This is the first study to provide a systematic comparison of the age-specific mortality burden of the 2009–10 A/H1N1 pandemic with the historical A/H3N2 pandemic, based on detailed comparison of mortality outcomes and modeling approaches. Our burden estimates are based on a rigorous model selection approach considering several indicators of influenza activity. We estimate that the first season of circulation of the novel A/H1N1 pandemic virus in France (May 2009-May 2010) was associated with 0.98 excess respiratory deaths (95% CI: 0.20; 1.9) per 100,000. Relative to seasonal influenza, individuals aged 5–24 yrs were the most severely affected by the pandemic and experienced excess respiratory death rates 10.6 fold higher than those of seasonal epidemics. In contrast, the pandemic mortality burden in seniors over 65 years old was 3.8 fold lower than that of typical inter-pandemic seasons. The 2009–10 A/H1N1 pandemic was substantially milder than the A/H3N2 pandemic in France, consistently across all age groups and mortality indicators.
The French authorities identified 264 laboratory-confirmed pandemic A/H1N1 influenza deaths between July 1st
, 2009 and February 28, 2010 
and 349 deaths listing an influenza code on the death certificate 
. Our relatively conservative estimate of 613 pandemic excess respiratory deaths suggests that nearly 1 in 2 influenza-related deaths was laboratory confirmed, in a period where testing was considerably strengthened. For comparison purpose, a Mexican study suggested that only 1 in 7 influenza-related deaths were captured in laboratory-confirmed surveillance 
. In France, the proportion of deaths occurring among seniors over 65 yrs was 26% based on laboratory-confirmed deaths 
, 44% based on deaths specifically coded as influenza 
and 68% based on excess mortality models. The age distribution difference could be explained by a lesser propensity to confirm and diagnose influenza-related deaths among seniors than among younger individuals, as was reported for Mexico 
. Alternatively, we could have overestimated excess respiratory death rates in seniors. However, a rise in P&I and respiratory mortality in this age group coincided with a period of intense pandemic activity in November-December 2009, lending validity to our estimates. In addition, we obtained a similar age distribution of influenza-related pandemic deaths by the Serfling approach (65% of excess respiratory deaths among seniors).
Overall, the mortality burden of the 2009–10 A/H1N1 pandemic in France was particularly mild, relative to the impact of seasonal influenza, except in the 5–44 yrs age group. Older populations did not experience significant excess cardio-respiratory and all cause mortality coinciding with the period of intense pandemic A/H1N1 activity. Most remarkably, the only age group to experience significant all-cause excess mortality rates was the 5–24 yrs, while all-cause estimates were negative in all other age groups, consistent with mortality patterns in other European countries 
. We therefore consider that P&I and respiratory deaths are the most reliable outcomes to model the age-specific mortality burden of the A/H1N1 pandemic in Europe and note that previously published all-cause mortality estimates from high-income countries in North America, Europe, and Asia may lack precision 
Our data confirm that the main waves of the A/H3N2 and A/H1N1 pandemics in France were characterized by mortality age shifts, with a significant increase in the proportion of excess respiratory deaths among people under 65 years during pandemics, relative to inter-pandemic seasons. Our results and those of others reinforce one of the signature features of influenza pandemics – a younger than usual age distribution of influenza-related deaths 
. In pandemic situations, a more appropriate estimation of disease burden may be the years of life lost approach, which integrates the age distribution of deaths with excess mortality estimates 
. We applied the years of life lost approach to French data, revealing a 2–2.6 fold higher impact for the 2009 pandemic than for seasonal epidemics in people under 45 yrs, corroborating findings from other settings (, 
Year of Life Lost associated with the 2009–10 A/H1N1 pandemic, the 1969–70 A/H3N2 pandemic, and seasonal epidemics, by mortality outcome.
There are several limitations to our study. First, we developed a Poisson regression model linking mortality with influenza activity data, which is a more specific approach to estimate influenza-related mortality than the traditional Serfling method 
. However, this approach may be less widely applicable than the Serfling method in that it requires several years of reliable weekly viral activity data collected with systematic surveillance criteria, which is not always the case in pandemic situations 
. Also, we obtained negative 2009–10 pandemic estimates for less specific outcomes in individuals over 45 yrs by the Poisson approach, while the Serfling method gave positive estimates, in particular for cardio-respiratory deaths. Another caveat relates to the use of ILI as a proxy of influenza activity in Poisson models. Sensitivity analyses identified ILI as the best covariate to model respiratory deaths, over any indicator relying on influenza virus surveillance data. The poorer statistical fit of models incorporating the influenza virus surveillance data, even after standardization, smoothing, and allowing for lags, could be related to age biases in influenza testing or lack of information on circulation of other respiratory viruses. Specifically, if the majority of virus specimens were collected from children, weekly viral surveillance activity could be asynchronous with weekly senior mortality rates and decrease model fit. Our models driven by ILI, which were selected based on statistical grounds, may not be the most biologically relevant in that they may include the contribution of various pathogens and overestimate the contribution of influenza. However, ILI data do not introduce an age bias if all age groups are represented and the probability of influenza infection given ILI is similar across age groups 
. Most importantly, models driven by viral activity data suggest a similar pandemic mortality burden as models driven by ILI, with an excess mortality rate estimated at 0.93 per 100,000.
