EDMG509Wk2

Learning_from_the_past_to_prep.pdf

CHANCE 55

Learning from the Past to Prepare for the Future: Modeling the Impact of Hypothetical Interventions During the Great Influenza Pandemic of 1918 Gerardo Chowell, N. W. Hengartner, Catherine Ammon, and Mac Hyman

R ecurrent outbreaks of avian H5N1 influenza (flu) in poultry and other birds in several regions around the

world during the last few years have highlighted the need to prepare for the next influenza pandemic. Although most H5N1 flu cases in humans have been attributed to close contact with infected poultry, limited human-to-human trans- mission cannot be ruled out. Should this novel flu virus be able to recombine with existing human flu strains or mutate into a form that is capable of efficient human-to-human transmission, a pan- demic of great magnitude could sweep the world.

Pandemics are global epidemics associated with a high morbidity and mortality burden. There have been four pandemic outbreaks in recent history: the Asiatic (Russian) flu (1889–90), the Spanish fl u (1918–19), the Asian flu (1957–58), and the Hong Kong fl u (1968–69). Some pandemics, such as the Asian fl u outbreak, have been relatively minor and quickly contained; others, such as the Spanish fl u, could not be contained.

The consequences of failing to con- tain a disease epidemic can be disas- trous. It is estimated that at least 20 million deaths were due to the Span- ish fl u. One million deaths have been attributed to the Asian fl u. In the United States, about 675;000 lives were lost to the Spanish fl u with an approximate case fatality rate of 2%; that is, about 2% of

fl u cases succumbed to the disease. This case fatality rate is an order of magni- tude larger than the case fatality rates observed in seasonal fl u epidemics in normal years.

“Regular” epidemics of influenza, which occur annually in temperate regions of the world during the winter months, are no less of a threat. Infl uenza can be deadly, and it claims roughly 36,000 lives each year in the United States alone. The virus causing infl uenza has an antigenic structure that mutates frequently. Most of the time, the muta- tions are minor. These mutations ensure the persistence of the virus in popu- lations that develop immunity against

Here’s to Your Health Mark Glickman,

Column Editor

The Infl uenza Epidemic of 1918: Crowded sleeping area at the Naval Training Station, San Francisco, California Courtesy of U.S. Naval Historical Center Photography

One of the many newspaper highlights on the 1918 pandemic infl uenza in Geneva, Switzer- land, announcing the end of the pandemic Courtesy: La Suisse (newspaper), February 01, 1919

56 VOL. 21, NO. 2, 2008

individual virus strains. However, major changes in the virus composition can also occur, mainly from the recombina- tion of different infl uenza viruses, espe- cially those in waterfowl. This often results in novel infl uenza subtypes to which the human population has little or no immunity and which, in turn, lead to infl uenza pandemics. The severity of fl u pandemics could depend on the population immunity background, the virus virulence, the human-to-human contact rate, and the effectiveness of control interventions.

Recently, scientists at the Armed Forces Institute of Pathology in Rock- ville, Maryland, reconstructed the virus from remains of victims of the 1918 pandemic preserved in the Alaskan permafrost. The reconstructed strain is an H1N1 virus, which likely origi- nated from a bird fl u. This virus strain and the emerging H5N1 avian fl u virus both seem to kill via an overly vigorous response of the victims’ immune sys- tems (“cytokine storm”), according to Michael Osterholm, who wrote “Prepar- ing for the Next Pandemic” for The New England Journal of Medicine in 2005.

The risk of a new bird fl u pandemic is real, as is attested by the World Health Organization’s warning of a substantial risk for an infl uenza pandemic outbreak over the next few years. This warn- ing highlights the need to prepare for a potential pandemic event. Hence, understanding the effectiveness of the various intervention strategies available to decisionmakers is critical to ensuring epidemic control.

Simulations and mathematical models are important tools to help prepare for epidemic events by making it possible to evaluate the effectiveness of various intervention strategies. Interventions are varied, but they aim at the following:

• Reducing the contact rate of humans with poultry and birds in general by altering farming practices and banning cock fi ghts

• Increasing public hygiene, such as the use of protective gear (e.g., face mask)

• Implementing isolation measures such as school closures, quarantine, effective isolation of infectious cases in hospitals

• Implementing pharmaceutical inter- ventions that include the use of anti- viral medications for both treatment and prophylaxis

Thus, by modeling the 1918–1919 Spanish fl u pandemic, we can evaluate the impact of hypothetical control inter- ventions on morbidity and mortality, as well as on the duration and peak of the main wave of infection of a pending infl uenza pandemic.

