The Spanish Influenza in Madrid: Excess Mortality by Age in Four - - PDF document

The Spanish Influenza in Madrid: Excess Mortality by Age in Four Consecutive Waves

Laura Cilek, Gerardo Chowell, Diego Ramiro Fari˜ nas ∗ 1 Background

Contemporary estimations claim the influenza pandemic events between 1918 and 1921, the so-called ”Spanish” flu, accounts for the deaths of more than 50 million people throughout the world [1]. The series of successive influenza virus outbreaks gripped the world beginning in early 1918, however the initial circulation of the virus from avian or swine and other mammal species to humans, according to various phylogenetic and molecular-clock analyses, may have occurred as early as 1911 [2] or just before the first epidemic outbreaks in early 1918 [3]. Moreover, the symptoms and mortality patterns associated with this particular flu pandemic are particularly unique. For example, young-adults often exhibited the highest excess mortality rates, in contrast to seasonal influenza epidemics, which primarily affect the very young and elderly [4, 5]. The difference in age-specific influenza-related mortality by year and flu sub-type is often examined by using age specific mortality rates and calculating the risk ratio of mortality between two age groups of the proportion of excess deaths in a given age-group, as Simonsen et al calculate for influenza seasons between 1918 and 1989 [5]. While the location of the first human infection remains unclear, the virus likely moved to Spain via Spanish and Portuguese labor migrants in Southern France during the first world war [6]. The name ”Spanish” flu comes from the first reports of symptoms in Madrid in the late spring of 1918. However, the flu gained its moniker after mentions of the virus were first reported on and published in Spain, a neutral country in the war [7]. Nonetheless, in the spring and summer months of 1918, many concurrent herald

• utbreaks featured a strain that, while highly contagious, contributed to fewer overall excess deaths than

the subsequent fall waves [8, 9]. In fact, the actual timeline and progression leading to the virus’s emergence is debated, though likely, the H1N1 strains responsible for the Spanish flu are related to those which caused the ”Russian” pandemic influenza events at the end of the 19th century and may have been present in both swine and humans more than 5 years before the first waves in 1918 [2]. While strains of the H1N1 virus continue to circulate in the form of seasonal influenza viruses, biological remnants of the particularly deadly 1918 strains are still found in avian species via the presence of specific encoded proteins [10]. In this manner, continued research into the unique aspects of the Spanish flu and its health and mortality impacts on the population are essential to understanding the potential effects that a virulent influenza strain could have on the global population today.

1.1 Age-Specific Mortality and Waves in the 1918 Influenza Epidemic

Following a 1998 conference that focused on the pandemic, Johnson and Mueller undertook a effort to re-estimate global mortality from the pandemic using available studies and new techniques, proposing that the virus claimed the lives of at least 50 million [1]. While the pandemic events associated with the Spanish flu are perhaps most well-known for the aggregate mortality burden inflicted on the world, the unique age-specific mortality patterns of the successive outbreaks are also an extremely important trait of the

• virus. While excess mortality in seasonal influenza outbreaks nearly-unilaterally affects young children and

those older than 65 (some longitudinal research using yearly cause- and age-specific death counts indicates additional, smaller influenza epidemics, such as in Canada in 1957, experience higher adult relative risk and mortality [11]), the epidemic waves beginning in the spring of 1918 uncharacteristically impacted young adults between the ages of 25 and 30 [12, 13]. Supporting evidence can be found in analyses employing a variety of methods and different types of data; for example, Viboud et al used individual death records in Kentucky from 1911-1919 to create a strong mortality baseline, then identified a peak mortality risk in

∗laura.cilek@cchs.csic.es

Center for Humanities and Social Sciences, Spanish National Research Council, Spain; Division of Epidemiology & Biostatis- tics, School of Public Health, Georgia State University

1

1918 (relative to the baseline years) for those aged between 24 and 26 [13]. Gagnon et al reviewed a slew of mortality data sources throughout the United States and Canada, including parish and civil registers, from September to December of 1918. Across these locations, they found peaks around age 28 in the percentage

• f deaths for both the percentage of deaths by age (across all ages, for pandemic-related causes) and for the

percentage of all deaths (ages 15-44, all cause mortality) [12]. A few studies documenting excess age-specific mortality rates during the 1918-20 influenza pandemic reported little to no excess mortality for the elderly in the US and European settings [14], but an analysis

• f pandemic excess mortality rates revealed that seniors greater than 65 years of age experienced 1.5-2.4

fold higher excess mortality rates than young adults during autumn 1918 wave [15], corresponding to a 12-fold elevation over their baseline mortality rate. During the 1918 pandemic, in these cases of high excess mortality for both young adults and seniors, a true W-shaped pattern of excess mortality risk by age and a pattern occurs, where excess mortality rates peak in infants and young children, young adults, and the elderly population. Yet in the appearance of this w-shaped mortality curve varies by location and perhaps exposure to pre- vious strains of a familiar virus [16, 15]. Analyses conducted with census data and raw death counts during epidemic periods (with little to no baseline mortality information) reveal conflicting results as to a general mortality pattern by age; some evidence in rural and ”geographically isolated” populations show a w-shaped mortality pattern [17, 18, 19]. However, similarly completed analyses in other remote areas find instead a v-shaped mortality curve, in which the heighted mortality rates for adults does not decrease after the young-adult peak [14, 20, 15]. Other ”typical” w-shaped mortality curves can be found in non-remote urban and rural locations throughout the world (for example, a study in the United States measuring country-wide annual excess mortality determined from an yearly five-year baseline [16] or analyses in Copenhagen using nine years of weekly surveillance and mortality data [21]). Together, these continued excess morbidity and mortality analyses continue to foster the idea that a place or individual’s exposure to circulating H3 and H1 influenza strains contributed to their acquired immunity and played a role in mortality during the Spanish flu pandemic [15]. In addition to research using historical records, a review of the effects of pathogenic responses to strains of influenza virus hypothesizes that previous exposure to different circulating strains could have effected an individual’s mortality risk to the 1918 virus [22]. Depending on particular strain(s) of exposure, either some immunity may have been provided or the body many have triggered an incorrect pathogenic response, explaining high flu- related mortality from subsequent respiratory infections [22]. These findings suggest substantial differences between countries in prior immunity to the 1918 influenza A(H1N1) virus. Both systematic reviews of seasonal and pandemic virus circulation [8] and mortality analyses that focus on the timing of each wave, as in a 2010 paper examining the 1918 influenza impact in Mexico [23], show such differences could result from a heterogeneous circulation of influenza viruses in the 19th century. The Reproduction number (R) is another key estimation often included in influenza epidemiology anal- yses to analyze the transmissibility of a disease in a given population. Most simply, R can be interpreted in terms of the demographic measurement of the Net Reproduction Rate; thus, the Reproduction number can be considered as the number of secondary cases (of influenza) that each infected individual (primary case)

