[Skip to content]

Sign up for our daily newsletter
The Actuary The magazine of the Institute & Faculty of Actuaries

Influenza pandemics: time for a reality check?

While mortality has generally
been improving for
many decades, life insurers
still face the risk that
an influenza pandemic could cause a one-time
mortality shock. It is not easy to specify in
advance the loss value from such an event, and
therefore the amount of capital to hold.
This makes it difficult for life insurers to quantify
the risk and to manage their capital efficiently
for the benefit of both policyholders and
shareholders. This is often because many companies
lack the tools to determine the loss value
and the amount of capital to hold.
With the move from a rules-based or formulaic
solvency requirement to a principles-based
approach, regulators are now showing an
increasing interest in the use of internal models.
At the same time there is a heightened general
awareness of a pandemic threat, with various
views being expressed on the possible impact.
In light of these developments, Swiss Re
has developed a sophisticated epidemiological
model to improve the level of understanding of
the potential range of outcomes from a pandemic.
This article summarises the workings of
the model, along with some key results.
Key outputs
The model results show that, in most developed
countries, a 1-in-200-year severity pandemic
would give rise to excess mortality of between
1 and 1.5 deaths per 1,000 lives within an insurance
portfolio. This result is particularly relevant
for insurers in setting mortality shock
assumptions in their own internal models.
The model also shows that the influenza pandemic
of 1918, a unique event in 420 years,
would have a much lower impact on mortality
today than it did in 1918.
These key outputs are discussed further below.
Inside the model
The model works by simulating many thousands
of hypothetical pandemics, with each
simulation producing an estimate of the resulting
excess mortality.
In generating its results, the model factors in
the different features of the three pandemics of
the last century. These include the rate of
spread, the ability of each pandemic to cause death (its lethality), and differences in infection
rates and lethality between age groups. The
model also allows for advances in pharmaceutical
and behavioural interventions.
? Pharmaceutical interventions include antibiotics,
vaccines, and antivirals:
? Antibiotics are highly effective in reducing
mortality from secondary infections, and this
effect is well understood. Because their
impact can be predicted so well the model
assumes that their effectiveness is generally
the same in each hypothetical pandemic.
? While the effect of vaccines is accounted for
in the model, they are largely ineffective
in reducing mortality in a pandemic’s first
wave owing to current production technology
and capacity.
? Antivirals, which are being stockpiled by
many governments, are expected to slow the
spread of influenza and substantially reduce
overall illness and mortality. However, their
effectiveness against a wide range of potential
viruses is not fully understood. The
model therefore assumes that antivirals
would not work at all in one in four simulated
? Behavioural interventions modelled include
contact modification (reduced mixing of infected
and uninfected people), and travel restrictions:
? Contact modification is modelled at varying
levels of intensity and duration, to recognise
that behavioural changes reflect the perceived
level of risk.
? Travel and travel restrictions have little effect
on the final level of mortality.
The effect of each potential pandemic produced
by the model on an insurer’s portfolio of
business can be estimated by applying weightings
for exposures by age group and by country
(within the model, the global population is subdivided
into 37 countries or regions).
This produces a complete distribution model
of potential additional mortality and of the
probability that each level of excess mortality
will occur. This enables those using the model
for capital management purposes to estimate all
relevant statistical measures, such as 99% value
at risk (VaR), 99% tail VaR, and losses for various
return periods.
Any pandemic simulated by the model produces
a particular level of excess mortality that
would occur in that hypothetical pandemic. If
these simulations are sorted from the lowest
excess mortality to the highest, it is possible to
examine the probability that the mortality over
a one-year period will exceed a certain level. figure 1 illustrates the results of this process of
simulation and sorting for 10,000 hypothetical
pandemics generated by the model. Because
excess influenza mortality varies with age, the
individual country populations have been
