Since pneumonia caused by coronavirus disease 2019 (COVID-19) broke out in Wuhan, Hubei province, China, tremendous infected cases has risen all over the world attributed to its high transmissibility. We aimed to mathematically forecast the inflection point (IFP) of new cases in South Korea, Italy, and Iran, utilizing the transcendental model from China.
Methods
Data from reports released by the National Health Commission of the People’s Republic of China (Dec 31, 2019 to Mar 5, 2020) and the World Health Organization (Jan 20, 2020 to Mar 5, 2020) were extracted as the training set and the data from Mar 6 to 9 as the validation set. New close contacts, newly confirmed cases, cumulative confirmed cases, non-severe cases, severe cases, critical cases, cured cases, and death were collected and analyzed. We analyzed the data above through the State Transition Matrix model.
Results
The optimistic scenario (non-Hubei model, daily increment rate of − 3.87%), the cautiously optimistic scenario (Hubei model, daily increment rate of − 2.20%), and the relatively pessimistic scenario (adjustment, daily increment rate of − 1.50%) were inferred and modeling from data in China. The IFP of time in South Korea would be Mar 6 to 12, Italy Mar 10 to 24, and Iran Mar 10 to 24. The numbers of cumulative confirmed patients will reach approximately 20 k in South Korea, 209 k in Italy, and 226 k in Iran under fitting scenarios, respectively. However, with the adoption of different diagnosis criteria, the variation of new cases could impose various influences in the predictive model. If that happens, the IFP of increment will be earlier than predicted above.
Conclusion
The end of the pandemic is still inapproachable, and the number of confirmed cases is still escalating. With the augment of data, the world epidemic trend could be further predicted, and it is imperative to consummate the assignment of global medical resources to curb the development of COVID-19.
Hinweise
Zhong Zheng, Ke Wu, Zhixian Yao and Xinyi Zheng contributed equally to this work.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Abkürzungen
NCP
Novel coronavirus pneumonia
COVID-19
Coronavirus disease 2019
CoV
Coronavirus
SARS
Severe acute respiratory syndrome
ACE2
Angiotensin-converting enzyme 2
IFP
Inflection point
HCHP
The Health Commission of Hubei Province
NHC
The National Health Commission of the People’s Republic of China
WHO
The World Health Organization
STM
State transition matrix
MO
Medical observation
Disc
Discharge
NS
Non-severe case
S
Severe case
Cr
Critical case
Cu
Cured case
D
Death
V
Vector
NCC
New close contacts
5DMA
5-day moving average
10DMA
10-day moving average
SEIR
susceptible - exposed - infectious – recovered
EEMD
Empirical mode decomposition
ANN
Artificial neural network
ML
Machine learning
Background
Since the first case of novel coronavirus pneumonia (NCP), caused by coronavirus disease 2019 (COVID-19), occurred in Wuhan, Hubei Province, China, the dreadful epidemic broke out during Dec 2019 to Mar 2020 under the pace of Chinese Spring Festival [1]. With the untiring efforts of the people and the selfless dedication of medical staff, a total of 59,897 cured patients were discharged [2]. By 24:00 on Mar 9, China has accumulated a total of 80,754 confirmed cases (including 4794 severe cases) and 3136 dead cases [3]. However, in January, when the large-scale outbreak in China began, the disease initiated to spread to other parts of the world [4, 5]. Up to Mar 9, a total of 7382 cases were confirmed in South Korea, 7375 cases in Italy, and 6566 cases in Iran [6].
Similar to another coronavirus (CoV) —SARS-CoV— COVID-19 is an RNA virus that contains particular spike proteins conjugating with angiotensin-converting enzyme 2 (ACE2) that widely expressed in different human tissues [7]. However, its doughty transmissibility in the community has a strong correlation to reasons such as long incubation period, mild early symptoms, and the like [8, 9]. Even though some studies have proved that Remdesivir designed for the Ebola virus may have a promising effect on COVID-19 [10], the kernel strategies for the prevention and treatment of NCP are still effective quarantine as in the case of SARS [11]. After the implementation of strict isolation methods, the most significant goal is to predict the arrival of the peak and inflection point (IFP) of new cases of NCP so that administrative departments can modify current strategies.
