Mathematical Models

of Epidemics
Christopher Lin, Math 89S Spring 2016

pidemics occur often, and result in in 1957 killed 2 million, and Hong Kong influenza

E both severe human and monetary losses.

However mathematical modeling of the
spread of infectious diseases can reveal impor-
in 1968 resulted in 1 million deaths (What are Epi-
demics). Every year, in the United States alone, more
than 200,000 people are hospitalized due to influenza,
tant insights into their behavior, and help sci- and anywhere from 5% to 20% of the population be-
entists and policy makers prevent or reduce comes infected. It is estimated that between 3,000 to
their adverse impacts. 49,000 people die from influenza each year (Seasonal
Influenza). Economic damage can also be severe; one
study estimated that the 2002 SARS outbreak in
What is an Epidemic? East Asia resulted in a 2.63% drop in GDP in Hong
Kong, and a 1.5% drop in GDP in mainland China
Generally, an epidemic is the spread of a disease to
(Lee and McKibbin, 2004).
a large number of people over a short period of time.
Because epidemics hold possibly dire consequences
Epidemics are divided into two major types: common
for entire countries, regions, or continents, it is ex-
source, and propagated outbreaks. Common source
tremely important for scientists and policy makers
outbreaks occur when individuals are exposed to
to prevent epidemics if possible. However, due to
the same toxin or infectious agent. The epidemic of
the complex and large-scale nature of epidemics, it
leukemia in Hiroshima following the atomic bomb
is basically impossible to perform non-retrospective
detonation, or an outbreak of E. coli from a batch of
studies. Furthermore, any sort of controlled experi-
infected vegetables would be examples of epidemics
ment in which a population is infected with a disease
caused by a common source. More relevant to our
would be incredibly unethical. Thus, the best way
discussion in this paper, are propagated outbreaks.
to study, predict, and experiment with the behavior
In a propagated outbreak, disease is spread from
of epidemics is through mathematical modeling.
person to person, such as in the case of influenza or
syphilis. A propagated outbreak can also be spread
by other vectors, such as mosquitoes in the case of Mathematical Modeling
malaria. When an epidemic spreads to encompass
multiple countries or continents, it is often reclassified One of the first to develop a mathematical model
as a pandemic (Principles of Epidemiology in Public of disease was the Swiss scientist and mathemati-
Health Practice). cian, Daniel Bernoulli, who was best known for his
Epidemics can cause massive human and economic work with the mathematics of fluid mechanics. In
losses. The Spanish Flu epidemic in 1918 caused 1766, Bernoulli, using a simple compartmental model
between 40 and 50 million deaths, Asian influenza and census data, was able to show that vaccinating

Mathematical Models of Epidemics • Math 89s Spring 2016 page 1 of 15

the population from smallpox would result in an in- one compartment to the next is modelled by a dif-
creased expected lifespan (Dietz and Heesterbeek, ferential equation, and when several compartments
2002). are present, the entire system can be modelled by a
The next major development came in the early system of differential equations.
twentieth century, with the work of Ronald Ross and
William Hamer. Ross was the first to discover the The SIR Model
malarial parasite, and was the first to prove that
mosquitoes were the vectors for malaria transmis- The SIR model, first developed by Kermack and
sion. Ross and and epidemiologist William Hamer McKendrick, is the most basic model, and is the basis
developed a compartmental method to model the for most compartmental epidemic models. It consists
spread of malaria (Weiss, 2013). of three compartments, susceptible (S), infectious (I),
Building upon the work of Ross and Hamer, Ker- and recovered (R), as shown in the following diagram.
mack and McKendrick developed the basis for what
is now known as the SIR model in the 1920’s. Using Figure 1: The SIR Model
their model, they were able to successfully explain
the behavior of bubonic plague in London and Bom-
bay, and also of cholera in 19th century London
(Weisstein).
Also in the 1920s, a very different model was devel-
oped by Reed and Frost. Rather than modelling an
epidemic with compartments and differential equa-
tions, they instead turned to a more statistical ap-
proach. By using a binomial distribution, they were
able to predict the number of infections in any given
time period. Reed and Frost, however, never thought
their model noteworthy, so the model was never offi- Individuals who start in the susceptible (S) become
cially published (Reed-Frost Epidemic Model). infectious and move into the infectious (I) compart-
ment at a rate proportional to the number of sus-
Types of Models ceptible, the number of infectious, and the infection
The most commonly used type of model in epidemiol- rate β. Individuals in the infectious (I) compartment
ogy are deterministic models. Generally speaking, a recover and move into the recovered (R) compart-
deterministic model is a mathematical model where ment at the recovery rate γ. It is important to note
there is no randomness involved, and the output is that this simple model does not account for vital
determined solely by the initial conditions. Deter- dynamics, that is, births or deaths, due to the dis-
ministic models work well for large populations, as ease in question or otherwise. By representing these
random fluctuations become negligible at such large movements as derivatives, we can derive the system
scale. Most deterministic models in epidemiology are of equations as follows:
based on the SIR compartmental model, which will
dS
we will explore in depth further on in this paper. = −βSI (1)
dt
Another type of model that could be used is the
stochastic model. Stochastic epidemic models ac- dI
= βSI − γI (2)
count for a certain degree of randomness, so that dt
the output of the model is not solely determined by dR
= γI (3)
the initial conditions. Stochastic models are usually dt
applied to small populations, where random fluctua- Looking at each term in the equations, we can see
tions can cause noticeable effects in outcome. that whenever one term appears as a negative in one
equation, it appears as a positive term in another.
Compartmental Models For example, in dS dI
dt , −βSI appears as βSI in dt .
This makes intuitive sense, because the model does
Compartmental models divide populations into com- not account for any vital dynamics, which means
partments, and simulate individuals moving from that the population remains the same for all time,
compartment to compartment. The movement from and individuals only move between compartments.

