You are on page 1of 34

ABSTRACT

This paper aims at studying how SEIR model


predict and evaluate the trajectory of the
pandemic in Hong Kong so as to provide the
government with a scientific basis and
reference for preventing the spread of the
epidemic.
REPORT ON THE Instructor: Dr. Hugo Wai Leung MAK
Author: CHAN, Wong Shan
MAED 6980

6980 PROJECT
Applications of The Modified SEIR Model in
Generating COVD-19 Prediction
Content
Introduction ……………………………………………………………………………………………….. 2
Research Background and Literature Review ………………………………………….. 2-4
Overview of SEIR Model …………………………………………………………………………. 4-7
Special Case ……………………………………………………………………………………………….. 7-9
Stability of ODE Systems …………………………………………………………………………. 9-11
Numerical Analysis …………………………………………………………………………………. 11-14
Data Source and Parameters …………………………………………………………………… 14-15
Geographical Region ……………………………………………………………………………….. 15-16
Simulations Results …………………………………………………………………………………. 16-19
Analysis …………………………………………………………………………………………………… 19-20
Discussion ……………………………………………………………………………………………….. 20-22
Limitations and Improvements ………………………………………………………………… 22-31
Conclusion ……………………………………………………………………………………………….. 31
References ………………………………………………………………………………………………. 32-33

1|Page
1.Introduction
Coronavirus disease 2019 (COVID-19), an infectious illness brought on by SARS-CoV-
2, was initially noted in Wuhan, China, and has now spread to all of Hong Kong since
its initial epidemic at the end of December 19. Currently, this virus has infected
2,894,776 persons in Hong Kong and has killed up to 13,605 people. According to
some reports, the pandemic's impacts caused Hong Kong's economy to lose an
estimated $27 billion in potential growth [1]. The economic growth of Hong Kong
and the lives of its citizens have already been profoundly and significantly impacted
by this novel kind of coronavirus pneumonia. Understanding the dynamics of COVID-
19 transmission and developing effective control strategies are urgently needed. This
paper aims at studying how SEIR model predict and evaluate the trajectory of the
pandemic in Hong Kong so as to provide the government with a scientific basis and
reference for preventing the spread of the epidemic.
In this study, the deadly COVID-19 pandemic will be analyzed using the SEIR
(Susceptible-Exposed-Infectious-Recovered) model, which divides the entire
population into four distinct compartments. The case-recovery-death count data
reported in Hong Kong on April 1, 2020 will be used to train this model, which will
then produce predictions for the next two years. Further, the special cases of this
model will be discussed. The policy effect with different contact rates will be
simulated to estimate the COVID-19 trends while the policy declared on different
stage will be simulated as well to depict the disease’s behavior. With the help of the
well-known computing program MATLAB Simulink, the results of this numerical
simulation of the SEIR model are clearly examined and the findings are also
subjected to theoretically rigorous numerical analysis. Our findings suggest that the
government should adopt the preventive policy such as subsidizing medical masks
and sanitizer, keeping social distance, locking down city, and banning travel as early
as possible to lower the contact rete between people and people so that Hong Kong
lives can be better saved.
2.Research Background and Literature Review
During the 2020 Chinese Lunar New Year, Hong Kong experienced its first COVID-19
epidemic [2], and it had a significant negative impact on the city's economic growth
as well as the lives of its residents. As a result of the COVID-19 epidemic, there was
either direct or indirect political unrest, economic misery, and even fatalities. The
2|Page
health and safety of the people of Hong Kong are seriously threatened. On February
5, the Hong Kong authorities shut down all border crossings save from those at the
airport, Shenzhen Bay, and the Hong Kong-Zhuhai-Macau Bridge. Beginning in late
January 2021, the government wanted to frequently lock down residential dwellings
in order to conduct extensive testing.
Researchers are adopting different disease transmission models to investigate the
coronavirus's growth patterns and conduct their research on how to halt COVID-19.
To assist people in responding to the COVID-19 pandemic, some well-known
epidemiological models were employed, including the baseline curve-fitting model,
extended SIR (eSIR) model, extended SEIR model (SAPHIRE and SEIR-fancy), and
mechanistic Bayesian hierarchical model (ICM) [3]. Differential equations are
typically thought to be the form that these models take. The SEIR model stood out as
a strong choice when comparing these five methods because of its publicly available
R-package, necessary flexibility, and accuracy. This may be explained by the fact that
the exposed (E) compartment was taken into account by the SEIR model since there
is a lengthy latency period during which sick individuals are not yet contagious. We
note that this SEIR model is much applicable for the reality world in the face of
COVID-19.
1. To discuss the unique ODE system situations in this paper, we build an SEIR
model based on nonlinear differential equations. Examples include the stability of
ODE systems, the generalized-endemic equilibrium point, and the disease-free
equilibrium point. To examine the stability of the non-linear equation systems,
the Jacobian matrix will be used to linearize the equation systems.
2. Later, we will discuss the numerical analysis about ode45 solver provided by
MATLAB and its basic principles Runge Kutta-4 behind ode45 solver.
3. Using COVID-19 case-recovery-death count data reported in Hong Kong on
April 1, 2020, we utilize the ode45 solver offered by MATLAB to carry out
numerical analysis and make predictions on April 1, 2020, in order to forecast
and assess the course of the pandemic in Hong Kong..
4. Lastly, we analyze the importance of early preventive policy effect as
flattening the ‘infectious’ curve will be beneficial for local hospitals to ease its
capacity at the peak of outbreak and beneficial for scientists to have enough time
to develop a vaccine.

We divided the paper into four sections for this use. The mathematical model,
stability analysis, and the possibility of an equilibrium point are covered in the first

3|Page
section. The simulation results and analysis will be presented in the second section,
which is for the case study. The third section is for discussion, we analyze the
importance of early preventive policy effect. The last section is for limitations and
improvements.
3.Methodology
• 3.1 Overview of SEIR Model
The SEIR model, which represents four compartments of the entire population, is the
most basic and straightforward model. In this model, there is compartment E
(exposed) that accommodates those who have been infected but are not yet
infectious because there is a latency period for some infections.

