The 1918–19 Influenza Pandemic in Boyacá, Colombia

Timing of pandemic onset and prior immunity of populations varied by region.

December 1920, were manually retrieved from the parish archives of 78 municipalities in the department of Boyacá. From these archival records, we extracted age, cause, and exact date of death. To estimate mortality rates, we compiled weekly numbers of deaths from all causes and from respiratory illness, stratifi ed into 5-year age groups (Figures 2, 3). To obtain precise estimates of the transmission potential, we compiled daily death time series, combining all age groups.

Census Data
We obtained age-specifi c estimates of population size for the department of Boyacá from a 1918 census report (15). In 1918, ≈70% of Boyacá's population was located in rural areas. During 1912During -1918, the average annual population growth rate in Boyacá was 1.7%; during 1918-1920, it was 3.8%.

Estimation of Excess Deaths
For characterization of mortality rates for the Boyacá pandemic, infl uenza-associated mortality rates must be separated from background mortality rates (deaths from respiratory illness other than infl uenza) and considered separately for each age group and cause of death (respiratory or all causes). To estimate pandemic mortality rate, we can defi ne a discrete period of pandemic infl uenza activity and estimate the number of deaths in excess of background deaths that occurred during the pandemic period. Because mortality rates tend to vary seasonally throughout the year, our background estimate must also vary seasonally. To fi nd the best estimate for baseline mortality rate in the absence of infl uenza activity, we applied regression methods, using harmonic terms and time trends, to mortality rate data (6,16,17) (online Technical Appendix, wwwnc.cdc.gov/ EID/pdfs/10-1969-Techapp.pdf ).
The regression model determines the extent to which observed weekly mortality rate fi t the expectation of background mortality rate. Periods of poor fi t indicate that observed mortality rate exceeds typical baseline levels, presumably because of increased infl uenza activity.
We defi ned pandemic periods as the weeks when deaths from respiratory illness exceeded the upper limit of the 95% CI of the background model. To estimate the mortality rate during the pandemic, for each age group we summed the weekly number of deaths from respiratory illness and from all causes that exceeded model baseline rates during each pandemic period during 1918-20.
To ensure that our estimates were not sensitive to modeling assumptions, we also estimated excess deaths by using an alternative approach to calculate background deaths. In this approach, background mortality rates for a given week are obtained by averaging mortality rates during the same week in previous years (online Technical Appendix).
Finally, we estimated a relative measure of the effects of pandemic-associated deaths for each age group, which considers the typical mortality rate experienced by that age group. We calculated relative risk for pandemic-associated death, defi ned as the ratio of excess deaths during pandemic periods to expected baseline deaths. Relative risk has been shown to facilitate comparison between age groups or countries, which have different background risks for death (17,18).

Comparing Patterns of Age-specifi c Deaths
We compared patterns of age-specifi c excess deaths from the 1918-19 Boyacá pandemic with those recently published for Mexico City (3). The estimates for Mexico City were based on excess-death rates obtained from monthly pneumonia and infl uenza records (1916)(1917)(1918)(1919)(1920), stratifi ed by 6 age groups (<5, 5-19, 20-29, 30-49, 50-69, and >70 years). Excess-death rates for Mexico City were calculated with a method similar to that used in this study.
We also reviewed key epidemiologic features of the pandemic in various locations as recently reported (1,3,4,(6)(7)(8)(9)19,20), focusing on comparisons of overall excess-death rates associated with the pandemic. We also reviewed epidemiologic evidence for early (herald) waves occurring before September 1918 and for death sparing among elderly persons. We limited the review to studies that provided monthly or weekly historical death data because such data enable identifi cation of herald waves and precise estimation of excess-death rates.

