Influenza or flu is a viral
infection that is readily transmitted through the air and causes
respiratory problems in humans and other animals. It occurs seasonally
and results in an average of 30,000 deaths in the U. S. annually
primarily among the young and old. There have been some epidemics, such
as the pandemic of 1918, which resulted in over 30 million
deaths around the world.
There are two primary types, A and B of influenza, which appear
seasonally, usually starting in the fall and continuing through the
winter months. These types mutate creating new strains each year. Flu
is an RNA virus with 11 genes that mutate regularly, making it
difficult to create general vaccines. Each year scientists anticipate
the most likely dominant strain and determine which mutant is most
likely to spread in order to create a vaccine to protect the
population. In addition to the many deaths, this disease results in
huge economic losses from lost work and treatment of fragile
patients.
The disease follows a classic mathematical model known as an SIR model.
The population can be divided into susceptible, infected, and recovered
individuals. Once a person has a particular strain of one of the types
of the influenza virus, then that individual develops immunity to that
specific strain, preventing reinfection. (This is why flu viruses need
to mutate to keep finding new hosts.) Since flu acts over a very short
period of time, we will simplify our model by assuming that the
population remains constant with a size of N. This means that it is
sufficient to simply keep track of two classes of individuals, the
susceptibles, Sn, and the infected, In, as the recovered
satisfy
Since we are ignoring births and deaths in the population, we only
consider the spread of the disease, which is based on successful
contacts (such as inhaling aerosol infected particles from the sneeze
of an infected individual into the lungs of a susceptible host). The
discrete mathematical model is given by the system of equations:
where bS/N represents the proportion of
contacts by an
infected individual that result in the infection of a susceptible
individual.
The parameter g
is the probability that an
infected person
recovers (enters class R of the SIR model). The
ratio 1/g is the average length of the
infectious period of the disease.
Below we provide data from the Center of Disease Control for one season
of flu. The table below gives the number of cases week by week for a
control set of individuals. In particular, we consider the data for a
particular strain of A for the flu season in 2004 and 2005. Assume
that the control population consists of N
= 157,759. The first week, n
= 0, corresponds to the last week
in September, which is near the seasonal
start
of the flu season in the U. S.
n (wk) | In | n (wk) | In | n (wk) | In |
0 | 3 | 17 | 1096 | 34 | 2 |
1 | 2 | 18 | 1354 | 35 | 0 |
2 | 7 | 19 | 1335 | 36 | 2 |
3 | 12 | 20 | 1109 | 37 | 1 |
4 | 9 | 21 | 936 | 38 | 6 |
5 | 10 | 22 | 627 | 39 | 0 |
6 | 27 | 23 | 476 | 40 | 0 |
7 | 21 | 24 | 295 | 41 | 1 |
8 | 36 | 25 | 164 | 42 | 0 |
9 | 63 | 26 | 94 | 43 | 0 |
10 | 108 | 27 | 37 | 44 | 0 |
11 | 255 | 28 | 26 | 45 | 1 |
12 | 472 | 29 | 15 | 46 | 0 |
13 | 675 | 30 | 8 | 47 | 3 |
14 | 580 | 31 | 5 | 48 | 0 |
15 | 844 | 32 | 3 | ||
16 | 974 | 33 | 1 |
a. We want to simulate the SIR model above, finding the best
parameters, b
and g, that match the data from the CDC. Insert the
week number, n,
and the data on infected individuals in the first two columns. Use
Columns C and D for the simulation of the discrete model, Sn, and In. For initial values, take I0 = 3 and S0 = 157,756 = N - I0. Define your parameters, b and g, and take your initial guesses for these
parameters to be b = 3.2 and g = 2.7. In Column E, compute the square error between
the number of infected individuals in the data and those found by the
model. Use Excel's solver to find the best fitting parameters b and g that minimize the sum of square errors
of the infected individuals. Also, give the least sum of square
errors.
Give the model prediction for people infected with influenza at n
= 15 and n
= 25 and find the percent error at each of
these times from the actual CDC data given. As noted above, the average
length of the infectious period is equal to 1/g. Find this period in units of days. (Recall
that n
is
in weeks.)
Epidemiologists often examine what is called the basic reproduction ratio given
by
R0 = b/g,
which provides a measure of how rapidly
a disease will spread and how much of the population will be affected
by a particular disease. Use your values of b and g to find R0.