• produces. Similar to its demographic equivalent, R also depends on the length of time required to generate

the disease, or the mean length of time the virus is present in an individual [24]. If the value of R equals 1, the size of the epidemic remains the same or reaches unity, and if the number becomes lower than 1, the number of cases of disease will decrease. However, an R value greater than one implies that an epidemic grows, and the larger the number is, the higher the likelihood that interventions will be more difficult to implement and the growth of the disease difficult to control. Estimates of the Reproduction number for pandemic events tend to be higher than those of seasonal influenza; a comprehensive review of published R estimates (calculated by various methods) by Biggerstaff et al found the mean R value of the 1918 and 1968 influenza outbreaks was 1.80, but seasonal influenza estimates for R were often lower, with a mean published value of 1.28 [25]. Chowell’s 2009 chapter on the calculation of R found more than 11 published estimates of R for the 1918 epidemic with values ranging from 1.4 to 5.4 [24], though he also notes that differences in location and size, demographic composition, and which wave(s) were considered, as well as the technique used to calculate the number, can affect the estimated value of R. In general, the timeline of the pandemic is broadly classified into three different wavesa short but intense 2

”herald” wave in the Spring of 1918, a longer wave in the late fall and winter of 1918-1919 which accounted for a majority of deaths, and a third, less lethal wave taking place in the following year. Both the first and third waves were not universally experienced; those locations that were more remote or practiced better preventative techniques often skipped the first wave but instead endured longer and more persistent second waves [26]. Yet the Spanish flu epidemic is unique in that the rapidity with which these waves reappeared was a new phenomenon compared to prior influenza pandemics. Traditionally, pandemic waves occurred annually, such as in the case of the Russian flu pandemic, which began with a strong, high-mortality wave in 1889 and continued with minimized intensity in yearly subsequent waves through 1893 [27, 28]. Unlike this typical pattern of epidemics, the initial influenza outbreaks during 1918 and 1919 occurred nearly consecutively; even if a location did not experience a noticeable herald wave in the spring or summer, successive fall and winter waves, or at the very least, a long, sustained outbreak in 1918-1919 occurred nearly everywhere, a distinct separation from the timing of outbreaks in previous influenza epidemics.

1.2 1st wave

The different timing and mortality patterns of pandemic Spanish flu waves throughout the world most likely also affected the mortality impact of each individual wave within a population, as well as how much

• f that population was at risk. In many urban centers that experienced a stronger herald wave in the spring
• r summer of 1918, the severity of the succeeding fall wave is less pronounced relative to the rest of the
• world. Often, the total effect of the epidemic was lower in these cities with herald waves. While the spring

wave lacked the virulence of the successive fall outbreaks, its overall transmission rates and corresponding Reproduction numbers were quite high, and the virus tended to affect much of the population [9, 21], mortality rates specific to the spring waves tend to be lower than in the fall. Using aggregate surveillance and death count data in Copenhagen from 1910-1919, Andreasen et al found that the summer wave in June and July of 1918 contributed to around one third of total excess cases of influenza-like cases and hospitalizations, but most of these illnesses did not lead to death [21]. The also paper found corresponding Reproduction numbers for the waves; estimates for the summer wave ranged from 2.0-5.4, but though the fall wave had a much higher case-fatality rate, R ranged between 1.2 and 1.6. An analysis of mortality and timing of the Spanish flu in New York City used New York City Health Department monthly cause-specific mortality records from 1911 to 1907-1931 [9]. Using this robust length of baseline mortality data, the authors identified that epidemic level influenza seasons occur in the winters of 1915-1916 and 1916-1917, and twice in 1917-1918 (January and March) before the characteristic pandemic wave beginning in the fall of 1918 [9]. However, a shift to the distinct ”w” shaped pattern of age-specific death rates, in which young adult mortality is disproportionately higher, occurs in the spring of 1918, suggesting New York also experienced a herald wave, though it is difficult to ascertain if the presence

• f these initial waves tempers the impact of the fall wave.

