¶ Membership of the GLaMOR Collaborating Teams is provided in the Acknowledgments.
LS has provided consultancy services to GlaxoSmithKline (GSK) and served on expert advisory boards for GSK, Roche, Pfizer, Merck, and Novartis. LS and RJT have an ownership interest in Sage Analytica, a consultancy in epidemiology and bioinformatics. DMF has provided consultancy services for GSK, Novartis, and MedImmune relating to influenza epidemiology and vaccine effectiveness and has been supported to attend international influenza meetings. WJP has served on an influenza advisory board for GSK.
Conceived and designed the experiments: LS WJP DMF. Performed the experiments: LS PS RL. Analyzed the data: LS PS RL RJT WJP DMF. Contributed reagents/materials/analysis tools: AWM MDVK. Wrote the first draft of the manuscript: LS RJT WJP MK PS. Contributed to the writing of the manuscript: LS RJT WJP MK DMF PS RL AWM MDVK.
Lone Simonsen and colleagues use a two-stage statistical modeling approach to estimate the global mortality burden of the 2009 influenza pandemic from mortality data obtained from multiple countries.
Assessing the mortality impact of the 2009 influenza A H1N1 virus (H1N1pdm09) is essential for optimizing public health responses to future pandemics. The World Health Organization reported 18,631 laboratory-confirmed pandemic deaths, but the total pandemic mortality burden was substantially higher. We estimated the 2009 pandemic mortality burden through statistical modeling of mortality data from multiple countries.
We obtained weekly virology and underlying cause-of-death mortality time series for 2005–2009 for 20 countries covering ∼35% of the world population. We applied a multivariate linear regression model to estimate pandemic respiratory mortality in each collaborating country. We then used these results plus ten country indicators in a multiple imputation model to project the mortality burden in all world countries. Between 123,000 and 203,000 pandemic respiratory deaths were estimated globally for the last 9 mo of 2009. The majority (62%–85%) were attributed to persons under 65 y of age. We observed a striking regional heterogeneity, with almost 20-fold higher mortality in some countries in the Americas than in Europe. The model attributed 148,000–249,000 respiratory deaths to influenza in an average pre-pandemic season, with only 19% in persons <65 y. Limitations include lack of representation of low-income countries among single-country estimates and an inability to study subsequent pandemic waves (2010–2012).
We estimate that 2009 global pandemic respiratory mortality was ∼10-fold higher than the World Health Organization's laboratory-confirmed mortality count. Although the pandemic mortality estimate was similar in magnitude to that of seasonal influenza, a marked shift toward mortality among persons <65 y of age occurred, so that many more life-years were lost. The burden varied greatly among countries, corroborating early reports of far greater pandemic severity in the Americas than in Australia, New Zealand, and Europe. A collaborative network to collect and analyze mortality and hospitalization surveillance data is needed to rapidly establish the severity of future pandemics.
Every winter, millions of people catch influenza—a viral infection of the airways—and hundreds of thousands of people (mainly elderly individuals) die as a result. These seasonal epidemics occur because small but frequent changes in the influenza virus mean that the immune response produced by infection with one year's virus provides only partial protection against the next year's virus. Influenza viruses also occasionally emerge that are very different. Human populations have virtually no immunity to these new viruses, which can start global epidemics (pandemics) that kill millions of people. The most recent influenza pandemic, which was first recognized in Mexico in March 2009, was caused by the 2009 influenza A H1N1 pandemic (H1N1pdm09) virus. This virus spread rapidly, and on 11 June 2009, the World Health Organization (WHO) declared that an influenza pandemic was underway. H1N1pdm09 caused a mild disease in most people it infected, but by the time WHO announced that the pandemic was over (10 August 2010), there had been 18,632 laboratory-confirmed deaths from H1N1pdm09.
The modest number of laboratory-confirmed H1N1pdm09 deaths has caused commentators to wonder whether the public health response to H1N1pdm09 was excessive. However, as is the case with all influenza epidemics, the true mortality (death) burden from H1N1pdm09 is substantially higher than these figures indicate because only a minority of influenza-related deaths are definitively diagnosed by being confirmed in laboratory. Many influenza-related deaths result from secondary bacterial infections or from exacerbation of preexisting chronic conditions, and are not recorded as related to influenza infection. A more complete assessment of the impact of H1N1pdm09 on mortality is essential for the optimization of public health responses to future pandemics. In this modeling study (the Global Pandemic Mortality [GLaMOR] project), researchers use a two-stage statistical modeling approach to estimate the global mortality burden of the 2009 influenza pandemic from mortality data obtained from multiple countries.
