Building a Basic Model COVID-19 Parameters Implementations Extensions Two-Group Model General Recursive Model

Simple Mathematical Models

Clayton Mizgerd
Max Enis
Hemal Patel
Alexandra Bonat

November 2020
Building a Basic Model COVID-19 Parameters Implementations Extensions Two-Group Model General Recursive Model

Do outbreak curves have a characteristic shape?

Salmonella outbreak

COVID-19 outbreak China SARS outbreak

Building a Basic Model COVID-19 Parameters Implementations Extensions Two-Group Model General Recursive Model

A simple model

Consider an infectious disease outbreak where each

infected individual infects two other individuals

Number of Infected Individuals t

What’s wrong with this model?
Building a Basic Model COVID-19 Parameters Implementations Extensions Two-Group Model General Recursive Model

Compartmental Modelling

The disease will eventually run out of people to infect.

We can split population into "compartments":

Susceptible, S
Infected, I

βSI
S −→ I
Observe the following
dS
= −βSI
dt
dI
= βSI
dt
Building a Basic Model COVID-19 Parameters Implementations Extensions Two-Group Model General Recursive Model

Compartmental Modelling

We no longer infect
Number of Infected Individuals

infinitely, however
all of the population
remains infected.

Thus, we can
introduce a new
compartment, R, to
indicate the recov-
t ered population.
Building a Basic Model COVID-19 Parameters Implementations Extensions Two-Group Model General Recursive Model

Refined Model

Our new model now looks like this:

βSI µI
S −→ I −→ R
We now get
dS
= −βSI
dt
dI
= βSI − µI
dt
dR
= µI
dt
Building a Basic Model COVID-19 Parameters Implementations Extensions Two-Group Model General Recursive Model

SIR Model
Number of Infected Individuals

We now have a
simple model which
only implements two
parameters β and µ.
βSI µI
S −→ I −→ R

t
Building a Basic Model COVID-19 Parameters Implementations Extensions Two-Group Model General Recursive Model

A Basic Extension

Infectious disease outbreaks are not smooth curves.

We can use stochastic modelling to capture the random

nature of outbreaks.
Building a Basic Model COVID-19 Parameters Implementations Extensions Two-Group Model General Recursive Model

Model Parameters

Effective Reproductive Number (Rt )

Fatality Rate
Length of Infection
Building a Basic Model COVID-19 Parameters Implementations Extensions Two-Group Model General Recursive Model

Reproductive Number

R0 : Basic Reproductive Number

COVID-19: Likely around 2.5, estimates range from 1.4-4.0
(Majumder and Mandl, 2020)
Depends on location
Rt : Effective Reproductive Number
Infection rate with public health measures initiated
Influenced by many factors
Building a Basic Model COVID-19 Parameters Implementations Extensions Two-Group Model General Recursive Model

Massachusetts Infection Rate

Building a Basic Model COVID-19 Parameters Implementations Extensions Two-Group Model General Recursive Model

Factors Influencing Rt : Undocumented Cases

40% of cases are asymptomatic (CDC)

Additional mild cases
Generally less infectious than confirmed cases (75%
for asymptomatic)
Ruiyun et al., May 2020
Building a Basic Model COVID-19 Parameters Implementations Extensions Two-Group Model General Recursive Model

Factors Influencing Rt : Non-Pharmaceutical Interventions

Social distancing
Reducing each person’s number of contacts
Dehning et al., July 2020
Building a Basic Model COVID-19 Parameters Implementations Extensions Two-Group Model General Recursive Model

Factors Influencing Rt : Masking

Rader et al. (2020) found that an increase in mask

wearing is associated with increased odds of
transmission control (Rt < 1)
Stutt et al. (2020) found that mask wearing can
decrease Rt even if masks are as little as 50%
effective
Building a Basic Model COVID-19 Parameters Implementations Extensions Two-Group Model General Recursive Model

Factors Influencing Rt : Facemask Effectiveness

Building a Basic Model COVID-19 Parameters Implementations Extensions Two-Group Model General Recursive Model