The following model results from assuming that the latency period has an
exponential distribution from E to I with parameter 𝜀 and birth rate of the population
Λ equals to the death rate 𝜇𝑁 :

𝛽𝐼𝑆
𝑆 ′ = Λ − 𝜇𝑆 −
𝑁
𝛽𝐼𝑆
𝐸′ = − (𝜇 + 𝜀)𝐸
𝑁
𝐼 ′ = 𝜀𝐸 − (𝛾 + 𝜇)𝐼
𝑅′ = 𝛾𝐼 − 𝜇𝑅

We make several assumptions so that the SEIR model is suitable to use:


1. There is a large enough population to use this SEIR model.
2. Human to human contact is the only method of transmission.
3. The population is constant since we assume that the death and birth rates are
equal.
4. We can ignore the low migration rate in the area we were thinking about.

4|Page
5. Personal features of an individual, such as age, ethnicity, and comorbidities, are
not related to susceptibility or recovery.
The aforementioned assumptions are applicable to Hong Kong scenario by following
reasons:
1. The Hong Kong population is around 7.481 million [4] derived from the Hong
Kong Census and Statistics Department in early 2020, this number is large
enough to run the SEIR model.
2. Hong Kong is a highly dense city and the chance of transmission is mainly from
human from human.
3. By the Public Health Information and Statistics of Hong Kong, the death rate is
6.8 per 1000 population while its birth rate is 5.6 per 1000 population in Hong
Kong in 2020 [4]. The minor error between the birth rate and death rate is
small enough to be ignored.
4. Due to the locking down city policy in 2020, the migration rate is low to be
ignored.
5. The assumption 5 is our limitation and we will discuss our improvement in the
Limitation and Improvement part.
For better analyzing the real scenarios, we modify the SEIR model by considering the
fatality compartment with fatality rate 𝛼. The transmission of a population with
various transition rates among the four compartments is shown in the picture below
(Figure 1):

Figure 1: Flowchart of the proposed SEIR model.

The population's birth rate is represented by the parameter Λ in Figure 1, whereas


the population's natural death rate is represented by the parameter 𝜇. Additionally,
people from the susceptible population are spreading infection-forcedly to the
exposed population with 𝛽. The parameter 𝜀 displays the rate at which a random
variable is transmitted from an exposed population to an infected population.

5|Page
Further, 𝛼 is the fatality rate and 𝛾 shows the recovery rate from infectious
population to recovered population.
Table 1: Definition of the compartments.
Compartments Brief definition
S population that is susceptible to sickness
E population at risk of exposure (infected but not infectious)
I population that is infected and infectious
R population that is recovered from COVID-19
F population where COVID-19 causes fatalities

Table 2: Precise parameter interpretation.


Parameters Interpretation
𝚲 Birth rate per person
𝜷 Probability of susceptible individuals joining exposed class
𝜺 Rate of exposed class members being infectious
𝜸 Recovery rate of infectious population
𝜶 Average mortality rate caused by viruses
𝝁 Natural death rate per person

The governing differential equations are now under consideration:


𝛽𝐼𝑆
𝑆 ′ = Λ − 𝜇𝑆 − (1)
𝑁
𝛽𝐼𝑆
𝐸′ = − (𝜇 + 𝜀)𝐸 (2)
𝑁

𝐼 ′ = 𝜀𝐸 − (𝛾 + 𝛼 + 𝜇)𝐼 (3)
𝑅′ = 𝛾𝐼 − 𝜇𝑅 (4)
𝐹 ′ = 𝛼𝐼 (5)

With 𝑁 = 𝑆 + 𝐸 + 𝐼 + 𝑅 ≤ 𝑁0 in this case, and each of parameters and initial


conditions are non-negative.
For this model, the reproduction number 𝑅0 which is simplified as:

6|Page
𝛽𝑆(𝑡)
𝑅0 (𝑡) = (*)
𝛾

Since we assume the population is constant, this SEIR model can be simplified
without considering the death rate, birth rate, and the fatality caused by COVID-19.
By definition, the reproduction number 𝑅0 is the condition in which the disease will
spread. In other words, we need compartments I and E to increase. Then,
𝐸 ′ = 𝛽𝐼𝑆 − 𝜀𝐸 > 0
𝐼 ′ = 𝜀𝐸 − 𝛾𝐼 > 0
Adding above-mentioned two equations, we have:
𝛽𝐼𝑆 − 𝛾𝐼 > 0
𝛽𝐼𝑆 > 𝛾𝐼
𝛽𝑆
𝑅(𝑡) = >1
𝛾
𝛽𝑆(0) 𝛽𝑁
𝑅0 = = (#)
𝛾 𝛾

• 3.2 Special Case


Based on the abovementioned ODE systems, there are two special cases such as an
Endemic Equilibrium (EE) and a Disease-Free-Equilibrium (DFE). The initial conditions
are shown below:

(𝑆0 , 𝐸0 , 𝐼0 , 𝑅0, 𝐹0 ) ∈ {(𝑆, 𝐸, 𝐼, 𝑅, 𝐹) ∈ [0, 𝑁]5 }


Where 𝑆 ≥ 0, 𝐸 ≥ 0, 𝐼 ≥ 0, 𝑅 ≥ 0, 𝐹 ≥ 0 𝑎𝑛𝑑 𝑆 + 𝐸 + 𝐼 + 𝑅 ≤ 𝑁0
If 𝐼 = 0, lim (𝑆, 𝐸, 𝐼, 𝑅, 𝐹) = (𝑁, 0,0,0,0)
𝑡→∞

If 𝐼 > 0, lim (𝑆, 𝐸, 𝐼, 𝑅, 𝐹) = 𝐸𝐸


𝑡→∞

• 3.2.1. The Disease-Free Equilibrium Point (DFE)


To find the equilibrium point of the ODEs system, it means that all the derivatives
should be zero:
𝛽𝑆𝐼
𝑆 ′ = Λ − 𝜇𝑆 − =0 (6)
𝑁

7|Page
𝛽𝑆𝐼
𝐸′ = − (𝜇 + 𝜀) = 0 (7)
𝑁

𝐼 ′ = 𝜀𝐸 − (𝛾 + 𝜇 + 𝛼)𝐼 = 0 (8)
𝑅′ = 𝛾𝐼 − 𝜇𝑅 = 0 (9)
𝐹 ′ = 𝛼𝐼 = 0 (10)
The physical interpretation of DFE is that no people getting infected in this case. In
other words, we have the condition that 𝐼 = 0 . Since we have the boundary domain
of (𝑆0 , 𝐸0 , 𝐼0 , 𝑅0 , 𝐹0 ) ∈ {(𝑆, 𝐸, 𝐼, 𝑅, 𝐹) ∈ [0, 𝑁]5 }, then considering the equations from
(6) to (10), we have:
∵ 𝐼=0
∴ 𝑁 = 𝑆, 𝐸 = 0, 𝑅 = 0, 𝐹 = 0
The disease-free equilibrium point is therefore (N, 0, 0, 0, 0) which is trivial.