The researchers obtained weekly virology data from the World Health Organization FluNet database and national influenza centers to identify influenza active periods, and obtained weekly national underlying cause-of-death time series for 2005–2009 from collaborators in more than 20 countries (35% of the world's population). They used a multivariate linear regression model to measure the numbers and rates of pandemic influenza respiratory deaths in each of these countries. Then, in the second stage of their analysis, they used a multiple imputation model that took into account country-specific geographical, economic, and health indicators to project the single-country estimates to all world countries. The researchers estimated that between 123,000 and 203,000 pandemic influenza respiratory deaths occurred globally from 1 April through 31 December 2009. Most of these deaths (62%–85%) occurred in people younger than 65 years old. There was a striking regional heterogeneity in deaths, with up to 20-fold higher mortality in Central and South American countries than in European countries. Finally, the model attributed 148,000–249,000 respiratory deaths to influenza in an average pre-pandemic season. Notably, only 19% of these deaths occurred in people younger than 65 years old.
These findings suggest that respiratory mortality from the 2009 influenza pandemic was about 10-fold higher than laboratory-confirmed mortality. The true total mortality burden is likely to be even higher because deaths that occurred late in the winter of 2009–2010 and in later pandemic waves were missed in this analysis, and only pandemic influenza deaths that were recorded as respiratory deaths were included. The lack of single-country estimates from low-income countries may also limit the accuracy of these findings. Importantly, although the researchers' estimates of mortality from H1N1pdm09 and from seasonal influenza were of similar magnitude, the shift towards mortality among younger people means that more life-years were lost during the 2009 influenza pandemic than during an average pre-pandemic influenza season. Although the methods developed by the GLaMOR project can be used to make robust and comparable mortality estimates in future influenza pandemics, the lack of timeliness of such estimates needs to be remedied. One potential remedy, suggest the researchers, would be to establish a collaborative network that analyzes timely hospitalization and/or mortality data provided by sentinel countries. Such a network should be able to provide the rapid and reliable data about the severity of pandemic threats that is needed to guide public health policy decisions.
Please access these websites via the online version of this summary at
The US Centers for Disease Control and Prevention provides information about
The World Health Organization provides information on
Public Health England provides information on
More information for patients about H1N1pdm09 is available through
More information about the
Recurring seasonal influenza epidemics impose a moderate, if variable, mortality burden every year. But when a new human-transmissible influenza virus emerges, the ensuing pandemic can be catastrophic; the 1918 Spanish influenza pandemic, for example, killed approximately 1%–2% of the global population. Understanding the global mortality impact of pandemic influenza—who died, where, and when—is fundamental to understanding how pandemics emerge and evolve, and will help to guide responses to future pandemics. And because so few pandemics have occurred in the modern era, it is essential that each one be studied thoroughly—even if, as was the case with the 2009 influenza A H1N1 pandemic (H1N1pdm09), the catastrophe failed to appear.
As of 31 August 2010 the World Health Organization (WHO) received reports of 18,449 laboratory-confirmed deaths from H1N1pdm09 infection
Statistical methods are therefore used to separate the influenza-attributable fraction of deaths from the background
The majority of deaths from seasonal influenza occur among people aged 65 y or older, but in a pandemic the proportion of deaths among the young increases
The Global Pandemic Mortality (GLaMOR) project aimed to make a conservative estimate of the global H1N1pdm09 mortality burden in 2009 using statistical models applied to mortality, virology, and other available data. The project was funded by WHO, which requested global and regional estimates of H1N1pdm09 influenza deaths for the year 2009; thus, all mentions of pandemic flu mortality refer specifically to deaths that occurred in the last 9 mo of 2009. We invited global collaborators to contribute national mortality data detailed by week, age, and cause of death for 2005 through 2009, at minimum.
Our novel method was inspired by a study that estimated the 1918–1920 global pandemic mortality burden using a two-stage statistical approach
A previous study of global H1N1pdm09 mortality, published in 2012 by Dawood et al.