Mathematical Models of Epidemics • Math 89s Spring 2016 page 2 of 15

 Because there is no way for individuals to leave
Figure 3: The SIR model initialized with conditions
the recovered compartment, the basic SIR model can
S(0) = 0.8, I(0) = 0.2, with R0 > 0 and
be used to represent a disease in which immunity is R0 < 0
permanent. Examining this model, we can see that
over time, all of the population that is infectious
will move to the recovered compartment, and the
epidemic will burn itself out.

Figure 2: The SIR model initialized with conditions β =

0.6, γ = 0.34, S(0) = 0.8, I(0) = 0.2

Table 1: R0 of common diseases

Measles 12-18
Diptheria 6-7
Smallpox 5-7
Polio 5-7
Rubella 5-7
Mumps 4-7
HIV/AIDS 2-5
R0 , the Basic Reproduction Number

A result that can be derived from this model is the

concept of R0 , or the basic reproduction rate. R0 In the example given in in Figure 3, if R0 is greater
N 1
is a very important quantity in epidemiology. It than S(0) = 0.8 = 1.25, then an epidemic will occur.
represents the number of people that will be infected For different types of models, the epidemic threshold
by one infectious individual. Since γ represents the will be calculated differently.
rate at which an individual recovers, γ1 represents Table 1 gives the R0 values for several common
the time during which he is infectious. The number diseases (Smallpox: Disease, Prevention, and Inter-
of people infected by an individual can be found by vention).
multiplying the rate of infection by the time interval
during which he is infectious, yielding
SIR Model Variants
β
R0 = (4) Despite the variety of insights it may offer, the basic
γ SIR model is only useful in very limited situations.
It is important to note that the R0 derived by However, the SIR model is a very useful starting
equation (4) is not the true basic reproduction rate point to create models tailored to each situation.
of the disease. Because the model makes enormous
simplifications, the R0 that is derived here is only SIR with vital dynamics
an approximation, and is more accurately described
as the ”epidemiological threshold”. A more realistic version of the SIR model accounts
By rewriting dI dI
dt as dt = (R0 S/N − 1)γI, we can
for births and deaths within the population. This
see that if R0 > N/S(0), then dI dt > 0, so the number
kind of model is more useful when examining diseases
of infectious will rise, creating an epidemic, whereas in longer time intervals, as births and deaths do not
if R0 < N/S(0), the number of infectious will im- have a significant impact on shorter time intervals.
mediately drop and the outbreak will come to an Given a birth rate Λ and death rate µ, we can now
end. modify the SIR to reflect vital dynamics.

Mathematical Models of Epidemics • Math 89s Spring 2016 page 3 of 15

 susceptible and infectious populations reach an equi-
Figure 4: The SIR model with vital dynamics
librium, and the disease becomes endemic to the
population.

Figure 6: The SIS model initialized with conditions β =

0.6, γ = 0.34, S(0) = 0.8, and I(0) = 0.2

From this, we can formulate the following system

of differential equations:

dS
= Λ − µS − βIS (5)
dt
dI
= βIS − (γ + µ)I (6)
dt
dR
= γI − µR (7) SIRS Model
dt
Since we now have a different set of equations, the Sometimes, a disease does not confer permanent
formula for R0 is now immunity, but also is not without immunity. In
limited immunity situations, individuals only gain
βΓ immunity for a limited amount of time.
R0 = (8)
µ(µ + γ) In order to model this, we modify the SIR model
so that there is movement from the recovered com-
SIS Model partment back to the susceptible compartment.
The SIR model is only valid if immunity is conferred
upon those who have recovered from infection. In Figure 7: The SIRS model
diseases such as influenza and the common cold, there
is no long lasting immunity after recovery. Thus, the
SIR model is simplified to only two compartments,
susceptible and infectious.

