Predicting COVID-19 Dynamics Using SEIR-PADC Model
Predicting COVID-19 Dynamics Using SEIR-PADC Model
Predicting COVID-19 Dynamics Using SEIR-PADC Model
1
School of Engineering, Australian College of Kuwait, Safat 13015, Kuwait; [email protected];
2
School of Economics and Business, Norwegian University of Life Sciences, 1430 Ås, Norway
3
John von Neumann Faculty of Informatics, Obuda University, 1034 Budapest, Hungary;
[email protected]
Abstract
There are a number of derivates of SIR type models developed in mathematics community with
hospitalized populations; but are mathematically complex and cumbersome. These methods
rarely used actual clinical data in details and usually fitted with one or maximum two major
clinical data. In this paper, we introduce SEIR-PADC model to include exposed, deceased,
super-spreader and critical populations and divide infected population to symptomatic and
best fit coefficients to 5 set of COVID-19 data in Kuwait. Our focus is to track trends of
COVID-19 in coming days. Initial conditions for 8 populations and initial guess values for 12
unknown coefficients are found in a way to best fit COVID-19 data. We used 136 days of
COVID-19 data in Kuwait and obtained solutions to cumulative populations rather than daily
prediction
1
Introduction
In 2012, the Middle Eastern respiratory syndrome coronavirus (MERS-CoV) was reported with
largest impact in Saudi Arabia, United Arab Emirates, and South Korea which caused more
than 35% mortality rate among infected people with fever, cough, and shortness of breath [1,
2].
This raised concerns in the largest gathering event (Hajj) due to super-spreaders population
from the locals infected by MERS-CoV. Alasmawi et al. [3] studied MERS-CoV by an
extended SIR model with 5 set of ordinary differential equations for two main subcategories:
local population and non-local pilgrims. The 5 set of population include susceptible (S),
infected (I), super-spreaders (P), recovered (R), and hospitalized (H). Their study focused on
finding mathematically the re-production number (R0) and they model coefficients merely
Kim et al. [4] studied MERS-CoV which caused the major outbreak in South Korea in 2015.
They used a similar extended SIR model as [3] except with 6 set of ordinary differential
transmission rates using estimation or literature. Only one set of clinical data on cumulative
incidence numbers were used for model validation suggesting that the super spreader
population caused initial large reproduction number. They concluded preventive measures such
as intensive lockdown and quarantine hospitalized cases had major impact on stopping spread
Ndaïrou et al. [5] have recently applied the same extended SIR type model introduced in [3, 4]
for studying outbreak of COVID-19 in Wuhan, China with emphasis in stability of their model
with 8 set of ordinary differential equations including mortality rates. They fitted their model
2
with only two set of clinical data including confirmed daily cases and confirmed daily death
cases in Wuhan, China. They estimated model coefficients with values so that a reproduction
number below the value of one is obtained. The SIR type models are usually designed for fitting
More recently, Xue et al. [6] developed a 7 set of equations SIR type model using a
transmission network to fit COVID-19 clinical data in Wuhan (China), Toronto (Canada), and
the Italy. They used an optimization algorithm (MCMC) to fit coefficients of 7 set of ODE
equations. They only their model with two set of clinical data, confirmed total infected cases
Observing the above studied literature, it is not clear why 5 to 8 set of ordinary differential
equations are used to only fit with one or maximum two clinical data. From recent COVID-19
data, it can be easily seen that at least 5 set of different clinical data including susceptible,
infected, recovery, critical and deceased populations can be used for validation of extended
SIR type models. In this paper, we introduce SEIR-PADC which is SIR type model with 8 set
of ODE equations. The aim of the paper is to computationally study the 5 set of clinical data
Methodology
SEIR-PADC Model
We introduce here SEIR-PADC model consist of susceptible (S), exposed (E), symptomatic
infected (I), recovered (R), super spreaders (P), asymptomatic infected (A), deceased (D), and
critical (C) populations. The model consist of 8 different cumulative populations and
3