We obtained weekly virology data from the WHO FluNet
We requested weekly national mortality time series based on the “underlying” cause-of-death determination from 1 January 2005 to 31 December 2009, stratified by at least two age groups (<65 and ≥65 y) and by four International Classification of Diseases–10 (ICD-10)–coded outcomes: all causes, cardiorespiratory (J and I codes), respiratory (J codes), and pneumonia and influenza (codes J10–J18) (
WHO Region (Number of Countries) | Country | Income Level |
Data Years | Outcome |
Detail |
Virology Source | Percent World Population |
South Africa | Upper middle | 2003–2009 | AC,CR,R | Monthly | FluNet | 0.7 | |
Denmark | High | 1998–2009 | AC | Weekly | FluNet | 0.1 | |
Israel | High | 2004–2009 | AC,CR,R |
Weekly | Israel | 0.1 | |
France | High | 1998–2009 | AC,CR,R | Weekly | FluNet | 0.9 | |
Germany | High | 1998–2009 | AC,CR,R | Weekly | Germany | 1.3 | |
Poland | Upper middle | 2003–2009 | AC,CR,R | Weekly | FluNet | 0.4 | |
Romania | Upper middle | 2005–2009 | AC,CR,R | Weekly | EuroFlu | 0.3 | |
Slovenia | High | 2003–2009 | AC,CR,R | Weekly | FluNet | 0.1 | |
Spain | High | 2000–2009 | AC,CR,R | Weekly | Spain | 0.7 | |
UK | High | 2000–2009 | AC,CR,R | Weekly | UK | 0.9 | |
Argentina | Upper middle | 2001–2009 | AC,CR,R | Monthly | FluNet | 0.6 | |
Chile | Upper middle | 2002–2009 | AC,CR,R | Weekly | FluNet | 0.2 | |
Mexico | Upper middle | 2000–2009 | AC,CR,R | Weekly | FluNet | 1.6 | |
US | High | 2000–2009 | AC,CR,R | Weekly | FluNet | 4.6 | |
Australia | High | 2003–2009 | AC,CR,R | Weekly | FluNet | 0.3 | |
China |
Lower middle | 2004–2009 | AC,CR,R | Weekly | China | 19.5 | |
Hong Kong | High | 1999–2009 | AC,CR,R | Weekly | Hong Kong | 0.1 | |
Japan | High | 1998–2009 | AC,CR,R | Weekly | FluNet | 1.9 | |
New Zealand | High | 2000–2009 | AC,CR,R | Weekly | FluNet | 0.1 | |
Republic of Korea | High | 2003–2009 | CR,R | Weekly | FluNet | 0.7 | |
Singapore | High | 2007–2009 | CR,R | Weekly | Singapore, FluNet | 0.1 | |
Brazil | Upper middle | NA | R | Weekly | NA | 2.8 | |
Canada | High | NA | R | Weekly | Canada | 0.5 | |
Peru | Upper middle | NA | R | Monthly | Peru | 0.4 | |
Netherlands |
High | NA | R | Weekly | Netherlands | 0.2 | |
Bangladesh |
Low | NA | R | NA | Bangladesh | 2.1 |
Income level in 2009
Underlying cause of mortality: AC, all cause; CR, cardiorespiratory (ICD-10 I and J codes); R, respiratory (ICD-10 J codes).
Standard request was for age groupings: 0–4, 5–14, 15–44, 45–64, 65–84, and ≥85 y of age.
Did not include influenza with pneumonia.
Data from multiple surveillance settings representing rural and urban areas across China
Respiratory mortality estimated by Netherlands team based on estimated all-cause pandemic mortality
Respiratory mortality estimated by Bangladesh team using a novel method combining virology surveillance and verbal autopsy data
NA, not available.
Country | Exclusions | Summer Definition (Week Numbers) | Secular Model Fit ( |
Full Model Fit ( |
Pandemic Parameter |
Argentina | None | 1–10,45–52 | 0.8290 | 0.9301 | <0.0001 |
Australia | None | 1–18,46–52 | 0.8657 | 0.9098 | 0.1206 |
Chile | None | 1–17,42–52 | 0.7032 | 0.9128 | <0.0001 |
China | None | — | — | — | <0.0001 |
France | Heat wavess | May–Sep | 0.6989 | 0.8830 | 0.3836 |
Germany | Heat wavess | 19–45 | 0.6684 | 0.9196 | 0.2918 |
Hong Kong | None | None | 0.6098 | 0.7266 | <0.0001 |
Israel | 2006 Lebanese war | 15–42 | 0.6832 | 0.9246 | 0.2718 |
Japan | None | 27–40 | 0.7926 | 0.8813 | 0.0527 |
Mexico | None | Jun–Aug | 0.7497 | 0.8500 | <0.0001 |
New Zealand | None | 1–19,43–52 | 0.7122 | 0.8320 | 0.0176 |
Poland | None | 18–41 | 0.7096 | 0.8660 | 0.1634 |
Republic of Korea | None | 23–43 | 0.6739 | 0.8320 | 0.0749 |
Romania | None | 19–45 | 0.8047 | 0.8738 | 0.0054 |
Singapore | None | None | 0.5306 | 0.6676 | 0.3705 |
Slovenia | None | 17–44 | 0.5101 | 0.7270 | 0.4362 |
South Africa | None | 1–14,39–52 | 0.8193 | 0.9254 | <0.0001 |
Spain | Heat waves |
21–39 | 0.6640 | 0.8973 | 0.8243 |
UK | None | NA | 0.7190 | 0.8756 | 0.1505 |
US | None | 23–39 | 0.8086 | 0.9408 | <0.0001 |
Europe experienced severe heat waves in 2003 and 2006; hence, summer weeks with elevated mortality were excluded.