Factors Influencing Rt : Non-Pharmaceutical Interventions

Imperial College Report

Building a Basic Model COVID-19 Parameters Implementations Extensions Two-Group Model General Recursive Model

Fatality Rate

Determines % of those moved from Infected to

Removed who died/recovered
Case Fatality Rate vs. Infection Fatality Rate
Case Fatality (US): 2.2%
Infection Fatality Rate:
Estimates range from 0.00% to 1.54% (median 0.27%)
(Ioannidis, 2020)
Varies based on age demographics (CDC):
0-19 years: 0.003%
20-49 years: 0.02%
50-69 years: 0.5%
70+ years: 5.4%
Building a Basic Model COVID-19 Parameters Implementations Extensions Two-Group Model General Recursive Model

Length of Infection

4-5 days average incubation period

10 days after onset of symptoms for mild to moderate
cases (CDC)
Length of required isolation for asymptomatic cases
For severe cases, likely no longer than 20 days
88% no longer shedding replication-competent virus after
10 days
95% after 15 days (van Kampen et al., 2020)
Building a Basic Model COVID-19 Parameters Implementations Extensions Two-Group Model General Recursive Model

Matrix SIR Model

Definition (Matrix SIR Model)

The matrix SIR model looks at the population as a vector
~vt = {St , It , Rt } and follows the equation ~vt+1 = A~vt for
some matrix A.
The simplest way to implement the SIR model is using a
matrix.
Building a Basic Model COVID-19 Parameters Implementations Extensions Two-Group Model General Recursive Model

Matrix SIR Model

Definition (Matrix SIR Model)

The matrix SIR model looks at the population as a vector
~vt = {St , It , Rt } and follows the equation ~vt+1 = A~vt for
some matrix A.

Example
Consider starting population ~v = {100, 10, 0}, and matrix
 
0.9 −0.1 0
A = 0.1 0.9 0 .
0 0.2 1

This is equivalent to the system of linear equations

St+1 = 0.9St −0.1It , It+1 = 0.1St +0.9It , Rt+1 = Rt +0.2It .
Building a Basic Model COVID-19 Parameters Implementations Extensions Two-Group Model General Recursive Model

Issues

Consider our last example. While the total population will

remain constant, the susceptible population will be
negative by time t = 14.
Adjusting parameters to prevent negative populations
results in a simple exponential curve without the
characteristic shape of an outbreak.
Building a Basic Model COVID-19 Parameters Implementations Extensions Two-Group Model General Recursive Model

Remedy

How can we fix this problem?

Example
A matrix model might give the linear system
St+1 = St − 0.3It , It+1 = 1.1It , Rt+1 = Rt + 0.2It .
Consider instead
St+1 = St − 0.3It St /N,
It+1 = 0.8It + 0.3It St /N,
Rt+1 = Rt + 0.2It ,
where N is the total population.
We now have St , It , Rt ≥ 0 for any nonnegative input. The
data will have the characteristic shape of an outbreak.
Building a Basic Model COVID-19 Parameters Implementations Extensions Two-Group Model General Recursive Model

Nonlinear SIR Model

Definition (Nonlinear SIR Model)

The nonlinear SIR model uses a system of nonlinear
equations to model the spread of disease.

Advantages:
no population can go negative
more accurate
characteristic shape
Building a Basic Model COVID-19 Parameters Implementations Extensions Two-Group Model General Recursive Model

Nonlinear SIR Model

Definition (Nonlinear SIR Model)

The nonlinear SIR model uses a system of nonlinear
equations to model the spread of disease.

Disadvantages:
cannot generally give closed-form expression
Building a Basic Model COVID-19 Parameters Implementations Extensions Two-Group Model General Recursive Model

Differential SIR Model

Definition (Differential SIR Model)

The differential SIR model uses a system of differential
equations to model disease.