• 3.2.2. The Generalized-Endemic Equilibrium Point (EE)

The physical interpretation of EE is that there are people getting infected in this case.
In other words, we have the condition that 𝐼 > 0 . Again, by considering the
equations from (7) and (8), we have:
(𝛾 + 𝜀 + 𝜇)(𝜇 + 𝜀)
𝑆= 𝑁
𝛽𝜀
By equation (6), we have:
𝑁
𝐼 = 𝛽(Λ − 𝜇𝑆)
𝑆
By equations (6) and (7), we have:
Λ − 𝜇𝑆
𝐸=
𝜇+𝜀
By equations (6) and (9), we have:
𝛾𝛽 𝑁
𝑅= (Λ − 𝜇𝑆)
𝜇 𝑆

8|Page
By equations (6) and (10), we have:
𝛼
𝐹= (Λ − 𝜇𝑆)𝑁
𝛽
All the above-mentioned equations are represented by the total population N and
susceptible population S which can be found by all given variables and total
population. Hence, it is easy to find the generalized-endemic-equilibrium point once
the variables are defined or estimated.
• 3.3 Stability of ODE Systems
Before we start to explore the stability of the given ODEs system, it is necessary for
us to recall the fundamental methodology applied to do the stability analysis.
For any two given differential equations:
𝑑𝑥
= 𝑎𝑥 + 𝑏𝑦
𝑑𝑡
𝑑𝑦
= 𝑐𝑥 + 𝑑𝑦
𝑑𝑡
It can be written in this manner:
𝑎 𝑏 𝑎
𝑇=( ) 𝑎𝑛𝑑 𝑋 = ( ) , 𝑠𝑜 𝑋 ′ = 𝑇𝑋
𝑐 𝑑 𝑏
Where 𝑋(𝑡) = 𝛼𝑒 𝑡𝜆1 𝑉1 + 𝛽𝑒 𝑡𝜆2 𝑉2 while 𝜆1 and 𝜆2 are eigenvalues and 𝑉1 and 𝑉2 are
eigenvectors.
With different roots of the equation systems, the stability of the systems can be
classified as following:
Table 3: Stability analysis
Roots Signs Stability
Two distinct real values Both + Unstable
Both - Stable
One + One - Unstable
Repeated values Both + Unstable
Both - Stable
Complex numbers 𝐑𝐞(𝛌) > 𝟎 Unstable
𝐑𝐞(𝛌) < 𝟎 Stable
𝐑𝐞(𝛌) = 𝟎 Oscillatory
9|Page
Now, we linearize the nonlinear systems by using Jacobian matrix:

By the Jacobian matrix of the systems equations from (1) to (5), we have:
𝛽𝐼 𝛽𝑆
−𝜇 − 0 − 0 0
𝑁 𝑁
𝛽𝐼 𝛽𝑆
𝐽(𝑆, 𝐸, 𝐼, 𝑅, 𝐹) = −(𝜇 + 𝜀) 0 0
𝑁 𝑁
0 𝜀 −(𝛾 + 𝛼 + 𝜇) 0 0
0 0 𝛾 −𝜇 0
( 0 0 𝛼 0 0)
Now, consider the DFE point where (𝑆, 𝐸, 𝐼, 𝑅, 𝐹) = (𝑁, 0,0,0,0). Then:
−𝜇 0 −𝛽 0 0
0 −(𝜇 + 𝜀) 𝛽 0 0
𝐽(𝑁, 0,0,0,0) = 0 𝜀 −(𝛾 + 𝛼 + 𝜇) 0 0
0 0 𝛾 −𝜇 0
(0 0 𝛼 0 0)

Since the eigenvalues are the diagonal elements of the matrix, and all the
eigenvalues are negative in this case. Hence, it is stable at DFE point.
(𝛾+𝜀+𝜇)(𝜇+𝜀) Λ−𝜇𝑆
Again, consider the EE point where (𝑆, 𝐸, 𝐼, 𝑅, 𝐹) = ( 𝑁, , 𝛽(Λ −
𝛽𝜀 𝜇+𝜀
𝑁 𝛾𝛽 𝑁 𝛼
𝜇𝑆) , (Λ − 𝜇𝑆) , (Λ − 𝜇𝑆)𝑁). Then:
𝑆 𝜇 𝑆 𝛽

10 | P a g e
𝛽𝐼𝐸𝐸 𝛽𝑆𝐸𝐸
−𝜇 − 0 − 0 0
𝑁 𝑁
𝛽𝐼𝐸𝐸 𝛽𝑆𝐸𝐸
𝐽(𝑆𝐸𝐸 , 𝐸𝐸𝐸 , 𝐼𝐸𝐸 , 𝑅𝐸𝐸 , 𝐹𝐸𝐸 ) = −(𝜇 + 𝜀) 0 0
𝑁 𝑁
0 𝜀 −(𝛾 + 𝛼 + 𝜇) 0 0
0 0 𝛾 −𝜇 0
( 0 0 𝛼 0 0)

Since the eigenvalues are the diagonal elements of the matrix, and all the
eigenvalues are negative in this case. Hence, it is stable at EE point.