NA, not available.
When choosing which outcome to use as our primary estimate of mortality, we had to make a trade-off between sensitivity and specificity while maintaining sufficient precision. Modeling all-cause mortality data would by definition ensure that all deaths are captured (100% sensitivity), but would sacrifice specificity and therefore precision. At the other extreme, pneumonia and influenza (P&I) is a specific influenza outcome but captures only a fraction of total pandemic deaths.
After much deliberation and advice from the Ad Hoc Advisory Committee on H1N1pdm09 Mortality Estimates
We developed a multivariate linear regression model of influenza burden based on correlations between laboratory surveillance and national mortality data
The GLaMOR Stage 1 model form was:
We computed the pandemic attributions as the sums of the products of the pandemic model parameter β9 multiplied by the H1N1pdm09 positive count. When negative parameter values were obtained, mortality burden estimates were set to zero (as negative burden is not biologically meaningful). The confidence intervals were derived from uncertainty on the pandemic model parameter estimate β9. We determined the Stage 1 95% confidence intervals from the standard error on the H1N1pdm09 parameter estimate. We did not address autocorrelation in the residuals, and therefore the confidence intervals are narrow. We evaluated the model fit by three criteria: (1) the adjusted
We explored four strategies to project our single-country estimates to the rest of the world before settling on multiple imputation, a Monte Carlo method that imputes values to missing data points and is often used to supply missing values in survey and census data
The survey method was a direct extrapolation of the average Stage 1 pandemic mortality rates using bootstrapping, in which the average excess mortality rate and upper and lower 95% confidence levels were used to calculate the global numbers and rates of pandemic deaths by age group and WHO region.
The limitation of this method is that it assumes that the average pandemic mortality experience in the Stage 1 countries is representative of the experience in all countries, and gives only a global estimate (no country- or region-specific estimates).
The gross national income (GNI)/latitude method was derived from the method Murray et al. employed to estimate the 1918 global pandemic burden
The limitation of this method is that the model assumes that GNI and latitude are sufficient proxies for the many variables that influence influenza mortality, and that mortality and the reasons for its variability in Stage 1 countries are representative for all countries. Importantly, there was an assumption that the relationship between GNI and excess mortality is exponential, so that the method yields very high mortality rates for low-income countries (e.g., countries in Africa).
The matching method was devised by the GLaMOR team. It obtains the missing data points by matching (as closely as possible) Stage 1 countries to non–Stage 1 countries based on a set of country indicators. It involves two steps: (1) a data creation step using the matching approach, and (2) a data analysis step where a hierarchical linear random effects model is used to provide a single estimate for each country.
The data creation step involves calculation of multiple estimates per country based on the indicators listed in
Indicator Number | Indicator |
1 | WHO region (Africa, the Americas, Eastern Mediterranean, Europe, South-East Asia, Western Pacific) |
2 | Age group all-cause mortality rates (0–14, 15–59, 60+ y) |
3 | Physician density (per 10,000 population) |
4 | Obesity (percent with body mass index >30 kg/m2) |
5 | Population density (per km2) |
6 | Major infectious diseases (percent HIV and percent tuberculosis prevalence) |
7 | GNI per capita (US dollars) |
8 | Rural population (percent) |
9 | Population age structure: percent <15 y and >60 y |
10 | Latitude (absolute value) |
BMI, body mass index; Resp. disease, respiratory disease.
The data analysis step of the matching method is similar to that of the multiple imputation method (below), with the same hierarchical linear random effects model but with the imputed datasets replaced by the matched datasets.
We chose this method to make our Stage 2 extrapolations. Like the matching method, the multiple imputation method involves two steps, a data creation step followed by a hierarchical regression modeling step to project the burden in all world countries.