Figure 5: The SIS model

From this, we derive the system of equations

dS
= −βSI + µR (11)
The following system of equations describe this dt
model: dI
= βSI − γI (12)
dt
dS
= −βSI + γI (9)
dt dR
= γI − µR (13)
dI dt
= βSI − γI (10)
dt Like the SIS model, because individuals can move
In a situation where the population cannot acquire back into the susceptible compartment, the system
immunity, the SIS model shows that eventually the will eventually reach an endemic equilibrium point.

Mathematical Models of Epidemics • Math 89s Spring 2016 page 4 of 15

Figure 8: The SIRS model initialized with β = 0.6, γ = Figure 10: The SEIR model
0.34, µ = 0.01, S(0) = 0.8, I(0) = 0.2

Here, we can derive a system of four differential

As shown in Figure 8, initially, the plot of the
equations:
SIRS model is quite similar to that of the basic SIR
model. However, by examining a larger time period, dS
we can see how it reaches an equilibrium point. = −βSI (14)
dt
dE
= βSI − αE (15)
dt
Figure 9: The SIRS model initialized with β = 0.6, γ = dI
0.34, µ = 0.01, S(0) = 0.8, I(0) = 0.2, over a = αE − γI (16)
dt
period of 1000 days
dR
= γI (17)
dt
The behavior of this model seems very similar to
that of the basic SIR model.

Figure 11: The SIRS model initialized with α = 0.14,

β = 0.6, γ = 0.34, µ = 0.01, S(0) =
0.8, I(0) = 0.2

SEIR Model

For many diseases, individuals who are infected do

not immediately become infectious. Instead, they go
through a latent or incubation period. To model this,
we can modify the SIR model to include a fourth However, upon closer inspection, we can see that
compartment, the exposed (E) compartment, with the introduction of a non-infectious latent phase
average incubation period α1 . Since the incubation to the system considerably dampens the spread of
period is α1 , the rate at which exposed become infec- infection. In fact, with all other initial conditions
tious occurs at rate α. being the same, the number of infectious in the SEIR

Mathematical Models of Epidemics • Math 89s Spring 2016 page 5 of 15

model never increases. However, the introduction destroyed through chemical or physical processes.
of a latent period has the effect of prolonging the These inactivated pathogens stimulate the body to
outbreak. to produce an immune response. Influenza, cholera,
and polio are examples of inactivated vaccines. At-
Figure 12: The infectious compartments of the SEIR
tenuated vaccines consist of live pathogens that have
and SIR models been made less virulent, or of similar but less harmful
pathogens. Examples of attenuated vaccines include
yellow fever, measles, and mumps.
In the SIR model, the best way to represent vac-
cination is a change in initial conditions. By vac-
cinating a portion of the population, we have ef-
fectively moved a portion of the population from
the susceptible compartment and into the recovered
compartment.
By testing out different initial values I(0) and R(0),
we can see that, as expected, vaccination decreases
suppresses the outbreak.

Figure 13: The SIRS model initialized with α = 0.14,

β = 0.6, γ = 0.34, µ = 0.01, and varying
Experimenting with the SIR model values of S(0) and R(0)

and Variants
Without mathematical models, scientists can only
examine the behavior of epidemics and the impact of
different outbreak controlling methods by examining
historical data. Because of the number of variables
that differ from epidemic to epidemic, relying on
historical data may not be a very effective method.
Also standing in the way of accurate predictions
based on historical data alone is the difficulty of
collecting information about a disease at large scales.
However, with the appropriate mathematical mod-
els, one can run ”experiments” by tweaking constants
and initial conditions. This can allow scientists to This result, however, as interesting as it might be,
test the efficacy of various disease-control methods, does not provide much insightful information into
such as vaccination and quarantine. the behavior of the model when vaccination is put
into effect. A much more useful insight would be
Effect of Vaccination on Epidemics finding at what percentage of vaccination does the
number of infected no longer increase. In order to
Vaccination is the process of activating the body’s ac- find this, we can conduct a phase plane analysis.
quired immunity system in order to prevent a person
For this example, we will use the SIRS model,
from contracting a disease in the future. The first to
with constants α = 0.14, β = 0.6, γ = 0.34, µ =
demonstrate the effects of vaccination was Edward
0.01. First, we observe that, when dealing with the
Jenner in the 1760s. Noticing that smallpox did not
population in terms of ratios, S(t)+I(t)+R(t) = 1 at
seem to effect dairy workers, he discovered that in-
all times. From this, we can make the substitution
oculating a person with a related disease, cowpox,
R = 1 − I − S, which enables us to simplify the
gave them immunity to smallpox. Vaccines for many
system of three equations into a system of two, as
diseases have since been formulated, most notably
such:
polio in the 1952 by Jonas Salk.
There are two main types of vaccines. Inacti- dS
vated vaccines consist of pathogens that have been = −βSI + µ(1 − S − I) (18)
dt

