Discussion
This study has generated estimates of influenza-attributable mortality in the USA by influenza type and subtype and an estimate of RSV-attributable mortality. We estimated that between 1997 and 2009, an annual average of 19 100 deaths was attributable to influenza, and an annual average of 11 300 deaths was attributable to RSV, using the respiratory broad definition that included any mention of ICD9/ICD10 codes for any respiratory illness, cough, breathing abnormalities, fever, and other viral infections.
As expected, influenza-attributable mortality increased as outcome definitions became more sensitive: 10 682 deaths for the P&I outcome definition, 19 100 deaths for the respiratory broad outcome definition, and 28 169 deaths for the cardiorespiratory outcome definition. However, because influenza infections are likely to have a lower impact on mortality for more sensitive definitions, use of these outcome definitions can lead to mis-attribution of influenza-attributable deaths. By focusing on those deaths with any mention of respiratory illnesses (i.e., the respiratory broad definition), we sought to make a reasonable trade-off between sensitivity and specificity.
Our estimates are in broad agreement with CDC findings for the period 1976–2007, which estimated an annual average of 23 607 influenza-attributable deaths with underlying respiratory and circulatory causes and 6309 with underlying P&I causes. In contrast, our estimates were lower than earlier CDC estimates from 1990 to 1999 for influenza-attributable deaths with underlying respiratory and circulatory causes (36 155 deaths), but higher than CDC estimates for influenza-attributable deaths with underlying P&I causes (8097 deaths) during the same period. More recent estimates using a very different model approach to ours and that of the CDC attribute an average annual influenza-attributable death rate of 1·7 per 100 000 due to P&I causes between 1997 and 2007 in the USA, compared with our rate of 3·7 deaths per 100 000 for the P&I outcome definition between 1997 and 2009.
Modeled estimates of influenza-related mortality are the preferred source; however, estimates are tied to the modeling choices. Here, we used a modeling approach that included viral surveillance data representing the relative importance of each pathogen on a regional and weekly basis and, conservatively, a stationary trigonometric baseline to ensure that regular non-specific seasonal drivers of mortality (other pathogens) were considered. These choices are in line with those adopted in several previous studies and allow verification of the consistency of our estimates against previously published findings. However, the extent to which they are robust to model variations, such as the recent suggestion of different baseline functions (e.g., cubic spline) or other types of influenza incidence proxies, remains to be determined. Percent positive proxies were adopted in previous analyses but not all authors agree with this approach; it has been argued that regressing mortality linearly against the percent positive proxies may overestimate the contribution of influenza. Nonetheless, the observation that our estimates are largely consistent with epidemiological expectations regarding the attribution of burden to different age and risk strata, despite the use of the same model structure and virology time series in all cases, adds confidence to our results. It is worth noting that CDC estimates for influenza-attributable mortality for the same season have varied widely according to the study period, model used, resolution of outcomes, and inclusion or not of RSV virology. The most recent estimate from the CDC covering a common period with our study (1997–2007) reported an annual average of 32 805 influenza-attributable deaths with underlying respiratory and circulatory causes, while our study estimated an annual average of 21 861 influenza-attributable deaths using the cardiorespiratory outcome definition during the same period (data not shown). However, the CDC estimates did not account for the circulation of RSV.
Centers for Disease Control estimates from 1990 to 1999 reported that mortality attributable to influenza B was higher than that attributable to influenza A/H3N2 only in one season (1990/1991). However, the present study, which covered the period 1997–2009, estimated that all-age mortality attributable to influenza B was higher than that of A/H3N2 in four seasons (2000/2001, 2002/2003, 2006/2007, and 2008/2009). The contrast between the studies is unsurprising, considering the lower circulation of influenza B in the 1990s compared with the first decade of the twenty-first century. In two seasons (2000/2001 and 2008/2009), the B Yamagata lineage strain circulated, while the B Victoria lineage strain, which appeared in the USA in 2001, circulated in the other two seasons. In these four seasons, between 51% (5610 deaths in 2006/2007) and 95% (11 303 in 2000/2001) of all influenza-attributable deaths were attributable to influenza B, reflecting the higher proportion of circulating influenza B in those seasons.
Since the unusually severe 2003/2004 Fujian A/H3N2 season, the US CDC has conducted national influenza mortality surveillance in children aged <18 years. During the five seasons from 2004/2005 to 2008/2009 for which the CDC surveillance and our study period overlapped, we attributed a seasonal average of 187 influenza deaths for the respiratory broad outcome definition and 203 deaths for the cardiorespiratory outcome definition in children aged <18 years. As expected, this was higher than the corresponding seasonal average of 76 laboratory-confirmed pediatric influenza deaths reported to the CDC surveillance system during this period, likely reflecting under-reporting associated with passive surveillance and its dependency on laboratory confirmation.
The mortality attributable to RSV was substantial in children aged <5 years, with approximately twice as many deaths attributed to RSV compared with influenza. This finding is in broad agreement with the 1990–1999 CDC estimate, which found that RSV-attributable P&I mortality was approximately 3·5-fold higher than influenza-attributable mortality in children aged <5 years. Our study also confirmed the recent recognition of the substantial mortality contribution of RSV in the elderly.
We generated mortality estimates according to risk status, with high risk status defined as the presence of any underlying medical condition known to increase the risk of complications of influenza (e.g., chronic respiratory or cardiovascular diseases). In all but pediatric age groups, persons at high risk had higher influenza-attributable mortality rates than persons at low risk. Underlying health conditions have been previously shown to influence influenza-associated mortality. The greatest difference in mortality between high and low risk individuals was in the 65–74 year age group. There was less impact of risk status in adults aged 75 + years compared with those aged 50–74 years. A possible explanation is that frailty and susceptibility to influenza complications increase with age independently of underlying chronic conditions. The pattern in infants (0–4 years) was the opposite of that in adults, with most influenza-attributable mortality recorded in infants considered to be at low risk. This may be partially explained by the fact that infants have less time than adults to develop a high-risk chronic disease and have it recorded in their medical records.
The study had several limitations. First, mortality data were only available by month, and therefore, for seasons, when there was substantial overlap between the influenza and RSV epidemics, the model may not have ascertained the burden of each pathogen precisely. Second, although the relative distribution of influenza A, influenza B, and RSV is known to vary by age, all-age composite virologic data for influenza and RSV were used to guide the timing of the epidemic in the model. Finally, limitations inherent to indirect statistical approaches based on retrospective data must be considered. Although cyclic terms were included in the model to adjust for the putative influence of non-infectious factors and pathogens other than influenza and RSV on respiratory mortality, lack of specific information on the nature and patterns of other drivers of mortality adds uncertainty to the model estimates due to possible co-variability of these terms especially with the RSV term in the model. We should also note that, as in the analysis of any data series with temporal structure, P-values and confidence intervals are likely to be affected by autocorrelation in the residuals. Previous efforts have tried to circumvent this issue by including autoregressive terms as predictors in the model. However, this approach can blur the interpretation of the remaining regression coefficients, which then account only for the anomalies in the time series.
These updated estimates of influenza and RSV mortality provide new insights into the changing burden of influenza and RSV during the last decade and critically inform public health policy decision-making on strategies to reduce this burden. These data also once again highlight the important burden of influenza in children, older adults, and at-risk groups and reinforce the rationale for public health measures to prevent and control influenza.