In the fall of 1918 the same pandemic influenza reached Philadelphia and St. Louis a few weeks apart, and the two cities acted on very different schedules. Philadelphia held a Liberty Loan parade on September 28 and began closing schools, churches and places of amusement on October 3. St. Louis closed its gathering places on October 7, within days of its first reported civilian cases. Analyses published in 2007 used that contrast, along with dozens of other cities, to argue for early, layered nonpharmaceutical interventions. This guide covers what each city did, what those analyses found, and how much weight the 1918 mortality data can actually carry.

September 1918: how the fall wave reached each city

The deadly fall wave in the United States is usually traced to military installations in the Boston area in late August 1918, and it moved outward along the routes that wartime traffic used. Philadelphia sat on one of the busiest of those routes. The Philadelphia Navy Yard received sailors from the Boston area in early September, and illness among sailors there appeared within days. The city was also one of the country’s major war production centers, with the Hog Island shipyard and other plants that had drawn in workers faster than housing could be built for them. Many of the city’s physicians and nurses were away in military service. None of that was a decision anyone made that September, and all of it shaped how fast the virus could move once it arrived.

St. Louis was farther from the eastern ports, and the wave reached it later. Military cases at Jefferson Barracks, south of the city, came before the first reported civilian cases, which Hatchett and colleagues date to early October. By then, newspapers and federal public health bulletins had carried weeks of reports from Boston, Philadelphia and other eastern cities, so St. Louis officials had an example in front of them that Philadelphia officials did not. That lead time comes from geography and travel patterns, and any honest comparison of the two cities has to count it, because the earlier city faced the same choices with less information.

The clinical picture matters for understanding why the measures available in 1918 were what they were. The pandemic struck young adults unusually hard. A 2008 analysis of autopsy material by David Morens, Jeffery Taubenberger and Anthony Fauci, published in the Journal of Infectious Diseases, concluded that most deaths were likely caused by secondary bacterial pneumonia following the viral infection. There were no antibiotics, no influenza vaccine and no mechanical ventilation, so a patient with that pneumonia had nursing care and little else. I find that sobering as a clinician, because it means the health officer’s toolkit consisted almost entirely of orders that changed how people gathered: closing schools, banning public meetings, isolating the sick, quarantining contacts, and in some cities requiring masks or staggering business hours.

September 28 in Philadelphia: the parade and the closure order five days later

The federal government was raising war money through a series of Liberty Loan bond drives, and the fourth drive opened in late September 1918. Philadelphia scheduled a parade along Broad Street for Saturday, September 28. The crowd is commonly given as about 200,000, and that number should be read as a contemporary estimate of a street crowd rather than a count. By parade day, influenza was established at the Navy Yard and civilian cases had been reported in the city. Wilmer Krusen, the city’s Director of Public Health and Charities, allowed the parade to proceed. Popular histories, including John M. Barry’s The Great Influenza (2004), describe physicians warning against the event beforehand and city officials offering public reassurance. I am paraphrasing those accounts rather than quoting anyone, because several lines attributed to Philadelphia officials that fall circulate without dependable sourcing.

Cases and deaths climbed steeply in the days after the parade, and the city’s hospitals were overwhelmed. On October 3 the city ordered schools, churches, theaters and other places of public amusement closed, and the state issued its own closure order shortly after. The worst of October followed anyway. Histories of the period describe the city morgue far beyond capacity, bodies waiting days for burial, shortages of coffins and gravediggers, and volunteers and religious orders taking on nursing work that the depleted medical workforce could not cover.

The parade is the most vivid single decision in the Philadelphia story, and it is often told as the cause of the city’s disaster. The record supports a narrower statement. Holding a mass gathering while transmission was already rising very plausibly accelerated spread, and the steep climb in illness over the following days is consistent with that, given influenza’s short incubation period. The virus was spreading in Philadelphia before September 28, though, and the city’s crowding and staffing shortages were already in place. Nobody can count the cases that the parade produced, so the parade is best used as an example of a gathering decision made late in the face of available warning, alongside the other conditions that made Philadelphia vulnerable.

