 Research
 Open access
 Published:
A novel method to detect the early warning signal of COVID19 transmission
BMC Infectious Diseases volume 22, Article number: 626 (2022)
Abstract
Background
Infectious illness outbreaks, particularly the coronavirus disease 2019 (COVID19) pandemics in recent years, have wreaked havoc on human society, and the growing number of infected patients has put a strain on medical facilities. It’s necessary to forecast early warning signals of potential outbreaks of COVID19, which would facilitate the health ministry to take some suitable control measures timely to prevent or slow the spread of COVID19. However, since the intricacy of COVID19 transmission, which connects biological and social systems, it is a difficult task to predict outbreaks of COVID19 epidemics timely.
Results
In this work, we developed a new modelfree approach, called, the landscape network entropy based on AutoReservoir Neural Network (ARNNLNE), for quantitative analysis of COVID19 propagation, by mining dynamic information from regional networks and shortterm highdimensional timeseries data. Through this approach, we successfully identified the early warning signals in six nations or areas based on historical data of COVID19 infections.
Conclusion
Based on the newly published data on new COVID19 disease, the ARNNLNE method can give early warning signals for the outbreak of COVID19. It’s worth noting that ARNNLNE only relies on small samples data. Thus, it has great application potential for monitoring outbreaks of infectious diseases.
Background
Infectious illness outbreaks, particularly the recent pandemic of the coronavirus disease 2019, which continue to affect humanity, have posed enormous challenges to socioeconomic progress. This dangerous infectious disease [1], whose mortality rate is very high, has brought a serious threat to human health. Various clinical trials and investigations [2,3,4] have shown that the COVID19 may cause severe damage to the kidneys, liver, heart, and almost all organ systems in humans. Even after recovery, it can bring serious sequelae, including longterm negative effects on the nervous system, mental health, and the human body metabolism. Additionally, the global outbreak of COVID19, resulting in absenteeism, indirectly caused incalculable economic losses, completely disrupted the world’s social and economic order [5]. Although some COVID19 vaccines have been developed, humans still have to combat COVID19 owning to the mutation of this dangerous virus.
There are numerous researches [6] demonstrate that preoutbreak measures, such as social isolation and vaccine development, can contain the outbreak of infectious diseases. However, the cost of developing new infectious disease surveillance systems may be prohibitive for most developing countries [7, 8]. Lack of effective surveillance or adequate response could enable the emergence of new epidemic or pandemic patterns [9] from an endemic infection of SARSCoV2. From a public health and economic perspective, if an early warning signal can be given before an outbreak, the health ministry can take measures in advance to block or slow the spread of infectious diseases to prevent a new coronavirus disease epidemic or at least reduce the scale of an epidemic outbreak. Consequently, numerous machine learning methods [10] have been used to predict the trend of the epidemic, and various statistical models [11] are also utilized to analyze the spread of COVID19. Nevertheless, predicting the outbreak of infectious diseases in realtime is still a challenge since coronavirus diseases are affected by many factors in the biological system and social system. Alternatively, the traditional machine learning method is difficult to deal with shorttime highdimensional data, and the deep learning method also needs lots of data. All these methods are easy to encounter the problem of overfitting. Hence, it is of great significance to develop a novel approach for early warning of the outbreak of COVID19.
To develop early warning methods, we can make simple assumptions that the transmission of an epidemic can be divided into three stages [12,13,14]: the normal stage, the preoutbreak stage, and the outbreak stage, as shown in Fig. 1. The spread of Coronavirus Disease 2019 can be regarded as the dynamic behavior of the dynamical system, the critical transition of COVID19 corresponds to the bifurcation of this dynamical system [12]. The principle for detecting the critical transition in this paper is based on the theory of dynamic network marker or biomarker [15] (DNM or DNB), by mining dynamic information from highdimensional historical data. The DNM theory is a generalized method for identifying the critical transition before a catastrophic event. This method has been applied to many biological processes with remarkable results, including identifying the critical points of cellular differentiation [16], detecting the critical periods of various biological processes [17], and predicting the tipping points of infectious disease outbreaks [14]. Since information entropy [18] is a method to measure the uncertainty of the system, it can be utilized combined with DNM theory to derive a quantitative index, for measuring or detecting the state of the transmission process of the COVID19 disease.
