All data used in this study were recorded for each influenza season (surveillance weeks 40–20) from 1994–95 through 2004–05. Weekly IPP cases were obtained from the Active Bacterial Core Surveillance for Georgia Health District 3. Cases of IPP were defined by isolation of Streptococcus pneumoniae from normally sterile sites, e.g., cerebrospinal fluid or blood. Only IPP case-patients (defined as persons with S. pneumoniae isolated from pleural fluid or persons with a clinical diagnosis of pneumonia and S. pneumoniae isolated from blood or another sterile body site) were included in this study. Influenza virus surveillance data were obtained from the World Health Organization. Weekly data were collected from each state from October through May of each influenza season. The influenza virus–positive isolate percent was defined as the percentage of influenza virus–positive isolates out of all influenza specimens. Percentages of isolates positive for influenza A (H1N1), A (H3N2), and B viruses were also calculated.

Weekly RSV data were obtained from the National Respiratory and Enteric Virus Surveillance System (www.cdc.gov/surveillance/nrevss/). For each season, hospitals and laboratories reported to the Centers for Disease Control and Prevention the numbers of specimens tested for RSV by antigen detection each week. The RSV detection percent was defined as the percentage of RSV-positive isolates out of all RSV specimens. The influenza virus and RSV data used in this study were from the US Census South Atlantic Region (www.census.gov/geo/www/us_regdiv.pdf). Data on daily mean temperature (in ^oF), total sunshine (in hours), and total precipitation (in inches) were obtained from the National Weather Service Atlanta regional weather center at Hartsfield-Jackson International Airport (9). The weekly mean temperature, total sunshine, and total precipitation were calculated from daily data.

For each of the 11 influenza seasons, we applied negative binomial regression models with multiple predictors to relate the weekly IPP rates with the indicators of influenza and RSV activities while adjusting for the weekly mean temperature, total sunshine, and total precipitation. The full model used to explore the association of IPP with influenza and RSV was as follows:

log(Y_t) = α + β₀X_t + β₁X_t−1 + β₂X_t−2 + γ₀Z_t + γ₁Z_t−1 + γ₂Z_t−2 + λW_t + ηU_t + θV_t + log(N_t)

where for week t, Y_t is the incidence of IPP, X_t is the influenza virus–positive isolate percent, Z_t is the RSV detections percent, W_t is the mean temperature, U_t is the total sunshine, V_t is the total precipitation, and N_t is the population size. The reduced model excluded the influenza and RSV terms. We calculated the difference in the log-likelihood ratio statistic between the full and reduced models and used it to test the hypothesis that IPP incidence is not associated with influenza and RSV activities when weekly temperature, sunshine, and precipitation were adjusted for.

For influenza seasons 1994–95 through 2004–05, the average annual incidence of IPP was 19.78 per 10 million persons (Table 1). Table 1 also shows the means of total number of IPP cases; influenza-positive isolates; RSV detections; and the average temperature, sunshine, and precipitation for each of these influenza seasons.

Table 2 shows the result of the negative binomial regression analysis for each influenza season. Significant associations between IPP and both influenza-positive isolate percent and RSV detection percent were found for 5 of the 11 seasons. Table 2 also shows the percentages of the influenza A (H1N1), A (H3N2), and B virus strains in each season. The predominant influenza virus strain (percentage >50%, [10]) varied across seasons. Influenza A (H3N2) virus was predominant in 8 seasons, and influenza B virus was predominant in 2 seasons. One season had no predominant influenza virus strain. Table 2 also shows a significant association between IPP incidence and influenza virus and RSV activities in 5 of the 8 seasons in which A (H3N2) was the predominant influenza virus strain; there was no significant association in any of the 3 seasons in which influenza A (H3N2) virus was not predominant (p = 0.12, 2-sided mid–p-value exact test).

To demonstrate the association of IPP incidence with influenza virus and RSV activities and the climate data, we displayed the weekly IPP, influenza virus, RSV, temperature, sunshine, and precipitation data for seasons 1998–99 and 2003/04 (Figure). Similar figures for the other 9 seasons can be found in the Technical Appendix.

To explore the temporal variability in the association of IPP incidence with influenza virus and RSV activities over a long period, we conducted a season-by-season analysis over 11 influenza seasons. We adjusted the association of IPP with influenza virus and RSV by temperature, precipitation, and sunshine because these factors are believed to be associated with pneumococcal and viral respiratory infections (5,7,8).

The results indicated substantial variability across seasons in the strength of association of IPP incidence with influenza virus and RSV activities. This variability explains why findings from previous studies (2–5,11), which were based on data from a single season or on combined data from several seasons, are inconsistent. We found significant associations of IPP incidence with influenza virus and RSV activities for 5 of 11 influenza seasons from 1994–95 through 2004–05. Notably, in each of the 5 seasons for which we found a significant association, influenza A virus strain H3N2 was predominant (this strain predominated in 8 seasons). Alternatively, we did not find a significant association in any of the 3 seasons in which strain A (H3N2) was not the predominant influenza virus strain. Although this difference was not statistically significant, it suggests that the association between IPP and influenza virus and RSV activities might be stronger in seasons in which strain A (H3N2) is predominant. This finding is consistent with the idea that excess neuraminidase expression of strain H3N2 compared with strain H1N1 influenza viruses may lead to an excess number of pneumococcal superinfections (12).

In summary, in this season-by-season analysis, we found substantial variation in the strength of the association of IPP incidence with influenza virus and RSV activities across seasons. We also found that this association may be associated with the predominant influenza virus strain. More studies, using data from more seasons and from several geographic areas, are needed to better explain the variation in the association between invasive pneumococcal diseases and respiratory viral infections.