Estimation of Transmission Potential
Transmissibility of an infectious pathogen is measured by the basic reproduction number (R 0 ), which is the average number of secondary infections generated by an infectious person in an entirely susceptible population (21). A related quantity is the reproduction number, R, which can be used for partially immune populations who have been vaccinated or previously exposed to similar pathogens (21).
We estimated R for the 1918 pandemic virus in Boyacá by using a simple method that relies on the epidemic growth rate, a measure of how fast the number of cases increases over time (online Technical Appendix). Briefl y, in the early ascending phase of an epidemic, the daily number of cases (or deaths) should follow an exponential function. By taking the log of daily deaths in the ascending phase, a straight line can be fi t to the data. R can be derived from the growth rate estimate r by a simple equation involving the duration of the latency and infectious periods (22) (online Technical Appendix).
Because of the uncertainty associated with duration of the latency and infectious periods for infl uenza, we considered periods of 1.5 and 2 days each (23,24). Latency and infectious periods can be summed into a single statistic called the generation interval, which measures the interval between disease onset in 2 successive cases. The generation intervals considered in this study were 3 and 4 days (23,24).
We defi ned the ascending phase as the period between the day of pandemic onset (defi ned as the fi rst day of the period of steadily increasing deaths) and the day immediately before the epidemic peak. We tested the robustness of R estimates to the choice of death indicator (deaths from respiratory illness or from all causes). We also compared estimates derived from crude numbers of deaths and excess deaths from respiratory illness that were above background rates.
The same approach and assumptions have been used to quantify Rs associated with the 1918 pandemic in 50 Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 18, No. 1, January 2012 Copenhagen, Denmark, and Mexico City, and hence the Boyacá estimates are directly comparable to estimates from these studies (3,6). For comparison with Boyacá, we also reviewed the literature for published estimates of R associated with the 1918 pandemic in the Americas (2)(3)(4).

Timing of Pandemic Waves and Age-specifi c Patterns of Death
The age-stratifi ed time series of deaths from respiratory illness or all causes in Boyacá indicated that a severe pandemic wave occurred during a 15-week period, October 20, 1918-January 26, 1919 (Figures 2, 3). The profi le of age-specifi c excess deaths from respiratory illness associated with the pandemic period formed a W-shaped pattern; peak mortality rates among infants (<5 years of age) were followed by peak rates among elderly persons (>60 years) and young adults (25-29 years) ( Table 1). Excess deaths were lowest among children 5-14 years of age and adults 50-59 years of age. Similar age patterns were found for all-cause deaths ( Figure 4); the correlation coeffi cient between respiratory and all-cause excess-death rates was >0.99 (p<0.01). Excess deaths from respiratory illness captured most infl uenza-related all-cause excess deaths across all age groups (95% on average, range 81%-100%). Confi dence intervals were larger for the most extreme age groups.
To facilitate the comparison between population age groups with different background risks for death, we calculated the risk for excess-death rates relative to baseline rates (Table 1). Although absolute excess-death rates were highest for young children (0-4 years of age) and elderly persons (>60 years), during the pandemic the relative risks were lowest for these age groups. Relative risk was highest for young adults 25-29 years of age; excess-death rates increased 51-fold above background death rates for respiratory causes and 6-fold for all causes.
Comparison of Boyacá and Mexico City shows that age-specifi c excess-death rates produced a W-shaped pattern for both locations ( Figure 5). However, excess-death rates among young adults (20-29 years) were substantially higher for Mexico City than for Boyacá. By contrast, excess-death rates among infants were 2-fold lower for Mexico City than for Boyacá. Excess-death rates for elderly persons were similar for both cities. Overall, we estimate that the October 1918-January 1919 pandemic period was associated with 47 and 40 excess respiratory deaths per 10,000 population in Mexico City and Boyacá, respectively.  A broader comparison of epidemiologic patterns associated with the pandemic at 12 locations on different continents highlights substantial variations in the timing, number of pandemic waves, and age-specifi c death rates ( Table 2). Europe and the United States generally experienced herald waves in during March-August 1918 (except for Paris) and low excess-death rates among elderly populations. In contrast, there was no evidence of death sparing among elderly populations in Latin America or Asia, and herald waves occurred at 4 of the 7 locations studied in these regions. Excess-death rates from respiratory illness were high for Iquitos, Peru; Toluca, Mexico; and Basque Provinces, Spain (121-288 deaths/10,000 population); intermediate in Taiwan (78-180 deaths/10,000 population); and lower elsewhere, including in Boyacá (29-67 deaths/10,000 population). Table 3 provides summary estimates for the R for the 1918 infl uenza pandemic in Boyacá, based on growth in daily rates for death from respiratory illness. R was estimated to be 1.4, assuming a short generation interval of 3 days, and 1.5-1.6, assuming a longer interval of 4 days. A sensitivity analysis, based on excess deaths from respiratory illness occurring above a background of expected deaths, generated slightly higher R estimates (1.4-1.5 for a generation interval of 3 days and 1.6-1.7 for a generation interval of 4 days). Different approaches for estimating background deaths resulted in R estimate differences of <0.06 (4%).

