Revisiting influenza deaths estimates--Learning from the H1N1 pandemic

A major contributing factor to some of the criticisms of the management of the H1N1 pandemic was that its mortality impact was considerably less than had been predicted. Pre-pandemic global estimates of death were often in the millions while the ultimate death toll has been reported in the thousands. The large discrepancy in the predicted and reported death tolls suggests that there is a potential problem in how influenza deaths are estimated, counted and compared. The impetus for developing a coordinated and robust response to an influenza pandemic was largely driven by estimates of its potential death toll. The WHO predicted that the next influenza pandemic could result in 2–7 million deaths in a ‘best case’ scenario.1 Other analyses suggested even higher death rates. An analysis based on vital registry data from 1918 to 1919 predicted that if the next pandemic behaved like the 1918 pandemic, there would be ∼62 million deaths.2 Another estimate by a leading expert suggested that there could be 180–360 million deaths.3 According to the Canadian influenza pandemic plan, the lower bound of the 95% confidence interval for estimated Canadian deaths from a pandemic with a 35% attack rate was 24 603 (Table 1).4 Sample of some quoted estimates of death from next influenza pandemic, seasonal influenza death estimates and reported deaths from H1N1 Sample of some quoted estimates of death from next influenza pandemic, seasonal influenza death estimates and reported deaths from H1N1 Yet, at the completion of the pandemic, the worldwide death toll was 18 156, the Canadian death toll 428.5,6 In other words, the most conservative estimates of Canadian deaths in one scenario was approximately 6000 greater than the worldwide death toll. What is further perplexing is how the death toll from a pandemic strain of influenza has seemingly caused far fewer deaths than that caused by seasonal influenza. The estimates of yearly deaths from seasonal influenza range from 250 000 to 500 000 globally.7 A novel pandemic influenza strain therefore killed 1/10th to 1/20th the number of individuals as seasonal influenza does worldwide. The discrepancy in predicted deaths and reported deaths is likely a combination of three factors. The first is that for clinical, environmental or biological reasons, H1N1 actually caused that many fewer deaths. Given that a percentage of the population may have been exposed to previously circulating H1N1 strains, these individuals may have had some latent immunity to the current virus. Those with this immunity would have been the elderly, typically those most at risk of death from seasonal influenza viruses. Also, in many countries public health measures, including mass immunization, were implemented which may have mitigated the death toll of the virus and medical care may have prevented additional deaths, particularly in the critically ill patient. The second possibility is that the death rates from H1N1 have somehow been underestimated. The third possibility is that the death rates from seasonal influenza and perhaps the past pandemics have been overestimated. While there has been considerable focus on biological/environmental reasons for why the pandemic was so mild, it is the difficulties with determining the death tolls from influenza that deserve further attention. The choice of methods for counting deaths from an exposure can be controversial. Different methodologies exist some of which err on the side of conservative estimates designed to identify deaths definitively caused by an exposure at the risk of missing cases that are attributable to the exposure. Other approaches try to capture all potentially caused deaths at the risk of capturing some deaths, which were only marginally attributable or not at all attributable to the cause. These methodologies can result in huge variations in estimates. Attempts to estimate influenza deaths have had to address these challenges. Counting reported laboratory positive influenza patients who subsequently die as a direct consequence of the infection would result in a gross underestimation of the true death toll. First, very few people are tested for influenza. Second direct influenza deaths are only a small fraction of all deaths caused by the virus. One Canadian analysis reported that confirmed laboratory positive deaths only made up 8% of model estimated total influenza-related mortality.8 Secondary pneumonias and cardiovascular complications that follow infections are responsible for a large proportion of influenza-related deaths. And, even if tested, in many of these instances, an individual may be influenza negative at the time that they develop the complication. How should these missed deaths be calculated? The standard methodology has been to calculate excess mortality using ecological analyses in which observed death rates are compared to a model prediction of what total or cause-specific mortality would have been in the absence of influenza i.e. a statistical baseline. All excess deaths occurring when there is evidence that influenza is circulating are then attributed to influenza. In examining these analyses, the spikes in mortality coincide temporally with influenza season. However, these analyses are susceptible to important limitations. One is the ecological fallacy. Even though there are increased deaths during a season where there are an increased number of influenza cases, we cannot know for certain if the excess deaths are occurring in individuals who had suffered from influenza. These analyses are also susceptible to confounding. While most analyses adjust for confounding by season and trends in mortality rates, what is more difficult to adjust for is that the conditions that contribute to the spread of influenza would also be the same conditions that can contribute to the spread of other pathogens. Therefore, at least some of the excess mortality may be due to the circulation of concomitant respiratory viruses that are not accounted for in the analysis. How were H1N1 deaths calculated? In most instances, these were laboratory confirmed deaths. There is a huge discrepancy between how these deaths were calculated and how seasonal influenza deaths are calculated. The H1N1 deaths do not include the laboratory negative secondary pneumonias and many of the cardiovascular deaths that would be captured in the seasonal flu mortality data. In the USA, the Centers for Disease Control estimated that for every recognized hospitalized case of H1N1, 2.7 cases were missed.9 Based on this multiplier, they dramatically adjusted upwards their estimates of death from H1N1. Another American analysis of H1N1 deaths attempted to count the deaths in a similar manner to how seasonal flu deaths are counted. The analysis found that the number of life-years lost at its lowest estimates is similar to seasonal influenza and its highest estimates are similar to the 1968 pandemic.10 Comparing the laboratory-confirmed death tolls from H1N1 with the seasonal influenza death estimates is inappropriate. Most likely the H1N1 death toll is a significant underestimate and the WHO has emphasized that it is inappropriate to make the comparisons with seasonal influenza deaths. However, it is important to recognize that the likelihood of not having tested a person for H1N1 is much less than the likelihood of not testing someone for seasonal influenza given how much more comprehensive laboratory viral surveillance was during the outbreak. Therefore, the possibility exists that the excess mortality estimates of seasonal influenza deaths as well as the estimates of deaths from previous influenza pandemics, which are partly based on excess mortality calculations, may be significant overestimates. A key lesson, therefore, from the H1N1 pandemic is that these estimates may have significant limitations. Re-evaluating these calculations should be a focus of any examination of the pandemic response. Dr Wilson is supported by a Canada Research Chair in public health policy. Conflicts of interest: K.W. has relationships with the World Health Organization and Public Health agency of Canada that might have an interest in the submitted work in the previous 3 years and K.W. has no non-financial interests that may be relevant to the submitted work. The documented worldwide mortality from H1N1 is 50–100 times lower than conservative pre-pandemic predictions and substantially lower than seasonal influenza deaths. Documented H1N1 deaths likely substantially underestimate total H1N1 deaths by not including missed cases and death from secondary causes. Excess mortality estimates of deaths from previous pandemics and from seasonal influenza are susceptible to bias and may be an overestimate.