A systematic review of literature considers the difference between pandemic and seasonal influenza, as well as the circulation of constantly and competing and evolving viruses summarizes a postulation regarding the effect of the first wave on remaining outbreaks in an area [8]. In this sense, the strain circulating during the spring and summer months may have also provided partial immunity from the fall waves, but few definitive conclusions have been made [8]. In this way, after the mutation of the virus from the spring strain into the fall version, much of a population in a spring-affected area possessed a form of protection that led to lower overall influenza-related mortality in the subsequent fall and winter waves. Morens notes that research asserts, rather conclusively, that the second wave, in the fall of 1918 provided immunity to those who also experienced a winter wave [27], yet he finds the connection between spring exposure and fall immunity difficult to prove. However, a study of the 1951 flu epidemic in the UK implies that the amount

• f antigenic drift in a virus leading to different strains may determine to what extent one wave may provide

immunity to the next [29]. It is possible that exposure to a spring wave provided protection against the fall if there was little change in the makeup of the attacking strain. 3

1.3 Influenza in Spain

Prior research provides abundant information regarding the timing and severity of the 1918-19 Spanish Flu epidemic in Spain [6, 7, 30], including some work to quantify the reproduction and spread of the virus within the city of Madrid [30, 31]. Echeverri [7] and Trilla [? ] present comprehensive backgrounds on the events, their timing, and some observed outcomes of the Spanish u pandemic in Spain. They present a descriptive picture of the events in Spain, primarily through the lens of period press reports and mid-century publications on the evolution of sanitation and health in Spain that include the 1918 outbreaks. To quanity the effects of the influenza virus in Spain, Chowell et al used four and a half years (January 1915 to June 1919) of cause of death coded, provincial level monthly death data to determine a cyclical baseline and excess mortality rates across the 49 Spanish provinces [30]. Another analysis of influenza outbreaks in the city of Madrid using the same data as in this analysis (see section 2), to calculate start and end dates of epidemic influenza waves (and their corresponding Reproduction numbers) using a breakpoint regression estimation technique [31]. However, this paper differs substantially from the present analysis in that it focuses primarily on the 1889-90 Russian influenza outbreak and the final wave of Spanish influenza in the winter of 1919-1920, the technique employed incorporates no baseline mortality information, and at the time of the analysis, only data from four of the ten districts was available for use[31]. From the above outlined prior analyses, there is evidence that Madrid experienced three waves of epi- demic influenza in 1918-1919, generally happening concurrently to the timing of outbreaks in the rest of Spain, in addition to a large fourth wave in the winter of 1919-1920. The first happened in May of 1918, when many, including King Alfonso XIII, are reported to have experienced u-like symptoms for less than a week creating fodder in the press [7]. While not universally present, this type of herald wave is primarily found in large cities and characterized by a high transmissibility and large reproduction rates [9, 1]. Sta- tistically present in 32 provinces in Spain, the mortality of the spring wave in Madrid is quite low overall when compared to other herald waves and the total impact of the succeeding waves [30]. The second wave, generally regarded worldwide as the most lethal [1], occurred primarily in fall of 1918. According to the Madrid Civil Register records on deaths, 1,767 deaths were recorded during the month of

• October. Other research indicates total mortality in Madrid from this wave may be less extreme than the

typical patterns of second wave mortality [31]. This is not dissimilar from other cities that experienced the strong herald wave [1]. A third wave affected the city from the end of December 1919 to the beginning of spring 1920. An extreme spike in influenza and respiratory related mortality, considered the 4th wave, can be exam- ined in the winter u season of 1919-1920 also [7]. Using weekly death counts, a prior comparison of mortality for 8 of 10 districts in Madrid during this fourth wave to the city’s 1889-1890 epidemic wave reveals similar, though slightly less extreme, patterns of mortality and timing during the winter of 1919 [31]. While seasonal influenza occurs on an annual basis, before the Spanish u epidemic events, large pandemics were not altogether uncommon. However, these pandemics generally produced annual waves (albeit with higher mortality) similar to seasonal events, such as in the case of the Russian u pandemic, which appeared in yearly waves between 1889 and 1893 [27]. The rapidity with which the waves in 1918-1920 reappeared was a new phenomenon compared to these prior u pandemics. Despite these efforts to identify and compare these general patterns of the epidemic in Madrid to the rest of Spain, less research has sought to compare the timing, strength, and age-specific morality patterns

• f each wave in context to both each other and other urban locations. Moreover, given the known typical

patterns of the Spanish influenza pandemic throughout the world, this paper seeks to quantify to what extent the city of Madrid follows these patterns, by looking at the timing and severity of each wave, as well as overall and respiratory excess mortality for key age groups. In this sense we hope to also compare the events of 1918 to 1920 with the citys 1890 pandemic to identify similar patterns and the ways in which they diverge.

2 Data

The Madrid Civil Register records on Deaths provides individual death records with which to examine the mortality of the 1918-20 epidemic. Each record provides specific details of the deceased, including age 4

at death, sex, address, and civil, foreign or native Madrile˜ no, and occupational statuses. Between 1917 and 1923, the register holds 103,500 records (an average 17,500 deaths per year). Table 2 provides information

• n the yearly number of deaths available as well as the median age of death, which may be relevant to

examine the presence of a ”w”-shaped mortality curve. Table 1: Madrid Civil Register Death Records, 1917-1922

Year 1917 1918 1919 1920 1921 1922 total Deaths 15415 18956 18279 17951 16171 16254 103026 Median Age 37 33 36 30 35 35 34

Cause of death information for each record allows us to subset influenza and respiratory mortality from

• ther causes. Further information regarding the population composition of Madrid comes from the citys

yearly population books. These act as a summary of the population and demographic events each year, and they allow the results to be standardized and studied in conjunction with the age structure and composition

• f each district.

Together, these sources provide a rich source of individual-level and aggregated data with sufficient information to determine baseline and excess mortality rates, as well as wave Reproduction numbers and excess mortality ratios for both the entire city and at more refined levels, such as by age group.

