1 Introduction
-
a detailed account of all available mathematical models used in pandemic modeling and prediction;
-
a comprehensive survey of state-of-the-art applications of available mathematical models in the modeling and prediction of COVID-19 infection transmission;
-
a comparative analysis of different mathematical models on the basis of their usage in COVID-19 infection transmission modeling and prediction;
-
an elaborate discussion of open challenges and required future research to fight against the COVID-19 pandemic using mathematical modeling.
2 A Phase-Specific COVID-19 Case Study
Phase | Model | Ref. | Country/Region/area |
---|---|---|---|
Phase#1 | SIR | [24] | Mainland China |
[25] | India | ||
[26] | India | ||
SEIR | [27] | Wuhan, China | |
[28] | Wuhan, China | ||
[29] | Wuhan, China | ||
[30] | India | ||
[31] | Wuhan, Hubei Province and nearby regions | ||
[32] | Korea | ||
[33] | Korea | ||
[7] | Wuhan, China | ||
SUQC | [34] | Hubei, Wuhan, China | |
M-SDI | [35] | China, (data-Chinese Sina-microblog) | |
BHRP | [36] | Wuhan City, China | |
SEIHR | [37] | Daegu and North Gyeongsang Province, Korea | |
Theta-SEIHRD | [38] | Chinese Mainland, Macao, Hong-Kong and Taiwan | |
SEIPAHRF | [23] | Wuhan, China | |
SQIR | [39] | Pakistan | |
Offspring distribution | [40] | Data used from 46 countries reported by WHO | |
Phase#2 | SIR | [41] | China |
[42] | Comparison(china, Italy) | ||
[23] | Wuhan, China | ||
[43] | Italy | ||
SEIR | [44] | The Republic of Kazakhstan | |
[45] | India | ||
[46] | Saudi Arabia | ||
SEIRD | [47] | London and Wuhan | |
FO (DECS) | [48] | N/A (Simulation) | |
FO (CFFD) | [49] | N/A (Simulation) | |
FO (KTFD) | [50] | N/A (Simulation) | |
ASM | [51] | USA, UAE and Algeria | |
FO (CS) | [52] | N/A (Simulation) | |
Phase#3 | Fractional order | [53] | Wuhan, China |
[54] | Simulation, Wuhan, China | ||
[55] | Saudi Arabia | ||
[56] | USA | ||
[57] | Nigeria | ||
[58] | Pakistan | ||
SIR | [59] | Comparative study (China, South Korea, India, Australia, USA, Italy) | |
[60] | WHO data | ||
[61] | Brazil | ||
[62] | Brazil | ||
[63] | Malaysia | ||
SEIR | [64] | Pakistan | |
[65] | USA | ||
[66] | Morocco | ||
[67] | Simulation, India | ||
[68] | Saudi Arabia | ||
[69] | Egypt & Oman | ||
[70] | Saudi Arabia | ||
[71] | India | ||
[72] | Comparative (China, South Korea, Italy and Iran) | ||
[73] | China | ||
SEIAQRDT | [74] | India | |
SEIHQRD | [75] | Kenya |
3 Overview of Mathematical Models used in Pandemic Modeling
3.1 Model 1: Susceptible-Infectious-Removed (SIR)
3.2 Model 2: Susceptible-Exposed-Infectious-Removed (SEIR)
3.3 Model 3: Multiple-Information Susceptible-Discussing-Immune (M-SDI)
3.4 Model 4: Susceptible-Unquarantined Infected-Quarantine Infected-Confirmed Infected (SUQC)
3.5 Model 5: Bats-Hosts-Reservoir-People (BHRP)
3.6 Model 6: Susceptible-Exposed-Symptomatic infectious-Hospitalized-Removed (SEIHR)
3.7 Model 7: Susceptible-Exposed-Infectious-Hospitalized- Recovered-Dead (Theta-SEIHRD)
3.8 Model 8: Susceptible class-Exposed class-Symptomatic and Infectious class-Super spreaders class-Infectious but Asymptomatic class-Hospitalized class-Recovery class-Fatality class(SEIPAHRF)
3.9 Model 9: Offspring Distribution
4 The State-of-the-Art Application of Mathematical Modeling: A COVID-19 Case Study
4.1 Model 1: Susceptible-Infectious-Removed (SIR)
4.2 Model 2: Susceptible-Exposed-Infectious-Removed (SEIR)
4.3 Model 3: Multiple-Information Susceptible-Discussing-Immune (M-SDI)
4.4 Model 4: Susceptible-Unquarantined Infected-Quarantine Infected-Confirmed Infected (SUQC)
4.5 Model 5: Bats-Hosts-Reservoir-People (BHRP)
4.6 Model 6: Susceptible-Exposed-Symptomatic infectious-Hospitalized-Removed (SEIHR)
4.7 Model 7: Susceptible-Exposed-Infectious-Hospitalized-Recovered-Dead (Theta-SEIHRD)