• 3.4 Numerical Analysis


ODE problems can be solved using a variety of techniques, including the Runge
Kutta-2 and Runge Kutta-4 methods, the analytical functions ode23 and ode45
offered by MATLAB, and the Euler method (forward and backward) [5]. Two factors
need to be taken into account in order to choose the best strategy for solving ODE
systems. First, the strategy enables us to obtain the most precise algorithm
calculation results. Second, we may achieve accurate findings at a low cost, with time
and storage costs mostly based on the number of runs and the volume of data used,
respectively. To determine a large class of equations, it is necessary to first
determine whether an equation is stiff or not based on the accurate requirement
with low cost, after which it is necessary to determine whether the numerical
calculation method has stability, consistency, and order convergence of truncation
error. Consider whether the numerical answer converges after calculation. After
completing this procedure, we can identify which numerical computation approach
solves ODE systems the quickest and with the greatest accuracy. Since the accuracy
of ode23 is much lower than that of ode45 in our situation, we would use ode45
from MATLAB to carry out the numerical analysis. Actually, the fourth and fifth
orders of the Runge Kutta method are the mathematical foundation for the MATLAB
solver known as ode45.
Consider the Runge Kutta-4:
𝐾1 = 𝑓(𝑥𝑛 , 𝑦𝑛 )
ℎ ℎ
𝐾2 = 𝑓 (𝑥𝑛 + , 𝑦𝑛 + 𝐾1 )
2 2

11 | P a g e
ℎ ℎ
𝐾3 = 𝑓 (𝑥𝑛 + , 𝑦𝑛 + 𝐾2 )
2 2
𝐾4 = 𝑓(𝑥𝑛 + ℎ, 𝑦𝑛 + ℎ𝐾3 )

𝑦𝑛+1 = 𝑦𝑛 + (𝐾1 + 2𝐾2 + 2𝐾3 + 𝐾4 )
6
Consider the nonlinear ordinary differential equations from (1) to (5) again:
Let,
𝛽𝐼𝑆
𝑆 ′ = Λ − 𝜇𝑆 − = 𝑓1
𝑁
𝛽𝐼𝑆
𝐸′ = − (𝜇 + 𝜀)𝐸 = 𝑓2
𝑁

𝐼 ′ = 𝜀𝐸 − (𝛾 + 𝛼 + 𝜇)𝐼 = 𝑓3
𝑅′ = 𝛾𝐼 − 𝜇𝑅 = 𝑓4
𝐹 ′ = 𝛼𝐼 = 𝑓5

With 𝑁 = 𝑆 + 𝐸 + 𝐼 + 𝑅 + 𝐹 and 𝑆(𝑡0 ) = 𝑆0 , 𝐸(𝑡0 ) = 𝐸0 , 𝐼(𝑡0 ) = 𝐼0 , 𝑅(𝑡0 ) =


𝑅0 , 𝐹(𝑡0 ) = 𝐹0 , we have:
𝐾1 = ℎ𝑓1 (𝑡0 , 𝑆0 , 𝐸0 , 𝐼0 , 𝑅0 , 𝐹0 )
𝐿1 = ℎ𝑓2 (𝑡0 , 𝑆0 , 𝐸0 , 𝐼0 , 𝑅0 , 𝐹0 )
𝑀1 = ℎ𝑓3 (𝑡0 , 𝑆0 , 𝐸0 , 𝐼0 , 𝑅0 , 𝐹0 )
𝑁1 = ℎ𝑓4 (𝑡0 , 𝑆0 , 𝐸0 , 𝐼0 , 𝑅0 , 𝐹0 )
𝑂1 = ℎ𝑓5 (𝑡0 , 𝑆0 , 𝐸0 , 𝐼0 , 𝑅0 , 𝐹0 )
ℎ 𝐾1 𝐿1 𝑀1 𝑁1 𝑂1
𝐾2 = ℎ𝑓1 (𝑡0 + , 𝑆0 + , 𝐸0 + , 𝐼0 + , 𝑅0 + , 𝐹0 + )
2 2 2 2 2 2
ℎ 𝐾1 𝐿1 𝑀1 𝑁1 𝑂1
𝐿2 = ℎ𝑓2 (𝑡0 + , 𝑆0 + , 𝐸0 + , 𝐼0 + , 𝑅0 + , 𝐹0 + )
2 2 2 2 2 2
ℎ 𝐾1 𝐿1 𝑀1 𝑁1 𝑂1
𝑀2 = ℎ𝑓3 (𝑡0 + , 𝑆0 + , 𝐸0 + , 𝐼0 + , 𝑅0 + , 𝐹0 + )
2 2 2 2 2 2
ℎ 𝐾1 𝐿1 𝑀1 𝑁1 𝑂1
𝑁2 = ℎ𝑓4 (𝑡0 + , 𝑆0 + , 𝐸0 + , 𝐼0 + , 𝑅0 + , 𝐹0 + )
2 2 2 2 2 2
12 | P a g e
ℎ 𝐾1 𝐿1 𝑀1 𝑁1 𝑂1
𝑂2 = ℎ𝑓5 (𝑡0 + , 𝑆0 + , 𝐸0 + , 𝐼0 + , 𝑅0 + , 𝐹0 + )
2 2 2 2 2 2
ℎ 𝐾2 𝐿2 𝑀2 𝑁2 𝑂2
𝐾3 = ℎ𝑓1 (𝑡0 + , 𝑆0 + , 𝐸0 + , 𝐼0 + , 𝑅0 + , 𝐹0 + )
2 2 2 2 2 2
ℎ 𝐾2 𝐿2 𝑀2 𝑁2 𝑂2
𝐿3 = ℎ𝑓2 (𝑡0 + , 𝑆0 + , 𝐸0 + , 𝐼0 + , 𝑅0 + , 𝐹0 + )
2 2 2 2 2 2
ℎ 𝐾2 𝐿2 𝑀2 𝑁2 𝑂2
𝑀3 = ℎ𝑓3 (𝑡0 + , 𝑆0 + , 𝐸0 + , 𝐼0 + , 𝑅0 + , 𝐹0 + )
2 2 2 2 2 2
ℎ 𝐾2 𝐿2 𝑀2 𝑁2 𝑂2
𝑁3 = ℎ𝑓4 (𝑡0 + , 𝑆0 + , 𝐸0 + , 𝐼0 + , 𝑅0 + , 𝐹0 + )
2 2 2 2 2 2
ℎ 𝐾2 𝐿2 𝑀2 𝑁2 𝑂2
𝑂3 = ℎ𝑓5 (𝑡0 + , 𝑆0 + , 𝐸0 + , 𝐼0 + , 𝑅0 + , 𝐹0 + )
2 2 2 2 2 2
𝐾4 = ℎ𝑓1 (𝑡0 + ℎ, 𝑆0 + 𝐾3 , 𝐸0 + 𝐿3 , 𝐼0 + 𝑀3 , 𝑅0 + 𝑁3 , 𝐹0 + 𝑂3 )
𝐿4 = ℎ𝑓2 (𝑡0 + ℎ, 𝑆0 + 𝐾3 , 𝐸0 + 𝐿3 , 𝐼0 + 𝑀3 , 𝑅0 + 𝑁3 , 𝐹0 + 𝑂3 )
𝑀4 = ℎ𝑓3 (𝑡0 + ℎ, 𝑆0 + 𝐾3 , 𝐸0 + 𝐿3 , 𝐼0 + 𝑀3 , 𝑅0 + 𝑁3 , 𝐹0 + 𝑂3 )
𝑁4 = ℎ𝑓4 (𝑡0 + ℎ, 𝑆0 + 𝐾3 , 𝐸0 + 𝐿3 , 𝐼0 + 𝑀3 , 𝑅0 + 𝑁3 , 𝐹0 + 𝑂3 )
𝑂4 = ℎ𝑓5 (𝑡0 + ℎ, 𝑆0 + 𝐾3 , 𝐸0 + 𝐿3 , 𝐼0 + 𝑀3 , 𝑅0 + 𝑁3 , 𝐹0 + 𝑂3 )
Hence,
1
𝑆1 = 𝑆0 + (𝐾1 + 2𝐾2 + 2𝐾3 + 𝐾4 )
6
1
𝐸1 = 𝐸0 + (𝐿1 + 2𝐿2 + 2𝐿3 + 𝐿4 )
6
1
𝐼1 = 𝐼0 + (𝑀1 + 2𝑀2 + 2𝑀3 + 𝑀4 )
6
1
𝑅1 = 𝑅0 + (𝑁1 + 2𝑁2 + 2𝑁3 + 𝑁4 )
6
1
𝐹1 = 𝐹0 + (𝑂1 + 2𝑂2 + 2𝑂3 + 𝑂4 )
6

