Deaths from flu outbreak, The Lancet..

[Guest guest]

From " The LANCET "

http://www.thelancet.com

To listen to the audio submission, got to the Lancet address. To read

all of this with the tables mentioned, you should sign-up and

register at that site.

Articles

Estimation of potential global pandemic influenza mortality on the

basis of vital registry data from the 1918–20 pandemic: a

quantitative analysis

Prof JL Murray DPhil a , Prof Alan D PhD b,

Chin ScB a, Dennis Feehan AB a and Prof H Hill

PhD c

See Comment

Summary

Introduction

Methods

Results

Discussion

References

Summary

Background

The threat of an avian influenza pandemic is causing widespread

public concern and health policy response, especially in high-income

countries. Our aim was to use high-quality vital registration data

gathered during the 1918–20 pandemic to estimate global mortality

should such a pandemic occur today.

Methods

We identified all countries with high-quality vital registration data

for the 1918–20 pandemic and used these data to calculate excess

mortality. We developed ordinary least squares regression models that

related excess mortality to per-head income and absolute latitude and

used these models to estimate mortality had there been an influenza

pandemic in 2004.

Findings

Excess mortality data show that, even in 1918–20, population

mortality varied over 30-fold across countries. Per-head income

explained a large fraction of this variation in mortality.

Extrapolation of 1918–20 mortality rates to the worldwide population

of 2004 indicates that an estimated 62 million people (10th–90th

percentile range 51 million–81 million) would be killed by a similar

influenza pandemic; 96% (95% CI 95–98) of these deaths would occur in

the developing world. If this mortality were concentrated in a single

year, it would increase global mortality by 114%.

Interpretation

This analysis of the empirical record of the 1918–20 pandemic

provides a plausible upper bound on pandemic mortality. Most deaths

will occur in poor countries—ie, in societies whose scarce health

resources are already stretched by existing health priorities.

Back to top

Introduction

The avian influenza epidemic in birds and the 258 cases recorded in

human beings (as of Nov 29, 2006) in several continents1 are

generating tremendous media coverage, public concern, and policy

debate.2 Governments and donor agencies have joined together to

pledge substantial funds to fight the spread of avian influenza;3 for

example, the US government has committed $3·8 billion for the USA4

and Australia has set aside AUD$555 million.5 This high degree of

concern is in part due to estimates of potential mortality from a

major influenza pandemic. Estimates from 2 million6 to 360 million7

and even up to 1 billion8 deaths have been proposed. These numbers,

combined with predictions of the inevitability of the next influenza

pandemic, are driving continued attention and policy focus.9

Various models of the effect of influenza pandemics on mortality have

been developed.10–12 These models make strong assumptions about

attack rates and case-fatality rates in influenza cases. Irrespective

of the modelling assumptions, however, the three pandemics of the

20th century—in 1918–20, 1957–58, and 1968–70—are the main source of

empirical evidence on the potential human toll of the next

pandemic.13–15 The 1918–20 Spanish flu pandemic caused the highest

mortality by far and is often used to set the upper bound on the

number of deaths caused by a future pandemic.16 Medical historians

have generated estimates of mortality in 1918–20 ranging from 20

million to 100 million.17–19 These estimates are based on reviews of

various historical documents, including national commissions, eye-

witness accounts, and local government reports. With some

exceptions,20–22 these analyses have not distinguished quantitative

analyses based on underlying high-quality vital registration data

from qualitative accounts.

Systematic analysis of all available vital registration data would

permit the calculation of pandemic mortality due to the major 20th

century influenza pandemics in a comparable manner. Here, our aim is

to assess vital registration data from the 1918–20 pandemic, since

this pandemic was overwhelmingly larger than other 20th century

pandemics and provides a clearly identifiable effect on mortality. We

aim to develop statistical models that relate annual pandemic

mortality to community attributes, and to use these models to

estimate the effect on mortality of an influenza pandemic in 2004,

the most recent year for which per-head gross domestic product in

international dollars is available.

Back to top

Methods

Data collection

We included all available vital registration data from 1915 to 1923—

ie around the period of the 1918–20 influenza pandemic—from

populations in which vital registration is believed to be 80% or more

complete.23,24 Data were mainly taken from the Berkeley Human

Mortality Database25 and B R 's International Historical

Statistics Series.26 These sources were supplemented with subnational

data for US states27–35 and what were then described as provinces of

pre-partition India.36,37

Epidemic and pandemic influenza mortality in previous studies were

calculated in one of two ways: with seasonal and inter-year

variations in mortality to calculate excess mortality in the

influenza season38,39 and with cause-specific mortality data.

