Deaths from Bacterial Pneumonia during 1918–19 Influenza Pandemic

A sequential-infection hypothesis is consistent with characteristics of this pandemic.

Deaths during the 1918-19 infl uenza pandemic have been attributed to a hypervirulent infl uenza strain. Hence, preparations for the next pandemic focus almost exclusively on vaccine prevention and antiviral treatment for infections with a novel infl uenza strain. However, we hypothesize that infections with the pandemic strain generally caused self-limited (rarely fatal) illnesses that enabled colonizing strains of bacteria to produce highly lethal pneumonias. This sequentialinfection hypothesis is consistent with characteristics of the 1918-19 pandemic, contemporaneous expert opinion, and current knowledge regarding the pathophysiologic effects of infl uenza viruses and their interactions with respiratory bacteria. This hypothesis suggests opportunities for prevention and treatment during the next pandemic (e.g., with bacterial vaccines and antimicrobial drugs), particularly if a pandemic strain-specifi c vaccine is unavailable or inaccessible to isolated, crowded, or medically underserved populations.
M any infl uenza experts, policy makers, and knowledgeable observers believe that a novel infl uenza A (H1N1) strain directly caused most deaths during the 1918-19 pandemic, often from a hemorrhagic pneumonitis that rapidly progressed to acute respiratory distress syndrome and death (1)(2)(3). Not surprisingly, plans and resources to respond to the next infl uenza pandemic focus almost exclusively on the virus, i.e., preventive vaccines and antiviral treatment of infections with a novel infl uenza strain (4). However, healthcare providers, medical experts, and published data from the 1918 period suggest that most deaths were caused by secondary bacterial pneumonias (5)(6)(7)(8)(9)(10)(11)(12); hemorrhagic pneumonitis that rapidly progressed to death was considered an alarming but uncommon clinical manifestation (8,(11)(12)(13).
Undoubtedly, the 1918-19 pandemic strain of infl uenza had unique pathophysiologic effects. In the wake of its worldwide spread, the number of deaths was unprecedented. However, contemporaneous reports suggest that the pathophysiologic effects of the virus, in and of themselves, did not directly cause most (or even many) of the deaths during the pandemic. If the pandemic strain was not inherently hypervirulent (i.e., if direct pathophysiologic effects of the virus were necessary but not suffi cient to cause death in a large proportion of immunologically susceptible hosts) and if bacterial infections were also necessary causes of most deaths during the pandemic, then preparations for the next pandemic should focus on more than preventing and treating infections with a novel infl uenza strain alone.
We have identifi ed epidemiologic and clinical characteristics of the 1918-19 pandemic that are not readily consistent with the view that most deaths were caused by the direct effects of an inherently hypervirulent virus and were clinically expressed as rapidly progressing, ultimately fatal pneumonitis. Our alternative hypothesis is consistent with known characteristics and fi rsthand accounts of the pandemic and contains implications for preparing for the next pandemic.

Epidemiologic and Clinical Characteristics of 1918-19 Pandemic Disease Usually Mild and Self-limited
The 1918-19 pandemic spread worldwide with remarkable speed. Over several months, a novel strain of infl uenza virus attacked communities worldwide; most persons were immunologically susceptible. However, most cases followed a mild or self-limited course. Had the pandemic strain been inherently hypervirulent, in the absence of modern lifesaving measures one would expect exceptionally high case-fatality rates for all affected pop- ulations. Yet during that pandemic, most infected persons had self-limited clinical courses and complete recovery (3,7,8,11,14). For most affected populations, the casefatality incidence was <2% and the overall mortality rate was <0.5% (3,7,8,13,15,16).

Clinical Courses of Fatal Cases Highly Variable and Often Prolonged
In most affected populations, <5% of deaths occurred within 3 days of illness onset, median time from illness onset to death was 7-10 days, and signifi cant numbers of deaths occurred >2 weeks after initial symptoms (5,17-22; Figures 1, 2). These fi ndings do not suggest that an inherently virulent virus caused fulminant disease and rapid progression to death in high proportions of infected persons-or even in most fatal cases. In the prominently cited experience of Sydney, Australia, most infl uenza-related deaths occurred within 3 days of hospital admission (2,23,24); however, only the sickest patients were admitted to Sydney hospitals (23). In New South Wales overall, only ≈10% of fatalities occurred within 3 days of illness onset ( Figure 1, panel F; Figure 2) (20).