3 Methods

The majority of influenza-related deaths resulted from secondary respiratory ailments following u con- traction (rather than influenza itself), most commonly bacterial pneumonia [27, 26]. Additional influenza- related deaths came from additional types of bacterial infections and severe-acute respiratory distress, often evidenced by blueish-gray skin shortly before death [27], and as such, estimates of respiratory related mor- tality also provide key information regarding the effects of influenza-specific mortality. Thus, we estimate figures using the following techniques for both all-cause and flu related mortality. When determining which deaths occurred due to influenza and influenza-related respiratory related causes, often called pneumonia and influenza” (or ”P&I” deaths), we found it most accurate to all records containing influenza, pneu- monia, bronchopneumonia, or bronchitis and from this new subset, removing those which also contained tuberculosis.

3.1 Baseline

Given seasonal variation in death rates, our estimation of the mortality baseline also needed to mimic the underlying season oscillations. First, we verified the basic mortality pattern in 1917 (from 15,536 individual registered death records and yearly population information published by the city) was a relatively similar shape and size to 1921 and 1922, the years following the conclusion of Spanish flu related epidemic waves, as in figure 3.1. In each of the years with no reported epidemic influenza activity, seasonal flu peaks (with variations in timing and strength) are visible at the beginning and end of the year, as well as a seasonal summer peak. In the years of data with reported pandemic activity, deviations from the ”normal” mortality patterns are evident. Then, using only pre-epidemic mortality data from 1917, we calculated weekly mortality rates which were used to fit a simple cyclical Serfling linear regression model estimating seasonal baseline mortality. Serfling [32]. However, we found this initial baseline did not account for the small but noticeable summer mortality peak (see figure 3.1a). We furthermore found that due to variation in the timing of seasonal mortality peaks, concatenation of our estimated baseline for the six years of data provided an unrealistic continuous mortality baseline. Figure 3.1b indicates the results of this concatenation; in 1917, both all- cause and respiratory mortality rates were much lower in the beginning of the year than in the end, meaning that our initial baseline estimation would have assumed the ”normal” mortality patter each year dropped drastically between the last week of December and the first week of January. Thus, to account for these mortality differences, we modified the initial Serfling model with additional parameters, as in Cobos et al’s estimation of the 1957 influenza pandemic in Maricopa County, Arizona [33]. The added coefficients in the model account for both time (α) and seasonal (β & γ) variations in normal u 5

Figure 1: 1a (left): all cause weekly mortality rates using yearly population data and deaths from the civil register. 1b

(right): Respiratory related (with tuberculosis deaths) mortality rates by year.

Figure 2: Estimated Serfling Regression Model to establish mortality baseline using 1917 data. 1a (left): estimated baseline

within a single year, 1b (right): regression baseline for all years.

6

activity, such that the oscillations (at time t) may be written as:

Deathsxt P opulationxt =

u + α ∗ (t) + α ∗ ( 100

t )2+

β ∗ sin(2 ∗

π 52.17 ∗ t) + β ∗ sin(4 ∗ π 52.17 ∗ t) + β ∗ sin(8 ∗ π 52.17 ∗ t)+

γ ∗ cos(2 ∗

π 52.17 ∗ t) + γ ∗ cos(4 ∗ π 52.17 ∗ t) + γ ∗ cos(8 ∗ π 52.17 ∗ t)

While visual anal- ysis reveals similarity in year-to-year baseline mortality, we felt that using only the one year of available death records (1917) may have ultimately provided an incorrect estimation of the baseline, as it assumes mortality in 1917 followed a normal pattern at all ages. Thus, we decided to employ a parametric bootstrap [34] method to account for some of the potential variability in the 1917 data from typical mortality. To employ this method, we first simulated data before fitting the above regression model, accounting for the possibility of aforementioned fluctuation in the annual timing of winter and summer mortality peaks. A single set of mortality points from which the bootstraped points were estimated consisted of six consecutive iterations of total weekly deaths in 1917, to mimic the six years of mortality data used in our analysis. 1 For each of these 312 week sets of weekly death counts, we simulated a Poisson-distributed number of expected deaths, as the number of deaths each week is a ”count” variable that must be 0 or greater. Our Poisson estimations assumed the mean and variance of a week were equal the observed total number of deaths in that week of 1917. From each of 500 simulated six-year datasets, α and β parameters are estimated according to the modified seasonal regression model above. We calculate our five year baseline from the mean values of the coefficients from 500 models and compute the upper baseline from the upper quantile value of the 95% Confidence Interval of coefficients. As in other papers using Serfling regression to estimate baseline mortality, we define weeks with mortality above the upper baseline as epidemic [33, 35, 36].

3.2 Excess Mortality and Relative Risk

A conscious decision to combine the fall and winter waves was made; thus, we defined three distinct wave periods as May-July 1918, August 1918- April 1919, and November 1919-February 1920. While there is evidence to suggest the city of Madrid experienced unique fall and winter waves, the break points become unclear when dissecting the data into smaller categories such as age group, sex, sec. For both this and in

• rder to compare with prior research concerning the 1918-1920 influenza pandemic waves in Spain using

different data, we examine the successive fall and winter waves as one [30, 28]. We characterize excess mortality for each wave by summing the total deaths above the baseline during the epidemic periods. To aid in the comparison of our results with other research, we also provide relative for each wave and age group to allow easier comparison between the groups with different underlying rates due to varied population size [30, 37]. For each wave, we define relative risk as the ratio of total excess mortality observed to expected baseline mortality. This aids in the direct comparison of the total mortality impact of the flu between study groups, as baseline mortality varies substantially by age group, geographic area, and other subsets [35].