The Spanish Flu Pandemic The 1918 influenza pandemic is the most severe pandemic in recent history. While it is referred to as the “Spanish”

Figure 1. Daily number of hospital notifi cations of infl uenza cases during the 1918–1919 infl uenza pandemic in the Canton of Geneva, Switzerland

flu because it was widely reported by the uncensored press of a nation not at war, this pandemic is thought to have origi- nated not in Spain, but in Kansas, and spread worldwide by mass movement of troops in the First World War. However, earlier accounts of a similar form of dis- ease to the 1918 pandemic were reported in 1916–1917 at Etaples military camp in France, raising the possibility that a similar variant of the pandemic strain was already circulating prior to 1918. In fact, evidence indicates the elderly population

1918/9 Influenza Pandemic in Geneva, Switzerland

D ai

ly n

um be

r of

h os

pi ta

l n ot

ifi ca

tio ns

100

200

300

400

500

Jul 01

Sep 10 Nov 14

Feb 28

Spring wave Fall wave

0 50 100 150 200 Time (days)

F61 Aircraft Hall at the Naval Aircraft Factory, Philadelphia, Oct. 19, 1918 Courtesy of www.usasearch.gov

CHANCE 57

A church service being held outdoors in Lau- sanne, Switzerland. Places of indoor assem- bly—including churches, theaters, and dance halls—were closed during the 1918 infl uenza pandemic as a measure to reduce transmis- sion of the disease. Courtesy: La Patrie Suisse (weekly paper), August 21, 1918

Chinese pandemic fl u: infl uenza patients by their beds. In July, an American soldier said that while infl uenza caused a heavy fever, it “usually only confi nes the patient to bed for a few days.” The mutation of the virus changed all that. Courtesy of www.usasearch.gov

Spanish fl u patients standing next to their beds Courtesy of www.usasearch.gov

was not as affected as young adults, prob- ably due to partial protection they had acquired from a similar virus long before the 1918–1919 pandemic.

The 1918–1919 pandemic swept the world in a series of up to three waves of varying severity. The fi rst pandemic wave took place around the spring and summer time. For example, the first wave (“herald” wave) reached France in March, England in June, and Swit- zerland in July 1918. The second wave (“fall” wave) was the most severe and occurred around October–November of 1918, whereas the third wave (“win- ter” wave) was of limited intensity. This winter wave is probably associated to the seasonal fl u, and took place in late 1918 and early 1919. The symptoms experienced during the second wave were typically more severe than those of the fi rst and third waves. Symptoms included high fever, coughing, a dis- tinctive dark violet coloration of face and fi nger tips due to a lack of oxygen supply to the blood, and nose bleeding. The last two conditions are medically referred to as heliotrope cyanosis and epistaxis, respectively.

While summary statistics about each wave are generally available, only a few regions have detailed records docu- menting the day-to-day evolution of the epidemic.

Thanks to a mandatory notifi cation of infl uenza hospitalizations in the Can- ton of Geneva in Switzerland, we were able to manually extract daily counts

58 VOL. 21, NO. 2, 2008

of infl uenza hospitalizations from the notifi cations registry archives in Swit- zerland for the July 1918–February 1919 period (Figure 1). The data reveal that in the Canton of Geneva, the infl uenza pandemic affected more than 50% of the population.

The overall case fatality was about 4.2%, and was, surprisingly, at its high- est in the 21–40 age group, particularly in males. These fi ndings are consistent with the theory that the virus kills via a cytokine storm, a positive feedback loop between cytokines that stimulates excessive immune cell reactions and the production of more cytokines.

During the pandemic, control mea- sures were implemented, but there is no evidence of their effectiveness because disruptions in the sanitary, medical, private, and public sectors were com- mon. Although there were unsuccessful efforts to develop a vaccine known as protovaccination, neither an infl uenza vaccine nor an antiviral medication was available for prophylaxis or treatment in 1918. Moreover, the social climate was one of insecurity and fear, as the popula- tion questioned the effectiveness of the control measures that were undertaken, including school and church closures, banning public events and hospital vis- its, and mandatory spraying of disinfec- tants on the streets.

Mathematical Model of Influenza Transmission Our data are the reported daily number of new hospitalizations. To model such data, classical mathematical epidemiol- ogy models classify individuals as either susceptible (S), infected and infectious (I), or recovered (R) and model the evolu- tion of the number of individuals in each of these categories through a system of coupled differential equations (Figure 2). For a given set of initial conditions (initial number of susceptible, infected individuals) and epidemiological param- eters (e.g., transmission rate, infectious period), these models can describe the expected course of an epidemic. The models can be extended by adding more “compartments” to better describe the dynamics of the disease, the impact of interventions, and variations in reporting rates depending on the severity of the dis- ease. The underlying assumption behind

these compartmental models is that the population is well mixed (i.e., that the average social distance—“degrees of physical separation”—between each pair of individuals is relatively short).