4.8 Model 8: Susceptible Class-Exposed Class-Symptomatic and Infectious Class-Super Spreaders Class-Infectious but Asymptomatic Class-Hospitalized Class-Recovery Class-Fatality Class (SEIPAHRF)
4.9 Model 9: Offspring Distribution
Basic model | Reference | Main findings | Strengths | Limitations |
---|---|---|---|---|
SIR | Vega 2020 | Provides an overview to enhance awareness of COVID-19 disease trends | Investigated the effectiveness of social distancing considering both social contact and age structuring | Emphasizes quarantines only |
Effect of the quarantine in decreasing infection | ||||
Proposed extended lockdown | ||||
Zhong et al. 2020 | Healthcaresystem could significantly shorten the outbreak period | Good ability to predict by historical datasuch as of the SARS 2003 | Usedshort period data (two weeks) | |
It could reduce one-half of the disease transmission. | It can also give a goodprediction of the limited COVID-19 data | |||
Singh & Adhikari 2020 | Accentuates the importance of both social contact and age structures | Estimates the contact structures | Insufficient data used in the asymptomatic case | |
Social distancing is effective for controlling andmitigating the virus | Large-scale socialdistancing is effective | Themodel is not resolved spatially | ||
SEIR | Kim et al. 2020a | Quantifying the school closure potential effect on the disease | Found schoolopening delay is effective | They did not considercross-population infection rise |
Considered isolation and behavior-changed susceptibleindividuals | The rate of child-to-childtransmission decreases | |||
Lin et al. 2020 | Captured thecourse of COVID-19 outbreaks | Considered: government actions | Considers small number ofconfirmed asymptotically infected transmission cases | |
Computed the reportedratio and future trends | Individual behavioral responses | |||
The method is applicable toother cities or other countries | Emigration oflarge portion of the people | |||
Zoonotictransmission | ||||
Chang et al. 2020 | Estimatedepidemic peak: In Wuhan and Hubei Province in the end of February2020 | To estimate the epidemic trend, theyapplied phase-adjusted and region-adjusted mathematical model | Assumed diseases transmission evenlyacross homogeneous population | |
Other regions in China on February 13, 2020 | Total cases might beunderestimated as the existence of asymptomatic and super-spreadersinfectors | |||
Outbreaks would decrease in March and April inChina | Data lag might exist | |||
Kim et al. 2020b | Investigatedpattern of local transmission dynamics | Predicted the time of end of the corona outbreaks | Mortality rate was not included | |
Found aper-capita infection transmissions rate 8.9 times higher in thelocal area (Daegu/Gyeongbuk) than nationwide (average). | ||||
Modnak & Wang 2019 | The effects of infection latency and humanvaccination | Virus can spread from birdsto humans | Considerbi-linear incidence | |
Human hosts | ||||
Tang et al. 2020a | Reproduction ratequantification for the evolution of interventions | Time-dependent contact and diagnose rates | Highlysensitive & depend upon available period data | |
Prem et al. 2020 | Physical distancing canreduce and delay the peak of the disease | Changes intransmission patterns decreased the number of cases in Wuhan | Individuals’ level heterogeneity is notcaptured in contacts | |
Climatic factor is not included | ||||
Large uncertainties over the estimation of reproductionand infectiousness duration | ||||
Mandal et al. 2020 | Found abasic reproduction rate of 1.5 the best case, and it reduces 62%cumulative incidence | Described rationalinterference to control the outbreaks | Used dataonly of airport entry individuals from China | |
In worst case, basic reproductionrate is 4 | Found potentialimpact of port entry screening | Ignoredtravelers from other countries | ||
A mitigation strategy ofsymptomatic cases | It may affectinfection duration; period of incubation and fatalityrate | |||