Progression to Death, No Difference between Early and Late Pneumonias
If most deaths resulted from primary infl uenza pneumonias that progressed rapidly, one might expect that fatal pneumonias that developed early in clinical courses would progress more rapidly than those that developed later. However, the fi ndings of Opie et al. suggest that primary infl uenza pneumonias did not progress unusually rapidly to death. Opie et al. conducted postmortem examinations and documented the clinical courses of 234 fatal cases that occurred during the epidemic at Camp Pike, Arkansas, USA (5). They found that the durations of pneumonia before death were similar among those in whom pneumonia developed early (0-2 days) versus later (3-5, 6-8, >8 days) after infl uenza onset ( Figure 3) (5).

Mortality and Case-Fatality Rates High for Young Adults and Other Unlikely Groups
During the pandemic, overall mortality and case-fatality rates were higher for young adults, indigenous and other relatively closed populations, and certain military and occupational subgroups than for their respective counterparts. Case-fatality and mortality rates were higher for those 25-40  Table 1 in [21]). F) New South Wales, Australia (n = 3,866) (20). G) US Army training camp, Camp Pike, Arkansas, USA (n = 234) (5). Horizontal bars indicate interquartile ranges; vertical lines indicate medians. years of age (particularly men) than for those younger or older (15,16). Explanations have included aberrant host immune responses to infections with the subtype H1N1 pandemic strain-increasing the risk for "cytokine storm" (1)and higher cardiac stroke volumes in young adults (24).
However, at US military training camps, recent arrivals had worse clinical outcomes than their similarly aged, male counterparts who had been in camps longer. For example, during wartime, 60% of all infl uenza-pneumonia deaths affected soldiers who had been in the service <4 months (total infl uenza-pneumonia deaths, 34,446; deaths of soldiers with <4 months of service, 20,837) (10). In the Australian Imperial Forces, mortality rates differed by 50-fold across units of similarly aged soldiers in France and the United Kingdom (G.D. Shanks, unpub. data). US soldiers and Marines who were being transported on ships had similar infl uenza case rates but higher case-fatality rates (infl uenza cases 11,385, case rate 8.80/1,000, deaths 733) than the sailors who were permanently assigned to the same ships (infl uenza cases 2,123, case rate 8.88/1,000, deaths 42) ( Figure  4, panel A) (9). Among Australians and Americans, sharply higher death rates were reported for civilian miners (6,25) and military tunnelers (G.D. Shanks, unpub. data) than for their similarly aged counterparts ( Figure 4, panel B).
During the pandemic in New Zealand, death rates were ≈7× higher for indigenous (Maori) populations (infl uenza deaths 2,160, mortality rate 42.3/1,000) than for other residents (infl uenza-related mortality rate 4.5/1,000) (28). Across other South Pacifi c islands, death rates were generally higher for indigenous populations than for others. For example, death rates in Fiji were ≈4× higher for indigenous Fijians (infl uenza cases 5,154, mortality rate 5.7%) than for Europeans (infl uenza cases 69, mortality rate 1.4%) (8). In Guam, where military and indigenous populations were both located, ≈4.5% of the indigenous population, but only 1 sailor assigned to the US Naval base, died (9). In Saipan, "practically all of the inhabitants contracted the disease"; however, the mortality rate was reportedly sharply higher for Chamorrans (12.0%) than for Caroline Islanders (0.4%) (29). In Western Samoa, an estimated 22% (deaths 7,542) of the entire population died (8,30).
In various communities of Canada, Sweden, Norway, and the United States, mortality rates were estimated to be 3-70× higher for indigenous than for nonindigenous populations (8,31). Across British colonial countries of the Caribbean, the difference in mortality rates was >45-fold between the least affected (Bahamas: deaths ≈60, mortality rate ≈0.1%; Barbados: deaths ≈190, mortality rate ≈0.1%) and the most affected (Belize: deaths ≈2,000, mortality rate ≈4.6%); in general, the highest mortality rates in the Caribbean affected East Indian workers, Native Americans, and the poor (32).
The fi ndings of sharply different clinical courses and outcomes in subgroups of infected persons of similar ages, sociocultural circumstances, and prior health states belie the importance of host immune intensity and cardiac stroke Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 14  O. Jordan concluded that "one of the chief reasons for the great variation in case-fatality in different groups is undoubtedly the nature and relative abundance of secondary invaders ... The excessively high mortality in certain army camps, on certain transports and in particular hospitals or barracks seems most readily explicable in this way" (6).
The bacteria most often recovered from the lungs of patients who died were all common colonizers of the upper respiratory tracts of healthy persons. Types III and IV pneumococci (ubiquitous colonizing strains) were often recovered from the lungs of patients who died during the 1918-19 pandemic but were not considered important pathogens otherwise. Opie et al. concluded, "Every patient with infl uenza must be considered a potential source of pneumococcus or hemolytic streptococcus infection for his neighbor ... Every person engaged in the care of patients with respiratory diseases must also be regarded as a potential source of danger" (5).