Mathematical Models of Epidemics • Math 89s Spring 2016 page 6 of 15

 dI
= βSI − γI (19)
dt
F or
Since we now have a system of two equations S(0) = 0.3, I(0) = 0.6,
that deals only with two variables, we can easily
dS (22)
draw a two dimensional plot, or phase plane that = −0.0980
dt
represents the behavior of the model at different dI
initial conditions. Furthermore, we note that since = −0.0960
dt
S(t) + I(t) + R(t) = 1, I(t) + S(t) = 1. Next, we wish
to find the points at which the number of infected
and susceptible are neither increasing nor decreasing, F or
or the nullclines. These are the points where dSdt = 0
S(0) = 0.2, I(0) = 0.2,
and dIdt = 0. Graphing I(t) + S(t) ≤ 1, dS
dt = 0, and dS (23)
dI = 0.0360
dt = 0, we generate the following plot: dt
dI
= −0.0440
dt
Figure 14: Phase plane of SIRS
By noting the sign of dS dI
dt or dt at each point, we
can draw arrows to approximate the behavior in each
quadrant.

Figure 15: Phase plane of SIRS

These nullclines separate the area bounded by the

aces and I(t) + S(t) = 1 into four quadrants. In each
of these quadrants, the model will behave differently.
By choosing values in each quadrant and plugging
into dS dI
dt and dt , we are able to ascertain the behavior
in each quadrant. From this, we are able to see that in order to
prevent the number of infectious from increasing at
all, the initial conditions must be in the bottom two
F or quadrants of the phase plane. Thus, the susceptible
S(0) = 0.6, I(0) = 0.075, population should be below the dI dt nullcline. Recall
dI
that the dt nullcline is given by the points where
dS (20) dI
= 0.0055 dt = 0, therefore, we have the equation
dt
dI dI
= 0.0015 = βSI − γI = 0 (24)
dt dt
which can easily be factored to

F or I(βS − γ) = 0 (25)
S(0) = 0.6, I(0) = 0.3, which in turn yields
dS (21)
= −0.0980
dt βS − γ = 0 (26)
dI
= 0.0060 Therefore, the line is given by
dt

Mathematical Models of Epidemics • Math 89s Spring 2016 page 7 of 15

 Effect of Quarantining Infected Individuals
γ 1
S= (27)
β
A common tactic employed in fighting disease out-
In this example, βγ = 0.5667. If the population is breaks is quarantining. Under such a policy, persons
vaccinated so that the susceptible population is below who have been infected by the disease will be iso-
0.5667, then the number of infectious individuals will lated, so that they are no longer able to infect more
not increase. Indeed, we see that with S(0) = 0.5, it people. Although, the practice of quarantining raises
does not matter what the initial value of I(0) may some ethical questions about forced confinement, it
be, the number of infectious will not increase. is still seen as a highly effective and common way of
controlling disease spread.
To model the effects of quarantining, we will make
Figure 16: SIRS model initialized with constants β = 0.6,
γ = 0.34 µ = 0.1, initial condition S(0) =
use of a modified SIR model. We will add a fourth
0.5, and varying values of R0 compartment, Q, to represent the quarantined popu-
lation. The Q compartment receives input from the
I and outputs to the R compartment, but does not
affect the rate at which people from the S compart-
ment become infected. Thus, we get the following
model:

Figure 17: The SIQR model

The quantity βγ should look quite familiar to the

reader. It is the inverse of the basic reproductive
rate, R0 . Recall that in our previous discussion of
R0 , we noted that the number of infectious would
rise if