Anzeige
Based on the previous data, we analyzed the epidemic situation in Hubei Province [12]. After the validation of the model in different datasets, we were able to analyze the world epidemic trends, and predict the arrival of peaks and IFPs of newly confirmed cases and provide references for NCP prevention and control strategies in various countries.
Method
Study population, data collection, and analysis
Data from reports, including medical observation, close contacts, confirmed cases, severe cases, critical cases, cured cases and death data and corresponding information, released by the Health Commission of Hubei Province (HCHP) (Dec 31, 2019 to Feb 8, 2020) were extracted as the training set. Primarily, the arrival of the IFP of new cases and epidemic trends in Hubei were deduced and testified in the validation set, whose data were extracted from HCHP (Feb 9, 2020 to Mar 5, 2020). Subsequently, another training set consisting of the data from the National Health Commission of the People’s Republic of China (NHC) and the World Health Organization (WHO) (Jan 20, 2020 to Mar 5, 2020) were established. Eventually, the data, including cumulative confirmed cases, cumulative cured cases, death data and corresponding information, from NHC and WHO (Mar 6 to 9, 2020) were collected and constructed the validation set. The data period starts from Dec 31, 2019 to Mar 9, 2020. Data is updated on the daily basis. All data were analyzed using Microsoft Excel (Microsoft Office 2016) and R studio (R Foundation for Statistical Computing, Vienna, Austria). The world epidemic situation was performed using the nCov2019 package of R [13]. Histogram was obtained using the ggplot2 packages of R.
State transition matrix model
State transition matrix (STM) modeling is a well-regarded approach widely applied in clinical decision analysis based on computer simulation. For estimating the IFP of newly confirmed cases and the scale of cumulative cases in the globe in subsequent days, we chose the Markov model cohort simulation.
Parameter selection and estimate
In order to estimate the risk metrics (infectivity, severity, lethality) of the NCP, we build a STM model as showing in the figure (Fig. 1).
×
Anzeige
We define the states in this model. Medical Observation (MO) is the close contact of confirmed cases and put into medical observation. In the subsequent days, outcome could be any of the three: confirmed cases, discharged without COVID-19 infection, or stay in MO. Discharge (Disc) is a terminal state for a close contact, until he or she becomes another incident of close contact again. Infected is an intermediate state, where the patient becomes a confirmed infected case. The outcome is binary: severe, or non-severe. And the outcome is revealed immediately. Non-Severe Case (NS) is the patient also has three possible outcomes in the next day: cure, severe case, or stay in non-severe case. Severe Case (S), the patient has three possible outcomes in the next day: critical case, non-severe case, or stay in severe case. Critical Case (Cr), the patient has three possible outcomes in the next day: cured case, severe case, or stay in critical case. Cured Case (Cu) and Death (D) are also the terminal states for the patient. So, at any moment, we can identify the close contact or patient’s state by utilizing a state vector, defined as the following:
Where each element of the vector (V) stands for one state in the same sequentially arranged order as mentioned above. Please note that the comfirmed itself is not an independent state, since the outcome is revealed instantaneously, so we combine confirmed case with Non-Severe, Severe, and Critical cases.
For each person, the state vector can only have one element with value of 1, and the other elements all have value of zero. For example, if a patient is currently in state “Severe Case”, the state vector for him is ‘[0 0 0 1 0 0 0]’. The next day, his state vector could become either ‘[0 0 0 0 1 0 0]’ (Critical Case), ‘[0 0 1 0 0 0 0]’ (Non-Severe Case) or stay the same.
For the sample population, the state vector is defined as the count of people in each state. For example, if there are 100 patients being treated today, out of which 10 are critical, and 90 are severe. The state vector for this sample population is [0 0 0 90 10 0 0]’.
Since the head count of a certain state comes from itself, all other possible transitions into the state (e.g. S has three possible income states, MO, NS, and Cr), minus the outcome states (NS, and Cr).