Example

dS
= −βS(t)I(t)
dt
dI
= βS(t)I(t) − µI(t)
dt
dR
= µI
dt
Building a Basic Model COVID-19 Parameters Implementations Extensions Two-Group Model General Recursive Model

Differential SIR Model

Definition (Differential SIR Model)

The differential SIR model uses a system of differential
equations to model disease.

Advantages:
most realistic
continuous
Building a Basic Model COVID-19 Parameters Implementations Extensions Two-Group Model General Recursive Model

Differential SIR Model

Definition (Differential SIR Model)

The differential SIR model uses a system of differential
equations to model disease.

Disadvantages:
can be more difficult to explain & understand
cannot be explicitly solved
Building a Basic Model COVID-19 Parameters Implementations Extensions Two-Group Model General Recursive Model

Euler’s Method

Euler’s method is one way mathematicians model

differential equations that cannot be solved.
Euler’s method treats each step of a differential equation
as a linear equation.
S

t2

t1 ∆t
t

dS
S(t2 ) = S(t1 ) + ∆t (t1 ).
dt
Building a Basic Model COVID-19 Parameters Implementations Extensions Two-Group Model General Recursive Model

Runge-Kutta

Runge-Kutta is a more accurate tool for estimating

differential equations.
Building a Basic Model COVID-19 Parameters Implementations Extensions Two-Group Model General Recursive Model

Runge-Kutta

Runge-Kutta is a more accurate tool for estimating

differential equations.
Given the system dS/dt, dI/dt, dR/dt, calculate
dS dS k1 m1 dS k2 m2
   
k1 = ∆t (S(t), I(t)) k2 = ∆t S(t) + , I(t) + k3 = ∆t S(t) + , I(t) +
dt dt 2 2 dt 2 2
dI dI k1 m1 dI k2 m2
   
m1 = ∆t (S(t), I(t)) m2 = ∆t S(t) + , I(t) + m3 = ∆t S(t) + , I(t) +
dt dt 2 2 dt 2 2
dR dR k1 m1 dR k2 m2
   
n1 = ∆t (S(t), I(t)) n2 = ∆t S(t) + , I(t) + n3 = ∆t S(t) + , I(t) +
dt dt 2 2 dt 2 2

dS k1 + 2k2 + 2k3 + k4
k4 = ∆t (S(t) + k3 , I(t) + m3 ) k = S(t + ∆t) = S(t) + k
dt 6
dI m1 + 2m2 + 2m3 + m4
m4 = ∆t (S(t) + k3 , I(t) + m3 ) m= I(t + ∆t) = I(t) + m
dt 6
dR n1 + 2n2 + 2n3 + n4
n4 = ∆t (S(t) + k3 , I(t) + m3 ) n= R(t + ∆t) = R(t) + n
dt 6
Building a Basic Model COVID-19 Parameters Implementations Extensions Two-Group Model General Recursive Model

Our Approach

We decided to use the nonlinear system of equations

model.
Euler’s method is effectively the same as the recursive
method with small step size.
Building a Basic Model COVID-19 Parameters Implementations Extensions Two-Group Model General Recursive Model

SIS Model (Susceptible-Infectious-Susceptible)

The SIS model is a model for diseases that can be

contracted multiple times by the same person.
Definition (SIS Model)
The SIS model is a variant of the SIR model with only two
compartments, Susceptible and Infectious. Persons who
would be removed are returned to susceptible.
Building a Basic Model COVID-19 Parameters Implementations Extensions Two-Group Model General Recursive Model

SIS Model (Susceptible-Infectious-Susceptible)

The SIS model is a model for diseases that can be

contracted multiple times by the same person.
Definition (SIS Model)
The SIS model is a variant of the SIR model with only two
compartments, Susceptible and Infectious. Persons who
would be removed are returned to susceptible.

Evidence so far suggests that COVID-19 reinfections

are uncommon
Nobody dies and is removed
Building a Basic Model COVID-19 Parameters Implementations Extensions Two-Group Model General Recursive Model

SEIR Model (Susceptible-Exposed-Infectious-Removed)

The SEIR model introduces the Exposed compartment.