13 | P a g e
4.Case Study
• 4.1 Data Source and Parameters
To simulate the SEIR model and forecast the COVID-19 in Hong Kong from 1 April
2020 to 1 April 2021, we will use data from the Census and Statistics Department,
the Department of Health in Hong Kong, and the World Health Organization (WHO).
According to the data provided by the Department of Health of Hong Kong on 1 April
2020, we have:
Table 4 [6]: Reported cases and Deaths in Hong Kong on 1 April 2020.
Total cases New cases Total death New death Total Recovery Active Cases
766 0 4 0 118 644

At that time, the Hong Kong population is around 7.481 million derived from the
Census and Statistics Department. So, we have the initial conditions:
𝑆0 = 7.481 × 106 − 766
𝐸0 = 0
𝐼0 = 644
𝑅0, = 118
𝐹0 = 4

The World Health Organization (WHO) published a report of parameter estimation


on March 3, 2020 (during the early stages of the COVID-19 outbreak), which claimed
that approximately 3.4% of known COVID-19 cases have died globally and that
infected individuals need 5.2 days to become contagious. In other words, the early
stage latent periods predicted by the WHO are approximately 5.2 days. Additionally,
it was stated that 2.2 persons in the overall susceptible population would be infected
by one infectious person, and that an infectious person would require around 7 days
to recover. In other words, the infectious individuals need seven days to go from
compartment I to compartment R, and the reproduction number 𝑅0 is around 2.2.
At that time, the Hong Kong population is around 7.481 million derived from the

14 | P a g e
Census and Statistics Department. Hence, we are able to define our parameters as
following:
Table 5 [7]: Parameters estimated by the WHO on 3 March 2020.
Parameters Defined Estimated values
𝟏 𝟏
𝜺
𝐋𝐚𝐭𝐞𝐧𝐭 𝐩𝐞𝐫𝐢𝐨𝐝𝐬 𝟓.𝟐
𝟏 𝟏
𝜸
𝐃𝐮𝐫𝐚𝐭𝐢𝐨𝐧 𝟕
𝜶 Fatality rate 𝟑. 𝟒%
𝑹𝟎 Reproduction number 𝟐. 𝟐
𝐑𝟎 𝟐.𝟐
𝜷 Contact rate =
𝐍×𝐃𝐮𝐫𝐚𝐭𝐢𝐨𝐧 𝟕.𝟒𝟖𝟏×𝟏𝟎𝟔 ×𝟕

Even though the estimated values declared by the WHO will change across time, it is
the best estimation for all parameters on 3 March 2020.

• 4.2 Geographical Region


Since its initial pandemic at the end of December 19, the SARS-CoV-2-related
sickness known as COVID-19, which was first reported in Wuhan, China, has spread
over the entire territory of Hong Kong. As of right now, this virus has killed up to
13,605 people and infected 2,894,776 people in Hong Kong. According to estimates,
the pandemic's effects cost Hong Kong's economy $27 billion in lost potential growth
[1]. This novel type of coronavirus pneumonia has already had a profound and
significant impact on the economic development of Hong Kong as well as the lives of
its residents. Understanding the mechanics of COVID-19 transmission and creating
efficient control techniques are essential needs for the Hong Kong authorities in
order to stop the disease's spread. To stop the pandemic from spreading, the Hong
Kong authorities closed all border crossing points on February 5, 2020, with the
exception of those at the airport, Shenzhen Bay, and the Hong Kong-Zhuhai-Macau
Bridge. The SEIR model simulation now serves as a scientific foundation and point of
reference to show that it is crucial for the Hong Kong government to adopt
preventative policies like city lockdowns, funding for medical masks and hand
sanitizer, maintaining social distance, and travel bans to lower the contact rete
between people as soon as possible in order to save more Hong Kong lives.
4.3 Simulations Results

15 | P a g e
Now, we consider the parameters as follow to conduct the SEIR model simulation
and predict the COVID-19 trend in one year.
Parameters Estimated values
𝟏
𝜺
𝟓.𝟐
𝟏
𝜸
𝟕
𝜶 𝟑. 𝟒%
𝑹𝟎 𝟐. 𝟐

2.2
In this scenario, we set the contact rate to be β = 7.481×106×7 . in other words, we
predict the COVID-19 trend without preventive policy effect. we get:

Figure 2: Simulations without preventive policy effect

According to Figure 2, if no preventive measures are implemented by the


government to reduce the rate of human contact, the infectious population will peak
around 1.4 million people on 1 July 2020 and then begin to decline, while the death
population will peak around 2.2 million people on 1 September 2020 and remain
unchanged. Even though the prediction number is not realistic due to statistical error
and parameter estimation error, both of which were reported by the WHO at the
early stage of COVID-19 on 3 March 2020, it is still crystal clear that COVID-19 will