4 Results 4.1 Baseline

The estimation technique for both the all-cause and respiratory mortality baselines created a general linear patter mimicking mortality patterns for the year 1917. In the years following the main epidemic waves, the all-cause baseline continues to mimic the general shape and pattern of the baseline, but the overall mortality rate falls relative to that of 1917. Exhibiting a different pattern, mortality due to respiratory diseases appears to shift rightward, in other words, seasonal influenza/respiratory activity peaks later in

1 We consider each year to have 52 weeks, and calculate the ”total” deaths in the 52nd week of the year as the 7 8 or 7 9 of

deaths in the final week and associated excess day(s) of the year (1920 was a leap year).

7

the year. Additional figures depicting overall and respiratory related excess mortality baselines for each age group may be found in the Appendix. Figure 3: Points represent actual weekly mortality rates from 1917-1922, and lines show mean and upper 95% bound baseline from simulated 1917 deaths data. Vertical lines indicate breakpoints from which excess mortality is generated.

4.2 Age Specific Death Rates by Wave

Tables 2 and 3 provide all-cause and respiratory-specific excess deaths, weekly mortality rates, and relative mortality risk for each of the three evaluated waves. When considering the total population, average weekly excess mortality is highest in the first and third waves, each with a weekly excess rate of about 1.51 deaths per 10,000. Weekly excess rates for the combined fall and winter waves in 1918-1919 are only slightly more than one third of the rates in the remaining waves, but they account for more total deaths than in the first wave and nearly as many as the final wave in the winter of 1919-1920. Infants and children aged up to 15 experienced the highest all-cause and respiratory excess mortality rates during the final wave in the winter of 1919-1920, nearly double the rates calculated from the first two

• waves. This pattern of higher excess rates in the last wave also holds for those 70 and older; relative to

the first two waves, seniors’ excess mortality rates jumped in the winter of 1919-20. The 15-24 year age group maintained similar excess rates through each successive wave, but their mortality risk changes slightly due to the seasonality in baseline rates. Only one age group, 25-50 year olds, saw a notable decrease in 8

excess mortality rates during the last winter wave. This age group’s excess mortality rates were highest in the first wave, slightly less in the combined fall and winter waves of 1918-1919, but, while still higher than expected mortality, dropped between one third (all cause) and one half (respiratory related) from the combined second wave to the final winter wave one year later. Concurrent with typical Spanish influenza age-specific mortality patterns, in each wave of influenza, the standardized risk of death relative to normal mortality rates was higher for younger ages, particularly in the 15-24 and 25-49 year-old age groups. With few exceptions, weekly all-cause excess mortality rates were much larger than excess respiratory mortality rates, but the standardized mortality risk, or is much larger when considering respiratory-only mortality. As such, during the first wave, the age groups between 15 and 49 were more than three times likely to die from respiratory causes than during the same time in a non-pandemic year. Figure 4 shows the age-specific excess death rates by wave for both all-cause and respiratory related

• mortality. In the spring wave, excess death patterns generally follow a normal age-specific structure of

mortality where the highest mortality is found in infants and elderly populations. Total excess death rates during the combined fall and winter waves of 1918-1919 exhibit similar patterns to other age-specific analyses

• f Spanish influenza mortality. Here, the three age groups between five and fifty years of age have higher

excess death rates than the youngest and eldest groups. The fall wave has the highest excess respiratory mortality rates for all age-groups except infants and young children but generally lower mortality risk relative to the same time in a non-epidemic year. This is because baseline mortality rates are higher during this period of the year.

5 Discussion and Conclusion

Curiously, our findings of overall excess mortality appear to differ slightly from those of a previous study of excess monthly mortality in all provinces of Spain [30]. It is difficult to ascertain what specifically contributes to this difference; the studies use different sources of data, and the previously published analyses looks at the entire population of the Madrid region, while we focus on the city itself. Our different technique for the estimation of the baseline may also play a roll in the differences, particularly the disparities in our weekly excess rates (0.59 per 10,000) and their monthly results (55 per 10,000). Both our analysis and the

• ther paper also find different excess mortality rates when considering only influenza and respiratory-related
• deaths. Alternatively, the baseline mortality rate for the city of Madrid may be higher than that of the

entire region. Thus, it is possible an implicit urban mortality penalty in the city moderated the presence of excess mortality caused by the influenza virus. Overall, the short but strong Spring wave contributed the most (on a weekly basis) to the deaths of individuals between the ages of 15 and 50, as they experienced 75% higher mortality risk than normal during this wave. The spring wave in 1918 shows an increased mortality risk for young adults; thus, this data indicates this mortality peak is indeed a ”herald” wave present in many places throughout the world. While these specific age groups were chosen for comparison with prior work on the 1890 Spanish flu [31, 28], it is expected that future detailed age-specific death analyses will reveal an even more prevalent heightened risk of death is for those between 25-30 [12, 13]. Furthermore, while those older than 70 typically had the highest excess morality rate (with the exception

• f those 0-5 years old), their overall elevated risk of dying during one of the epidemic waves relative to a

period of baseline mortality was lower than younger counterparts, revealing a higher impact of the Spanish influenza on ages with a lower risk of death. Interestingly, those over 70 and under 5 experienced fewer

• verall deaths during the fall wave than might have been expected in baseline mortality circumstances.