October 7 in St. Louis: Starkloff’s order and the reopening fight

Max C. Starkloff was the St. Louis health commissioner in 1918, and Mayor Henry Kiel backed his use of emergency powers. On October 7, with the first civilian cases only days old, the city closed schools, theaters, churches and other places where crowds gathered, and it banned public gatherings. The order came before the city had a large number of cases on the books, which was the whole point of acting then, and it meant that Starkloff was asking businesses and congregations to accept losses for an epidemic most residents had not yet seen. The University of Michigan Center for the History of Medicine publishes a city-by-city account of the 1918 orders in its online Influenza Encyclopedia, and that is the place to check the exact scope and dates of each St. Louis measure.

The orders were contested. Theater owners and merchants pressed for relief as the weeks went on, and the arguments in St. Louis are familiar to anyone who has sat in a policy group during a long event: the measures were costing money every day, the case counts were falling, and the people paying the costs wanted a date. The city eased its restrictions in November as reported cases declined. Cases then rose again, and the city reimposed some measures, with school closures among the most significant. The pattern of relaxing and reimposing shows up in many of the cities studied later, and it is one of the most useful parts of the record for planners, because it shows what happened when measures came off while the virus was still circulating.

St. Louis was not spared. It had a substantial death toll, and its epidemic ran longer than Philadelphia’s, with a lower and flatter curve instead of a single towering peak. The meaningful comparison between the two cities concerns the shape of the epidemic and the total burden over a defined window, and a version of the story in which one city escaped and the other did not misstates both. For a hospital system, the difference in shape is the difference between a surge that can be absorbed with strain and one that overwhelms every bed, every nurse and the morgue in the same two weeks.

The 2007 analyses: Hatchett, Bootsma and Ferguson, and Markel

The modern use of the Philadelphia and St. Louis comparison comes mostly from three papers published in 2007, when federal pandemic planners were trying to work out what communities could do before a vaccine existed. Two appeared in the same issue of the Proceedings of the National Academy of Sciences. The first, by Richard Hatchett, Carter Mecher and Marc Lipsitch, examined the timing of interventions across a set of large American cities against weekly excess pneumonia and influenza death rates. They reported that cities which put multiple interventions in place early in their local epidemics had peak death rates around half those of cities that acted late, and they used Philadelphia and St. Louis as the illustrative pair, with Philadelphia’s peak several times higher. They also reported that the association with total excess mortality was weaker than the association with the peak, and that cities which lifted measures early tended to see transmission return.

The second paper, by Martin Bootsma and Neil Ferguson, fitted transmission models to weekly mortality from American cities. They concluded that the interventions reduced transmission while they were in force, that the effect on total deaths was more modest than the effect on the peak, and that measures were often lifted too early, which allowed second peaks in cities where much of the population was still susceptible.

The third, by Howard Markel and colleagues in JAMA in August 2007, assembled newspaper, health department and other records of the nonpharmaceutical interventions used in 43 cities and compared them with excess death rates over the fall and winter of 1918 to 1919. The most common categories were school closure, cancellation of public gatherings, and isolation and quarantine, and most cities used some combination of them. The authors found that earlier, longer and combined use was associated with lower excess mortality. They also stated the limits plainly: the study was observational and retrospective, compliance could not be measured, and the cities differed in many ways the analysis could not control. Philadelphia sits among the higher excess death rates in their data and St. Louis among the lower, and readers who want the city figures should take them from the paper’s own tables rather than from summaries.

Layered measures: what the record supports and where it stops

The 1918 record supports a claim about combinations and timing. Cities did not close schools in isolation or ban gatherings in isolation. They usually issued a bundle of orders, sometimes within the same day, and then kept them in place for some period before relaxing them. That makes it close to impossible to separate the effect of any one measure from the others, because the measures moved together. Markel and colleagues framed their result around early, sustained and layered use for exactly this reason, and anyone who cites the 1918 data to argue that school closure alone, or masking alone, or gathering bans alone explained a city’s outcome is claiming more than the data can show.

The mechanism behind layering is straightforward even though the 1918 data cannot quantify it measure by measure. Influenza spreads in households, schools, workplaces, churches, theaters, streetcars and shops, and each closure order removes or reduces contact in some of those settings while leaving the rest untouched. Closing schools does nothing about a crowded shipyard, and banning theater audiences does nothing about a crowded streetcar. A set of measures that reaches several settings at once reduces total contact more than any single order can, and the timing question is about how many people are already infected when the contact drops. An order issued when a small fraction of the population is infected can slow the epidemic before the curve steepens. The same order issued at the peak arrives after most of the transmission that will fill the hospitals has already happened.