Reproduction Number Estimates
Comparison of estimates derived from different locations in the Americas revealed some geographic variations in the transmission potential of the 1918-19 pandemic wave (Table 4). Although R estimates were 1.3-1.8 in most locations in the Americas, assuming a 3-day generation interval, the transmissibility of infl uenza during the autumn wave might have been particularly high in Toluca, Mexico (estimated R = 2.0-2.5).

Discussion
Our study makes use of extensive archival death records covering before and during the 1918-19 infl uenza pandemic in Boyacá, Colombia, and confi rms the substantial number of deaths caused by the pandemic in this region. The main epidemiologic features of the pandemic in Boyacá include a single wave of excess deaths during October 1918-January 1919; high excess-death rates among infants and elderly  persons; and a moderate R (estimated at 1.4-1.5, assuming a 3-day generation interval). We did not identify a herald wave of deaths from pandemic infl uenza in the early part of 1918 in Boyacá. According to epidemiologic data, herald waves of mild pandemic activity have been reported for the spring and summer of 1918 in other regions of the world, including New York City (1), Mexico (3), Lima (4) Geneva (25,26), Copenhagen (6), military camps in the United States (6), the United Kingdom (27), and Singapore (9). The absence of a herald wave in Boyacá could be explained by late introduction of the pandemic infl uenza virus; alternatively, a mild fi rst wave may have occurred without causing many deaths. Thus, we cannot rule out early pandemic activity, which might have been associated with mild illnesses, before October 1918 in Boyacá. For instance, the summer pandemic wave of 1918 in Denmark was clearly evident only from time-series case data (6). These epidemiologic fi ndings suggesting early pandemic virus activity have recently been confi rmed by sequencing of pandemic infl uenza virus specimens isolated from Army camp populations in the United States as early as May 1918 (28).
Although substantial postpandemic waves have been reported for 1919-20 in New York City (1), Mexico City (3), Lima (4), Japan (5), and Taiwan (8), we could not identify a clear recrudescent pandemic wave in 1920 in Boyacá. A 3-week period in January 1920 and a 4-week period in April-May 1920 were associated with a small increase in deaths from respiratory illness, mostly affecting elderly persons, but we cannot with certainty attribute these deaths to pandemic infl uenza. Early public health warnings and effective implementation of control interventions in large cities such as New York City, Mexico City, Lima, and Taiwan, could have contributed to maintaining a large pool of susceptible persons, which could fuel subsequent pandemic waves (29). In Japan, postpandemic waves were somewhat limited to regions that escaped earlier waves (5). Given that Boyacá was a relatively small rural area, pandemic activity in 1918 might have proceeded unabated, with no particular interventions, medical or nonmedical. Alternatively, Boyacá could have escaped the recrudescent pandemic wave in 1920 because of its remote location. Overall, the main wave of deaths from pandemic infl uenza that occurred during October 1918-January 1919 in Boyacá is reminiscent of the single wave of pandemic infl uenza A (H1N1) 2009 wave that occurred in the Southern Hemisphere during the winter of 2009 (e.g., Chile [30], Australia [31], and New Zealand [31]). Additional data from the 1918 pandemic in other Southern Hemisphere locations are warranted before these fi ndings can be generalized.
The W-shaped age-specifi c pattern of deaths during the 1918-19 pandemic wave in Boyacá is in agreement with recent reports from the Mexico City area (3) and Peru (4). These reports suggest a lack of death sparing among elderly populations of urban and rural areas of Latin America, although data from additional locations would be useful for generalizing these conclusions. This pattern is also in agreement with anecdotal evidence from aboriginal populations in Alaska in 1918 (32). In contrast to reports for Latin America and Alaska, reports for the United States and Europe suggest that elderly populations were substantially protected from infl uenza-associated death in 1918 (1,5,6). Previous studies have hypothesized that childhood exposure to infl uenza A (H1N1) viruses before 1870 might account for prior immunity among elderly persons during the 1918 pandemic. A similar phenomenon has been noted for pandemic (H1N1) 2009, during which risk for clinical infection and death was lower during the pandemic than during seasonal epidemics for persons >60 years of age (13,33).
