Why the 1918 flu hit some cities harder than others | by Eric Jaffe | Sidewalk Talk
Chicago officials inspect street cleaners during the 1918 flu. (Bettmann via Getty)
An evidence-based list of several lessons worth remembering in the time of coronavirus.
For years, it was commonly believed that the 1918 flu indiscriminately destroyed everything in its path. But subsequent research has found that, in fact, some cities were hit much harder than others — as were some urban populations within the same cities. With the novel coronavirus now officially a pandemic, and city leaders starting to act, it’s worth looking back to understand which factors made a difference in urban health outcomes during the last big outbreak.
To that end, I reviewed seven studies from the past 15 years that focused on factors explaining the variation in 1918 flu severity across U.S. cities. The encouraging news (a relative term, for sure) is that many of these factors are within a city’s immediate control, such as the nature and timing of social distancing interventions. Other key factors typically change over longer timelines, such as air pollution and social health disparities, but remain important to know.
Perhaps the most intuitive factor — population size — doesn’t actually seem to have played a big role. For example, New York City fared relatively well during the 1918 flu, largely thanks to strong and swift interventions. (That’s not an invitation to go coughing on the subway.)
Sometimes it’s best to put the caveats up top: It’s far too soon to say the coronavirus is on par with the 1918 flu. The world of 2020 is also very different from that of 1918, particularly with regard to epidemiological insight and medical technology. Since these studies focus on the U.S., it’s unclear how well the findings might explain flu severity in other global cities a century ago. And of course, many factors influence individual and population health outcomes alike.
With those cautions in mind, here are some key takeaways:
Social distancing and quarantine measures worked.
Some cities have started to take severe measures to slow the spread of infection — buying important time for medical workers and hospitals to prepare for new cases, and ultimately for the deployment of a vaccine (though that’s far away). Here in the U.S., a containment area has been set up around New Rochelle, just outside New York City, and both Seattle and San Francisco have put a halt to large public gatherings.
For the 1918 flu, the nature of such local interventions went a long way toward determining health outcomes across different cities. That was the main finding of a study published in a 2007 issue of the Journal of the American Medical Association (JAMA). The research team — led by Howard Markel of the University of Michigan — studied “non-pharmaceutical” interventions taken by 43 U.S. cities of at least 100,000 people from September 1918 to February 1919,* at the height of the outbreak.
Markel and company focused on three interventions in particular: closing schools, cancelling public gatherings (defined as closing bars, public entertainment venues, sporting events, and other types of large indoor gatherings), and implementing quarantines (defined as mandatory isolation). Each of these measures was legally enforced.
All 43 cities adopted at least one of the measures. But cities that fared best imposed two of the measures concurrently: closing schools and restricting public gatherings. Doing those things in tandem led to significant reductions in weekly “excess” deaths attributable to the 1918 flu (defined as more deaths than occurred on average in the preceding years of 1910 to 1916).
The researchers conclude:
These findings contrast with the conventional wisdom that the 1918 pandemic rapidly spread through each community killing everyone in its path. Although these urban communities had neither effective vaccines nor antivirals, cities that were able to organize and execute a suite of classic public health interventions before the pandemic swept fully through the city appeared to have an associated mitigated epidemic experience.
But early intervention mattered — a lot.
Markel and colleagues also found that the timing of city interventions mattered as much as the nature of them. Cities did better when they implemented measures earlier. The earlier a city acted, the longer it took to reach its peak mortality rate, the lower its peak mortality rates, and the fewer its total deaths.
Here the cases of New York, St. Louis, and Pittsburgh prove illustrative:
New York. Given its size and density, New York would have seemed a terrible target for the 1918 flu. But the city’s health department was all over the problem, implementing behavioral interventions—including a strict quarantine and staggered business hours—11 days before excess flu deaths spiked (defined as the moment the weekly excess death rate exceeded twice the city’s baseline pneumonia and influenza death rate). Partly as a result of this early and sustained response (more on response duration in the next section), New York had the lowest death rate on the Eastern seaboard, with 452 excess flu deaths per 100,000 people.
