To our knowledge, our study is the first to quantify detailed changes in the age patterns of influenza-related deaths in the decades following the 1918-20 pandemic, using a consistent approach throughout time to estimate influenza mortality burden. Based on comprehensive data from Copenhagen, Denmark, we have shown evidence for a shift in the age pattern of influenza-related mortality in fall 1918 and winter 1919, where the population under 65 years of age experienced nearly 7 times higher excess mortality rates than older individuals. An intermediate age pattern of deaths was found in the following winter in 1919-20, with nearly equal excess mortality rates in people under and over 65 years. While young and middle-aged adults stopped experiencing increased respiratory mortality by 1921-22, the age distribution of influenza-related mortality did not fully return to inter-pandemic patterns until 1925-26, and the first noticeably severe epidemic in older people occurred in 1928-29. Overall, these results suggest a relatively rapid post-pandemic adaptation to the 1918 A/H1N1 pandemic in the Copenhagen population, within 3-10 years of emergence of the novel virus.
There are several strengths and limitations of our study. We took advantage of the high quality Copenhagen surveillance system and its thorough records, which provide mortality and morbidity information by week and age group, resulting in a unique dataset covering the early part of the 20th century. However, these data are limited to a relatively small city of ~500,000 inhabitants in 1918, introducing demographic noise. Aggregating the data by 4-week periods stabilized excess mortality estimates, while maintaining adequate temporal resolution. In addition, by using respiratory mortality rather than total mortality, we produced more specific and robust estimates of influenza mortality burden, especially in age groups with high background rates of deaths (infants and seniors), and during moderate influenza seasons. Excess mortality models fit reasonably well in all age groups, except in 5-14 year olds, a group that did not experience much seasonality or high death rates in normal years. Reassuringly, we found evidence of excess mortality in this age group in the 1918-20 pandemic period, suggesting that our estimates for 5-14 yr old are relatively accurate.
An important caveat in our study is the poor age resolution in the Copenhagen data, especially in the adult population (15-64 yrs, and 65 yrs and over), limiting our ability to finely characterize the age distribution of influenza-related deaths. Another caveat is related to variations in socio-economic conditions and healthcare, which have been shown to affect influenza-related mortality rates [19
]. Long-term declines in background mortality rates during the study period, due to improvements in socio-economic conditions and healthcare, are expected to drive a decline in excess mortality rates attributed to influenza In particular, baseline respiratory mortality declined in the early 20th
century in children under 5 years in Copenhagen. However, the mortality patterns evidenced in this study were so different across age groups, and post-pandemic changes occurred so rapidly and dramatically, that slow trends in background mortality would not affect overall conclusions.
Another important caveat is related to reporting rates of medically-attended influenza-like-illnesses, which we suspect may have increased in the post-pandemic periods with increasing influenza awareness. Alternatively, the high influenza morbidity rates observed during 1918-37 may truly reflect high influenza transmission rates in the post-1918 H1N1 era, compared to those of viruses circulating before 1918. Given the potential reporting bias in morbidity data, and the lack of an appropriate pre-pandemic baseline, we were unable to quantify post-pandemic changes in the age distribution of cases. Nevertheless we were able to show a relatively good agreement between high excess mortality and excess morbidity seasons in the post-pandemic period 1918-37 (correlation=0.72, P<0.001).
One limitation with the applied statistical approach is that we can not disentangle competing causes of respiratory excess mortality, e.g. effects from co-circulating virus or bacteria and effects of extreme weather conditions. This is not likely to be a limitation for the assessment of excess mortality in the period of 1918-20 where the evidence of the effect of influenza A/H1N1 is overwhelming, in Copenhagen but also in Europe, Asia and the Americas [6
]. However, knowledge of the etiologic agent associated with the respiratory epidemics in 1921-22, 1923-24 and 1928-29 would be important to shed light on this issue.
