Severe mortality impact of the 1957 influenza pandemic in Chile

Introduction Epidemiological studies of the 1957 influenza pandemic are scarce, particularly from lower‐income settings. Methods We analyzed the spatial–temporal mortality patterns of the 1957 influenza pandemic in Chile, including detailed age‐specific mortality data from a large city, and investigated risk factors for severe mortality impact across regions. Results Chile exhibited two waves of excess mortality in winter 1957 and 1959 with a cumulative excess mortality rate of 12 per 10 000, and a ~10‐fold mortality difference across provinces. High excess mortality rates were associated with high baseline mortality (R 2=41.8%; P=.02), but not with latitude (P>.7). Excess mortality rates increased sharply with age. Transmissibility declined from R=1.4‐2.1 to R=1.2‐1.4 between the two pandemic waves. Conclusions The estimated A/H2N2 mortality burden in Chile is the highest on record for this pandemic—about three to five times as severe as that experienced in wealthier nations. The global impact of this pandemic may be substantially underestimated from previous studies based on high‐income countries.

The first local outbreaks of pandemic influenza A/H2N2 virus activity were reported in continental China during February-March 1957, and the emerging virus quickly reached Hong Kong and other parts of Asia in a matter of weeks. 10 The pandemic reached Europe in June 1957 and South American countries of the Pacific coast, New Zealand, and South Africa in July of that year. 10,11 In the United States, sporadic outbreaks were reported during June-August 1957, but widespread transmission was not observed until September as schools reopened.
In this study, we collected and analyzed archival epidemiological data from Chile, a low-income country at the time that was hit particularly hard. 7 We characterize geographic variation in influenza-related mortality patterns across Chilean provinces in the first 3 years of pandemic virus circulation and investigate age mortality patterns and transmissibility estimates.

| National-and Province-level Vital statistics, 1953-1959
Chile had a population size of 6.9M in 1957 and was an emerging economy with excellent vital statistics records, a life expectancy at birth of just under 60 years, and a high infant mortality rate of 110 deaths per 1000 live births. 12 We obtained monthly national all-cause and respiratory mortality statistics for Chile from 1953 to 1959 from the Statistical Yearbooks of Chile. The choice of these study years allowed for estimation of a pre-pandemic mortality baseline, and covers two waves of pandemic influenza A/H2N2 activity in 1957-59.
To further explore geographic variation in pandemic-related mortality in Chile, we also compiled monthly all-cause mortality statistics from 1953 to 1959 for the 25 administrative provinces 12 (Fig S1 in Appendix). Of note, because Chile has undergone changes in administrative divisions over the years, the geographic areas referred to in our study only approximately translate to contemporary administrative divisions in this country. We also obtained province-level population size estimates and infant mortality rates in 1956. 12 We compiled latitude coordinates for province-specific population centers to explore pandemic timing and impact along Chile's latitudinal gradient (17°S to 56°S 13 ).

| Individual death certificates from the city of Concepcion, 1953-1959
Given the lack of age detail in the official province-level vital statistics, separate data sources were used to study the age patterns of pandemic mortality. The City of Concepcion, located in the southern central part of Chile, is the third largest city in this country and has preserved unique records of individual death certificates. 13 We manually retrieved from the city's civil registry all 15 736 individual death certificates for the period of January 1953-December 1959. For each record, we tabulated the age, cause, and date of death. We then created weekly all-cause and respiratory mortality time series stratified into six age groups (<5 years, 5-14 years, 15-24 years, 25-49 years, 50-64 years, and >=65 years). We used a broad definition for respiratory deaths which included deaths from influenza, pneumonia, bronchopneumonia, or bronchitis as underlying causes. Less than 1% of records lacked age or cause of death information.
We obtained age-specific population data for Concepcion from the Instituto Nacional de Estadísticas for year 1955 (total population of 170 457).

| Weekly mortality counts, Santiago and Concepcion, 1956-1957
In order to characterize the reproduction number of the pandemic, we also obtained weekly all-cause and respiratory mortality counts for 1956-57 for the capital city, Santiago, from an official report. 14,15 Based on daily individual death certificates collected for Conception, we compiled weekly mortality time series for this city as well.