β N
R0 = > (28)
γ S(0)
The average time it takes for an infectious person
Since we define N = 1, we can take the inverse of
to be quarantined is given by ζ1 , so the rate at which
the entire equation.
infectious move into quarantined is proportional to ζ.
γ S(0) From this model, we are able to derive the following
> (29) system of four differential equations:
β 1
dS
Thus, the we arrive at the same conclusion as = −βSI (30)
discussed before via the nullclines of the phase plane. dt
Here, we can see an illustration of the concept dI
= βSI − ζI (31)
of herd immunity. Herd immunity is the idea that dt
when a large portion of the population is immunized dQ
against a disease, the entire population is effectively = ζI − γR (32)
dt
immune. In this example, once the susceptible pop- dR
ulation is reduced to 0.5667, herd immunity takes = γR (33)
dt
effect, and outbreaks are quickly contained.
1
Although the general premise of the SIQR model was found
online, the analysis was performed solely by the student

Mathematical Models of Epidemics • Math 89s Spring 2016 page 8 of 15

 As intuition may suggest, decreasing the average
Figure 19: The I compartment of the SIQR model ini-
time before quarantine ζ1 and thereby increasing rate
tialized with conditions β = 0.6, γ = 0.34,
ζ dampens the spread of disease. S(0) = 0.8, I(0) = 0.2, and varying values
of ζ

Figure 18: The I compartment of the SIQR model ini-

tialized with conditions β = 0.6, γ = 0.34,
S(0) = 0.8, I(0) = 0.2, and varying values
of ζ

This means, that in this situation, the average

length of time before an individual is quarantined
must be below ζ1 = 0.48
1
= 2.0833 in order to prevent
a rise in infectious.

Case Study: The Haitian Cholera

Again, it is useful for us to find at what values of
ζ will the number of infectious increase or decrease.
Outbreak
dI
To do this, we examine the dt equation, and see that On January 12, 2010, a massive 7.0 magnitude earth-
it can easily be factored. quake struck Haiti just west of Port-au-Prince. Fol-
lowing the main quake, at least 52 aftershocks oc-
curred. It is estimated that the death toll ranged
dI from 100,000 to 160,000 people, and over three mil-
= βSI − ζI = I(βS − ζ) (34)
dt lion Haitians were affected by the quake. The damage
to infrastructure was massive; over 250,000 residences
and 30,000 commercial structures were destroyed.
By setting dI About ten months after the earthquake, the
dt = 0, we can see that the epidemic
threshold in relation to ζ is given by Haitian Ministry of Public Health and Population
was notified of a sudden rise in patients with watery
diarrhea and dehydration, trademark symptoms of
cholera. On October 22, 2010, the first outbreak
ζ = βS (35)
of cholera in Haiti in over a hundred years was an-
nounced. Due to poor sanitation infrastructure, es-
pecially in water, cholera spread quickly throughout
Thus, if ζ > βS, the number of infectious will the country. Over 470,000 cases of cholera have been
initially decrease, and if ζ < βS, the number of reported to date, and over 6,500 deaths have been
infectious will initially increase. For this example, attributed to the disease.
we obtain the threshold value
The Biological Basis of Cholera
Cholera is caused by the bacteria Vibrio cholerae.
ζ = (0.6)(0.8) = 0.48 (36)
Vibrio cholerae is transmitted through contaminated
food or water. After being consumed, the Vibrio
cholerae that have survived the acidity of the stom-
This result is verifiable by our MATLAB plot. ach enter the small intestine, and inhabit the the

Mathematical Models of Epidemics • Math 89s Spring 2016 page 9 of 15

intestinal wall. The Vibrio cholerae then begin to Cholera may also be prevented through the usage
multiply, and after an incubation of about 2-3 days,of cholera vaccines. Most commonly, cholera vac-
begin to release cholera toxin (CTX). CTX causes cines consist of inactivated Vibrio cholerae, which
the body to undergo severe, watery diarrhea. If the are administered orally. Trials of the oral vaccine
diarrhea containing Vibrio cholerae enters the food have shown 85% protection for 4-6 months, and 65%
or water supply, Vibrio cholerae is able to once again
effectiveness up to five years. Although rarely used,
undergo its life cycle. there also exists an injectable version of the cholera
vaccine. Despite the effectiveness of these vaccines,
the WHO does not recommend using vaccines in
Figure 20: Vibrio cholerae viewed under an scanning
place of conventional cholera prevention methods,
electron microscope
such as building a sanitary water infrastructure.
The World Health Organization, however recom-
mends focusing resources on the prevention of, rather
than the treatment of cholera. Cholera outbreaks
can be prevented by proper sterilization practices,
treatment of sewage, and purification of drinking
water sources.