Comparisons of influenza-specific mortality, however, are confounded

by changes in influenza coding across several revisions of the

International Classification of Diseases40 and variation in influenza

certification and coding between and within countries.41 We opted to

use excess all-cause mortality because it avoids coding issues,

captures the effect of influenza on other causes of death, and avoids

inflating death figures when influenza merely hastens the deaths of

already sick individuals, an occurrence known as harvesting.42 Since

mortality data by day or week are not available for the period from

1915 to 1923 for most countries, we are only able to calculate excess

mortality with inter-year comparisons.

Statistical analysis

We used Stata version 9.2 for statistical modelling and analysis. For

both influenza and all-cause mortality, we calculated excess

mortality by comparison of annual death rates during the pandemic to

the average of annual death rates before and after the pandemic.

Because influenza pandemics might increase mortality not only in the

year of peak incidence, but also in the following year or two, based

on evidence on the time course of the pandemic, we compared death

rates in a 3-year pandemic window with those in surrounding years.

This approach also reduces short-term harvesting effects since deaths

that would have taken place during the 3-year window in any case will

not inflate the pandemic mortality rate.

For the 1918–20 pandemic, we calculated the average mortality rate in

1915–17 and 1921–23, and subtracted this from mortality in 1918–20.

Formally,

where PM is pandemic mortality. Observations where mortality is known

to have been increased by World War 1 or civil war are presented in

the tables, but were excluded from the statistical analysis. We

calculated 95% CI for excess mortality by simulation.

To study the age-pattern of the pandemics, we calculated excess

mortality for 1918–20 by use of the same approach but for every 5-

year age-group and both sexes for countries for which such data were

available. To increase our sample size to a maximum, we used all-age

mortality for the regression analysis and to estimate mortality in

2004. We used ordinary least squares regression models to test the

relation between the log of pandemic mortality and the log of per-

head income and the absolute value of latitude. Absolute latitude was

tested because of arguments in the 1920s that environmental factors

such as diurnal temperature fluctuations were a key determinant of

mortality.43 Per-head income was measured in real international

dollars (corrected for price changes).44–47 The regression model

generated a predicted all-age pandemic mortality rate. Age-specific

estimates were generated with the relative age-pattern of pandemic

mortality by age observed in 1918–20 for countries with age-specific

data available. Confidence limits were estimated with 1000 draws from

the multivariate normal approximation to the distribution of the

parameters estimated in the model. The estimated number of deaths was

calculated by country for every draw, and the quantities of interest

(ie, predicted number of deaths by country, region, and age-group)

derived from the distribution of simulated deaths for all countries.

The simulation results capture both uncertainty in the model's

parameter estimates (the result of having fewer than an infinite

number of observations), and the variance in the excess mortality

rate that is not explained by the independent variables, which is

sometimes called the fundamental uncertainty.48 2004 population

estimates were based on data from the 2006 World Health Report.49

Role of the funding source

The sponsor of the study had no role in study design, data

collection, data analysis, data interpretation, or writing of the

report. The corresponding author had full access to all the data in

the study and had final responsibility for the decision to submit for

publication.

Back to top

Results

Table 1 shows pandemic mortality calculations for 27 countries for

1918–20, 24 US states with data available for the period, and nine

Indian provinces. Pandemic mortality rates for the UK, France, and

Finland for 1918–20 are based on females only, since male mortality

is confounded by deaths due to war. In this sample of countries that

are heavily weighted to higher income, median pandemic excess

mortality was 0·75 deaths per 100 people (henceforth indicated as %)

and average excess mortality was 1·06%. Excess mortality ranged from

0·2% in Denmark to 4·4% in India. Since there was some under-

registration of mortality in India, total pandemic mortality could

have been even higher.

Click to view table

Table 1.

Pandemic excess mortality calculations, based on vital registration

data from 1918–20

Huge variation in the mortality rate from the 1918–20 pandemic is

born out by subnational data. Table 1 shows that, for nine provinces

in India, pandemic mortality ranged from 2·1% in Burma (covered by

the Census of India at the time) to 7·8% in the Central Provinces and

Berar. Commentators at the time attributed this huge variation to

differences in nutritional status and diurnal fluctuations in

temperature.50 During the period 1915–23, vital registration systems

were complete in 24 states in the USA.51 Across these states,

pandemic mortality ranged from 0·25% in Wisconsin to 1·0% in

Colorado. Thus, from Wisconsin to the Central Provinces and Berar in

India, the death rate from the 1918–20 pandemic varied 31-fold.

Figure 1 shows median excess mortality by age and sex for the 1918–20

pandemic. These data confirm the well-known observation that, unlike

the 1957–58 and 1968–70 pandemics, mortality was concentrated in

young adults, not elderly individuals.17–19 In this set of countries,

mortality was higher in males than in females, although sex-specific

(but not age-specific) data in India showed excess female mortality

in five of nine provinces (data not shown). The precise age-pattern

varies considerably across the 13 countries, with some having almost

no excess mortality in individuals aged over 60 years, and others

having substantial mortality in the same age-group (data not shown).