| Review of local newspapers
To support the identification of pandemic periods in mortality data, we gathered anecdotal information on the temporal course and severity of the pandemic by reviewing the most prominent daily news-

| Excess mortality estimates
To quantify the mortality burden associated with the influenza pandemic, we estimated the cumulative number of deaths occurring in excess of a model baseline during the period 1957-59. Because mortality fluctuates seasonally throughout the year during this time period in Chile, we fit cyclical regression models with temporal trends and harmonic terms [16][17][18] to monthly mortality in the pre-pandemic period 1953-56 to estimate baseline mortality. Pandemic periods were defined separately for each study population as the months when allcause or respiratory mortality exceeded the upper 95% confidence limit of the baseline. The absolute mortality burden of the pandemic was estimated as the sum of excess deaths during each pandemic period of 1957-1959.
We also calculated the relative mortality burden of the pandemic, defined as the ratio of excess mortality during the pandemic periods over the expected baseline mortality for these periods. This relative approach facilitates comparison between countries, regions, and age groups with different background mortality risks. 2,17 Further, we performed sensitivity analyses to check the robustness of our excess mortality estimates to the choice of the mortality baseline and epidemic periods (Supplement).

| Geographic patterns
We analyzed geographic variation in timing and severity of the pandemic across the 25 provinces of Chile (Fig S1 in Appendix). For each geographic area, pandemic peak timing was defined as the month with maximal all-cause mortality elevation in each winter during 1957-1959. Severity was defined as absolute and relative rates of excess mortality for each pandemic wave during 1957-1959, and cumulatively. We used univariate analyses and multivariate stepwise regression models to explore the association between province-level estimates of peak timing and severity, and covariates such as latitude, population size, baseline and infant mortality rates.

| Estimation of transmission characteristics (reproduction number)
The reproduction number is an important parameter for pandemic preparedness as it quantifies the transmissibility potential of a pandemic virus and directly informs the strength of mitigation efforts required to interrupt transmission. Reproduction number estimates were derived from the weekly growth rate in respiratory deaths 19,20 in the two large cities included in our study, Santiago and Concepcion.
The pandemic growth rate was estimated by fitting an exponential function to the increase in weekly number of respiratory deaths at the start of each pandemic wave, 21 assuming exponentially distributed latent and infectious periods or a fixed generation interval 19 of three or four days. 22,23 We also tested the robustness of estimates to the choice of mortality outcomes, ranging from highly specific (excess respiratory deaths) to highly sensitive (all-cause deaths).

| Newspaper information
The earliest reports of localized pandemic activity were on July 24, 1957, in Valparaiso, a large seaport north of Santiago, and influenza outbreaks were reported in Santiago 2 days later. The introduction of the pandemic virus was traced back to a US Navy vessel originating from an American port that had stopped in Valparaiso; an influenzalike-illness outbreak had been reported among the crew. Reports of a recrudescent influenza wave in September and October of 1959 were published in all three newspapers. In contrast, there was no mention of pandemic activity in the winter of 1958.  (Table 1).

| Geographic patterns, pandemic waves, and excess mortality
While no significant elevation in mortality rates was detected from national time series data in winter 1958, province-level data revealed a minor pandemic wave in five Chilean provinces located in the north (n=3) and central regions (n=2) ( Table 1) (Table 1). There was no association between excess mortality rates in successive pandemic waves, so that the severity of the second wave was not predictive of the se-  Table S1 and Fig S5).

| Pandemic age mortality patterns (Concepcion)
Age-specific analyses were limited to Concepcion due to data availability. Based on weekly mortality time series in Concepcion, the brunt of pandemic mortality occurred in July-December 1957, while relatively minor mortality elevations were observed in winter 1958 and 1959 ( Figure 3). All-age excess respiratory death rates were estimated at 15.3, 1.5, and 4.3 per 10 000 for these three putative pandemic periods, for a total pandemic burden of 21.2 per 100 000. The all-cause excess mortality estimate was similar (22.2 per 10 000 for the total pandemic period), indicating that Concepcion ranked relatively high among provinces with respect to pandemic severity.
Age-specific excess mortality rates displayed a J-shape in all three pandemic periods ( There was a shift toward older ages in the third pandemic wave in 1959, with 50-to 64-year-olds experiencing the highest relative burden.