Kucharski et al. 2020 | Estimated day-to-day reproduction number | Dynamics of transmission in Wuhan& risk of infections | Simple model | |
Reproductionnumber declined from 2\(\cdot \) 35 (95% CI 1 \(\cdot \)15-4\(\cdot \) 77) to 1\(\cdot \) 05(0\(\cdot \)41-2\(\cdot \)39) within one week | Transmission more homogeneous | |||
Found SARS-likevariations | ||||
Tang et al. 2020b | Calculatedthe effective daily ratio of reproduction | Used current revised data and information to estimatesoutbreaks trend | Needed to update parameters | |
Re-estimated disease transmission risk | ||||
Evaluated theoutbreaks trend | ||||
Estimated disease peak phase | ||||
Yang and Wang 2020 | Foundinfection transmission remain endemic | The reproduction rate was 4.25 | Ecological, pathological andepidemiological aspects not clearly considered | |
Long-term diseaseprevention and intervention programs are needed | Predicted the epidemic peak of the virusinfection | |||
M-SDI | Yin et al. 2020 | Reproductionratio decreased from 1.7769 to approximately 0.97 | Predictedthe multiple-information propagation trend | Used limiteddata for the estimation of parameters |
Public discussion peak was passed | ||||
SUQC | Zhao & Chen 2020 | Predictedtrends of transmission dynamics | Quantifying variables and parameters | Did not consider demographicfactors such as death |
Effects of quarantineor confirmation procedures on the disease | Able to provide guidance for other countries to controlthe outbreaks | |||
BHRP | Chen et al. 2020 | Reproduction estimated fromreservoir to person and it is lower than from person to person | Used many parameters to quantify transmissibility | Used limited data |
Parameterassumptions | ||||
Does not reflect the realresults | ||||
SEIHR | Choi & Ki et al. 2020 | Estimated the size of theoutbreak and the reproduction number | Evaluated theeffects of different preventive measures | Did not consider natural deaths and births |
Latency andasymptomatic infections, and re-infected cases were notconsidered | ||||
SEIHRD | Ivorra et al. 2020 | Calculatedbasic reproduction number | Estimated basic reproduction rate and percentage ofundetected cases | Spatial distributionwithin the territory is omitted |
The effective reproductiondecreases due to the different control measures taken | Between-countrytransmission was not considered | |||
Officially releaseddata was not of high quality due to severaluncertainties. | ||||
SEIPAHRF | Ndairou et al. 2020 | Investigated thesensitivity by considering the variations of its parameters | Considered many parameters to quantify transmissibilityand computed the basic reproduction number | Limited datawere studied |
Offspring distributions | Endo et al. 2020 | Better estimation of moderate uncertaintylevels with limited data resources | Moderate uncertainty levels | Highly over dispersed due toa very small fraction of individuals |
Provided a lowerboundary of the basic reproduction number |
5 Observable Behavior of the Different Models
5.1 Model 1: Susceptible-Infectious-Removed (SIR)
-
SIR is the basic compartmental model used to assess the pandemic size, and transmission peak and to predict the trends of the current COVID-19.
-
Identified effectiveness of lockdowns/quarantines and different social distancing measures through parameter estimation.
-
This model can be used for disease mitigation policies according to numerical illustrations in the context of COVID-19.
5.2 Model 2: Susceptible-Exposed-Infectious-Removed (SEIR)
-
The SEIR model is also applicable to assess the transmission dynamics of contagious disease such as COVID-19.
-
In SEIR the exposed compartment as well as an additionally adopted parameter (the movement from the exposed compartment to the infected compartment) were used to describe the transmission dynamics of exposed individuals in the context of COVID-19.