Modeling the Cholera Situation in Haiti

In 2011, Tulte et. al published ”Cholera Epidemic
in Haiti, 2010: Using a Transmission Model to Ex-
plain Spatial Spread of Disease and Identify Optimal
Control Interventions”, which used a highly complex
variant of the SIR model to mathematically describe
CTX causes a variety of symptoms in those who the spread of cholera throughout Haiti. Their model
have contracted cholera. The primary symptom is consists of two main components, the SIRW model,
intense, watery diarrhea, sometimes characterized as and the patch model.
”rice water stool”. The onset of diarrhea is often sud-
den, and can cause as much as 1 liter of fluid loss per The SIRW Model 2
hour. During the early stages of infection, an individ-
ual may experience persistent vomiting and nausea, Because individuals who have recovered from a
which can last for hours. Both of theses symptoms cholera infection do not lose their immunity at a
lead to dehydration, which can be severe or deadly. significant rate, the SIRW model is based upon the
Dehydration can cause dangerous symptoms such as basic SIR model. In order to reflect the biological
low blood pressure and arrhythmia. The low blood mechanism through which cholera is transmitted,
pressure, in turn, may lead to hypovolemic shock, that is, its water-borne nature, a new compartment,
which can cause death in minutes. Furthermore, the water (W) is added to the model. The water compart-
rapid rate of dehydration may lead to an electrolyte ment is ”infected” by the infectious compartment,
imbalance. Loss of sodium, chloride, and potassium and in turn can infect the susceptible compartment.
may cause muscle cramps. Children infected with
cholera are particularly susceptible to dehydration Figure 21: The SIRW model
induced hypoglycemia, which may cause seizures or
coma.
Treatment for cholera is usually supportive. Pa-
tients may undergo oral rehydration therapy, which
entails drinking water mixed with sugar and elec-
trolytes. Oral rehydration therapy is highly effective,
and very inexpensive to implement, a quality which
leads to its widespread use in developing nations,
where cholera is most often a problem. Antibiotic 2
Some of the variable and constant names have been changed
treatment can shorten the course of the disease and so that the equations remain somewhat consistent with the
reduce symptoms, but is not necessary for recovery. models previously described in this paper

Mathematical Models of Epidemics • Math 89s Spring 2016 page 10 of 15

 From this model, we derive the following system
Figure 22: The SIRW model with constants given by the
of equations:
paper, and initial conditions S(0) = 0.8 and
I(0) = 0.2, over a 90 day period.
dS
= µ − αS − µS (37)
dt

dI
= αS − γI − µI (38)
dt

dR
= γI − µR (39)
dt

dW
= ξ(I − W ) (40)
dt

In this model, vital dynamics are taken into ac-

count, with birth and death rate both being equal
The Patch Model
to µ. The concentration of Vibrio cholerae in the
water is given by W , and scaled by a parameter ξ, One of the major flaws of the SIR model is the lack
which reflects the decay rate of the bacteria in the of any sort of spatial or geographic considerations.
water. The most significant difference between the The SIR model assumes that the entire population
SIR model and the SIRW model is the way that is homogeneously mixed, and everyone has an equal
the infection rate is calculated. In the SIRW model, chance of coming in contact with everyone else. Of
the infection rate, given as α is determined by the course, this is not at all the case, as humans tend
following equation to gather into highly concentrated areas, separated
by sparsely populated areas. Furthermore, the rate
α = βI I + βW W (41) at which people move between these areas differ
due to differences in variables such as geography.
A rural town connected by only a few roads would
Where βI is the person-to-person transmission rate, experience far fewer travellers than a city connected
and βW is the waterborne transmission rate. The by a freeway.
basic reproductive rate, R0 , given by:
The patch model attempts to address these short-
comings. The patch model divides a country or
βI + βW region into distinct ”patches”. The number of in-
R0 = (42)
γ+µ fected in one patch influences the number of infected
in the others through a ”gravity” relationship, that
However, it is very unlikely for an individual to be is, it is proportional to the populations of the two
infected with cholera from person-to-person contact. patches and inversely proportional to a power of the
Also, the effect of vital dynamics, given by µ is quite distance between them.
small. Thus, Thus, the influence of the jth patch on the ith
patch is given by the equation
βW
R0 = (43)
γ (Pi Pj )
θij = k (44)
dn
Since this system is too complex to perform any
sort of phase plane analysis on, we input the model where k is some ”gravity constant”, and n is some
into MATLAB, and input the constants given by the power.
paper. The paper does not give a value for µ, so we In this paper, Haiti is divided into ten patches
find it by using the average life expectancy in Haiti. based on the ten administrative departments. The
Thus, we obtain µ = 1/62.7 = 0.0159. distance is calculated from their capitals.