The record also shows where layering stops helping. Measures that are lifted while most of the population remains susceptible delay transmission without preventing it, and both the Bootsma and Ferguson paper and the city histories show resurgence after early relaxation. Delay still has value, because a lower peak means fewer people needing care at the same moment, and in a modern pandemic it can buy time for a vaccine or for treatment capacity. In 1918 there was no vaccine coming, so the benefit of delay rested mostly on keeping the surge within what hospitals, nurses and the dead-care system could handle. That is a planning point about surge, and it should be presented as one rather than as a promise that closures reduce total deaths by a fixed amount.

The data underneath: death registration, coding and denominators

Every 1918 city comparison rests on mortality data, because case data from that fall is close to useless for analysis. Influenza was not a nationally reportable disease, cities varied in whether and when they required physicians to report it, and physicians who were working around the clock had little time for paperwork. The analyses therefore used weekly deaths from pneumonia and influenza, or deaths from all causes, as compiled from city registration records by the Census Bureau for cities in the federal death registration system. Excess mortality is then calculated by comparing those deaths with an expected baseline built from earlier years. That method is sound, but each step introduces choices and errors that differ from city to city.

Cause of death is the first problem. A physician signing a certificate in October 1918 might record influenza, pneumonia, bronchopneumonia or something else for the same clinical picture, and practice differed between cities and between individual doctors. Analysts reduce this problem by combining pneumonia and influenza or by using all-cause deaths, which captures the burden but also captures anything else that changed that fall. Registration is the second problem. In a city where bodies waited days for burial and the registrar’s office was overwhelmed, deaths could be recorded late, which shifts them between weeks and can distort the shape of the weekly curve even when the total is right.

Denominators are the third problem. Rates require a population figure, and the nearest censuses were 1910 and 1920, so analysts interpolate between them. Philadelphia’s wartime industries had pulled in workers whose numbers are hard to pin down, and a city whose real population differs from the interpolated figure will have its rate overstated or understated accordingly. The baseline is the fourth. Researchers who choose different reference years, or different methods for the expected seasonal pattern, will get different excess figures for the same city. None of this makes the comparison worthless, because the difference in the shape of the Philadelphia and St. Louis epidemics is large, but it means that small differences between cities should not be read as meaningful and that precise rankings deserve skepticism.

The tidy version of the story and the parts to drop

The popular version runs roughly as follows: Philadelphia ignored the warnings and paid with thousands of lives, while St. Louis locked down and came through almost untouched. Several parts of that need correcting. The claim that Philadelphia did nothing is a myth, because the city issued broad closure orders on October 3 and kept them in place for weeks, and its failure was one of timing, after a parade it should have reconsidered. The claim that St. Louis was nearly untouched is also wrong, because the city had a substantial death toll and a long epidemic with a resurgence after its first relaxation.

A second myth appears in some retellings, which describe St. Louis as having the lowest death rate of any American city. I have not seen that supported in the 2007 analyses, which place St. Louis among the lower-mortality cities rather than at the bottom of every ranking, and given the data problems described above, a single-city ranking claim should not be repeated without checking the specific source and its method. A third problem is the habit of treating the pair as a controlled experiment. The two cities differed in arrival date, in warning time, in population density, in war industry, in medical staffing and possibly in prior exposure, and the 2007 authors used many cities precisely because one pair cannot isolate the effect of policy.

The correction also runs in the other direction. Some commentary since 2020 has treated the uncertainty in the 1918 data as proof that the closures did nothing, and that overreaches as well. The consistent finding across the three 2007 papers, using different methods, is an association between earlier combined action and lower peaks, and the city histories show resurgence after relaxation in many places. The accurate statement is that the 1918 evidence is observational, that it supports early and layered action as a way to lower the peak, and that it cannot assign credit to any single measure or settle how much total mortality was prevented.