This groups people who are in the incubation period many
viruses have.
Definition (SEIR Model)
The SEIR model is a variant of the SIR model with four
compartments. The exposed category marks people who
cannot spread the disease, but have been exposed to it
and will become infectious.
Building a Basic Model COVID-19 Parameters Implementations Extensions Two-Group Model General Recursive Model

SEIR Model (Susceptible-Exposed-Infectious-Removed)

The SEIR model introduces the Exposed compartment.

This groups people who are in the incubation period many
viruses have.
Definition (SEIR Model)
The SEIR model is a variant of the SIR model with four
compartments. The exposed category marks people who
cannot spread the disease, but have been exposed to it
and will become infectious.

SARS-CoV-2 is infectious during the incubation

period
Building a Basic Model COVID-19 Parameters Implementations Extensions Two-Group Model General Recursive Model

SIRD Model (Susceptible-Infectious-Recovered-Deceased)

The SIRD model allows easy tracking of how many

people have died of the disease.
Definition (SIRD Model)
The SIRD model is a variant of the SIR model with four
compartments. The Recovered compartment marks only
those who recovered, and Deceased marks those who
died of the disease.
Building a Basic Model COVID-19 Parameters Implementations Extensions Two-Group Model General Recursive Model

SIRD Model (Susceptible-Infectious-Recovered-Deceased)

The SIRD model allows easy tracking of how many

people have died of the disease.
Definition (SIRD Model)
The SIRD model is a variant of the SIR model with four
compartments. The Recovered compartment marks only
those who recovered, and Deceased marks those who
died of the disease.

Useful for displaying value of flattening the curve

Deaths can also be calculated later as proportion of
Removed
Building a Basic Model COVID-19 Parameters Implementations Extensions Two-Group Model General Recursive Model

Two-Group Model


Iu,n Su,n ku Ie,n Su,n ke
Su,n+1

 = Su,n − − − jSu,n


 N N

 Iu,n Se,n ku Ie,n Se,n ke

 Se,n+1 = Se,n − − + jSu,n
N N



Iu,n Su,n ku Ie,n Su,n ke
Iu,n+1 = Iu,n + + − Iu,n m
N N





 Iu,n Se,n ku Ie,n Se,n ke
Ie,n+1 = Ie,n + + − Ie,n m


N N




Rn+1 = Rn + Iu,n m + Ie,n m


where Su,n + Se,n + Iu,n + Ie,n + Rn = N and

Su,0 = N − 1, Se,0 = 0, Iu,0 = 1, Ie,0 = 0, R0 = 0.
Building a Basic Model COVID-19 Parameters Implementations Extensions Two-Group Model General Recursive Model

Parameters

Two-group SIR model has four parameters: ku , ke , j,

and m
Note: Disease progresses from nth stage to (n + 1)st
stage after one "time period"
ku represents the quantity of people that an
uneducated, infected individual interacts with times
the probability that a susceptible individual who
interacts with this person gets sick
ke represents the same thing, but for educated,
infected individuals
Building a Basic Model COVID-19 Parameters Implementations Extensions Two-Group Model General Recursive Model

Excel Demonstration
Building a Basic Model COVID-19 Parameters Implementations Extensions Two-Group Model General Recursive Model

Parameters

m represents 1 divided by the number of time periods

for the average person to recover
j represents the proportion of uneducated who
become educated after each time period
Under this model, the R0 value for early stages of the
disease to be represented as ku /m for uneducated
individuals and ke /m for educated individuals
Building a Basic Model COVID-19 Parameters Implementations Extensions Two-Group Model General Recursive Model

Limitations

Does not allow for the handling of multiple NPIs

Does not allow for separate treatment of individuals
by age
Parameters cannot be determined to sufficiently great
accuracy
Model cannot be used to assess the spread of the
virus to particular locations within the USA
Model does not admit a closed form (most likely)
Building a Basic Model COVID-19 Parameters Implementations Extensions Two-Group Model General Recursive Model