Regional differences in prior immunity to infl uenza might result from heterogeneous circulation of infl uenza viruses during the 19th century, when long-distance travel was much less common than it is today (3). In 1918, Colombia's population of 5.8 million was heterogeneously distributed and relatively isolated from the rest of the world (34); this isolation could explain the lack of exposure to infl uenza viruses during the middle of the 19th century. Also in 1918, transportation was underdeveloped in Colombia, consisting mostly of horse-or mule-drawn street cars, waterways, and sparse railroads that did not connect with Boyacá (34). Remoteness could have affected the probability of introduction and of local dissemination of infl uenza viruses in the Boyacá region. A similar phenomenon could also explain the apparent lack of a herald pandemic wave in the spring of 1918, when pandemic virus activity was not yet globally widespread. Of note, the capital city, Bogota, was the fi rst area in Colombia to report increased infl uenza activity in October 1918; the virus quickly spread to other Colombia locations (34).
Excess-death rates among young adults were lower in Boyacá than in Mexico City (3). The reasons for this difference are unclear but could be associated with a more sporadic distribution of the population in Boyacá, resulting in lower overall infl uenza attack rates; however, we do not have epidemiologic evidence to support this assumption. Alternatively, the unidentifi ed factors that made young adults particularly susceptible to infl uenza-related death in Europe, the United States, and Mexico in 1918 (1,5,6) might have been less common among young adults in Colombia. Despite these geographic differences in absolute risk for death from pandemic infl uenza, in all locations with suffi cient data the relative risk for death consistently peaked among adults 20-29 years of age when compared with baseline death rates during nonpandemic years. Hence, our study confi rms the universal atypical severity of this virus in young adults, as previously reported for the United States (1), Mexico (3), Europe (6,7), and Taiwan (8). We also note that data from Boyacá and Mexico City do not support the pessimistic hypothesis that populations lacking prior immunity to the 1918 virus would experience a V-shaped age-associated risk for death, in which risk would rapidly and continuously rise past teenage years (35).
Our excess-deaths approach warrants some caveats. The regression model used to estimate background deaths poorly fi t the Boyacá data during the nonpandemic period, probably because of weak seasonality. However, our estimates of excess deaths from pandemic infl uenza based on deaths from respiratory illness and all causes were highly correlated, similar to those from other temperate countries, where baseline death rates are more seasonal (1,3,6). The sensitivity analysis that we conducted by using an alternative approach to estimate background deaths did not make assumptions about seasonality (20). This analysis produced excess-death estimates highly correlated with those derived from the regression approach (correlation = 0.97; p<0.01; mean difference 4%-7%).
Transmissibility estimates derived from 1918-20 pandemic illness and death data are 1.5-5.4 for communitybased settings in several regions of the world (2,6,36,37) ( Table 4). Our transmissibility estimates for Boyacá, Colombia, assuming a generation interval of 3 days, are in close agreement with those reported for the wave in autumn in Mexico City (3), Lima (4), England and Wales (27), and Copenhagen (6) and slightly lower than estimates reported for the city of Toluca, Mexico (3), and US cities (2,38). Boyacá's sparsely distributed population could explain why the estimated disease transmissibility is relatively low. It remains unclear whether differences in reproduction number estimates across locations and pandemic waves refl ect true differences attributable to variation in attack rates or local factors affecting transmission or merely illustrate diffi culties in measuring this parameter with precision (38). In previous studies focused on reproduction number estimates in which we used similar data and approaches, we have shown that inclusion of a delay between disease onset and death has little effect on the estimates (39).
In conclusion, historical studies from understudied areas are especially helpful for documenting the global death rates and transmission patterns of the 1918 pandemic and for revealing substantial variations among locations. In particular, the lack of death sparing for elderly persons in Colombia and Mexico differs markedly from contemporaneous observations in the United States and Europe. During the 19th century, the Latin American region was relatively isolated (and still is today) (40), which would affect the circulation of historical infl uenza viruses and baseline population immunity to infl uenza. We believe that this fi nding suggests recycling of infl uenza viruses as the best explanation for death sparing among elderly persons in the United States and Europe in 1918. Preservation and interpretation of archival epidemiologic data are crucial for a better understanding of past pandemics and for better preparedness against future pandemics.
*Values are based on a range of estimates provided by considering different distributions of the generation interval (exponentially distributed latent and infectious periods or fixed generation interval). R, reproduction number. †R estimates are based on the mean of the initial growth rates across 45 US cities.