If we want to predict for N period, the equation becomes the following:
If the population is limited and the transition matrix is stationary, the above formula will be sufficient in predicting all future outcomes. In our case, the population is not fixed, so we need to introduce the additional input into the population: new close contacts (NCC).
Every day, new close contacts are added to the medical observation pool, as people already in the pool will gradually be discharged or confirmed of infection.
Using this STM model, we will be able to predict when the inflection peak time as well as IFP of newly confirmed cases (the maximum open infection cases) in Hubei Province or non-Hubei will occur. Moreover, after verifying this matrix model in China, it could be utilized to evaluate the world epidemic development especially in the major epidemic areas.
Although there is an intermediate state during the above hospitalization: severe cases (the new standard is broken down into mild and normal), critical cases (which can also be divided into general critical and critical), due to the lack of intermediate state transfer probability, we combine the entire hospital period into a in-patient state, for the sake of keeping the model simple. This minimizes the need for only the following five parameters.
Increment of New Close Contacts (NCC), defined as ln (NCC(t)/NCC(t-1));
Discharge Rate from Medical Observation (MO), defined as Discharged(t)/MO(t-1)
Transitional Probability of Medical Observation - > Confirmed cases, defined as Newly confirmed cases (t)/MO(t-1)
Transitional Probability of Treatment - > Death, defined as New Death Incidents(t) / Treatment(t-1)
Transitional Probability of Treatment - > Cured, defined as New Cured Incidents (t) / Treatment(t-1)
In order to estimate the count of open non-severe cases, severe cases, and critical cases, we need three more parameters:
Ratio of Non-Severe Cases
Ratio of Severe Cases
Ratio of Critical Cases
Anzeige
Scenario setup and prediction
After validation of the STM model in Hubei Province, we set up three different scenarios derived from China for matching and fitting the major epidemic areas comprising South Korea, Italy, and Iran, in order to control for model error, including optimistic scenario, cautiously optimistic scenario, and relatively pessimistic scenario (Table 1).
The situation of Hubei Province, China, and the historical prediction model verification of Hubei Province in the beginning of march
According to the data of NHC [14], as of Mar 5, there were 67,592 cumulative confirmed cases, 41,966 cumulative cured cases, 126 newly confirmed cases, 29 new deaths, and 1478 new cured cases, and 19,758 in-patient cases in Hubei Province (Fig. 2a). The number of new close contacts in Hubei Province has gradually decreased, and the cumulative number of close contacts is currently 271,959 (Fig. 2b). The increment of new close contacts has crossed the IFP (Fig. 2c). Based on data from Dec 31, 2019 to Feb 8, 2020 in Hubei Province, we built a prediction model through the STM model, and the cautiously optimistic scenario could consummately predict the arrival of the IFP and several peak dates in Hubei (Table 2), which undoubtedly validate the predictive efficacy of this mathematic model.
Table 2
Training set and Validation Set of the Epidemic Trend in Hubei Province
Actual
Forecast
Key Metrics
S1
S2
S3
Inc of Confirmed Cases
−9.0%
− 10%
−10%
− 5%
−5%
−1%
− 1%
MO Release Rate
16.0%
17.0%
10.50%
17.0%
10.50%
17.0%
10.50%
Peak of Active Cases
50,633
39,612
47,148
44,082
55,150
62,041
85,502
Peak Date
2020/2/16
2020/2/23
2020/2/28
2020/3/1
2020/3/7
2020/4/6
2020/4/14
Peak of Severe Cases
9289
5753
6845
6400
8004
9000
12,402
Peak Date
2020/2/16
2020/2/23
2020/2/28
2020/3/1
2020/3/7
2020/4/6
2020/4/14
Peak of Critical Cases
2492
1786
2124
1986
2484
2793
3849
Peak Case
2020/2/21
2020/2/23
2020/2/28
2020/3/1
2020/3/7
2020/4/6
2020/4/14
Total Cases at Feb. End
66,907
54,189
64,064
60,192
71,596
68,045
81,284
The actual data were extracted from HCHP, and the forecast data in the three scenarios were deduced by the STM model based on the data before Feb 9, 2020. S1: optimistic scenario; S2: cautiously optimistic scenario; S3: relatively pessimistic scenario; MO Medical observation, Inc. Increment
×
Epidemic situation in training set and the epidemic trend fitting model
As of Mar 5, there were 23,784 confirmed cases, 53,726 cumulative cured cases, 3042 cumulative deaths, 80,552 cumulative confirmed cases, and 670,854 cumulative close contacts in China. Through the analysis, the 5-day moving average (5DMA) and 10-day moving average (10DMA) increment of the confirmed case in Hubei and non-Hubei suggested that the IFP in China was from Feb 6 to Feb 13 (Fig. 3a and b).