| Transmissibility estimates
Reproduction number estimates ranged from 1.7 to 2.0 for the 1957 pandemic wave in Santiago, assuming a three-day serial interval (Table 3; see  for the fit of the estimation procedure). Estimates were somewhat lower for the city of Concepcion, ranging from 1.4 to 1.7 (Table 3). Estimates were robust to use of different death outcomes, such as all-cause mortality or excess mortality. The reproduction number for the 1959 recrudescent wave at 1.2-1.3 in Concepcion was lower than for the first pandemic wave in 1957 (Table 3). We relied on excess mortality models applied to monthly allcause deaths to estimate pandemic burden at the province level, in line with prior studies. 24,25 Due to the severity of the 1957 pandemic in Chile, use of nonspecific mortality indicators such as monthly allcause mortality provides reliable estimates of influenza burden. 24,26 The validity of this approach is reinforced by the close agreement between all-cause estimates and more specific estimates derived from respiratory deaths for national-or city-level data, where both out-  proportionately from past influenza pandemics. [30][31][32] The detailed mortality estimates provided here for Chile reinforce the findings from an earlier global study, pointing to a particularly high impact of the 1957-59 pandemic in Chile. 7 Other studies have F I G U R E 3 Age-stratified weekly respiratory mortality rates per 10 000 in the city of Concepcion, Chile, 1953-1959 (black curve). Shaded areas highlight three winter periods (Jun-Dec) during 1957-1959. The Serfling seasonal regression model baseline (blue curve) and corresponding upper limit of the 95% confidence interval of the baseline (red curve) are also shown. Excess deaths are above the upper limit of the baseline mortality curve calibrated using mortality levels prior to the 1957 influenza pandemic T A B L E 2 Estimates of pandemic excess mortality rates, by pandemic wave, age groups, and cause of death, Concepcion, Chile, 1957-59. Excess mortality estimates were based on a seasonal regression model applied to weekly respiratory and all-cause mortality and presented as rates per 10 000 population Analysis of death certificates from Concepcion allowed for the estimation of age-specific burden. Excess mortality rates were lowest in school-age children and increased monotonously with older ages, following a J-shape reminiscent of US patterns. 2,34 This is indicative of lack of mortality protection among people older than 65 years, suggesting no prior exposure to antigenically related A/H2N2 viruses.

| DISCUSSION
In contrast, a Dutch study reported the presence of pre-pandemic protective antibodies to A/H2 in seniors over ~70 years. 35  and these findings are in line with recrudescent waves reported in all major influenza pandemics. 38 A final caveat relates to our excess mortality calculation approach, which was relatively simple, due to the lack of contemporaneous virologic surveillance and relatively weak temporal resolution of the mortality data. However, methods for estimating excess mortality data are not free of limitations. 43 For instance, the excess methodology employed in our study does not guarantee that excess mortality estimates are additive by age group. 44 Sensitivity analyses, however, indicate that our estimates are robust to modeling assumptions (Supplement).
F I G U R E 5 Weekly all-cause and respiratory death counts in Concepcion during weeks 28-38 in 1957 and the corresponding mortality baseline based on data from 1953 to 1957 during the same period (top). The basic reproduction number was estimated from the exponential growth fit to the initial phase of the excess mortality curve, which was computed by subtracting the baseline mortality curve to the 1957 pandemic period (bottom) In conclusion, our historical-epidemiological analysis reveals unusually high mortality during the 1957-59 influenza pandemic in Chile, the southernmost country of Latin America, with the highest excess mortality estimate published for this pandemic. We found a ~10fold difference in pandemic excess mortality rates between Chilean provinces, with overall baseline mortality rates explaining a moderate amount of variation (40%). Excess mortality rates increased with age, indicating a lack of measurable mortality sparing among senior populations. Overall, our study reinforces the importance of geographic variation in pandemic mortality burden between and within countries, echoing recent work on past pandemics. 7,9,29 Such variability needs to be recognized, especially in light of WHO's anticipated inclusion of a severity indicator as an element of future pandemic responses. We recommend that pandemic severity assessment should be based on multiple "sentinel" countries representing several world regions representing a variety of health and income conditions.