Click to enlarge imageFigure 1. Median excess mortality by age and

sex for the 1918–20 pandemic, based on data from 13 countries with

available complete age-specific mortality data

Table 2 summarises the results of two regression models. For both,

the dependent variable is the log of pandemic mortality. In the first

model, the independent variable is the log of per-head income, and

the second model adds the absolute value of latitude. Nearly 50% of

the variance in pandemic mortality is explained by per-head income

alone. Pandemic mortality is strongly negatively related to this

variable. The coefficient for income ranges from –0·88 in the model

with income alone to –0·97 in the model with absolute latitude. This

means that a 10% increase in per-head income was associated with a 9–

10% decrease in mortality. The coefficient for absolute latitude is

not significant; this model was, therefore, not used to estimate

mortality in 2004.

Click to view table

Table 2.

Results from two ordinary least squares regression models on the log

of the pandemic excess mortality rate 1918–20 for countries with

complete vital registration data

Table 3 shows estimates of global mortality in 2004 if a pandemic

strain of influenza with similar severity as the 1918 strain were to

emerge, taking into account population size, age composition of

populations, and changes in per-head income for the world, regions,

and major countries. The webtable shows estimates for all countries.

The range of these estimates is based on both parameter uncertainty

in the regression model and the variation in pandemic influenza

recorded in 1918–20 that is not explained by per-head income. The

median estimate of the number of deaths worldwide is 62 million (10th–

90th percentile range 51 million–81 million). If these deaths were

concentrated in a single year, global mortality would increase by

114% (10th–90th percentile range 93–147). Most deaths would occur in

15–29-year-old individuals, followed by those aged 0–14 years and 30–

44-year-olds (figure 2). 96% (95% CI 95–98%) of the estimated number

of deaths would take place in the developing world (figure 2 and

table 3).

Click to view table

Table 3.

Estimated number of deaths caused by the emergence of a pandemic

influenza strain in 2004

Click to enlarge imageFigure 2. Distribution of deaths due to the

emergence of a pandemic influenza strain in 2004 by region and age-

group

OECD=Organisation for Economic Co-operation and Development.

Back to top

Discussion

Were a strain of influenza much the same as that which caused the

1918–20 pandemic to emerge in 2004, we estimate that it could kill 51–

81 million individuals. This estimate is based strictly on recorded

patterns of mortality in countries with nearly complete vital

registration systems, rather than on theoretical models or

assumptions about attack rates and case-fatality rates.

Our results indicate that deaths would be concentrated in the 0–14,

15–19, and 30–44 years age-groups. Various theories have been

proposed for the unique pattern of mortality by age exhibited by the

1918–20 pandemic strain.52,53 Our results suggest that deaths in the

15–19 and 30–40 years age-groups would probably be a result of high

age-specific death rates, whereas those in individuals aged 0–14

years would most likely be due to the large population size and

moderate mortality in this age-group.

Most of the strong relation that we observed between per-head income

and pandemic mortality must be mediated through factors such as

nutritional status, co-morbidity, community characteristics

associated with poverty, and the effect of supportive care, since

therapeutic interventions had little or no effect on mortality in

1918–20. This income effect is consistent with a contemporary

observation of a relation in 1918–20 between household income and

mortality.50

The 1918–20 mortality rates that we calculated from vital

registration data are extremely high, reaching nearly 8% in one

province of India. By comparison, global mortality from all causes of

death was 0·92% in 2000. The more than 30-fold variation in mortality

across communities within the 1918–20 pandemic can be explained by

individual host and community factors. Variations in mortality could

perhaps be due to the timing of the epidemic; countries with

epidemics that began earlier might have had higher mortality than

those that succumbed later. Pandemic mortality is a function of both

the influenza attack rate and the influenza case-fatality rate.

Available mortality data do not allow us to determine how much of the

within-pandemic variation in outcome is due to transmission factors

or case-fatality factors. Individual factors such as current immune

function, nutritional status, acquired immunity through previous

influenza infection, co-morbidity, and community and environment

factors such as population density and mixing rates, access to health

care, quality of care, and the physical environment could all have a

role. Ultimately what matters is the effect on individuals that is

captured by the overall pandemic mortality rate.

Our estimates of deaths by country exhibited wide confidence limits

because, although per-head income explains half the variance in the

pandemic mortality, half the variance remains unexplained.