×
Applying the STM model again to establish a 10DMA increment of confirmed cases model in non-Hubei, the fitting line of the trend in non-Hubei could be obtained, which is
According to the derivatives taken from fitting lines, the epidemic trend in non-Hubei was set as an optimistic scenario with increment of − 3.87%, and the epidemic trend in Hubei as a cautiously optimistic scenario with increment of − 2.20%, and set a relatively pessimistic scenario with increment of − 1.50% (Table 1), which could forecast the situation outside China.
International epidemic situation and prediction
Data from WHO shows that there were 2232 new cases worldwide on Mar 5, the cumulative number of confirmed cases reached 95,324, and a total of 85 countries have suffered this epidemic (Fig. 4a) [6]. Starting from the cumulative 50 confirmed cases (T50), the cumulative confirmed case trends were compared in different countries with China, and it showed that the trends of France, Germany, United Kingdom, the United States, and Spain stayed steady, while the trends of newly confirmed cases in Korea, Italy, and Iran laid between Hubei and non-Hubei, which have been identified as the major epidemic areas in the globe (Fig. 4b and c).
×
Then the established STM model was implemented to the three countries. The results showed that the IFP in South Korea would arrive from Mar 6 to 12, 2020 (Fig. 5a and b); the IFP in Italy would arrive from Mar 10 to 24, 2020 (Fig. 5c and d); the IFP in Iran would come from Mar 10 to 24, 2020 (Fig. 5e and f). After completing the model and training set establishment, we compared the cumulative case prediction with the actual data on Mar 6 and Mar 9, which was validation set, and the results overtly testified the efficacy of this prediction model all in Korean, Italy, and Iran (Table 3). By utilizing this model, the approximate number of confirmed cases in the three countries at the end of March, April, and May could be predicted (details show in Fig. 6), which could instruct the international medical resources allocation.
Table 3
Training Set and Validation Set of the Epidemic Trends in the Major Epidemic Areas
The verification of STM model by the data updating after prediction
With the time going by almost 3 months, we reran the STM model in South Korea, Italy, and Iran. The results showed that the STM model well predicted the trend in South Korea and Italy, however, not in Iran (Fig. 7a to f). In South Korea, the line of the total confirmed cases laid between the optimistic scenario and cautiously optimistic scenario before April, and fitting the cautiously optimistic scenario after that (Fig. 7a), and so did the IFP of the confirmed cases in South Korea (Fig. 7b). In Italy, the line of the total confirmed cases fitting the relatively pessimistic scenario (Fig. 7c), and so did the IFP of Italy (Fig. 7d). Nevertheless, in Iran, the line of the total confirmed cases laid between the optimistic scenario and the relatively pessimistic scenario (Fig. 7e), and the increment of confirmed cases is still relatively high, which means the IFP of Iran has not come yet (Fig. 7f).
×
Discussion
The concept of state transfer matrix was put forward by Russian mathematician Markov. In the early twentieth century Markov found in the early twentieth century that for some factors of a system in the transfer, the result is only affected by the n-1 result, that is, it is only related to the current state, and has nothing to do with the past state. Thus in Markov analysis, the concept of state transition is introduced. The so-called state refers to the state in which objective things may appear or exist; State transition refers to the probability of objective things being transferred from one state to another. In this study, to estimate the IFP of new confirmed cases in the future and the global cumulative case size, we chose markov model cohort simulation. We define the different course states of COVID-19 as state vectors, through which state vectors are used to identify the state of close contacts or patients, and apply line equations to deduce the state transfer equation under N cycles. Using this STM model, we were able to predict the arrival date of IFP for newly confirmed cases in Hubei or non-Hubei provinces, and set three preset scenarios from China to match and fit major endemic areas, including Korea, Italy, and Iran, in order to control model errors.