To determine the impact of a particular flu season, we want to know the
total number of individuals who were infected by the influenza virus.
Since we are assuming a constant population N and because the number of infected individuals
is essentially zero at the end of the simulation. We estimate the total
number of cases of flu by computing the number in the recovered class,
so
Rn = N - Sn,
for n large. Estimate the number of Influenza A
cases for the 2004 - 2005 flu season. What percent of the original
population
ultimately got this strain of influenza?
b. In your Lab report, create one graph with the data and the model of
the individuals infected with influenza. Create another graph showing
the number of susceptible individuals for this particular strain of
influenza over the 48 weeks. Write a brief paragraph discussing how
well the model fits the data. Also, discuss if the predicted value
of g gives a reasonable estimate of the infectious
period for the flu.
c. The CDC is interested in minimizing the impact of flu on the
population, so uses a number of different controls. We will examine
three different
controls that are employed to fight outbreaks of the flu.
The first line of defense of which you are undoubtedly aware is the
annual flu vaccine. The effect of a vaccine is to lower the population
of susceptible individuals, which is terms of our model is simply to
lower S0 by moving a number of individuals to R0. (In fact, many individuals already have
immunity to a given strain of the flu because of earlier contact with a
related strain.) Suppose we vaccinate 5% of the total population
at the very beginning of the flu season. This immediately shifts some
of the population to the recovered individuals in the model. We
calculate the initial susceptible population by
Assuming no other changes for this
epidemic, then the values of g and b remain the same as above. Simulate the SIR
model with this new initial condition. Find the number of
infecteds and susceptibles at n
= 20 and n
= 30. Also, simulate the
model sufficiently long, so there are essentially no new infecteds and
determine the total number of people who would have suffered from the
flu if 5% percent of the population is vaccinated. This
is
given by the number of individuals in the recovered class (subtracting
out the ones protected by the vaccine). Since the flu epidemic has
faded out (I
= 0), Rn ~ N - Sn for n large. An estimate of the number of people
getting the flu is given by Rn - R0. What percent of the population becomes
infected?
d. A second method of control is the quarantine of infected individuals
or the education of the public on how to lessen the contact between
infected
and susceptible individuals. This lowers the value of b.
Suppose we lower the value of b by 5%, which
is equivalent to multiplying the value of b found in Part a by 0.95. Use the original initial
conditions and best fitting value of g and this new b value to simulate the SIR model. Find the
number of infecteds and susceptibles at n
= 20 and n
= 30. Also, simulate the
model sufficiently long, so there
are essentially no new infecteds and determine the total number of
people
who would have suffered from the flu if this control was used. (Recall
that
since the flu epidemic has faded out (I
= 0), the people who
got the flu
are the people in the recovered class. Thus, Rn ~ N - Sn for n large.
What percent of the population becomes infected?
e. Oseltamivir or Tamiflu is a drug that shortens the symptoms of flu
for many people. If we assumes that this drug shortens the length of
the
period of infectivity of the infected individuals, then this can be
modeled by
increasing g.
Suppose we increase the value of g
by 5%, which is
equivalent to multiplying the
value of g
found in Part a by 1.05.
Use the original initial conditions and best fitting value of b to simulate the SIR model. Find the
number of infecteds and susceptibles at n
= 20 and n
= 30.
Also, simulate the model sufficiently long, so there are essentially no
new infected individuals and determine the total number of people who
would have suffered from the flu if this drug was used. What percent of
the population becomes infected?
f. In your Lab report, reproduce the graph with the original data and
the best fitting SIR model. Add graphs of the infected individuals from
the
SIR model using the three means of controlling the disease discussed
above.
Be sure to label each type of control on the graph. Describe what you
observe in your graphs both quantitatively and qualitatively. Compare
and
contrast the different approaches to controlling a flu outbreak. Give
advantages
and disadvantages of each of the controls. Include a discussion of the
practicality and financial burden of each of these approach. Give a
couple of strengths and weaknesses for this SIR model. Select another
disease that satisfies the SIR model and discuss how this lab applies
to treatment of the disease you are considering. How does this study
apply to elimination of other serious diseases?
[1] CDC Flu website - www.cdc.gov/flu/ (last visited Sept. 2009).
[2] Wikipedia - en.wikipedia.org/wiki/Influenza (last visited Sept.
2009).