Our study confirms that seniors over 65 years of age had very low mortality during the pandemic period 1918-19 relative to surrounding seasons in Copenhagen (see also, [7
]) -- a pattern of mortality sparing consistent with previous studies of the pandemic impact in New York City, USA, and Japan [6
], but inconsistent with historical Mexican data [20
]. It is intriguing that seniors in Europe, Asia, and the US experienced higher death rates in epidemic years like 1915-16 (pre-H1N1 era) or 1928-29 (H1N1 era), than during the 1918 pandemic. This finding suggests that seniors might have had protection from previous exposure to similar antigens during childhood [1
], and that the level of protection may have varied between geographical locations depending on the circulation of historical influenza viruses.
If one assumes that seniors were spared from mortality during the 1918-19 pandemic waves because of earlier exposure to antigenically similar viruses, then the disappearance of such protection over time may provide indirect evidence of drift in H1N1 viruses circulating in the 1920s and 1930s. Seniors over 65 experienced moderately high influenza mortality burden during the 4th wave of the pandemic in the winter of 1919-20, with excess respiratory mortality rates nearly equal to those in younger individuals. This could be taken as evidence of an early antigenic drift in the pandemic virus, albeit a minor one. Using the pre-pandemics seasons of 1904-1917 to establish a baseline of excess mortality rates during seasonal epidemics suggests that the first markedly severe post-pandemic season in seniors occurred in 1928-29. This increase in excess mortality could be evidence of a large antigenic drift in influenza A/H1N1 viruses around 1928 in Copenhagen. Given that the 1928-29 season was also particularly severe in the US [21
], it is possible that an antigenically-novel influenza H1N1 variant could have spread globally around that time. At this stage, our inferences about putative years associated with antigenic drift remain relatively hypothetical given the paucity of studies formally matching epidemiologic and antigenic data [22
], especially following the emergence of a pandemic virus. A key area for future research is to systematically test age-related mortality and morbidity indicators that may help identify antigenic drift in influenza viruses in inter-pandemic periods [5
]. Such indicators could be developed by combining age-specific epidemiological data with measures of inter-annual variation in antigenic characteristics of circulating strains [16
], as in recent efforts [23
]. Once identified, such indicators could then be applied retrospectively to historical time periods where mortality statistics have been collected but no virus specimen is available.
It has now become established that the 1918 pandemic was associated with mortality sparing in people over 65 years of age in Europe and the US, and a similar phenomenon may have occurred during the 1968-70 pandemic in people over 77 years of age in the US [10
]. In the case of the 1968-70 pandemic, a return to normal baseline age mortality patterns occurred within 4-5 years, based on the ratio of excess mortality rates in people under and over 65 years [5
]. The first virological evidence of antigenic shift occurred 7 years after pandemic virus emergence, in 1975-76, a season also associated with a severe outbreak in seniors [5
]. By contrast, the 1957 pandemic was not conclusively associated with mortality sparing in seniors, although there was a shift towards younger age of death, with people under 65 years accounting for 36% of all excess influenza-related deaths [5
]. Over the next decade, this proportion was reduced 28-fold [5
]. Overall, based on all three pandemics in the 20th
century, we conclude that the period of “adaptation” to a new pandemic virus may take between 3-10 years. This is important to consider in light of the current influenza A (H1N1)pdm 2009 pandemic, as younger populations may be at increased risk of severe influenza-related disease for a few years, relative to seasonal epidemics occurring before 2009.
In terms of pandemic preparedness, the rich historical data from Copenhagen indicate that intervention strategies should be focused on young adults and children during the early pandemic waves. When the age pattern returns to normal, with highest mortality rates experienced among senior populations, the focus should return to mitigating disease burden in older age groups. The upcoming seasons of H1N1-pdm virus circulation may see a similar change from the current pattern of most deaths in people younger than 60 years to one of most deaths occurring in seniors. It will be important to keep monitoring the age pattern of influenza-related deaths in coming years to adjust vaccination and mitigation strategies accordingly.