In this study, the STM model of Korea, Italy and Iran was established in March, and the model data was continuously updated in the background daily. So far, the number of confirmed cases in the world has soared to more than 8 million [15]. As shown in Fig. 7, we successfully predicted the trend of the outbreak in South Korea and Italy, but Iran did not meet expectations. Considering that South Korea and Italy formed good medical experience exchange and medical resources support with Chinese medical experts at the early stage of the outbreak, it is understandable that the model based on Chinese data has a good prediction of the development trend of the epidemic in South Korea and Italy through the absorption of China’s experience and response mode. In the Lancet’s the Healthcare Access and Quality Index for 195 Countries and Territories and Selected Subnational Locations, Italy, South Korea and China all reach a good score, which can also be seen as having comparable national healthcare and epidemic prevention and control capabilities. The successful prediction of South Korea and Italy can prove that the development trend of newly and accumulatively diagnosed patients is consistent with the prediction of this model and changes according to our preset prediction model when medical resources are guaranteed and diagnostic capacity is sound. The model in this study did not predict the trend of the outbreak in Iran as expected. The main reason may be that Iran’s diagnostic capacity is limited by the unreasonable allocation of international medical resources, the shortage of PCR kits and the lack of CT scanners. The presence of people with the virus, those with atypical symptoms, and those with mild and asymptomatic symptoms reveals a huge risk of missed diagnosis and recurrence of outbreaks due to inadequate diagnostic conditions [16‐18]. In general, the model successfully predicted the outbreak trend of major covid-19 developing countries in the first half of 2020, and had a high guidance effect on the allocation of international medical resources during the epidemic.
Given the punchy transmissibility of COVID-19 [19], isolation and quarantine are undoubtedly the primary options [11]. And currently, predicting models of built for epidemic sprouted out a lot. Ziff et al. established a model of death cases and reported that death cases follow three patterns: exponential growth, power-law behavior, and then exponential decline in the daily rate [20]. Nevertheless, deaths are affected by many factors, such as age [21, 22]. More attention should be paid to the number of new cases, and the rate of increment, attributed to the effect of epidemic prevention and control, can be evaluated to guide the date of return to work. Based on the epidemiological data of 186 county-level administrative units in the UK, Davies et al. established a random inter-compartmental model, in which individuals were divided into susceptibility, exposure, infection (preclinical, clinical, or subclinical) and recovery status (removed from the model). The model is stratified by the age of 5 years, and the impact of various basic interventions on R0 is evaluated [23]. Scheiner et al. adjusted the classic epidemiological model, i.e., SEIR model. It’s based on the transmission characteristics of coronavirus, and concluded that the rule of delay from infection to death was more representative of the actual situation than the classical death dynamics rule, so the traditional SEIR model could be more applicable to the prediction of the transmission of COVID-19 epidemic [24]. Hasan et al. proposed a hybrid model of integrated empirical mode decomposition (EEMD) and artificial neural network (ANN) to predict COVID-19 outbreaks, using window period real-time COVID-19 time series data from 22 January 2020 solstice on 18 May 2020. EEMD is used to decompose the time series data, generate sub-signals, denoise the original data, establish neural network structure to train the de-noised data, and obtain a prediction model superior to the traditional statistical analysis [25]. Tuli et al. use machine learning (ML) and cloud computing to track disease and predict epidemic growth, and deploy an improved model based on MLS on cloud computing platforms to more accurately predict epidemic growth behavior in real time [26].