16 | P a g e
spread quickly and result in significant loss of life if no preventive measures are put
in place.
Now we consider the Hong Kong government adopt preventive policy such as locking
down city, subsidizing medical masks and sanitizer, keeping social distance and
1 1 2.2
banning travel to lower the contact rate namely βP = β= on 1 July
2 2 7.481×106 ×7
2020 which is around 90 days after the 1 April 2020. In this scenario, we predict the
COVID-19 trend with preventive policy effect. we get:

Figure 3: Simulations with preventive policy effect

According to figure 3, once the government adopt the preventive policy to lower the
contact rate between people and people on the 1 July 2020, the dead population will
reduce from 2.2 million derived from figure 2 to 2 million. In other words, policy
declared on the peak time of 1 July 2020 will save around 0.2 million lives. The
simulation highlights the importance of preventive policy as it can save people lives.
Similarly, if the preventive policy is declared on early stage, it will save much more
lives and lower the infectious population curve. We will simulate theses at the
following part.
Now, we compare the simulation results in terms of policy effect declared on different
days as following:
Policy declared day Lowered contact rate
17 | P a g e
𝟏 𝟏 𝟐.𝟐
𝟏 𝑨𝒑𝒓𝒊𝒍 𝟐𝟎𝟐𝟎 𝛃𝐏 = 𝛃=
𝟐 𝟐 𝟕.𝟒𝟖𝟏×𝟏𝟎𝟔 ×𝟕
𝟏 𝟏 𝟐.𝟐
𝟏 𝑶𝒄𝒕𝒃𝒆𝒓 𝟐𝟎𝟐𝟎 𝛃𝐏 = 𝛃=
𝟐 𝟐 𝟕.𝟒𝟖𝟏×𝟏𝟎𝟔× 𝟕
𝟏 𝟏 𝟐.𝟐
𝟏 𝑨𝒑𝒓𝒊𝒍 𝟐𝟎𝟐𝟏 𝛃𝐏 = 𝛃=
𝟐 𝟐 𝟕.𝟒𝟖𝟏×𝟏𝟎𝟔 ×𝟕

18 | P a g e
Figure 4: Comparison of policy declared on different days

Comparing 3 simulation results in figure 4, we can see that not only the preventive
policy will matter, but also the timing to declare the policy will matter. By the 3
simulation results, we can see the most early preventive policy will result in the
greatest number of lives saved and the greatest reduction in the infectious
population curve. The first simulation result show that if the government adopt the
preventive policy at once on 1 April 2020, it is able to lower the infectious population
peak from 1.4 million in figure 2 to 0.4 million while the dead population will drop
from 2.2 million in figure 2 to 1.4 million. In other words, the immediate preventive
policy will prevent the COVID-19 virus from spreading further while simultaneously
saving more lives. Hence, the simulation results highly convince the government to
adopt the preventive policy to halt the COVID-19 outbreak as soon as possible.

• 4.4 Analysis
Figure 2 shows that, starting on April 1, 2020, the COVID-19 outbreak will become
apparent after about 50 days and peak after about 90 days. Given that this is a
deterministic model, there will be no infectious individuals because they will all have
died, recovered from the illness, or otherwise not been impacted by the outbreak.
Figure 3 shows, however, that after 90 days from the first of April 2020, when the
government adopts preventive measures, the number of deaths attributed to the
COVID-19 will decline. Figures 2 and 3 show how the policy effect will save more lives

19 | P a g e
since preventive measures will decrease the rate of interaction between persons. If
the Hong Kong government adopts the preventive policy at once, it would not only
reduce the number of dead people but also the number of infected populations, as
shown in figure 4 which compares the preventive policy stated on different three
stages. To save the lives of Hong Kong residents and stop its future spread, it is
crucial that the Hong Kong government adopt the preventative policy as soon as
possible.
5.Discussion
Actually, we want to regulate the yellow "infectious" curve because, once it hits its
top, local hospitals will become congested. In other words, local hospitals won't be
able to accommodate so many contagious patients at once. Therefore, as it could be
advantageous for hospitals, it is crucial for the Hong Kong government to flatten the
yellow "infectious" curve. The following figures visualize the effect:

20 | P a g e
Figure 5: Comparison of flattening the curve

The graphs demonstrate that the infectious population curve will flatten and that its
peak will take a long time to attain rather than take a short time to attain. As a
result, local hospitals have more time to accommodate the contagious population,
and this impact is known as the slowed effect. We can also refer to the reason for
the slowed effect as a policy effect. The government adopted preventive measures
to minimize the contact rate between people and stop the COVID-19 from spreading
in a short amount of time. The slowed effect would reduce the strain on hospitals
and give researchers enough time to develop a vaccine. Using this model, it is easy to
determine how many people must be immunized in order to end the COVID-19
outbreak. Now, consider the equations (#) and (*) again:
𝛽𝑆(0) 𝛽𝑁 𝛽𝑆(𝑡)
𝑅0 = = and 𝑅𝑡 =
𝛾 𝛾 𝛾

𝑆 𝛽𝑁 𝑆
( ) 𝑅0 = ( ) = 𝑅𝑡
𝑁 𝛾 𝑁
Now, we need 𝑅𝑡 < 1 to stop the spreading. By the above computation, we have:
𝑆 1
<
𝑁 𝑅0
𝑆 1 𝑆 1 1 𝑆
− − < − −
𝑁 𝑅0 𝑁 𝑅0 𝑅0 𝑁
1 𝑆
− <−
𝑅0 𝑁

21 | P a g e
1 𝑆
1− <1−
𝑅0 𝑁
According to the WHO estimation about 𝑅0 on 3 March 2020, it was reported 𝑅0 =
2.2, so we have:
1 6
1− =
2.2 11

Hence, we need around 55% populations to get vaccinated so as to stop the disease
from spreading. In other words, 4.08 million Hong Kong people need to get
vaccinated to stop the disease spreading.
6.Limitations and Improvements
Despite being useful for running simulations and comprehending the COVID-19, the
modified SEIR model is still far from ideal for predicting real scenarios because it
makes the assumptions that the population is evenly mixed, the total population is
constant, and the parameter values remain constant over time. However, it is not
possible to maintain constant population and constant parameter values in the real
world. For instance, while we have listed the death rate as 3.4%, we can see that it
fluctuates with people's ages. Stratifying Hong Kong's people into several age groups
is one method we may improve our model. Instead of using a single contact
parameter, we require a contact matrix termed a WAIFW matrix (short for Who
Acquires Infection From Whom). The WAIFW matrix is more accurate since young
people frequently interact with one another while older individuals do not. The
following figures visualize the WAIFW matrix [8]:

22 | P a g e
Figure 6: WAIFW matrix

The matrix should be like the following:


00-10 10-20 20-30 70+
𝑃1 𝑃2 𝑃3 𝑃𝑛
𝐶11 𝐶12 ⋯ ⋯ 𝐶1𝑛
00-10
𝐶21 𝐶22 ⋯ ⋯ 𝐶2𝑛
10-20
𝐶31 𝐶32 ⋯ ⋯ 𝐶3𝑛
20-30
⋮ ⋮ ⋮ ⋮ ⋮

𝐶 𝑛1 𝐶 𝑛2 ⋯ ⋯ 𝐶𝑛𝑛
70+

Figure 7: Explanation of WAIFW matrix

23 | P a g e
Where the matrix input 𝐶𝑛𝑛 collected by the survey stands for the corresponding
contact rate with respect to different age groups.
To put it simply, we can consider to stratify Hong Kong ‘s people into the young
generation and old generation first.

Figure 8: Explanation of WAIFW matrix in 2 groups

𝑃1 in orange denotes the younger generation, and 𝑃2 in orange, the older


generation. The contact that participants in the first young age group had with
persons in the same age group and with people in the other age group is shown by
the first column entries 𝐶11 and 𝐶21 on the left-hand side. The terms 𝑚11 and 𝑚21
refer to the survey's average contact among all 𝑃1 participants. The contact in the
overall population 𝑁1 in the young age group is shown in the first column entries on
the right-hand side, 𝑚11 𝑁1 and 𝑚21 𝑁1 .
Now, we stratify Hong Kong's people into the young generation (0-40 years old)
labelled as 𝑃1 and old generation (>41 years old) labelled as 𝑃2 by considering the
following table:

24 | P a g e
Table 6 [9]: Social contact pattern in Hong Kong in 2017.

Table 7 [10]: Hong Kong age structure in 2020.

25 | P a g e
Using the table 5 data, we are able to find the corresponding average contact rate
with respect to young and old generation groups.
00-40 >40
𝑃1 𝑃2
8.08 6.63
00-40

5.4 8.25
>40

The off diagonal entries must be equal because the number of contact that people in
age group 1 make with people in age group 2 must be equal to the number of
contact that people in age group 2 make with people in age group 1. for this
inconsistence, we will use algebra to fix it. The population in young age group is
around 3166300 while the population in old age group is 4315500. Now we calculate
the contact in total population.
00-40 >40
𝑃1 𝑃2
8.08 × 3166300 6.63× 4315500
00-40

5.4× 3166300 8.25× 4315500


>40

1
Now, we replace the entries by inputting the function = (𝐴 + 𝐴𝑇 ), A stands for the
2
above-mentioned matrix and 𝐴𝑇 stands for transpose matrix.
We have 𝐴𝑇 : 00-40 >40
𝑃1 𝑃2
8.08 × 3166300 5.4× 3166300
00-40

6.63× 4315500 8.25× 4315500


>40

26 | P a g e
1
Replacing the entries by the function = (𝐴 + 𝐴𝑇 ):
2

00-40 >40
𝑃1 𝑃2
8.08 × 3166300 22854892
00-40

22854892 8.25× 4315500


>40
Then, divide all entries by the population number of its age group.
00-40 >40
𝑃1 𝑃2
8.08×3166300 22854892
00-40 3166300 4315500

22854892 8.25×4315500
>40
3166300 4315500

So, we have the calibrating contact rate in total population.


00-40 >40
𝑃1 𝑃2
8.08 5.30
00-40

7.22 8.25
>40

From the above result, we know the young generation (00-40 years old) in Hong
Kong have 8.08 contact rate with respect to their own generation (00-40 years old)
and 7.22 contact rate with respect to old generation (> 40 years old). Meanwhile, the
old generation (> 40 years old) in Hong Kong have 5.30 contact rate with respect to
young generation and 8.25 contact rate with respect to their own generation (> 40
years old).

27 | P a g e
With different ages group having different contact rate, we are able to modify our
ODE systems into young and old generations as following:
𝛽11 𝑆1 𝐼1 𝛽12 𝑆1 𝐼2
𝑆1 ′ = 𝑝Λ − 𝜇𝑆 − −
𝑁 𝑁
𝛽21 𝑆2 𝐼1 𝛽22 𝑆2 𝐼2
𝑆2 ′ = (1 − 𝑝)Λ − 𝜇𝑆 − −
𝑁 𝑁

𝐸1 ′ = 𝛽11 𝑆1 𝐼1 + 𝛽12 𝑆1 𝐼2 − (𝜇 + 𝜀1 )𝐸1


𝐸2 ′ = 𝛽21 𝑆2 𝐼1 + 𝛽22 𝑆2 𝐼2 − (𝜇 + 𝜀2 )𝐸2
𝐼1 ′ = 𝜀1 𝐸1 − (𝛾1 + 𝜇 + 𝛼1 )𝐼1
𝐼2 ′ = 𝜀2 𝐸2 − (𝛾2 + 𝜇 + 𝛼2 )𝐼2
𝑅1 ′ = 𝛾1 𝐼1 − 𝜇𝑅1
𝑅2 ′ = 𝛾2 𝐼2 − 𝜇𝑅2
𝐹′1 = 𝛼1 𝐼1
𝐹′2 = 𝛼2 𝐼2
Table 8: New Definition of the compartments.
Compartments Brief definition
𝑆1 young population that is susceptible to sickness
𝐸1 young population at risk of exposure (infected but not infectious)
𝐼1 young population that is infected and infectious
𝑅1 young population that is recovered from COVID-19
𝐹1 young population where COVID-19 causes fatalities
𝑆2 old population that is susceptible to sickness
𝐸2 old population at risk of exposure (infected but not infectious)
𝐼2 old population that is infected and infectious
𝑅2 old population that is infected and infectious
𝐹2 old population where COVID-19 causes fatalities

Table 9: New Precise interpretation of the parameters.