Furthermore, the method we used to calculate excess mortality rates

in 1918–20 could exaggerate the effect of influenza because events

such as wars, natural disasters, or other epidemics might also have

increased mortality in pandemic years. However, observations where

mortality is known to have been increased by World War 1 or civil war

were excluded from the statistical analysis. Harvesting during the

pandemic could also remove frail individuals from the population who

would have died in the 3 years after the pandemic in any case. In

other words, mortality after the pandemic window could be

artificially lowered and thus increase the estimated excess

mortality. However, because we used a 6-year average to establish the

baseline mortality rate, we believe this effect will be small.

Another potential source of bias is that global mean income is now

much higher than it was in 1918, and most countries in the

Organisation for Economic Co-operation and Development (OECD) have

levels of per-head income beyond any recorded in 1918. Thus our

extrapolations from data from 1918–20 must be viewed cautiously.

In most discussions of influenza, the 1918–20 pandemic sets the upper

limit, in terms of mortality, on what might occur in future

pandemics. However, there is no logical or biological reason why that

pandemic—albeit very severe—should represent the maximum possible

mortality in a future pandemic. Random genetic mutation could, in

principle, produce a more lethal virus, although pathogens that are

too lethal might not survive long enough in the host to effectively

transmit to different populations.54 In addition to this uncertainty

about what is genetically possible, future mortality could be larger

if the 1918–20 pattern of low older adult mortality were in fact due

to some acquired immunity from the pandemics of the mid-19th

century.15 Concerns about increased travel and mixing, which lead to

larger epidemics,55 might not alter our extrapolations, since the

historical record suggests that nearly all human populations were

eventually exposed to the 1918–20 influenza virus.

Despite these fears, there are many cogent reasons to expect that the

emergence today of a pandemic strain much the same as that which

caused the 1918–20 pandemic strain would lead to much lower mortality

than estimated here. First, symptomatic medical management is better

now than in 1918–20. However, although individuals with access to

health care in high-income and middle-income countries might benefit,

health-care systems could become overwhelmed, which would attenuate

this effect. Second, antivirals such as zanamivir and oseltamivir

phosphate might have a positive effect on the reduction of

transmission56,57 and case-fatality rates.58 Because we have not yet

seen the next pandemic virus, the magnitude of this effect cannot be

quantified. Third, vaccination with a lag of 4–6 months from the

onset of a pandemic could reach a large fraction of the high-income

populations.59 The speed of the epidemic, perhaps affected by various

efforts at quarantine, will determine the potential benefit of

vaccination. Strict quarantine in American Samoa seems to have

avoided the 1918–20 pandemic;60 quarantine efforts in Australia are

thought to have delayed but not avoided the pandemic,61 but strict

quarantine measures in other settings failed.62 Mathematical models

suggest that quarantine could be beneficial if highly effective and

if administered in combination with prophylaxis under certain

circumstances.63–65 In view of the restricted vaccine production

capacity and the reality of health system coverage, vaccination would

have little or no effect on the poorest populations. Fourth, in 1918–

20, a large proportion of deaths was due to secondary bacterial

pneumonia after primary viral pneumonitis.66 Antibiotics for

pneumonia could have a substantial effect on case-fatality rates. In

middle-income and low-income settings, prompt access to antibiotics

could be the most affordable strategy that has the largest effect on

mortality. One should note that all of these factors will lower

mortality more in richer nations than in those with lower per-head

income, which tends to strengthen the already observed inverse

relation with per-head income.

Our results indicate that, irrespective of the lethality of the

virus, the burden of the next influenza pandemic will be

overwhelmingly focused in the developing world, as has been suggested

for the 1918–20 pandemic.67,68 Symptomatic treatment, antivirals,

vaccination, and antibiotics for secondary bacterial pneumonia,

combined with the underlying relation between per-head income and

mortality, perhaps mediated through nutritional status, will reduce

the effect of the pandemic in OECD countries. By contrast, the

countries and regions that can least afford to prepare for a pandemic

will be affected the most. The potential risk to populations of sub-

Saharan Africa, south Asia, and other developing regions presents a

policy dilemma. When resources to tackle the health problems already

present in the community—including HIV, tuberculosis, malaria,

cardiovascular diseases, and road traffic accidents—are already

scarce, how much can these populations afford to spend on preparing

for a potentially very harmful but also very uncertain threat?

This analysis of a worst-case scenario based on the 1918–20 pandemic

provides no insight into the probability of an influenza pandemic in

the next 1, 5, or 10 years. In the past century, only the 1918–20

pandemic qualifies as a dramatic change in human health. However,

this does not mean that the threat of such a pandemic is 1% next year

or any other year. There is also no way to revise estimates of the

probability of a major pandemic because of the current H5N1 avian

influenza outbreak. However, to prepare for such a possibility,

especially focusing on practical and affordable strategies for low-

income countries where the pandemic will have the biggest effect, is

clearly prudent.