Mathematical Models of Epidemics • Math 89s Spring 2016 page 11 of 15

Figure 23: Map of the departments and capitals of Haiti dSi
= µ − αSi − µSi (46)
dt
dIi
= αSi − γIi − µIi (47)
dt
dRi
= γIi − µi R (48)
dt
dWi
= ξ(Ii − Wi ) (49)
dt
Now that we have developed our model of the
cholera outbreak in Haiti, we can begin to experiment
with the effects of different policies.

Effect of Vaccination on Haiti

Because the cholera outbreak started before anyone
could be vaccinated, we cannot represent vaccination
Combining the Models as a change in the initial conditions. Therefore, we
introduce the movement from the susceptible to the
Combining these two models, we develop a model
recovered compartments into the actual model. Our
in which the SIRW model takes place in each patch,
SIRW model now looks like this:
and influences the transmission rates in the other
patches.
Figure 25: The SIRW model with vaccination

Figure 24: Diagram showing the combined SIRW and

patch model, taken from Tulte, et. al, 2011

We can now rewrite our system of equations as

dSi
= µ − αSi − vi S − µSi (50)
dt
dIi
= αSi − γIi − µIi (51)
dt
dRi
= γIi − +vi S + µi R (52)
dt
dWi
Now, we can incorporate the patch model equationi = ξ(Ii − Wi ) (53)
into the SIRW equations. The infection rate α is dt
now given by the equation where vi represents the probability of being vacci-
nated within a department.
Using this model, Tulte et. al was able to three
X
10
different vaccination strategies, equally allocating
αi = βIi Ii + βW i Wi + θij Ij {i : i 6= j} (45)
vaccines to each department, allocation based on the
j=1
population of each department, and an optimized
Here, we have added a term which is the summa- allocation. The optimized allocation focused vaccina-
tion of the influences of every other patch on the tion in the department of Ouest, because of its high
ith patch. Now, we can rewrite our entire system of population as well as its central location. The high
equations with the patch model in mind. population and short distance to other patches lead

Mathematical Models of Epidemics • Math 89s Spring 2016 page 12 of 15

to Ouest having the highest impact on each of the
Figure 27: Vaccination vs Clean Water Distribution,
other patches.
taken from Tulte, et. al

Figure 26: Different allocation schemes, taken from

Tulte, et. al

Tulte et. al found that although water was some-

what effective in reducing cases, vaccination was still
a far better option.

Effect of Rainfall
Effect of Clean Water Provision It has been a long held observation that rainfall seems
to correlate with the spread of cholera. During rainy
The authors of this paper also explored using the seasons, the incidence of new cholera cases seems
provision of clean water as a substitute for vaccina- much higher.
tion. Providing clean water eliminates the possibility
of waterborne transmission, but does not change the
Figure 28: Cholera cases and rainfall, taken from Eisen-
person-to-person transmission rate. To reflect this
berg et. al, 2013
in our model, we subtract the individuals who were
infected due to waterborne transmission from the
infectious compartment, and add them back to the
susceptible compartment.
Thus, we arrive at the equations:

dSi
= µ − αSi + βwi wi ci si − µSi (54)
dt

dIi
= αSi − γIi − βwi wi ci si − µIi (55)
dt
In Eisenberg, Kujbida, Tuite, Fisman, and Tien’s
2013 paper ”Examining rainfall and cholera dynam-
dRi
= γIi − µi R (56) ics in Haiti using statistical and dynamic modeling
dt
approaches”, evidence is provided for a strong rela-
tionship between rainfall and cholera. We will not
discuss the statistical models employed in this paper,
dWi
= ξ(Ii − Wi ) (57) but we will look at the dynamic model.
dt
Eisenberg et. al employs the same SIRW model
previously discussed. However, they modified the
where ci represents the probability of clean water model so that the infection rate would be propor-
within a patch. tional to the output of a function. The rain function

Mathematical Models of Epidemics • Math 89s Spring 2016 page 13 of 15

is a function of time, independent of S, I, W, and R. initialization of the model, each agent being either
Thus, the rate of infection is now given as designated workers or students, commute to a work-
place or school, following a path determined by a
α = βI I + βW W frain (t) (58) transportation network supplied by a GIS.
With this model, they were able to simulate a
Combining the SIRW, Patch, and Rainfall measles outbreak in Burnaby, Canada.
Models 3
Figure 29: Measles Outbreak in Burnaby, taken from
In order to create the most robust model of the Haiti
Perez and Dragicevic, 2009
situation, we can combine the rainfall model and the
patch model. Doing so is quite easy; all we have to
do is change the rate of infection α. In the combined
model, α is given by

