Boyacá is located in the central part of Colombia within the Andes Mountains at latitude ≈5.5°N (Figure 1). In 1918, the population of Boyacá was 659,947 and <50% of the area was occupied. Hygienic conditions were poor. A centralized disease notification system was lacking; however, death records were maintained by parishes.

The climate in Boyacá varies from high humidity and high mean temperature (≈40°C) in low areas near the Magdalena River (altitude 600 m) to cold mean temperature (<6°C) and permanent snow in the Cocuy Mountains (altitude 5,500 m). The 2 rainy seasons, April–May and October–November, produce ≈1,000 mm³/rainfall/year.

For characterization of mortality rates for the Boyacá pandemic, influenza-associated mortality rates must be separated from background mortality rates (deaths from respiratory illness other than influenza) and considered separately for each age group and cause of death (respiratory or all causes). To estimate pandemic mortality rate, we can define a discrete period of pandemic influenza 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 find the best estimate for baseline mortality rate in the absence of influenza activity, we applied regression methods, using harmonic terms and time trends, to mortality rate data (6,16,17) (Technical Appendix).

The regression model determines the extent to which observed weekly mortality rate fit the expectation of background mortality rate. Periods of poor fit indicate that observed mortality rate exceeds typical baseline levels, presumably because of increased influenza activity.

We defined 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 (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, defined 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).

We compared patterns of age-specific 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 influenza records (1916–1920), stratified 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 (1,3,4,6–9,19,20), 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 identification of herald waves and precise estimation of excess-death rates.

Transmissibility of an infectious pathogen is measured by the basic reproduction number (R₀), 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 (Technical Appendix). Briefly, 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 fit 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) (Technical Appendix).

Because of the uncertainty associated with duration of the latency and infectious periods for influenza, 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 defined the ascending phase as the period between the day of pandemic onset (defined as the first 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 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 (2–4).

The age-stratified 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 (Figure 2, Figure 3). The profile of age-specific 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 coefficient between respiratory and all-cause excess-death rates was >0.99 (p<0.01). Excess deaths from respiratory illness captured most influenza-related all-cause excess deaths across all age groups (95% on average, range 81%–100%). Confidence 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-specific 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-specific 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 influenza 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%).

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 influenza during the autumn wave might have been particularly high in Toluca, Mexico (estimated R = 2.0–2.5).