Infl uenza infection is characterized by a latent period of about two days, during which individuals are not able to infect others until they become infec- tious for a period lasting about four days. Some individuals become infected but remain subclinical, that is, they are either asymptomatic or develop only a mild reaction to the virus. Because these individuals are not directly observed as infected, it is diffi cult to quantify the bur- den of infl uenza in terms of number of cases. Nevertheless, our modeling of the dynamic evolution of the epidemic can uncover the fraction of subclinical cases because, once recovered, they become immune to further infections.

Because of the novelty of the 1918 infl uenza virus, we assume the entire population is susceptible at the onset of the pandemic. Anecdotal evidence supports our assumption that recovered individuals from the fi rst (spring) wave in Geneva were immune to infection in the second (fall) wave. A percentage of individuals in the period of latency prog- ress to the clinically infectious class, and the rest progresses to the asymptomatic class. As disruptions in the sanitary and medical sectors were common, hospital- ized individuals were considered infec- tious in our model. Clinically infectious individuals were hospitalized or recov- ered without hospitalization (e.g., mild infections, hospital refusals). Hospital- ized (reported) individuals either recov- ered or died from infl uenza. We also account for subclinical cases of infl uenza that showed mild symptoms or none at all (asymptomatic). Given that subclini- cal cases only shed a small amount of virus, their transmission rate is a fraction of that of clinical cases.

The above description translates into the compartmental model (Figure 4) for the transmission dynamics of pandemic infl uenza. This model places individuals in the following epidemiological cat- egories: susceptible (S), exposed but not yet infectious (E), clinically ill and infectious (I), asymptomatic and par- tially infectious (A), hospitalized (H), recovered (R), and dead (D). We con- sidered birth and natural death rates

to have a common value (average life expectancy of 60 years in 1917). How- ever, population demographics could have been neglected because the epi- demic process occurred at a much faster time scale than births and deaths. Moreover, given the war, the Swiss sol- diers stayed in their country to protect the border.

The parameters in this model are the transmission rate, recovery rate, diag- nostic rate, relative infectiousness of asymptomatic cases, proportion of clini- cal cases, and initial numbers of exposed and infectious individuals. We estimated these parameters by minimizing the squared differences in the cumulative number of hospital notifi cations during the 1918 infl uenza pandemic in Geneva, Switzerland, to the values predicted by the model. The advantage of using the cumulative over the daily number of new notifications is that the former somewhat smoothes out known report- ing delays on weekends and national holidays. However, using the daily curve of case notifi cations did not generate signifi cant differences in parameter esti- mates. Figure 3 shows that there is a good agreement between the observed and fitted epidemic curves. Both the estimated variances of the parameters and sensitivity analysis confi rmed the model was well fi t by the data.

Our model predicts that most of the cases were either asymptomatic or experienced only mild infections. The fraction of symptomatic cases among infected individuals in the first wave was 10% and increased to 36% during the second wave. Our model also pre- dicts that the fraction of symptomatic cases diagnosed in hospitals was about 60% for the spring wave and 83% for the fall wave. We explain these fi ndings by noting that during the 1918–1919 pandemic, the number of sick people was such that some who did not war- rant attention were refused admission in overcrowded hospitals.

Moreover, it is unlikely that individu- als with mild symptoms sought medical attention. While misdiagnosis may be common for annual epidemics of infl u- enza due to the limited reliability of clin- ical diagnosis (nonspecifi c symptoms), clinical diagnosis for pandemic infl uenza should be more reliable because of the severity of symptoms.

CHANCE 59

Figure 3. Model fi t to the daily number of hospital notifi cations during the fi rst two waves of the 1918 infl uenza pandemic in the Canton of Geneva, Switzerland, and the effects of reducing the transmission rate from 10% to 50% within hospital settings

Figure 2. Schematic representation of the transition of individuals (indicated by arrows) among the different epidemiological states during an infl uenza pandemic

Births

Natural

deaths

Natural

deaths

Natural

deaths

Natural

deaths

Natural

deaths

Susceptible Latent Infectious Hospitalized Dead

Asymptomatic Recovered

Natural

deaths

60 VOL. 21, NO. 2, 2008

Figure 4. The combined effect of reductions in the transmission rate in the general population and reductions in transmissibility from hospital settings (nosocomial transmission) on the fi nal number of deaths

next infl uenza pandemic include popu- lation demographics, highly connected transportation systems, and an improved public health system. However, these uncertainties are probably of smaller magnitude than the pathogenicity and virulence associated to the next infl u- enza pandemic virus strain.

Our approach is based on the calibra- tion of a reasonable model of infl uenza transmission using retrospective epi- demic data. We then explore different hypothetical “what if” scenarios by adjusting model parameters appropri- ately. This allows us to explore, for example, the effectiveness of hypotheti- cal control strategies in terms of fi nal epidemic size, peak size, and duration (see supplemental material at www.amstat. org/publications/chance). We have shown that mathematical models can be useful tools to increase our understanding of historical epidemics and guide public health offi cials in effectively planning and controlling future epidemics.