However, this trend is not present when considering only respiratory deaths. while the standard mortality risk is lower for these age groups relative to others, both this ratio and the weekly excess mortality rate show higher than expected baseline mortality in this wave. With specific regard to elderly mortality, our findings for Madrid are somewhat inconclusive; in all waves, we find the mortality ratio of pandemic to normal rates to be higher than 1, which does not support Crosby’s findings of no excess mortality in seniors [14]. However, we also do not see cases of extremely high elderly mortality as Mamelund finds in the fall wave [15]. In fact, the standardized risk of dying for seniors is lowest in our combined fall and winter waves. 9

Relative to much of the rest of the world, Madrid also faces a mild fall/winter wave, in which the two must be combined to determine the total mortality effect of the virus in the late summer of 1918 through spring of 1919. While the mortality impact of the flu during this time cannot be ignored, the overall excess mortality rates found for all ages, especially those above 70, in this period are much smaller than in places such as Mexico or Colombia [23, 35]. In each of these two countries, daily death reports, epidemiological bulletins and mortality records (mortality records only in Colombia) were collected from one (Colombia) and three (Mexico) years prior to 1918 in order to estimate baseline and excess mortality through seasonal regression and an approach using average non-epidemic weeks as a reference point to mortality during epidemic periods. Chowell argues the results of these analyses indicate a general remoteness of a location could mean elderly populations had not been exposed to this type of influenza virus before, and therefore lacked immunity. Under this general assumption, it is likely the Russian flu pandemic of 1889-90 in Madrid played a role in exposing much of the adult and elderly population to another strain of influenza virus [31]. However the role that this earlier epidemic played in the immunity or lack thereof in the 1918-20 pandemics is also difficult to understand. Rather than asserting that the H2N2 strain of influenza in the late 19th century created immunity in the population, the ”Original Antigenic Sin” theory would instead assert that the circulation of the Russian Influenza in Madrid led to the spike of young adult mortality [3, 12]. Most generally, the OAS ”doctrine” states that young children experience a critical stage of development during which their immune systems learn to ght viruses. Individuals at or around age 28 during the Spanish u (H1N1 strain) were exposed to the Russian u (likely an H3 strain [38]) in 1888-90, and thereby, their immune systems developed the methods necessary to defend the body from the different strain. The ”incorrect” immune defense triggered by exposure to the H1N1 strain, led to a weakened immune system and body susceptible to the secondary bacterial infections contributing to additional respiratory causes of death[3, 12, 26]. However, this theory remains highly debatable, as discussions and reviews of literature remind researchers that acquired immunity through exposure to influenza and other viruses varies within both the broader contexts of geography and circulating strains, as well as individual contributing factors [39]. In this sense, these current analyses provide evidence to support our future work within the historical circumstances of Madrid and the individual characteristics of our study population. Beyond long-term immunity, we also find evidence to support the idea of a protective effect of the spring wave against the more virulent and deadly strains that circulated in the fall of 1918. For all ages, the highest transmission rates of the first waves in 1918 and 1919 are found in the herald wave of May and June of 1918 as the virus spread through the city, attracting media coverage and attention around Spain and the world [7]. Yet, as in other cities with strong first waves, the succeeding waves of influenza in Madrid had a much smaller effect on excess mortality when considered on a weekly basis. While the rest of the world suffered from the deadly virus circulating in the fall waves, Madrid escaped relatively unscathed. Similar to the patterns found in other large cities in Europe and the U.S., the muted presence of fall and winter waves in Madrid supports the idea that populations exposed to an initial herald wave received a type of protective effect from the virulent fall strains [9, 21]. While some, but not all spring waves provided partial protection later in the year, the case of Madrid implies that their was little antigenic drift between the strains of virus that circulated in the city in the first and seconds waves [29]. Considering the pandemic events collectively known as the Spanish influenza, the case of Madrid pro- vides additional insights into how, in a large urban environment, individual waves and their progression contributed to the overall mortality impact on the city. While other analyses look at herald waves and ques- tion the impacts of acquired immunity from spring to fall [21, 9], the force of the spring wave in Madrid, relative to he successive fall and winter outbreaks, appears to indicate some type of protective effect, perhaps due to a small amount of antigenic shift in the virus between the two waves. Currently, only strains from the fall wave have been studied, meaning to what extent earlier and later strains differed cannot be affirmed [29, 2]. Yet, further analyses of successive waves should be undertaken in an attempt to better understand acquired immunity and the protection it may provide against successive outbreaks. Using contemporary and historic demographic mortality and surveillance data of recent and past epidemics, further insights into the ways early outbreaks affect the immunity and transmission can affect the way officials respond to contain future outbreaks. 10

Figure 4: Excess mortality rates per 10,000 for all-cause (left) and respiratory related (right) deaths plotted by age group for each wave. 11

Table 2: Age Specific Excess Mortality By Wave Weekly Excess Age Group Total Mortality Rate Standardized Excess Deaths (Per 10,000) Mortality Risk Spring Wave, 1918 Overall 1240 1.51 1.44 [00, 04) 308 0.3934 1.27 [05, 14) 75 0.0538 1.49 [15, 24) 137 0.0913 1.67 [25, 49) 435 0.1353 1.72 [50, 69) 140 0.1669 1.20 [70, Inf) 80 0.6013 1.22 Fall and Winter, 1918 -1919 Overall 1554 0.59 1.18 [00, 04)

• 303
• 0.1225

0.91 [05, 14) 262 0.0594 1.40 [15, 24) 298 0.0630 1.40 [25, 49) 1037 0.1019 1.43 [50, 69)

• 40
• 0.0152

0.99 [70, Inf) 24 0.0561 1.01 Winter Wave 1919-1920 Overall 1612 1.87 1.40 [00, 04) 616 0.6302 1.37 [05, 14) 190 0.1090 1.64 [15, 24) 159 0.0848 1.49 [25, 49) 314 0.0781 1.30 [50, 69) 88 0.0844 1.06 [70, Inf) 325 1.9571 1.32 12