In the data creation step, we used statistical correlations between the same set of country indicators that we used in the matching method (
In the analysis step, we applied a hierarchical linear random effects regression model
Y = imputed individual rate
μ
ε
Estimated rates were then given by:
World = β0
WHO region = β0+β
Country = β0+β
Statistical 95% confidence intervals for these estimates were calculated using standard methods. We performed the imputation procedures with the Amelia II software package
We evaluated the performance of each of the four candidate Stage 2 methods. We rejected the survey method because the results could not show regional variation, a major disadvantage given that the Stage 1 results showed considerable variation among the regions. We rejected the GNI/latitude method for two reasons. First, the estimates for our validation countries did not match the validation estimates the method produced (
Country | Source | Estimate | 95% CI | Projection Method | |||
Matching | Multiple Imputation | GNI/Latitude | Survey | ||||
Bangladesh | Verbal autopsy |
4.0 | NA | 2.1 | 4.0* | 0.4 | 2.0 |
Brazil | GLaMOR; Serfling | 4.3 | NA | 2.8 | 3.5* | 0.1 | 2.0 |
Canada | Lab-confirmed deaths |
2.1 | 1.6–2.6 | 1.8* | 3.1 | 1.0 | 2.0 |
Netherlands | Poisson regression |
0.9 | 0.3–1.5 | 0.9* | 0.9* | 0.7 | 2.0 |
Peru | GLaMOR; Serfling | 6.8 | NA | 2.5 | 3.6* | 0.1 | 2.0 |
Bangladesh | Verbal autopsy |
3.0 | 1.6–3 | 1.2 | 1.8* | 2.6 | 1.2 |
Brazil | GLaMOR; Serfling | 3.1 | NA | 2.0 | 2.5* | 0.6 | 1.2 |
Canada | Lab-confirmed deaths |
1.1 | 0.9–1.3 | 1.1* | 2.0 | 0.7 | 1.2 |
Netherlands | Poisson regression |
0.2 | 0.2–0.4 | 0.5* | 0.9 | 0.6 | 1.2 |
Peru | GLaMOR; Serfling | 5.2 | NA | 1.4 | 2.5* | 0.5 | 1.2 |
The asterisk indicates for each country and category which of the four tested Stage 2 methods was in best agreement.
NA, not available.
After eliminating the first two methods, only the matching and the multiple imputation methods remained. We first compared the multiple imputation and matching data creation steps for a single country in each WHO region. The results (
Eastern Med, Eastern Mediterranean; Imp, multiple imputation method; Match, matching method.
Reliability, or internal consistency, is an important consideration. The reliability coefficient ranges from 0 to 1 (zero indicates no systematic effect)
We assessed the difference between the Stage 2 single-country estimates obtained with the matching and multiple imputation methods and the original Stage 1 estimates. The differences, expressed as standard deviations, were smaller for the multiple imputation method.
We compared the Stage 2 estimates to country reports of the number of laboratory-confirmed H1N1pdm09 deaths (
We compared the highest and lowest 20% of national estimates (both for all ages and for persons <65 y of age) derived from the multiple imputation and matching methods. For all ages, the matching method distributes the highest burden over Africa and the Americas, with 88% of the countries in the highest quintile in the Americas and 18% in Africa. The multiple imputation method distributes the highest burden not only over the Americas and Africa, but also over South-East Asia. As we have no Stage 1 estimates for South-East-Asia and only one for Africa, we cannot be sure which method is better, and this test was inconclusive.
Five of the 26 Stage 1 countries (Bangladesh, Brazil, Canada, Peru, and the Netherlands) could not be analyzed with the GLaMOR Stage 1 model for various reasons. Country collaborators in Bangladesh and Canada had each generated estimates using their own methods, as national vital statistics could not be provided. For Brazil and Peru the virological data did not align with pneumonia pandemic mortality spikes (see also Schuck-Paim et al.
The validation tests for the most part yielded only small differences between the multiple imputation and matching methods. We selected the multiple imputation method because it produced estimates that were more consistent with those generated by country collaborators in the five GLaMOR validation countries than the estimates produced by the matching method, and because multiple imputation is an established method while the matching method was one we developed. The choice of multiple imputation had two important consequences: it resulted in systematically higher country mortality rates compared to the matching method, and it placed a higher burden in people aged 65 y and over, especially in Asia and Africa.