transmission from each population starts similar to original SIS and SIR model reported by
The model have 12 unknown transmission coefficients x=[x1 x2 …x12], which all are positive
or negative constants and are obtained by an optimization algorithm to fit clinical data. In
Figure 1, x’1=x1E/N is selected as introduced in [8] and flow in and out of each population is
𝑑𝑆 𝑥1 (1)
= − 𝑆𝐸
𝑑𝑡 𝑁
𝑑𝐸 𝑥1 (2)
= 𝑆𝐸 − 𝑥2 𝐸 − 𝑥3 𝐸 − 𝑥4 𝐸
𝑑𝑡 𝑁
𝑑𝐼 (3)
= 𝑥3 𝐸 − 𝑥5 𝐼 − 𝑥6 𝐼 − 𝑥7 𝐼
𝑑𝑡
𝑑𝑃 (4)
= 𝑥4 𝐸 − 𝑥8 𝑃 − 𝑥9 𝑃 − 𝑥10 𝑃
𝑑𝑡
𝑑𝐴 (5)
= 𝑥2 𝐸
𝑑𝑡
𝑑𝐶 (6)
= 𝑥5 𝐼 + 𝑥9 𝑃 − 𝑥11 𝐶 − 𝑥12 𝐶
𝑑𝑡
𝑑𝑅 (7)
= 𝑥6 𝐼 + 𝑥8 𝑃 + 𝑥11 𝐶
𝑑𝑡
4
𝑑𝐷 (8)
= 𝑥7 𝐼 + 𝑥10 𝑃 + 𝑥12 𝐶
𝑑𝑡
𝑁 = 𝑆 + 𝐸 + 𝐼 + 𝑃 + 𝐴 + 𝐶 + 𝑅 + 𝐷 = 𝑐𝑜𝑛𝑠𝑡𝑎𝑛𝑡 (9)
Kuwait has a population around 4,270,571 based on Worldometer 2020 [9]. A set of initial
conditions are applied for the 8 populations and 12 coefficients to solve the set of 8 ODE
𝑆0 = 1200,000; 𝐸0 = 5; 𝐼0 = 0; 𝑃0 = 5; 𝐴0 = 0; 𝐶0 = 0, 𝑅0 = 0 ; 𝐷0 = 0 (10)
𝑥0 = [1 1 1 0 0 1 0 0 2 0 0 1] (11)
The SEIR-PADC model outlined above are solved using ode45 in MATLAB [10]. The
optimization tool (fminsearch) [11] is used to find best fitted ODE coefficients by applying the
following convergence criteria for 5-set of available data on goodness of fit (GOF):
+ 𝑅𝑀𝑆𝑅(𝐶))
In equation (12), the root mean-square ratio (RMSR) of computed population versus available
The root mean-square ratio (RMSR) is defined here using the coefficient of determination
which is widely used on goodness of fit when predicted variables are compared with actual
5
data particularly for situations that future outcomes are sought. The coefficient of determination
(R2) evaluates the predicted values (y) against actual data (𝑦𝑒𝑥𝑝 ) and is rewritten to provide
2
2 ∑(𝑦−𝑦𝑒𝑥𝑝 )
RMSR = 1 − R = 2 (13)
∑(𝑦𝑒𝑥𝑝 −𝑦
̅̅̅̅̅̅̅)
𝑒𝑥𝑝
In equation (13), 𝑦
̅̅̅̅̅̅
𝑒𝑥𝑝 is the average of COVID-19 clinical data value for certain population.
Better fitted functions provide R2 value close to unity and RMSR close to zero.
Using the optimization algorithm (fminsearch) in MATLAB and initial conditions given in
equations (10-11), the following values are found for the model coefficients:
6
(b) Infected population (f) Susceptible population
Figure 2: Fitting 5 set of COVID-19 data simultaneously with SEIR-PADC model for
Figure 2 shows that 5 set of COVID-19 data including critical, infected, deceased, recovered,
and susceptible populations are simultaneously fitted with SEIR-PADC model. Goodness of
fit value (GOF) is obtained 1.5756 for 5-set of data in equation (12).
Important dates including peak number and day of infectious, peak value and day of people in
critical condition, maximum recovered and deceased populations, and end of pandemic in
Kuwait can be feasibly obtained from the Figure 2 and computed values from SEIR-PADC
model.
Conclusions
7
We have developed SEIR-PADC dynamic model for COVID-19 in Kuwait. The aim is to
computationally fit simultaneously any available set of COVID-19 data for better prediction of
trends of COVID-19 dynamics in Kuwait. From the results of this study, it can be concluded
that:
identified.
• Important peaks of population sizes and dates are predicted successfully indicting that
Kuwait has passed all important peaks of critical hospitalization and peak of infectious
populations.
SEIR-PADC model is a simple and robust model to correctly pick up various well-defined
population in SIR type models. The model has successfully fitted 5-set of COVD-19 data in
Kuwait.
References
2. Centers for Disease Control and Prevention. A Novel Coronavirus Called MERS- CoV
corona virus during the Hajj season using simulation of the existing data. Int J Stat Med
8
4. Kim Y, Lee S, Chu C, Choe S, Hong S, Shin Y. The characteristics of middle eastern
5. Faïçal Ndaïrou, Iván Area, Juan J. Nieto, Delfim F.M. Torres, Mathematical modeling of
COVID-19 transmission dynamics with a case study of Wuhan, Chaos, Solitons & Fractals,
6. Ling Xue, Shuanglin Jing, Joel C. Miller, Wei Sun, Huafeng Li, José Guillermo Estrada-
Franco, James M. Hyman, Huaiping Zhu, A data-driven network model for the emerging
2020.
9. Worldometer, https://www.worldometers.info/world-population/kuwait-population/,
2020.
July 2020.
12. Devore, Jay L. (2011). Probability and Statistics for Engineering and the Sciences (8th ed.).
13. Mahmoudi, M.R., Heydari, M.H., Qasem, S.N., Mosavi, A. and Band, S.S., 2020. Principal
component analysis to study the relations between the spread rates of COVID-19 in high