Mortality Rates More Strongly Correlated with Pneumonia Rates than with Clinical Case Rates
If the pandemic strain had been inherently hypervirulent and had directly caused most infl uenza-related deaths, one would expect strong correlations between clinical case rates and mortality rates across affected populations. Yet in affected communities in general, correlations were stronger between mortality and pneumonia rates than between mortality and clinical case rates (15,16).
In general, age-related mortality rates and pneumonia rates-but not clinical case rates-were W-shaped with sharp peaks for young adults. Infl uenza-related mortality rates peaked sharply for young adults 25-40 years of age. Data from household surveys throughout the United States suggest that pneumonia case rates also peaked for young adults ( Figure 5) (15,16). In contrast, infl uenza case rates were highest for school-aged children, plateaued at a lower level for young adults, and continuously declined through older age groups ( Figure 5) (15,16).
After reviewing US household survey data, a senior statistician of the US Public Health Service concluded that "... these relations indicate that the mortality is determined primarily by the incidence of pneumonia. The cause of the high mortality in young adult life evidently lies in the 1196 Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 14, No. 8, August 2008 complicating pneumonia. All of the relations ... bear this out ..." (16)

Nonpharmaceutical Interventions Associated with Lower Overall Mortality Rates
Systematic analyses of mortality data from large US cities have shown that nonpharmaceutical interventions (e.g., isolation, quarantine, closing schools, banning public gatherings) were associated with lower infl uenza-related mortality rates during the autumn of 1918 (33). Given the rapidity of spread of the pandemic, reductions of mortality rate associated with nonpharmaceutical interventions are unlikely to have been primarily related to reductions of infl uenza transmission (particularly in large US cities during wartime).
On the basis of their extensive studies in US Army camps during the 1918-19 pandemic, Opie et al. concluded that "Secondary contact infection may be responsible for the development of pneumonia in patients with infl uenza. ... It is probable that secondary contact infection can be effectively prevented only by individual isolation and strict quarantine of every patient." (5) Perhaps the reduction in mortality rate after isolation, quarantine, and other social distancing measures were implemented resulted from decreased exposures of persons with infl uenza to bacterial respiratory pathogens to which they were transiently highly susceptible.

Firsthand Accounts and Reviews: Most Deaths Caused by Secondary Bacterial Pneumonias
During the pandemic, medical journals contained hundreds of detailed reports of local infl uenza epidemics. In addition, during and after the pandemic, remarkably detailed reviews of relevant epidemiologic and clinical records and population-based surveys were conducted by government and academic institutions worldwide. Care providers and experts of the day in epidemiology, pathology, bacteriology, and infectious diseases clearly concurred that pneumonias from secondary bacterial infections caused most deaths during the pandemic (5)(6)(7)(8)(9)(10)(11)(12)(13)(14). In his classic review, Jordan summarized the key factors involved in the production of infl uenza-related pneumonia during the pandemic as follows: "(1) The infl uenza virus weakens the resistant power of the pulmonary tissues so that various bacteria are able to play the role of secondary invaders; (2) the precise nature of the secondary-and tertiary-invaders is largely a matter of accident, dependent on the occurrence of particular bacteria in the respiratory tract of persons at the time of infection, and in the case of group outbreaks, on their occurrence in contacts; (3) the character of the resulting pneumonia, clinical and pathologic, is largely determined by the nature of the secondary invaders, whether Pfeiffer bacillus, streptococcus, pneumococcus, or other organisms; (4) there seems little doubt that the infl uenza virus, besides depressing the general pulmonary resistance, also acts directly on the pulmonary tissues, causing capillary necrosis, edema, and hemorrhage; (5) it seems to be true, therefore, that the fatal outcome of infl uenza pneumonia is determined partly by the degree to which the infl uenza virus depresses local and general pulmonary resistance, and partly by the virulence and nature of the bacteria which invade the tissues in the wake of the specifi c virus" (6).