We computed the average global seasonal influenza burden (type A plus type B) for the years immediately prior to the pandemic. Specifically, we calculated the average pre-pandemic seasonal influenza mortality for each Stage 1 country using model parameter values from each pre-pandemic season, then projected these estimates to global and regional values using our Stage 2 multiple imputation procedure (
Because of large background mortality in the elderly, it was difficult to measure all-age influenza-related mortality with precision in lower-burden countries. For example, in some European countries the H1N1pdm09 mortality impact was so subtle that the model applied to all-age time series produced a negative point estimate for the H1N1pdm09 burden, with confidence intervals that at times excluded the “ground truth” minimum of the reported number of laboratory-confirmed H1N1pdm09 deaths from that country. Modeling the data for the <65-y age group, however, almost always resulted in estimated H1N1pdm09 mortality rates that were comparable to, or far higher than, the laboratory-confirmed mortality count.
We therefore elected to generate Stage 2 all-age burden projections in two ways: one based on Stage 2 all-age estimates, and the other based on the <65-y Stage 2 estimates, which we proportionally projected to all ages using data from laboratory surveillance indicating that 85% of confirmed H1N1pdm09 deaths occurred in the younger group (
We investigated the sensitivity of our global Stage 2 burden estimates to changes in the Stage 1 sample by successively removing one Stage 1 country from the Stage 2 input dataset and rerunning the Stage 2 model. Because the range from this analysis was always wider than the 95% confidence intervals derived from the Stage 1 and 2 statistical procedures, we chose to report this range as a more realistic view of the uncertainty (see
Region | <65 y, Stage 1 | All Ages, Stage 1 | All Ages (from Stage 1 <65 y) |
|||
Estimate | Range |
Estimate | Range |
Estimate | Range |
|
World | 117,130 | 104,450–132,080 | 188,660 | 175,280–203,250 | 137,800 | 122,882–155,388 |
Africa | 17,922 | 15,408–21,172 | 25,476 | 22,431–28,447 | 21,085 | 18,127–24,908 |
Eastern Mediterranean | 11,108 | 10,092–12,564 | 14,911 | 13,592–17,718 | 13,068 | 11,873–14,781 |
Europe | 8,463 | 6,686–8,894 | 11,223 | 10,557–13,883 | 9,956 | 7,866–10,464 |
Americas | 22,975 | 20,768–28,328 | 35,298 | 29,107–38,461 | 27,029 | 24,433–33,327 |
South-East Asia | 30,412 | 25,829–36,861 | 73,449 | 50,012–83,346 | 35,779 | 30,387–43,366 |
Western Pacific | 20,179 | 17,023–25,259 | 30,554 | 28,427–41,862 | 23,740 | 20,027–29,716 |
Calculated assuming 85% of all deaths occurred among persons <65 y, as was the case with laboratory-confirmed pandemic deaths identified in seven countries; see
The confidence range was derived from a sensitivity analysis in which we successively removed one Stage 1 county at a time from the Stage 2 input set and recalculated the global and regional burden.
The GLaMOR Stage 1 countries experienced one to three pandemic waves during 2009. Most H1N1pdm09 deaths occurred in winter months: November–December in the northern hemisphere and July–August in the southern hemisphere. Several Asian countries experienced an H3N2 epidemic in the months immediately before their major H1N1pdm09 wave.
Data are grouped into four geographical regions.
The various outcomes and age stratifications each provided a different balance between sensitivity and specificity. In high-burden countries such as Mexico and Argentina, the pandemic impact could be modeled with precision even for all-age time series of all-cause and cardiorespiratory mortality outcomes, as evidenced by tight 95% confidence intervals and agreement with a published Mexico study using a different Serfling regression modeling approach
In Mexico, a substantial H1N1pdm09 respiratory mortality burden (red areas above gray background mortality) occurred among children, young adults, and middle-aged persons (<65 y) of age but not among seniors (≥65 y). In France, however, there was a far less dramatic pandemic impact that, despite the similar population size, was captured only in the <65-y age group model. Seasonal influenza burden (blue areas) was also generated by the Stage 1 model. The vertical black line represents the start of the pandemic.
Overall, Stage 1 respiratory mortality rates were consistently higher in the Americas, with the highest measurements in Central and South American countries. South Africa's pandemic burden was moderate and on par with that of the US and China, suggesting that Africa may have experienced a lower pandemic burden than Central and South America. In Europe, the pandemic burden was generally low and on par with national numbers of laboratory-confirmed H1N1pdm09 deaths. Spain, France, and Germany averaged a rate of just 0.3/100,000, while Romania, the lowest-income European country included in this study, felt an approximately 6-fold greater impact. In South and Central America, however, we did not find a consistent relationship between H1N1pdm09 mortality and country income group: Argentina and Mexico had particularly high pandemic death rates (∼5/100,000), whereas Chile, a country with a similar economy, had a pandemic death rate that was more than 3-fold lower.