From the 1918 record to the 2007 and 2017 federal guidance

The 2007 analyses fed directly into federal planning. In February 2007 the CDC issued interim planning guidance titled “Community Strategy for Pandemic Influenza Mitigation in the United States: Early, Targeted, Layered Use of Nonpharmaceutical Interventions.” It introduced a Pandemic Severity Index and tied recommendations such as school dismissal and social distancing to severity, with the explicit premise that measures work best when started early and combined. In 2017 the CDC replaced that document with community mitigation guidelines published in its Morbidity and Mortality Weekly Report recommendations series, which updated the severity assessment approach and the menu of measures. Anyone writing a local annex should work from the current CDC guidance and from the state health department rather than from the 2007 document, which has been superseded.

The doctrinal lesson that transfers best is about decision timing under uncertainty. St. Louis acted when its own case count was small and its information came mostly from other cities, and Philadelphia acted after its local evidence was overwhelming. Every emergency manager has seen the same tension in other hazards: the decision that does the most good has to be made while the local picture still looks manageable, and the official who makes it will be criticized for overreacting if it works. A plan that names the indicators, the authority and the decision point in advance makes that early call easier to defend and faster to make.

The second transferable lesson concerns authority and communications. In 1918, city health officers issued closure orders under their own powers, and enforcement depended on police, on public cooperation and on how clearly the orders were explained in the newspapers. Today the legal authority to close schools, cancel permitted events or restrict gatherings varies by state and is usually divided among the governor, the state health officer, local health officials and school boards, and those lines were tested and in some states changed after 2020. An emergency manager should not assume who holds which power. The county attorney and the state health department are the people who can say, and the answer belongs in the plan before the next event starts rather than in a conference call during it.

What to do at your agency

• The county emergency manager asks the county attorney for a written note identifying which official holds legal authority to close schools, cancel a permitted public event and restrict gatherings for public health reasons in your state, and the job is finished when that note, with its legal citations as the attorney provides them, is filed with the infectious disease or public health annex of the existing emergency operations plan.
• The emergency manager asks the local health department for its copy of the current CDC community mitigation guidance and its own pandemic plan, and the job is finished when the emergency operations plan’s public health annex references the current CDC document instead of the superseded 2007 interim guidance.
• The emergency manager and the local health department epidemiologist agree on which surveillance indicator would prompt a recommendation to the policy group about community mitigation, and the job is finished when the indicator, the person who watches it and the person who receives the call are written into the existing annex in one paragraph.
• The special events coordinator or permitting office confirms in writing who can cancel or postpone a permitted mass gathering on public health advice and how much notice the event organizer is owed, and the job is finished when that answer is attached to the standard special event permit review checklist.
• The exercise officer adds one inject to the next exercise already on the calendar in which the health officer recommends cancelling a large permitted event two days before it happens, and the job is finished when the after-action report records who made the decision and how long it took.
• The public information officer drafts and stores one holding statement announcing the postponement of a public event on health advice, naming the authority and the date of the next update, and the job is finished when the health department’s PIO has reviewed it.

Takeaways

• Philadelphia held its Liberty Loan parade on September 28, 1918, with a crowd commonly estimated at about 200,000 while influenza was already spreading in the city, and it began broad closures on October 3.
• St. Louis closed schools, theaters, churches and other gathering places on October 7, within days of its first reported civilian cases, and it had more warning from eastern cities than Philadelphia did.
• St. Louis eased its measures in November, saw cases rise again and reimposed some of them, so its epidemic was longer and flatter than Philadelphia’s rather than absent.
• The 2007 analyses by Hatchett, Mecher and Lipsitch, by Bootsma and Ferguson, and by Markel and colleagues associated earlier, longer and combined interventions with lower mortality, with the strongest effect on the peak rather than on total deaths.
• Because 1918 cities issued their measures in bundles, the record cannot show that any single measure such as school closure or masking explained the difference between two cities.
• The 1918 mortality data carries real limits from inconsistent cause-of-death coding, delayed registration, interpolated population figures and differing baselines, so small differences between cities should not be treated as meaningful.
• The parade is best described as a late gathering decision in a city that was already vulnerable, and the version in which the parade alone caused Philadelphia’s epidemic is a myth.
• The CDC’s 2007 community mitigation strategy built on these analyses and was replaced by updated guidance in 2017, which is the version a local plan should cite today.