A third limitation relates to the fact that we did not include weekly indicators of respiratory syncytial virus (RSV) activity and other pathogens proxies in our models. Lack of RSV information could explain why our mortality estimates for seasonal influenza lack significance in children under 4 years, an age group where RSV predominates 
. RSV-coded deaths were concentrated in January-March 2010 in France, suggesting that RSV activity did not overlap with the period of intense pandemic A/H1N1 virus activity in late fall. It is possible however that RSV circulation was displaced by pandemic activity 
, potentially affecting our estimate of baseline mortality in younger age groups. Similarly, the contribution of other seasonal factors such Streptococcus Pneumoniae and cold spells would be more important in winter, several weeks and months after pandemic activity had peaked 
. Overall, many countries lack age-specific information on various respiratory pathogen activity; such data will be useful to improve excess mortality models in the future 
A final caveat of our study is the lack of adjustment on time trends in health care and treatment, which matter for comparison of mortality rates associated with recent and historical pandemics. During the historical A/H3N2 pandemic, pediatric intensive care units did not exist in some countries and some of the children who survived in 2009 would have probably succumbed during the earlier pandemic. Treatment of secondary bacterial infections 
and complications has greatly improved since the 1960 s, together with widespread use of antivirals in the community and hospital settings, and occasional use of extracorporeal membrane oxygenation. Although our analyses did not adjust for these factors, the order of magnitude difference between mortality estimates for the two pandemics suggests factors that go beyond healthcare and treatment. The substantial severity of the A/H3N2 pandemic in France, relative to the 2009 A/H1N1 pandemic, could be explained by differences in virus virulence or prior immunity. The burden of the 2009 pandemic was likely mitigated by substantial cross protective immunity with previously circulating A/H1N1 viruses 
and/or cross-subtypic immunity from previous exposure to A/H3N2. By contrast, immunity from vaccination was minimal in the first wave of the 2009 pandemic in most countries due to delays in vaccine production and delivery. Pandemic vaccination was initiated on November 2nd
2009 in France, just 3 weeks before peak viral activity, and only 7.9% of the population was vaccinated by February 2010 
Important between-country variation in the mortality burden of historical pandemics has been reported 
. Our study reveals the particularly high excess mortality impact of the A/H3N2 pandemic in France, relative to more recent epidemic or pandemic seasons. Previous work has highlighted the severity of the A/H3N2 pandemic in Western Europe, as compared with the Americas 
. To gauge geographical variation in 2009 pandemic mortality burden, we reviewed published national studies providing age-specific mortality estimates for the 2009–10 A/H1N1 season (). Despite between-country variations in absolute estimates, most countries reported lower all-age excess mortality burden for the 2009 pandemic than for recent seasonal epidemics. School-age children and young adults experienced atypical mortality increase in most countries, including the UK, US and Mexico 
, consistent with our data. Similar results were found by pooling all-cause mortality data from eight European countries or regions 
. In contrast to other European studies, the Netherlands reported unusually high pandemic burden in children under 4 yrs 
. All countries reported lower than expected mortality burden among seniors, with a ratio of pandemic to epidemic excess mortality rates ranging from 0.16–0.34 across 6 countries on 3 continents. We note however that of the 7 studies providing excess mortality estimates based on vital statistics 
, 5 studies focus on all-cause mortality. This is unfortunate because our study suggests that all-cause mortality may not provide the most precise estimates of 2009 pandemic burden, especially in Europe where the pandemic was particularly mild. Moreover, not all studies provide 95% confidence limits on the estimates, and methodologies differ between studies. Finally, little information on the mortality burden of past and recent pandemics is available from low-income countries of Asia and Africa, which is a key area for future research.
Multinational comparison of excess mortality rates and age distribution of deaths associated with the 2009 A/H1N1 pandemic and seasonal epidemics, based on a literature review.
Our study adds to the growing body of evidence on the burden of the 2009 A/H1N1 pandemic and suggests a relatively low excess mortality impact in France compared to seasonal influenza, except in school-age children and young adults. Our results suggest that it is important to analyze cause-specific mortality outcomes, such as respiratory deaths, to accurately capture the burden of influenza in mild seasons, such as the 2009 pandemic. All-cause mortality models provided negative estimates for the pandemic in France, which is clearly imprecise as at least 264 laboratory-confirmed deaths are directly attributable to pandemic influenza A/H1N1. Estimates from many more settings, including low and middle income countries, are needed before the full spectrum of geographical variation in pandemic mortality burden can be established and the socio-economic and population determinants of disease burden are fully elucidated.