Assumptions

The model assumes that

educated individuals are equally likely to be infected
from the virus as uneducated individuals
educated infected individuals are less likely to spread
the virus than uneducated infected individuals
the USA is a closed and localized system (there is no
travel to and from the USA)
within the USA, location has no impact
the only artificial inhibitor to the spread of the virus is
"education"
age does not affect the spread of the virus
. . . and more
Building a Basic Model COVID-19 Parameters Implementations Extensions Two-Group Model General Recursive Model

General Recursive Model

Suppose that the population is split into p distinct groups,

so that we have groups S1 , S2 , . . . , Sp , I1 , I2 , . . . , Ip , and
R1 , R2 , . . . , Rp . Let A be the m × m matrix of interaction
coefficients k1,1 , k1,2 , . . . , k2,1 , . . . , kp,p representing the
number of people belonging to a group that a person from
another group may infect. Then,
k1,1 k1,2 · · · k1,p
 
k2,1 k2,2 · · · k2,p 
A=  ... .. ... ..  .
. . 
kp,1 kp,2 · · · kp,p
Building a Basic Model COVID-19 Parameters Implementations Extensions Two-Group Model General Recursive Model

General Recursive Model

Furthermore, define Xn to be the p × p diagonal matrix

with diagonal elements equal to S1,n , S2,n , . . . Sp,n , so that
S1,n 0 · · · 0
 
 0 S2,n · · · 0 
Xn =  .
 .. .
.. . .. ..  . Finally, denote Sn to be the
. 
0 ··· 0 Sp,n
S1,n
 
 S2,n 
N-dimensional vector Sn = 
 ... , and define In and Rn


Sm,n
similarly.
Building a Basic Model COVID-19 Parameters Implementations Extensions Two-Group Model General Recursive Model

General Recursive Model

Then, the general

 recursive SIR model for a population of
Sn+1 = Sn − Xn AIn

N



N groups is X AI
n n
In+1 = In + − mIn
N




Rn+1 = Rn + mIn

where m represents the frequency of recovery of the
virus, and N is the total population.
Note that if the p groups are split into uneducated and
educated groups, an additional "education" term may be
added to the first equation which takes some proportion of
the people in the uneducated group and moves them into
their respective educated group.
Building a Basic Model COVID-19 Parameters Implementations Extensions Two-Group Model General Recursive Model

Parameters

Interaction coefficients ka,b determine the rate at

which infected people from group b will infect
susceptible people from group a
The m parameter is the same between the two-group
model and the general model: represents the
frequency of which infected people are removed from
the population, or, in other words,
1/(length of infection)
Building a Basic Model COVID-19 Parameters Implementations Extensions Two-Group Model General Recursive Model

Excel Demonstration
Building a Basic Model COVID-19 Parameters Implementations Extensions Two-Group Model General Recursive Model

Benefits over Two-Group Model

The model can account for the drastically varying

effects of COVID based on age
The model can predict COVID effects in a partially
vaccinated population by creating a new population
which rarely infected
The model can also predict certain NPIs other than
"education" by creating a new population
The model achieves more generality while
maintaining simple calculations
Building a Basic Model COVID-19 Parameters Implementations Extensions Two-Group Model General Recursive Model

Limitations

The quantity of variables is p2 + 2, which is quadratic

in the number of population groups
The interaction coefficients are practically infeasible
to calculate so the model likely has limited application
to reality
Model still cannot account for location effects
Building a Basic Model COVID-19 Parameters Implementations Extensions Two-Group Model General Recursive Model

Assumptions

The model assumes that

as with the two-group model, the USA is a closed and
localized system (there is no travel to and from the
USA)
within the USA, location has no impact
there are no societal effects that exist temporally
across multiple populations (for example, legal
mandates)
the interaction coefficients between any two
populations remain constant over the lifetime of the
virus
Building a Basic Model COVID-19 Parameters Implementations Extensions Two-Group Model General Recursive Model

Thank You