St. Louis. Similarly, St. Louis implemented an early and combined set of interventions, including school closures and public gathering cancellations, beginning one day after excess flu deaths spiked over baseline. It also sustained these measures for 10 weeks. Again, partly in response to these efforts, the city ranked 7th of 43 in the 2007 study in terms of excess deaths, with just 358 per 100,000 people.
Pittsburgh. In contrast, Pittsburgh imposed a gathering ban a week after its flu deaths spiked over baseline, and it waited several more weeks to close schools. Compounding the problem, it actually reopened public gatherings shortly thereafter. Partly as a result of the delayed — and unsustained — interventions, Pittsburgh ranked last in the 2007 study of excess deaths, with 807 per 100,000 people.
As the researchers put it:
Late interventions, regardless of their duration or permutation of use, almost always were associated with worse outcomes.
Sustained intervention mattered, too.
So timing mattered. But so did duration. As the above cases suggest, the 2007 JAMA study found a significant link between keeping city measures active for longer and overall mortality during the 1918 flu.
Here the case of Denver is instructive. The city started late — nine days after its initial flu spike occurred. Some early success led to a false sense of security, and the city let up on some of its measures, ending school closures and easing bans on public gatherings. That led to subsequent flare-ups and a second spike in deaths. Partly as a result of this inconsistent approach, Denver’s excess death rate ended up at 631 out of 100,000 people.
Denver initially slowed the spread of the 1918 flu by closing schools and restricting gatherings, but when those policies let up, the virus returned and the city suffered another death spike. (via Markel et al., 2007, JAMA)
Another 2007 study — this one published in the journal PNAS — reached similar conclusions about the need for an early and sustained city intervention program. Applying an epidemic model to a sample of 16 cities, the researchers found that early and sustained interventions reduced transmission rates by up to 30 to 50 percent in some cities (including St. Louis). Places that introduced measures too late or lifted them too early experienced more moderate reductions (10 to 30 percent).
All told, this study found that the timing of an intervention explained nearly half (44 percent) of the variance in disease severity across cities during the 1918 flu, even after controlling for factors such as baseline health disparities from place to place. But those measures had to stay in place a long time; the research doesn’t say just how long, but suggests keeping them in place until a vaccination is ready, or “perhaps as long as 6 months.” They conclude:
The most important conclusion from this work is that the timing of public health interventions had a profound influence on the pattern of the autumn wave of the 1918 pandemic in different cities. Cities that introduced measures early in their epidemics achieved moderate but significant reductions in overall mortality. Larger reductions in peak mortality were achieved by extending the epidemic [measures] for longer.
Air pollution made things worse.
Bleak as that picture can seem today, it’s encouraging to know that the nature and timing of local interventions matter, because these things are under pretty direct and immediate city control. But another factor that had a significant impact on health outcomes in 1918 is harder to change on a moment’s notice: air pollution.
It’s true that cities do often lead on clean air policy (in particular green building laws), and that some clean air initiatives can have immediate pollution impacts (such as “car-free” days). But it’s also true that reducing air pollution often happens slowly over many years of policy intervention and technology advance, and that the cleanliness of a regional power grid is typically outside a city’s control.
To study the impact of pollution on the severity of the 1918 flu, a research team led by economist Karen Clay of Carnegie Mellon gathered “excess” flu death data on 183 U.S. cities with at least 10,000 people. They analyzed these flu deaths against several potential contributing factors, including pollution. While there’s an intuitive link between pollution and respiratory illness, scientists have only recently started to draw empirical connections.
Air quality data was sparse at the time of the 1918 flu (a clear study limitation). The best proxy the researchers could find was whether or not a city was within a 30-mile radius of a large coal-fired power station. As a control measure, they also analyzed cities near hydroelectric plants with no emissions.
