We have demonstrated that the 2009 H1N1 influenza pandemic period was associated with a significant increase in pneumococcal pneumonia hospitalizations among older children and young to middle-aged adults, but there was no such increase among persons over the age of 65, consistent with the pandemic sparing of seniors (Supplementary Figure 1
). The magnitude of the increase in disease incidence among the middle-aged adults was comparable to what has been seen in other recent severe seasonal influenza seasons. However, there was an unusual increase among the 5–19-year-old population, which has relatively low baseline levels of pneumococcal disease. Variations between states in the timing of pneumococcal pneumonia increases and variations between age groups in the magnitude of these increases matched the observed variation in influenza pandemic dynamics, further reinforcing the interaction between these 2 pathogens.
For the 20–39 and 40–64-year age groups, the magnitude of the increase in pneumococcal pneumonia during the pandemic was comparable to that seen during the most severe influenza epidemic in recent years, the influenza A/H3N2/Fujian-dominated 2003–2004 season [34
]. This suggests that although the timing of the 2009 pandemic was unusual and the age distribution of influenza cases was unusually shifted toward children and middle-aged adults [35
], the effect on bacterial disease was not substantially worse in this age group than in the most severe interpandemic influenza seasons. By contrast, the 5–19-year-olds, who typically have a low baseline incidence of pneumococcal pneumonia hospitalizations, experienced 2–3 times more excess hospitalizations during the pandemic period compared with any of the previous 6 influenza seasons. This unusual increase in pneumococcal pneumonia incidence in school-aged children could be explained by the high influenza attack rate in this age group during the pandemic [36
]. Immunity from prior experience with influenza A/H1N1 viruses in childhood had led to a relative sparing of the elderly [35
]; this agreed with our observation that this age group had no significant increase in pneumococcal pneumonia during the pandemic period. Additionally, children <5 years did not show a significant increase in pneumococcal pneumonia during the pandemic, reflecting the lower impact of the influenza pandemic in this age group (Supplementary Figure 1
Despite the clear relationship between influenza and pneumococcus reported in this study and by others [6
], influenza still explains only a small percentage of all invasive pneumococcal disease incidence. Recent estimates from the Centers for Disease Control and Prevention suggest that, overall, just 4.5%–6% of all invasive pneumococcal pneumonia cases can be attributed to influenza, and our estimates largely agree with these figures. Clearly, then, other risk factors can play an important role in determining susceptibility to pneumococcal pneumonia. Environmental risk factors [23
], social behaviors [24
], and other viruses [2
] could influence pneumococcal seasonality patterns. However, given that pneumococcal seasonal incidence was remarkably similar between baseline prepandemic years, and given the consistency in geographic and age patterns in the observed influenza and pneumococcal pneumonia time series in 2009, influenza is the most likely cause of the increase in pneumococcal pneumonia hospitalizations during the pandemic period.
There was a decrease in hospitalization rates in our database in most age categories and syndromes toward the end of 2009. We used a discharge database, so if a patient was admitted toward the end of 2009 but discharged in 2010, that patient would not be registered in the database until the 2010 data becomes available. On the basis of hospitalizations for all causes, we note that 97% of patients admitted before week 50 were discharged by the end of 2009 and thus included in our analysis (with substantially lower rates of 2009 discharge in those admitted after week 50, hence the week 50 cutoff in our study). Additionally, the mean length of hospital stay in 2009 for pneumonia and septicemia was 5.3 days and 8.8 days, respectively. Thus, the decreases seen late in the pandemic period might represent a real drop in pneumococcal disease incidence.
There have been suggestions that influenza pandemics or epidemics may be associated with a change in the distribution of pneumococcal serotypes causing disease. For instance, during the 1918 pandemic, specific serotypes increased as causes of disease [10
], and an epidemic of serotype 5 and serogroup 12 coincided with the influenza epidemic of 1968–1969. Unfortunately, we did not have serotype data available to explore this possibility.
It is not clear from our data, which ended in mid-December 2009 due to incomplete information for the end of the year, whether the increase in pneumococcal hospitalizations in the fall would be offset by a sustained decrease in pneumococcal pneumonia for the remainder of the 2009–2010 winter period, consistent with the unusual absence of influenza. For the over-65 population, which did not have an increase in pneumococcal disease above baseline in the fall, this could result in a net decline in the amount of disease for the 2009–2010 season. These short-term trends should be explored when more data become available.
One limitation of our study is the reliance on ICD-9
disease codes, which is subject to bias if there are changes in diagnostic testing or changes in coding practice during the pandemic. We tested for such biases by looking at changes in the incidence of E. coli
septicemia, a bacterial infection that should be unaffected by the pandemic. We saw no evidence of systematic, sustained increases in E. coli
incidence over baseline during the pandemic period compared with the rest of the year, suggesting that there was not an increase in testing for septicemia. However, we cannot exclude the possibility that clinicians were more aware of the possibility for pneumococcal coinfections during the pandemic and thus were more likely to diagnose pneumococcal pneumonia. Additionally, there were clearly increases in ICD-9
coding for influenza, even during the summer of 2009 when influenza activity was lower. Due to this issue, we allowed the effect of influenza in our model to vary during the 2009 pandemic period compared with the prepandemic years. The 2 statistical approaches that we employed—the baseline method and the attribution model—gave similar results, suggesting that this was a valid modification to the model. The similarity between the estimates from these 2 approaches also supports the validity of both statistical approaches, which have been used previously [6
We focused on pneumococcal pneumonia hospitalizations, the most common pneumococcal-specific ICD-9
diagnosis in our State Inpatient Discharge data set. This is perhaps a more sensitive but less specific diagnostic criterion than “invasive pneumonia,” which is frequently reported and requires the isolation of pneumococcus from a normally sterile site such as blood or cerebrospinal fluid. The similar patterns seen with pneumococcal septicemia diagnoses (more specific but less sensitive than pneumococcal pneumonia [39
]) suggest that the results are not sensitive to the diagnostic criteria (Supplementary Figure 2
It has been suggested that because of the risk for secondary bacterial diseases during a pandemic, stockpiling antibacterial, as well as antiviral, medications would be prudent public health policy [18
]. The 2009 pandemic was not as severe as had been initially feared, and as we show here, the number of bacterial infections was not much above what one would expect in a typical winter season. However, given a more severe pandemic, a rapid rise in subsequent bacterial infections does have the potential to cause major public health problems, and having sufficient antibacterial drugs might have the potential to prevent some severe disease cases. Also, the new formulation of the pneumococcal conjugate vaccine, which promises to further reduce disease incidence and transmission of highly invasive serotypes [41
], could help to reduce the impacts of future pandemics.
In summary, the findings of this study provide further support for the link between influenza and pneumococcal disease. During the autumn 2009 pandemic, we observed a significant spike in pneumococcal pneumonia hospitalizations, which temporally and spatially corresponded to the peak influenza pandemic periods. In young and middle-aged adults, the magnitude of the increase in pneumococcal pneumonia hospitalizations was comparable to what has been seen in previous years, while the increase was unusually high in school-aged children.