Across all Stage 1 countries, the model placed an average of 66% of respiratory pandemic deaths in persons <65 y. However, that proportion varied widely, from 100% in several European countries, to 70%–90% in Argentina and Mexico, to less than 10% in Hong Kong and Japan. Our Stage 1 estimates for seniors (≥65 y) were considerably more uncertain than those for the <65-y age group; this was due in part to the difficulty in precisely measuring the small H1N1pdm09 burden in seniors against their high background mortality, and partly because Hong Kong and Japan were outliers with a far higher burden among seniors than all the other countries.
In the few high-burden countries where we could measure Stage 1 all-cause mortality with confidence (e.g., Argentina and Mexico), the ratio of all-cause to respiratory mortality ranged from 1.6 to 2.3.
Using the Stage 1 model results for all ages (sum of <65-y and ≥65-y age group estimates) as input, the Stage 2 model projected a global pandemic respiratory mortality during 2009 of 189,000 (range, 175,000–203,000) deaths. These estimates correspond to an incidence rate of 2.77 (range, 2.57–2.98) per 100,000 population. Globally, 62% of these deaths were estimated to occur in persons <65 y, varying from 41% to 75% among WHO regions.
Because the pandemic virus caused considerable excess mortality in the <65-y age group, Stage 1 estimates for this age group were more reliable. To take advantage of that greater reliability, we computed the Stage 2 global and regional projections based on these estimates alone, by adjusting the <65-y estimates to all ages. To accomplish this we compiled data on the age distribution of laboratory-confirmed H1N1pdm09 deaths from mortality surveillance efforts in seven countries; these data indicated that an average of 85% of all pandemic deaths occurred in persons <65 y (
We found substantial regional heterogeneity in H1N1pdm09 mortality rates (
(A) Under 65 y and (B) all ages. Numbers in map legend are pandemic mortality rates per 100,000 persons.
In the sensitivity analysis in which one Stage 1 country at a time was removed from the Stage 2 analysis, the global H1N1pdm09 mortality estimates were stable, with all-age point estimates ranging from 2.6 to 3.0 per 100,000 population from all-ages data (
The Stage 2 model was run multiple times, each time removing one Stage 1 country, for (A) all ages and (B) <65 y. The global estimates (black diamonds) were relatively stable, but some regions were sensitive to the removal of individual countries.
We generated an estimate of the average global seasonal influenza mortality burden, based on Stage 1 average seasonal influenza attributions across the pre-pandemic period. The Stage 2 multiple imputation method projected a global seasonal influenza burden of 210,000 influenza-related respiratory deaths per influenza season; of these only 19% occurred in persons <65 y of age. The range of seasonal influenza mortality estimates in the sensitivity analysis (148,000 to 249,000 deaths per year) was again wider than the 95% confidence intervals; thus we believe the range to be a better measure of uncertainty. The effect of removing individual Stage 1 countries on the seasonal influenza mortality estimates was more pronounced than for the pandemic estimates (
East.Med, Eastern Mediterranean.
Our analysis suggests that between 123,000 and 203,000 H1N1pdm09 respiratory deaths occurred globally in 2009. This range is derived from the all-age burden computed in two different ways: the higher figure is the upper bound of the all-age Stage 2 projection as given by the sensitivity analysis, while the lower estimate is the lower sensitivity analysis bound of the more reliable <65-y age group estimate projected to all ages.
That range places H1N1pdm09 mortality below that of previous influenza pandemics, which varied from ∼1 million deaths in 1968 to ∼50 million deaths in 1918
When the H1N1pdm09 virus first emerged, early assessments of the threat level were mixed. Events in Mexico and Argentina in April and May 2009 suggested a severe pandemic on par with the 1957 pandemic or worse
Many factors might have contributed to the regional differences in H1N1pdm09 mortality impact, including the previous influenza exposure history of the population, use of antiviral drugs, the number and duration of pandemic waves during 2009, influenza vaccination coverage in preceding seasons, access to intensive care, and use of public health mitigation strategies. Use of pandemic vaccine is not on the list as it became available too late to play a role in 2009
Our global estimates were in reasonable agreement with those of Dawood et al.
But at the regional level, the two approaches produced entirely different patterns (
GLaMOR all-age respiratory mortality estimated directly from all-age multiple imputation (open circles) and by proportional extrapolation of the <65-y age group estimate to all ages using the age distribution of laboratory-confirmed mortality surveillance (black circles), compared to estimates by Dawood et al.