Clay and company found “sizeable” effects of pollution on 1918 flu severity, leading to upwards of 6,000 excess infant deaths and upwards of 23,700 excess deaths across all ages. Critically, the effect of pollution on mortality showed up in 1918 but not in any other years of the study (1915–1925), suggesting a strong connection. Moreover, the researchers found no relationship between mortality in 1918 (or any year) and hydroelectric capacity.
The researchers conclude in a 2018 issue of the Journal of Economic History:
From a policy perspective, the presence of these complementarities implies that there may be considerable co-benefits of pollution abatement policies that are not accounted for by conventional cost-benefit evaluations.
In a follow-up study — published in 2019 in Economics & Human Biology — the researcher confirmed their findings in a larger sample of 438 U.S. cities. That work concluded that cities in the top third of proximity to coal plants had 9 excess deaths per 10,000 residents more than those in the bottom third. This effect remained significant even after controlling for things like pre-pandemic baseline health rates (such as infant mortality), further suggesting a strong connection.
Social disparities also worsened the impact.
Research has also found a link between 1918 flu outcomes and existing social disparities — both at a whole city level and an individual neighborhood level. That’s unfortunately not a big surprise given urban inequality and poverty levels, but it’s important to document with regard to pandemics, both for planning long-term prevention and deploying short-term measures.
The city-level finding comes from the 2019 work of Clay and collaborators. In addition to linking flu severity and pollution, the researchers found more excess flu deaths in cities that already had high infant mortality rates and low literacy rates — indicators of health and other social disparities — at the time the pandemic struck. For each factor, cities in the top third experienced about 21 excess deaths compared to those in the bottom third.
As the researchers put it: “it appears that much of the relationship between poverty and pandemic mortality can be explained by the poor health in low-income populations.”
A 2016 study, published in PNAS, took an even more local look at the same relationship as it played out in Chicago. Evaluating data from 496 census tracts across a seven-week period in the fall of 1918,* the researchers found that flu and pneumonia mortality increased an average of 32 percent for every 10 percent increase in illiteracy rate — even adjusting for population density, employment, and age.
Meanwhile, the study found a negative relationship between flu severity and homeownership rates (often a proxy for high income). In other words, Chicagoans who owned homes experienced less severe flu outcomes than those who didn’t — further underscoring the role of structural social disparities on pandemic impact. The authors conclude:
This study found that, despite the highly virulent nature of the virus, influenza did not behave in a wholly democratizing fashion at the within-city scale. …
This study suggests that people living in underprivileged neighborhoods, in fact, experienced significantly higher mortality and that the outbreak spread in a spatiotemporally dependent manner.
Perhaps surprisingly, population size didn’t seem to play a role.
While it’s of course easier for communicable disease to spread among crowds of people, the connection between population size and 1918 flu severity is not as direct as one might expect. In fact, among big cities, smaller ones seem to have suffered worse than bigger ones.
That finding comes from a 2011 study published in the journal PLoS One. The research looked at both flu and pneumonia deaths in the 66 largest U.S. cities circa 1918 (as measured by the 1920 census). Across this sample, population size had no significant impact on pneumonia death rates. When it came to flu deaths, however, “smaller” big cities “suffered a disproportionately large mortality burden due to influenza in 1918, as compared with larger cities.”
Other studies have reached similar conclusions. The 2007 PNAS study mentioned earlier found no significant connection between excess 1918 flu deaths and population size or density. The study of Chicago disparities found mixed evidence, linking density and flu transmission but not density and mortality. A 2007 study of British cities found no relationship between flu rates and residential crowding (though it found that cities were hit harder than rural areas).
Just why population size wasn’t strongly connected to 1918 flu mortality is unclear. It’s possible the biggest cities, for all their potential health hazards of crowding, had more resources to direct at the problem: more efficient hospitals, more access to medical care, more information networks. That’s worth a study of its own, but it suggests, at least, that density isn’t destiny—and quick-acting, long-lasting interventions can make a difference.
*These dates mistakenly read 2018 instead of 1918 when this post was first published. The typos have been fixed.
Follow Sidewalk Labs with our weekly newsletter and our podcasts.