SIR | SEIR | M-SDI | SUQC | BHRP | |
---|---|---|---|---|---|
SIR
|
\(\begin{array}{l} \hbox {E absent}\\ \hbox { Fewer parameters}\end{array}\)
|
\(\begin{array}{l} \hbox {D and I absent }\\ \hbox { Different parameters}\end{array}\)
|
\(\begin{array}{l} \hbox {U, Q and C are absent}\\ \hbox {Different parameters}\end{array}\)
|
\(\begin{array}{l} \hbox {E, SI and AI are absent}\\ \hbox {Differently parameterize}\end{array}\)
| |
SEIR
|
\(\begin{array}{l}\hbox { E added}\\ \hbox { More parameter appended}\end{array}\)
|
\(\begin{array}{l} \hbox {E, I and R are different} \\ \hbox {Different parameter approximation}\end{array}\)
| E, I and R are different from U, Q and C |
\(\begin{array}{l} \hbox {I is different from SI and AI }\\ \hbox {Simple than RP}\end{array}\)
| |
M-SDI
|
\(\begin{array}{l}\hbox { I and R absent}\\ \hbox {Public opinion based}\end{array}\)
|
\(\begin{array}{l} \hbox {E, I and R absent}\\ \hbox {Parameterized public opinion}\end{array}\)
|
\(\begin{array}{l} \hbox {U, Q and C are absent}\\ \hbox {Multiple-information based}\end{array}\)
|
\(\begin{array}{l} \hbox {Different approach }\\ \hbox {Public opinion data}\end{array}\)
| |
SUQC
|
\(\begin{array}{l} \hbox {U, Q and C are different}\\ \hbox {R absent}\end{array}\)
|
\(\begin{array}{l} \hbox {U different from E }\\ \hbox {R absent}\end{array}\)
| Different approach |
\(\begin{array}{l} \hbox {E and R absent}\\ \hbox {U, Q and C appended}\end{array}\)
| |
BHRP
|
\(\begin{array}{l} \hbox {E appended}\\ \hbox {SI and AI different from I}\end{array}\)
| SI and AI different from I | Different approach |
\(\begin{array}{l} \hbox {E and R appended}\\ \hbox {SI and AI different from U and Q}\end{array}\)
| |
SEIHR
| E and H appended | H appended | Different model |
\(\begin{array}{l} \hbox {E, H, R appended}\\ \hbox { I not same as U, Q and C}\end{array}\)
|
\(\begin{array}{l} \hbox {H appended}\\ \hbox {Isi different from SI and AI}\end{array}\)
|
Theta-SEIHRD
|
\(\begin{array}{l} \hbox {H, R and D appended }\\ \hbox {SI(=I) is different}\end{array}\)
|
\(\begin{array}{l} \hbox {H appended}\\ \hbox {SI(=I) is different}\end{array}\)
| Different model |
\(\begin{array}{l} \hbox {I different}\\ \hbox {H, R and D appended}\end{array}\)
|
\(\begin{array}{l} \hbox {I different }\\ \hbox {H appended}\end{array}\)
|
SEIPAHRF
|
\(\begin{array}{l}\hbox { Isi(=I) different from I }\\ \hbox {E, P, A and H appended}\end{array}\)
|
\(\begin{array}{l} \hbox {Isi(=I) different from I }\\ \hbox {P, A and H appended}\end{array}\)
| Different model | E, I, P, A and H differently appended | P and A appended |
Offspring distributions
| Different model Assumed Stm | Different model Assumed Stm |
\(\begin{array}{l} \hbox {Different model}\\ \hbox {Assumed Stm}\end{array}\)
|
\(\begin{array}{l} \hbox {Different model}\\ \hbox {Assumed Stm}\end{array}\)
|
\(\begin{array}{l} \hbox {Different model}\\ \hbox {Assumed Stm}\end{array}\)
|
SEIHR | Theta-SEIHRD | SEIPAHRF | Offspring distributions | ||
---|---|---|---|---|---|
SIR
|
\(\begin{array}{l} \hbox {E and H absent}\\ \hbox {Fewer parameters}\end{array}\)
|
\(\begin{array}{l} \textit{Theta as fraction and E and H absent}\\ \textit{Fewer parameters}\end{array}\)
|
\(\begin{array}{l} \hbox {E,SI, AI,SS,H and F absent}\\ \hbox { Fewer parameters}\end{array}\)
| Different Approach | |
SEIR
|
\(\begin{array}{l} \hbox {I is different from SI and H }\\ \hbox {Parameters estimations different}\end{array}\)
|
\(\begin{array}{l} \hbox {H is absent}\\ \hbox {Parameter approximation different}\end{array}\)
|
\(\begin{array}{l} \hbox {H absent}\\ \hbox {I in simplest form }\\ \hbox {Parameter representation different}\end{array}\)
| Different Approach | |
M-SDI
| Different approach | Analyzed users propagation | Different Approach | ||
SUQC
|
\(\begin{array}{l} \hbox {E, H and R absent}\\ \hbox {U, Q and C appended}\end{array}\)
|
\(\begin{array}{l} \hbox {E, H, R and D absent }\\ \hbox {U, Q, C appended}\end{array}\)
|
\(\begin{array}{l} \hbox {E, I, P, A, H, R and F are different and absent }\\ \hbox {U, Q and C appended}\end{array}\)
| Different Approach | |
BHRP
|
\(\begin{array}{l} \hbox {SI and AI are different from I}\\ \hbox {H absent}\end{array}\)
|
\(\begin{array}{l} \hbox {SI and AI are different from I} \\ \hbox {H absent}\\ \hbox {Fraction Theta absent}\end{array}\)
|
\(\begin{array}{l} \hbox {SI and AI are different }\\ \hbox {H absent}\end{array}\)
| Different Model | |
SEIHR