X
10
αi = βIi I+βW i Wi frain (t)+ θij Ij {i : i 6= j} (59)
j=1

This model lets us not only determine the optimal

allocation vaccines, but also the optimal time to
administer them.
Through this case study, we can see the power
of compartmental models of epidemics. By simply Not only were they able to track the number of
adding more compartments as necessary and chang- agents in each state, they were also able to examine
ing constants, we can create a mathematical model their geographic concentration.
that is representative of the underlying biological
reality. Figure 30: Measles Outbreak in Burnaby, taken from
Perez and Dragicevic, 2009

Agent Based Modeling

Another way to represent an epidemic is through
agent based modeling. Agent based modeling is
a computational model, rather than a traditional
mathematical one. This means that they require a
computer to be run.
An agent based model is made up of many individ-
ual ”agents”, who act according to simple rules. The
behavior of an epidemic as a whole is ascertained
through the behavior of groups of agents.
A 2009 paper by Liliana Perez and Suzana Drag-
icevic, ”An agent-based approach for modeling dy-
namics of contagious disease spread” make use of
an agent bade model to model the spread of disease
within a single city.
In their model, an agent represents a single person.
Following the SEIR model, each agent can either be
in the S, E, I, or R state. Using geographical infor-
mation system (GIS) data, agents are then placed
randomly around the map of the city. At each time
interval, an infectious agent may infect other agents
With computers growing increasingly faster, com-
within a one meter radius of them. Following the
putational models such as this agent based model
3
This is original work from the student become more and more promising due to their ability

Mathematical Models of Epidemics • Math 89s Spring 2016 page 14 of 15

to accurately represent the effect of spatial relation-
ships on disease spread. "Smallpox: Disease, Prevention, and Intervention".
The CDC and the World Health Organization. Slide 16-

Conclusion Tuite, A. R., Tien, J., Eisenberg, M., Earn, D. J.

Cholera Epidemic in Haiti, 2010: Using a Transmissio
Mathematical models are essential to the field of
Spatial Spread of Disease and Identify Optimal Contr
epidemiology. Without them, there is no way to
Annals of Internal Medicine Ann Intern Med, 154(9),
conduct experiments or make predictions about the
doi:10.7326/0003-4819-154-9-201105030-00334
future. Whether compartmental, stochastic, or agent
based, mathematical models all give us deeper insight
into the behavior of epidemics. Furthermore, they
allow us to make better choices regarding disease
control and prevention.

Bibliography
Cholera. (n.d.). Retrieved April 26, 2016,
from https://www2.health.vic.gov.au/public-health/
infectious-diseases/disease-information-advice/cholera

Cholera in Haiti: One Year Later. (n.d.). Retrieved April 26, 2016,
from http://www.cdc.gov/haiticholera/haiti_cholera.htm

Dietz, K., & Heesterbeek, J. (2002). Daniel Bernoullis epidemiological model revisited.
Mathematical Biosciences, 180(1-2), 1-21. doi:10.1016/s0025-5564(02)00122-0

Eisenberg, M. C., Kujbida, G., Tuite, A. R., Fisman, D. N., & Tien, J. H. (2013).
Examining rainfall and cholera dynamics in Haiti using statistical and
dynamic modeling approaches. Epidemics, 5(4), 197-207.
doi:10.1016/j.epidem.2013.09.004

Epidemics, Pandemics, and Outbreaks of Contagious Diseases Like the Flu. (n.d.).
Retrieved April 26, 2016, from
http://www.webmd.com/cold-and-flu/what-are-epidemics-pandemics-outbreaks

Kermack-McKendrick Model. (n.d.). Retrieved April 26, 2016,

from http://mathworld.wolfram.com/Kermack-McKendrickModel.html

Lee, J. (n.d.). Retrieved April 26, 2016,

from http://www.ncbi.nlm.nih.gov/books/NBK92473/

Lesson 1: Introduction to Epidemiology. (2012).

Retrieved April 26, 2016, from
http://www.cdc.gov/ophss/csels/dsepd/ss1978/lesson1/section11.html

Perez, L., & Dragicevic, S. (2009). An agent-based approach for

modeling dynamics of contagious disease spread.
International Journal of Health Geographics Int J Health Geogr, 8(1), 50.
doi:10.1186/1476-072x-8-50

Seasonal Influenza Q&A. (2015). Retrieved April 26, 2016,

from http://www.cdc.gov/flu/about/qa/disease.htm

Mathematical Models of Epidemics • Math 89s Spring 2016 page 15 of 15