weighted by age to better represent the age
profile of an insurance portfolio.
Based on current intervention capabilities, the
model estimates that a 1-in-200-year pandemic
event (0.5% annual probability) would cause
excess mortality of about 0.7 per thousand (‰)
in Canada, 1.0‰ in Switzerland, 1.1‰ in the
United Kingdom, and 1.0‰ in the United States.
Variations between countries are largely the
result of differences in the underlying health of
the population, the robustness of the healthcare
system, stocks of antivirals, and the
capacity to implement successful non-pharmaceutical
interventions such as contact modification
(the success of which depends on a
country’s population density).
The 1918 pandemic
Is the experience of 1918 an appropriate level
for a mortality shock assumption? In Swiss Re’s
opinion, no it isn’t. The 1918 pandemic was
exceptional among all the pandemics that have
been recorded since 1580. The unusual virulence
of the influenza strain meant that
mortality was extremely high, and it was concentrated
in a much younger age group than in
other pandemics. Three waves of infection
rapidly followed one another.
A key factor that made 1918 so severe
compared with today is that no antibiotics,
vaccines, or antivirals were available. Also
among the factors that are likely to have contributed
to its severity, compared with other
pandemics, was World War I. Wartime restrictions
on media reporting led to widespread
unawareness of the virus, so communities were
unprepared for its re-emergence in the autumn
of 1918. The crowding together and movement
of troops provided the ideal environment for
disease outbreak and spread, and the military’s
need for doctors and nurses led to a rapid
decline in medical care for civilians.
Another contributor was the underlying disease
burden. Outbreaks of contagious disease,
including tuberculosis, were widespread in
1918. Research suggests that many apparently
healthy young adults who died at that time
were infected with tuberculosis, explaining the
unusually high mortality rates in this age range,
especially in males. Tuberculosis and many
other diseases of the time are now successfully
treated with antibiotics and vaccines.
What has changed since 1918?
? The population age structure is older than in
1918; older people are less contagious than
younger people, and they mix less, making
it harder for the virus to spread.
? Antibiotics are now available to treat complications
(penicillin was discovered in 1928).
? Virological research and knowledge has
grown rapidly; the influenza virus was first
isolated in 1933.
? The World Health Organisation’s Global
Influenza Surveillance Network was established
in 1952.
? Influenza vaccines have been available since
the 1950s.
? Antiviral drugs for treatment of influenza
were first approved in the1970s, and many
governments now have stockpiles of Tamiflu.
? The population density today is greater than
in 1918 (in running the model this was
found not to make a significant difference to
excess mortality).
? The volume and speed of travel has increased.
While this will give rise to a faster spread it
will not affect0 the overall level of mortality
over the duration of the entire pandemic.
The model takes account of these factors
when producing simulations of a modern-day
pandemic, after which their impact can be
The model indicates that the probability of a
pandemic with the same transmissibility and
lethality as 1918 is about 0.2%, ie a 1-in-500-year
event. However, as figure 2 shows, even if such
a pandemic were to occur today, the mortality
impact would be far less than it was in 1918.
What about H5N1?
The H5N1 virus that is currently infecting birds,
and occasionally people, has been unable to
transmit efficiently and sustainably between
humans. Given that no H5 subtype has ever
caused a human influenza pandemic, this virus
may never be able to do so. However, the model
does make an allowance for such a development,
but as a remote possibility. Figure 1 shows the annual probability that an influenza pandemic will cause mortality greater than the number shown
on the x-axis. In Canada, for example, there is a 0.5% annual probability (ie a 1-in-200-year likelihood) that mortality
will be above approximately 0.7 per thousand. Pandemics are assumed to occur every 30 years annual probability is
therefore deemed to be 3.33% (1 in 30) or less. Figure 2 shows the cumulative effect on mortality rates in the general population of selected material changes since
1918. The ‘base’ equates to the best simulation of 1918 that could be achieved by Swiss Re’s model using conditions
and interventions relevant to that time. Figure 1 Excess mortality due to pandemic influenza Figure 2 Relative change in total deaths per country: 1918 compared with 2006