Table 3: Age Specific Excess Respiratory Mortality By Wave Weekly Excess Age Group Total Mortality Rate Standardized Excess Deaths (Per 10,000) Mortality Risk Spring Wave, 1918 Overall 589 0.7203 2.74 [00, 04) 190 0.2433 2.07 [05, 14) 16 0.0128 1.70 [15, 24) 46 0.0354 3.06 [25, 49) 154 0.0532 3.41 [50, 69) 89 0.1058 1.72 [70, Inf) 34 0.2550 1.54 Fall and Winter, 1918 -1919 Overall 1355 0.5145 1.51 [00, 04) 150 0.0606 1.18 [05, 14) 66 0.0172 1.48 [15, 24) 155 0.0374 2.62 [25, 49) 396 0.0432 1.93 [50, 69) 265 0.1000 1.37 [70, Inf) 179 0.4257 1.37 Winter Wave 1919-1920 Overall 973 0.9121 1.66 [00, 04) 401 0.4104 1.82 [05, 14) 50 0.0326 1.64 [15, 24) 80 0.0489 2.58 [25, 49) 105 0.0289 1.43 [50, 69) 173 0.1652 1.42 [70, Inf) 158 0.9508 1.57 13

References

[1] N. P. Johnson and J. Mueller, “Updating the accounts: global mortality of the 1918-1920” spanish” influenza pandemic,” Bulletin of the History of Medicine, vol. 76, no. 1, pp. 105–115, 2002. [2] G. J. Smith, J. Bahl, D. Vijaykrishna, J. Zhang, L. L. Poon, H. Chen, R. G. Webster, J. M. Peiris, and Y. Guan, “Dating the emergence of pandemic influenza viruses,” Proceedings of the National Academy of Sciences, vol. 106, no. 28,

• pp. 11709–11712, 2009.

[3] M. Worobey, G.-Z. Han, and A. Rambaut, “Genesis and pathogenesis of the 1918 pandemic h1n1 influenza a virus,” Proceedings of the National Academy of Sciences, vol. 111, no. 22, pp. 8107–8112, 2014. [4] M. A. Miller, C. Viboud, M. Balinska, and L. Simonsen, “The signature features of influenza pandemicsimplications for policy,” New England Journal of Medicine, vol. 360, no. 25, pp. 2595–2598, 2009. [5] L. Simonsen, M. J. Clarke, L. B. Schonberger, N. H. Arden, N. J. Cox, and K. Fukuda, “Pandemic versus epidemic influenza mortality: a pattern of changing age distribution,” Journal of infectious diseases, vol. 178, no. 1, pp. 53–60, 1998. [6] A. Trilla, G. Trilla, and C. Daer, “The 1918 spanish flu in spain,” Clinical infectious diseases, vol. 47, no. 5, pp. 668–673, 2008. [7] B. Echeverri, “Spanish influenza seen from spain,” The’Spanish’Flu pandemic of, 1918. [8] J. D. Mathews, J. M. Chesson, J. M. McCaw, and J. McVernon, “Understanding influenza transmission, immunity and pandemic threats,” Influenza and other respiratory viruses, vol. 3, no. 4, pp. 143–149, 2009. [9] D. R. Olson, L. Simonsen, P. J. Edelson, and S. S. Morse, “Epidemiological evidence of an early wave of the 1918 influenza pandemic in new york city,” Proceedings of the National Academy of Sciences of the United States of America, vol. 102,

• no. 31, pp. 11059–11063, 2005.

[10] T. Watanabe, G. Zhong, C. A. Russell, N. Nakajima, M. Hatta, A. Hanson, R. McBride, D. F. Burke, K. Takahashi,

• S. Fukuyama, et al., “Circulating avian influenza viruses closely related to the 1918 virus have pandemic potential,” Cell

host & microbe, vol. 15, no. 6, pp. 692–705, 2014. [11] J. Ma, J. Dushoff, and D. J. Earn, “Age-specific mortality risk from pandemic influenza,” Journal of theoretical biology,

• vol. 288, pp. 29–34, 2011.

[12] A. Gagnon, M. S. Miller, S. A. Hallman, R. Bourbeau, D. A. Herring, D. J. Earn, and J. Madrenas, “Age-specific mortality during the 1918 influenza pandemic: unravelling the mystery of high young adult mortality,” PLoS One, vol. 8, no. 8,

• p. e69586, 2013.

[13] C. Viboud, J. Eisenstein, A. H. Reid, T. A. Janczewski, D. M. Morens, and J. K. Taubenberger, “Age-and sex-specific mortality associated with the 1918–1919 influenza pandemic in kentucky,” The Journal of infectious diseases, vol. 207,

• no. 5, pp. 721–729, 2012.

[14] A. W. Crosby Jr, “Epidemic and peace, 1918. 1976.” [15] S.-E. Mamelund, “Geography may explain adult mortality from the 1918–20 influenza pandemic,” Epidemics, vol. 3,

• no. 1, pp. 46–60, 2011.

[16] J. Luk, P. Gross, and W. W. Thompson, “Observations on mortality during the 1918 influenza pandemic,” Clinical Infectious Diseases, vol. 33, no. 8, pp. 1375–1378, 2001. [17] S.-E. Mamelund, “Spanish influenza mortality of ethnic minorities in norway 1918–1919,” European Journal of Popula- tion/Revue europ´ eenne de D´ emographie, vol. 19, no. 1, pp. 83–102, 2003. [18] R. C. Schmitt and E. C. Nordyke, “Influenza deaths in hawai’i, 1918-1920,” 1999. [19] N. Wilson, G. Rice, G. Thomson, and M. Baker, “Re-evaluating a local public health control measure used in new zealand for the pandemic influenza of 1918,” The New Zealand Medical Journal (Online), vol. 118, no. 1224, 2005. [20] P. Curson and K. McCracken, “An australian perspective of the 1918–1919 influenza pandemic,” New South Wales public health bulletin, vol. 17, no. 8, pp. 103–107, 2006. [21] V. Andreasen, C. Viboud, and L. Simonsen, “Epidemiologic characterization of the 1918 influenza pandemic summer wave in copenhagen: implications for pandemic control strategies,” The Journal of infectious diseases, vol. 197, no. 2,

• pp. 270–278, 2008.