Moreover, we must strictly follow the coping strategy and learn the Chinese model for dealing with NCP outbreaks. Li et al. developed a simple regression model, and based on this model, they estimated that about 34 founder patients outside of China were not observed in the early stage of transmission, and the global trend approximated an exponential increase, tenfold increase in 19 days [27]. This study reproduced the initial spreading mode to the world, yet made no prediction for the future trend, and exponential growth will be curbed immediately after the attention of local governments, and the IFP will come. Milan Batista proposed an estimate of the final size of the COVID-19 epidemic, the logistic growth model and classic susceptible-infected-recovered dynamic model are used to estimate the final size of the coronavirus epidemic, being approximately 83,700 (±1300) cases and that the peak of the epidemic was on Feb 92,020 [28]. However, as of Mar 5, the number of global cases has reached 95,333, and the IFP for growth in South Korea, Italy, and Iran has not yet arrived, which means the global size will be even more colossal.
Our model is based on the fitting of real data from standard authorities. Through the STM Model, based on data from Hubei and non-Hubei, we predict the IFPs in Korea, Italy, and Iran, while there are still some limitations. Due to the large outbreaks started at different times lines all over the world, the effects of seasonal and geographical factors have not been taken into account. Although the fitting with the Chinese model can better predict the situation around the world, through reference and learning, the response strategies of other countries may be more mature. As China resumes work, the production capacity of various medical resources will gear up rapidly, which will impose a positive impact on the world, and it could be more optimistic that the IFP will come soon.
Local governments, regardless of the speed of outbreaks, should learn from China’s primary response strategy, such as stopping working, reducing gathering, preventing contact transmission, wearing masks, and implementing quarantine. After the NCP being under control, the production and output of medical resources should be intensified, the production of coronavirus detection kits should be accelerated, existing cases should be summarized. More accurate diagnostic criteria should be compiled to prevent massive missed diagnoses in countries lacking the kit. Even if it currently causes some global economic regression, the recovery will swiftly come after holding the throat of NCP and COVID-19.
Conclusion
Based on data from China, we utilized the State Transition Matrix Model to predict the IFP of disease in countries currently experiencing outbreaks worldwide. If properly controlled, the IFP in South Korea and Italy will come in early March, and the IFP in Iran will come in mid-March. And through almost 3 months, our model fitted well in South Korea and Italy, however, not Iran, partly because of the irrational international medical resource allocation. During this period, countries around the world should work together to fight the epidemic.
Acknowledgements
We acknowledge the support from Youth Science and Technology Innovation Studio of Shanghai Jiao Tong University School of Medicine.
Ethics approval and consent to participate
There is not patient or animal who participated in this study, and informed consent was not applicable. This study was approved by the Ethical Committee of Shanghai Jiao Tong University according to the Chinese Ethical Regulations.
Consent for publication
Not applicable.
Competing interests
The authors declare that they have no competing interests.
Open AccessThis article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Bei Reizdarmsyndrom scheinen Diäten, wie etwa die FODMAP-arme oder die kohlenhydratreduzierte Ernährung, effektiver als eine medikamentöse Therapie zu sein. Das hat eine Studie aus Schweden ergeben, die die drei Therapieoptionen im direkten Vergleich analysierte.
Ob bei einer Notfalloperation nach Schenkelhalsfraktur eine Hemiarthroplastik oder eine totale Endoprothese (TEP) eingebaut wird, sollte nicht allein vom Alter der Patientinnen und Patienten abhängen. Auch über 90-Jährige können von der TEP profitieren.
Wenn unter einer medikamentösen Hochdrucktherapie der diastolische Blutdruck in den Keller geht, steigt das Risiko für schwere kardiovaskuläre Ereignisse: Darauf deutet eine Sekundäranalyse der SPRINT-Studie hin.
Insektenstiche sind bei Erwachsenen die häufigsten Auslöser einer Anaphylaxie. Einen wirksamen Schutz vor schweren anaphylaktischen Reaktionen bietet die allergenspezifische Immuntherapie. Jedoch kommt sie noch viel zu selten zum Einsatz.
Update Innere Medizin
Bestellen Sie unseren Fach-Newsletter und bleiben Sie gut informiert.