Hypothesis
We endorse a sequential-infection hypothesis. This hypothesis is consistent with the known epidemiologic and clinical characteristics of the 1918-19 infl uenza pandemic, refl ects the consensus views of fi rsthand observers and contemporaneous experts, and incorporates current knowledge regarding the effects of infl uenza on physical and immune respiratory tract defenses and physiologic interactions between infl uenza and respiratory bacteria (12,13,(34)(35)(36).
A novel strain of infl uenza spread rapidly throughout the world in 1918. For most patients, infection with Figure 5. A) Estimated age group-specifi c infl uenza case rates (15,16). B) Estimated age group-specifi c pneumonia rates and mortality rates, based on household surveys of 10 communities throughout the United States (15,16).
the virus was clinically expressed as an "infl uenza-like illness" that was transiently debilitating but rarely fatal. In addition, however, the virus induced aberrant immune responses, including excessive and prolonged production of interferons, proinfl ammatory cytokines, and chemokines, particularly among young adults (34). The pathophysiologic effects included infl ammation and destruction of respiratory epithelium; immune cell infi ltration of lung tissue with edema and hemorrhage; and ultimately, degradation or destruction of virtually all physical and immune defenses of the lower respiratory tract (34). Increased susceptibility of the lower respiratory tract enabled invasion by preexisting or newly acquired colonizing strains of bacteria (12,(35)(36)(37)(38). The synergistic effects of infection with the virus, aberrant immune responses to the virus, and secondary opportunistic bacterial pneumonias were severe and often fatal.
Finally, for brief periods and to varying degrees, affected hosts became "cloud adults" who increased the aerosolization of colonizing strains of bacteria, particularly pneumococci, hemolytic streptococci, H. infl uenzae, and S. aureus (39). For several days during local epidemicsparticularly in crowded settings such as hospital wards, military camps, troop ships, and mines-some persons were immunologically susceptible to, infected with, or recovering from infections with infl uenza virus. Persons with active infections were aerosolizing the bacteria that colonized their noses and throats, while others-often in the same "breathing spaces"-were profoundly susceptible to invasion of and rapid spread through their lungs by their own or others' colonizing bacteria.

Implications
Why is it important to determine the major pathophysiologic pathways that led to deaths during the 1918-19 infl uenza pandemic? After all, the effective prevention and treatment of infl uenza infections during a future pandemic would prevent all secondary effects, including opportunistic bacterial pneumonias. Yet concerns exist that an effective strain-specifi c vaccine and effective antiviral drugs may not be produced and distributed to all at-risk populations in time to mitigate the effects of the next pandemic. In the absence of an effective infl uenza vaccine and antiviral drugs, circumstances during a modern infl uenza pandemic could resemble those in 1918-19, with the notable exception of the availability of bacterial vaccines and antibacterial drugs. The exclusive focus on the prevention and treatment of a novel strain of infl uenza virus is risky because it unnecessarily limits options and opportunities for other potentially effective prevention and treatment methods, especially in medically underserved populations in lessdeveloped countries.
We suggest that preparations for the next infl uenza pandemic should focus on more than preventing and treating infl uenza virus infections. A modifi ed infl uenza pandemic plan might include the following components: 1) Before a pandemic, expand indications for and decrease barriers to receipt of vaccination against S. pneumoniae (36)(37)(38)40). 2) During a pandemic, in communities not yet affected, universally vaccinate with a safe and effective strain-specifi c infl uenza vaccine, if available. 3) During local epidemics, treat all serious clinical cases with an antibacterial agent that is effective against S. pneumoniae, S. pyogenes, H. infl uenzae, and S. aureus (including methicillin-resistant S. aureus); isolate patients with clinical cases from other patients and as many others as possible (35,(37)(38)(39). 4) Conduct pandemic-related surveillance that tracks the incidence, nature (e.g., species, affected sites, antimicrobial drug sensitivities), and outcomes of bacterial infections that complicate infl uenza cases.
Given highly variable colonization and drug-sensitivity patterns across populations and locations, stockpiles of antibacterial drugs should be tailored to their intended uses. Plans for providing medical care should include evidencebased triage and treatment algorithms and home-care treatment guidelines (including prepackaged antiviral and antibacterial drugs) to minimize hospitalizations and maximize home care. Perhaps most important, pandemic-related research activities (including laboratory animal studies, statistical models, and clinical trials) should elucidate the determinants and effects of bacterial pneumonias that occur secondary to infl uenza. Ultimately, research activities should determine the most effective uses of antibacterial drugs and bacterial vaccines (e.g., indications, agents, doses, and timing for prophylaxis and treatment) in preparation for and during pandemic infl uenza, particularly for medically underserved and other high-risk populations. G.D.S. received fi nancial support from the US Department of Defense Global Emerging Infections Surveillance and Response System. Dr Brundage is a physician-epidemiologist at the Armed Forces Health Surveillance Center. He has conducted epidemiologic studies and public health surveillance of illnesses and injuries, primarily in military populations and settings. His interests include the epidemiologic and clinical effects of interactions among cocirculating infectious agents.
Dr Shanks is director of the Australian Army Malaria Institute. He has conducted studies throughout the world of the epidemiology, prevention, and treatment of tropical infectious diseases, particularly malaria.