[22] G. D. Shanks and J. F. Brundage, “Pathogenic responses among young adults during the 1918 influenza pandemic,” Emerging infectious diseases, vol. 18, no. 2, p. 201, 2012.

14

[23] G. Chowell, C. Viboud, L. Simonsen, M. A. Miller, and R. Acuna-Soto, “Mortality patterns associated with the 1918 influenza pandemic in mexico: evidence for a spring herald wave and lack of preexisting immunity in older populations,” The Journal of infectious diseases, vol. 202, no. 4, pp. 567–575, 2010. [24] G. Chowell and F. Brauer, “The basic reproduction number of infectious diseases: computation and estimation using compartmental epidemic models,” in Mathematical and Statistical Estimation Approaches in Epidemiology, pp. 1–30, Springer, 2009. [25] M. Biggerstaff, S. Cauchemez, C. Reed, M. Gambhir, and L. Finelli, “Estimates of the reproduction number for seasonal, pandemic, and zoonotic influenza: a systematic review of the literature,” BMC infectious diseases, vol. 14, no. 1, p. 480, 2014. [26] D. M. Morens, J. K. Taubenberger, and A. S. Fauci, “Predominant role of bacterial pneumonia as a cause of death in pandemic influenza: implications for pandemic influenza preparedness,” Journal of Infectious Diseases, vol. 198, no. 7,

• pp. 962–970, 2008.

[27] D. M. Morens and A. S. Fauci, “The 1918 influenza pandemic: insights for the 21st century,” Journal of Infectious Diseases, vol. 195, no. 7, pp. 1018–1028, 2007. [28] G.-F. Sara, “La gripe de 1889-1890 en madrid,” 2017. [29] E. C. Holmes, E. Ghedin, N. Miller, J. Taylor, Y. Bao, K. St George, B. T. Grenfell, S. L. Salzberg, C. M. Fraser, D. J. Lipman, et al., “Whole-genome analysis of human influenza a virus reveals multiple persistent lineages and reassortment among recent h3n2 viruses,” PLoS biology, vol. 3, no. 9, p. e300, 2005. [30] G. Chowell, A. Erkoreka, C. Viboud, and B. Echeverri-D´ avila, “Spatial-temporal excess mortality patterns of the 1918– 1919 influenza pandemic in spain,” BMC infectious diseases, vol. 14, no. 1, p. 371, 2014. [31] J. Oeppen, D. Ramio Fari˜ nas, and S. Garcia Ferrero, “Estimating reproduction numbers for the 1889-90 and 1918-20 influenza pandemics in the city of madrid.,” ESPA ˜ NA, SALUD Y CIUDADES EN. [32] R. E. Serfling, “Methods for current statistical analysis of excess pneumonia-influenza deaths,” Public health reports,

• vol. 78, no. 6, p. 494, 1963.

[33] A. J. Cobos, C. G. Nelson, M. Jehn, C. Viboud, and G. Chowell, “Mortality and transmissibility patterns of the 1957 influenza pandemic in maricopa county, arizona,” BMC infectious diseases, vol. 16, no. 1, p. 405, 2016. [34] B. Efron and R. J. Tibshirani, An introduction to the bootstrap. CRC press, 1994. [35] G. Chowell, C. Viboud, L. Simonsen, M. A. Miller, R. Acuna-Soto, J. M. Ospina D´ ıaz, and A. F. Mart´ ınez-Mart´ ın, “The 1918–19 influenza pandemic in boyaca, colombia,” 2012. [36] S.-E. Mamelund, “A socially neutral disease? individual social class, household wealth and mortality from spanish influenza in two socially contrasting parishes in kristiania 1918–19,” Social Science & Medicine, vol. 62, no. 4, pp. 923– 940, 2006. [37] R. E. SERFLING, I. Sherman, W. J. HOUSEWORTH, et al., “Excess pneumonia-influenza mortality by age and sex in three major influenza a2 epidemics, united states, 1957-58, 1960 and 1963.,” American journal of epidemiology, vol. 86,

• no. 2, pp. 433–41, 1967.

[38] A.-J. Valleron, A. Cori, S. Valtat, S. Meurisse, F. Carrat, and P.-Y. Bo¨ elle, “Transmissibility and geographic spread of the 1889 influenza pandemic,” Proceedings of the National Academy of Sciences, vol. 107, no. 19, pp. 8778–8781, 2010. [39] A. Gagnon, J. E. Acosta, J. Madrenas, and M. S. Miller, “Is antigenic sin always original? re-examining the evidence regarding circulation of a human h1 influenza virus immediately prior to the 1918 spanish flu,” PLoS Pathog, vol. 11,

• no. 3, p. e1004615, 2015.

15

Appendix

For all figures: Points represent actual weekly mortality rates from 1917-1922, and lines show mean and upper 95% bound baseline from simulated 1917 deaths data. Vertical lines indicate breakpoints from which excess mortality is generated.

Figure 5: All Cause and Respiratory-Related Mortality Baselines, Ages 0-4 16

Figure 6: All Cause and Respiratory-Related Mortality Baselines, Ages 5-14 17

Figure 7: All Cause and Respiratory-Related Mortality Baselines, Ages 15-24 18

Figure 8: All Cause and Respiratory-Related Mortality Baselines, Ages 25-49 19

Figure 9: All Cause and Respiratory-Related Mortality Baselines, Ages 50-69 20

Figure 10: All Cause and Respiratory-Related Mortality Baselines, Ages 70+ 21