Neither Dawood et al. nor our study had many data from Africa and South-East Asia, however, so what actually happened in these largely tropical regions remains unclear. However, a recent study demonstrated that the H1N1pdm09 mortality impact was far greater in temperate and wealthy southern regions of Brazil than in tropical and less-wealthy northern regions
The H1N1pdm09 burden estimates from a few GLaMOR participating countries have been published. As shown in
Asterisks indicate significance at the
Asterisks indicate significance at the
Although our conservative range of global respiratory mortality estimates is an order of magnitude greater than the reported global number of WHO laboratory-confirmed deaths, it likely substantially understated the total H1N1pdm09 mortality burden. First, our study missed deaths that occurred late in the 2009–2010 winter as well as those occurring in later pandemic waves. For example, one-third of Germany's laboratory-confirmed first-wave deaths occurred in early 2010, while substantial waves of H1N1pmd09 mortality were observed later in the UK in 2010–2011
Our approach had several strengths. Because the analysis was based on the pandemic “excess” mortality that actually occurred in 20 Stage 1 countries in 2009, we were able to map large and important regional differences in the H1N1pdm09 mortality burden that had not been captured in a previous study
We recognize several caveats of our study, however. First, we lacked good representation of low-income countries and countries in South-East Asia, the Eastern Mediterranean, and Africa. Second, we were unable to explain the substantial pandemic mortality attributions among seniors that we and others measured in Hong Kong
Health care policy decisions depend on reliable and timely data whereby the risks and cost-effectiveness of interventions can be evaluated. In the GLaMOR study we developed methods whereby we can make robust and comparable mortality estimates in any future pandemic. But the lack of timeliness of such reports must be remedied. Ideally, a set of sentinel countries with timely hospitalization and/or mortality data could form a global sentinel system measuring severity, provided a common protocol was in place to allow comparisons across settings. EuroMOMO (European Monitoring of Excess Mortality for Public Health Action), which collects timely age-specific all-cause mortality data in European countries, is a big step in the right direction
(TIF)
(TIF)
(DOCX)
(DOCX)
(DOCX)
(DOCX)
This work would not have been possible without the generosity of the GLaMOR Collaborating Teams.
The authors also thank the Ad Hoc Advisory Committee on H1N1pdm09 Mortality Estimates: Isabelle Bonmarin, Mark Chen I-Cheng, Cheryl Cohen, Benjamin J. Cowling, Jean-Claude Desenclos, David N. Durrheim, Luzhao Feng, Neil M. Ferguson, Peter G. Grove, Bryan Grenfell, El Bushra Hassan, Pernille Jorgensen, Francis Kasolo, Gabriel Leung, Marc Lipsitch, Fatima Marinho, Colin Mathers, Anne Mazick, Kåre Mølbak, Anthony Mounts, Angus Nicoll, Yasushi Ohkusa, Otavio Oliva, Richard Pebody, Caterina Rizzo, Colin Russell, David Shay, Kumnuan Ungchusak, C. C. van den Wijngaard, Maria D. Van Kerkhove, Cecile Viboud, Sirenda Vong, Lara Wolfson, Joseph T. Wu, and Hongjie Yu.
Thanks also to the European Centre for Disease Prevention and Control (Angus Nicoll), the Pan American Health Organization/WHO Regional Office for the Americas (Fatima Marinho), the WHO Regional Office for Europe (Caroline Brown and Pernille Jørgensen) for facilitating access to national datasets in the Americas and Europe, and the WHO headquarters in Geneva(Julia Fitzner) for providing access to the FluNET data; to Lewis Kim, Megan McDonough, and Danae Spencer of the George Washington University School of Public Health and Health Services; and to Francois Schellevis and Liana Martirosyan at the Netherlands Institute for Health Services Research for expert editorial and technical assistance.
We thank the public health institutes in local governments of Japan, Germany, Spain, Canada, Hong Kong, and the Netherlands for providing additional laboratory data.
And finally we warmly thank the awesome researchers with the Multinational Influenza Seasonal Mortality Study Group influenza modeling network at the US National Institutes of Health Fogarty International Center, especially Cecile Viboud, Gerardo Chowell, Vivek Charu, and Cynthia Schuck, who worked alongside the GLaMOR core team to develop the methodology. Without their generous support this project would not have been possible.
Global Pandemic Mortality
gross national income
2009 influenza A H1N1 pandemic
International Classification of Diseases–10
World Health Organization