| Fraction Theta is absent |
\(\begin{array}{l} \hbox {Isi(=I) is same as SI(=I) }\\ \hbox {P and A absent}\end{array}\)
| Different Model | ||
Theta-SEIHRD
| I different from Isi |
\(\begin{array}{l} \hbox {I different }\\ \hbox { P, A absent}\end{array}\)
| Different Model | ||
SEIPAHRF
| P and A appended |
\(\begin{array}{l} \hbox {Fraction Theta absent }\\ \hbox {P A appended}\end{array}\)
| Different Model | ||
Offspring distributions
|
\(\begin{array}{l} \hbox {Different model}\\ \hbox {Assumed Stm}\end{array}\)
|
\(\begin{array}{l} \hbox {Different model }\\ \hbox {Assumed Stm}\end{array}\)
|
\(\begin{array}{l} \hbox {Different model }\\ \hbox {Assumed Stm}\end{array}\)
|
5.3 Model 3: Multiple-Information Susceptible-Discussing-Immune (M-SDI)
-
This approach is different from the studied models in this research as the model is developed based on information propagation on social media especially on the topic of public discussions in the context of COVID-19.
-
In this study, the authors focused on the characteristics of users choosing to discuss certain topics related to COVID-19.
-
To predict public discussion trends they approximated parameters using the LS method and estimated the public discussion peak by calculating the reproduction ratio.
5.4 Model 4: Susceptible-Unquarantined Infected-Quarantine Infected-Confirmed Infected (SUQC)
-
This model was developed in [34] to describe the COVID-19 transmission dynamics.
-
The interference effects of control measures are especially parameterized by analyzing the disease outbreak.
-
This model was adapted to daily released confirmed case data to analyze the outbreaks in Hubei, Wuhan, and four other first-tier cities in China.
-
They considered multiple characteristics to predict the transmission trends.
5.5 Model 5: Bats-Hosts-Reservoir-People (BHRP)
-
The BHRP model might be an advanced version of both the SIR and SEIR models.
-
The BHRP mathematical computational part involves many differential equations.
-
The most challenging part is the estimation of many parameters.
-
More data are needed to better understand the transmission dynamics in the context of COVID-19.
5.6 Model 6: Susceptible-Exposed-Symptomatic Infectious-Hospitalized-Removed (SEIHR)
-
The SEIHR model estimates the size of the outbreak and the reproduction number, and they evaluate the effects of different preventive measures.
-
Results demonstrated that wearing masks, maintaining social distance and other intervention measures can reduce the fast transmission of the COVID-19 virus.
-
In this study, they did not consider natural deaths and births, infections during latency and asymptomatic infections, and re-infected cases.
5.7 Model 7: Susceptible-Exposed-Infectious-Hospitalized-Recovered-Dead (Theta-SEIHRD)
-
Theta-SEIHRD was developed to assess the dynamics of COVID-19 by considering its major known characteristics.
-
In this method, the authors added ‘Theta’, the fraction of total detected cases (infected) to study the importance of this fraction on the influence of COVID-19.
-
They fitted and estimated many parameters with reported data in the model which could be useful to estimate the spread of COVID-19 in some other countries.
-
The reported data that used was homogeneously distributed and weak in quality (uncertainty in both undetected infections and a number of characteristics of the virus).
5.8 Model 8: Susceptible Class-Exposed Class-Symptomatic and Infectious Class-Super Spreaders Class-Infectious but Asymptomatic Class-Hospitalized Class-Recovery Class-Fatality class (SEIPAHRF)
-
The SEIPAHRF model gives a feasible approximation by studying different important aspects of COVID-19 transmission.
-
The model was illustrated mathematically with nonlinear ODEs, and sensitivity was investigated by considering the variations of its parameters.
-
They also fitted the model with real, daily confirmed case data and considered many parameters to quantify transmissibility.
-
According to the theoretical findings and numerical illustrations, the model was well adapted to the actual data and it reflected the real scenarios in Wuhan, China.
5.9 Model 9: Offspring Distribution
-
Offspring Distribution performs better for uniform and homogeneous data.
-
Computationally, this model is powerful.
-
It might be challenging to include additional parameters and distinguish stages that are quite easily applicable and identifiable in other mentioned models mentioned above.