Parameters Interpretation
𝚲 Birth rate per person
P Young population percentage in total population
𝜷𝟏𝟏 Effective contact per unit time per capita rate at young and young generations

28 | P a g e
𝜷𝟏𝟐 Effective contact per unit time per capita rate at young and old generations
𝜺𝟏 Rate of exposed class in young generations being infectious
𝜸𝟏 Recovery rate of infectious population in young generations
𝜶𝟏 Average mortality rate caused by viruses in young generations
𝝁 Natural death rate per person
1-P Old population percentage in total population
𝜷𝟐𝟏 Effective contact per unit time per capita rate at old and young generations
𝜷𝟐𝟐 Effective contact per unit time per capita rate at old and old generations
𝜺𝟐 Rate of exposed class in old generations being infectious
𝜸𝟐 Recovery rate of infectious population in old generations
𝜶𝟐 Average mortality rate caused by viruses in old generations

To make our modified SEIR model more practical in real scenarios, it is suggested to
remove the assumption that ‘Human to human contact is the only method of
transmission’ and build a new model considering the transmission between human
and animals. The new mathematical modeling for zoonotic strength of infectivity on
COVID-19 will be discuss later.

Figure 9: Route of transmission from animal to human sources

29 | P a g e
Figure 10: Zoonotic infectivity's cyclic transmission, COVID-19

The SEIR model is still adopted to analyze the animal influences on the COVID-19 and
analyze all incoming and outgoing ai passengers worldwide. In this new model, we
extend the population from SEIRF compartments to SEI𝐿1 𝐿2 𝐿3 𝐿4 𝐷1 𝐷2 𝑅0
compartments.
Table 10: New definition of the compartments.
Compartments Brief definition
S Population that is susceptible to sickness
E Population at risk of exposure (infected but not infectious)
I Population of infected people
𝑳𝟏 Population of international arrivals by air
𝑳𝟐 Population of domestic arrivals by any transportation
𝑳𝟑 Population of outbound international passengers
𝑳𝟒 Population of domestic outbound passengers
𝑫𝟏 Infectious period
𝑫𝟐 Mean latent period
𝑹𝟎 Reproduction number
𝒁𝟎 The zoonotic power of infection
𝑺𝟎 Population of newly births
𝑬𝟎 Population of exposed individuals prior to coronavirus infection
𝑬𝒍 Population of exposed individuals from infection

30 | P a g e
𝑰𝟎 Population of infected individuals prior to coronavirus infection
𝑰𝒍 Population of infected individuals from infection

The SEIR model now has the following mathematical structure [11]:
𝑆 𝑅0 𝐿3 𝐿4
𝑆′ = − ( 𝐼 + 𝑍0 ) + 𝐿1 + 𝐿2 − ( + )𝑆
𝑁 𝐷1 𝑁 𝑁
𝑆 𝑅0 𝐸 𝐿3 𝐿4
𝐸′ = ( 𝐼 + 𝑍0 ) − − ( + )𝐸
𝑁 𝐷1 𝐷2 𝑁 𝑁
𝐸 𝐼 𝐿3 𝐿4
𝐼′ = − ( + )𝐼
𝐷2 𝐷1 𝑁 𝑁
When boundary conditions are present:

7.Conclusion
In this report, the modified SEIR model gives us an insight on epidemic by analyzing
its mathematical properties such as generalized-endemic-equilibrium point, disease-
free-equilibrium point, and stability. With the help of MATLAB, we are able to use
ode45 to analyze the ODE systems and predict the COVID-19 trend with different
contact rate and test the policy effect with the policy declared on different stage.
The simulation results show that the Hong Kong government can save more lives and
reduce the number of infectious populations by implementing preventive policies
early on. In other words, if the 'infectious' curve flattens, it will ease the local
hospitals' capacity and provide scientists more time to create a vaccine.
In the future work, we could integrate regional heterogeneity, age structure, and
time-varying characteristics to enhance the model's accuracy.

31 | P a g e
Refenrences:

[1] Disis, J.E. (2023) Hong Kong Economy lost estimated $27 billion due to Covid
restrictions, Bloomberg.com. Available at:
https://www.bloomberg.com/news/articles/2023-01-04/hong-kong-s-isolation-
estimated-to-cost-economy-27-billion#xj4y7vzkg (Accessed: 05 June 2023).

[2] B. Mohan and N. Vinod, “Covid-19: An insight into SARS-cov2 pandemic


originated at Wuhan City in Hubei Province of China,” Journal of Infectious
Diseases and Epidemiology, vol. 6, no. 4, 2020. doi:10.23937/2474-
3658/1510146

[3] Purkayastha, S., Bhattacharyya, R., Bhaduri, R. et al. A comparison of five


epidemiological models for transmission of SARS-CoV-2 in India. BMC Infect Dis 21,
533 (2021). https://doi.org/10.1186/s12879-021-06077-9

[4] (No date) Year-end population for 2020. Available at:


https://www.info.gov.hk/gia/general/202102/18/P2021021800322.htm
(Accessed: 05 June 2023).

[5] Butcher, J.C. (2016) Numerical methods for ordinary differential equations.
Chichester: West Sussex.

[6] Coronavirus cases: (no date) Worldometer. Available at:


https://www.worldometers.info/coronavirus/ (Accessed: 05 June 2023).

[7] The true death toll of COVID-19: Estimating global excess mortality (2020) World
Health Organization. Available at: https://www.who.int/data/stories/the-true-
death-toll-of-covid-19-estimating-global-excess-mortality (Accessed: 05 June
2023).

[8] Caro JJ, Briggs AH, Siebert U, et al. Modeling good research practices—overview: A
report of the ISPOR-SMDM modeling good researchpractices task force-1. Value
Health 2012;15:796–803
[9] Leung, K., Jit, M., Lau, E.H.Y. et al. Social contact patterns relevant to the spread
of respiratory infectious diseases in Hong Kong. Sci Rep 7, 7974 (2017).
[10] “UNSD — Demographic and Social Statistics,” unstats.un.org, 2020.
https://unstats.un.org/unsd/demographic-social/products/dyb/dyb_2020/

32 | P a g e
[11] Kandasamy R, Bekele S, Belete T. Mathematical Modeling: Zoonotic Strength of
Infectivity on COVID-19. Proc. Natl. Acad. Sci., India, Sect. A Phys. Sci.
2021;91(4):723–32. doi: 10.1007/s40010-021-00765-6. Epub 2021 Dec 26. PMCID:
PMC8710259.

33 | P a g e

You might also like