the lack of direct measures for the number or proportion of children attending day care per county, the prevalence of underlying diseases that predispose to invasive pneumococcal disease per county, and more-specific socioeconomic factors (crowding, access to health care) per county. County sales of antibiotics and blood culture bottles may not precisely reflect their use among residents of a county, because people may receive medical care outside of their county of residence. In addition, the only county-level information available for vaccination coverage (i.e., the proportion of Medicare-eligible persons who had received vaccine) was likely not accurate enough to allow an assessment of the impact of coverage on disease burden. Furthermore, although isolate serotypes may have added valuable information, isolate serotype data were not available for this analysis

Our analysis indicates that limiting the spread of multidrug-resistant pneumococci could result in fewer cases of invasive disease in children. Strategies to decrease the spread of antimicrobial-resistant S. pneumoniae have been based on promotion of the appropriate use of antimicrobial agents. In the late 1990s, several guidelines for antimicrobial therapy for otitis media and meningitis in children were developed to address the increasing prevalence of antimicrobial-resistant S. pneumoniae isolates [7, 53–55]. These guidelines were written to provide antimicrobial regimens that would prevent the failure of treatment of infections due to antimicrobial-resistant S. pneumoniae as well as to educate medical providers about appropriate antimicrobial use for upper respiratory infections

However, vaccines remain the mainstay of prevention of pneumococcal disease. Strategies to prevent invasive pneumococcal disease have traditionally been based on promoting the use of the polysaccharide pneumococcal vaccine in older persons and in persons aged 2–64 years who have certain chronic illnesses. A protein conjugate pneumococcal vaccine that is effective in preventing invasive and noninvasive pneumococcal infections in young children has recently been licensed and is now available for use among children <5 years of age [56, 57]. The findings of this study emphasize the need for supporting the use of the antimicrobial therapy guidelines as well as pneumococcal vaccines

Acknowledgments

We are grateful for the support and cooperation of Paul Cieslak, Carolyn Wright, Ling Hsu, Jane Phelan, John Beltrami, Ellen Caldeira, Cynthia Hickman, and William Trick

Present affiliation: Chicago Department of Public Health, Chicago, Illinois