Recently, a short time series forecasting method [19] proposed by Chen et al., namely, AutoReservoir Neural Network (ARNN), to achieve accurate predicting future multistep information. Based on the theory of DNM and ARNN, we recently proposed a new scientific method, called the network landscape entropy based on AutoReservoir Neural Network (ARNNLNE). The algorithm can be described as follows. Firstly, a regional network [12] can be constructed to correlate the confirmed data of daily new cases in each region, where the daily new case data can be simply combined into highdimensional shortterm data. Secondly, the future information of these data can be predicted by the ARNN method [19]. Finally, a network entropy method [20] combined with the future information is used to obtain the critical early warning signal. The specific content of the ARNNLNE method is described in “Methods”. Unlike the existing methods, this method can determine the COVID19 contagion’s preoutbreak stage, in which there is no obvious abnormality but a high risk of turning into an irreversible outbreak stage.
Methods
Autoreservoir neural network
Many conventional forecasting algorithms have been used to predictability [9, 10] including autoregressive and autoregressive integrated moving average (ARIMA) and support vector regression (SVR). However, these approaches require sufficient training samples or data, such as high dimensional shortterm time series or longterm time series, so it’s extremely difficult to predict the future evolution reliably only by using shortterm timeseries data. Theoretically, some neural network techniques including recurrent neural networks [21] (RNN) and longterm and shortterm memory networks [22] (LSTM) could learn nonlinear dynamics from training data. But when there are a few samples are available for training networks, these algorithms encounter overfitting challenges frequently. Moreover, training neural networks might take a long time and require lots of computing resources.
To address these problems, a forecasting method called AutoReservoir Neural Network [19] (ARNN) was proposed. This network framework, as illustrated in Fig. 2a, translates the observed highdimensional dynamic information into the reservoir and maps the highdimensional spatial data to the target variable’s future time information. Specifically, assuming that there is an Hdimensional vector \(I^{t} = (i_{1}^{t} ,i_{2}^{t} ,...,i_{H}^{t} )^{\prime}\) for each of \(t = 1,2,...,m.\) A onedimensional delayed vector \(O^{t} = (o^{t} ,o^{t + 1} ,...,o^{m + L  1} )^{\prime}\) matching to \(I^{t}\) can be generated by the delayembedding theory [23]. By combining reservoir computing (RC) [24] and the spatial–temporal information transformation (STI) [25, 26], an ARNN framework can be obtained, as shown in formula (1).
where \(MN = I\), M is an \(L \times H\) matrix and N is an \(H \times L\) matrix and I represents an \(L \times L\) identity matrix. The nonlinear function F in Eq. (1) can be provided by a multilayer Feedforward neural network, which takes a hyperbolic tangent function \(y = \tanh (x)\) as the activation function. The weights of the neural network F are random values that obey the Gauss distribution, so it’s not necessary to train the neural network. Through the ordinary least square method, we can solve the conjugate Eq. (1) iteratively, and obtain the future information of the target variable \((o^{t} ,o^{t + 1} ,...,o^{m + L  1} )\) as well as the unknown weight matrices M and N. The prediction target variable o can be any of the highdimensional observation variables, such as \(o^{t} = i_{k}^{t} ,\;k = 1,2,...,H\). Moreover, L is the prediction step size, H is the number of observed variables, and m is the length of the observed data.
The daily new cases of COVID19 for each region can be regarded as onedimensional data and then the original data of multiple regions can constitute highdimensional data, which contains important information about the dynamic system. Naturally, we could predict any region’s daily new cases by the ARNN method and the inputting data of ARNN is the highdimensional data mentioned above.
Dynamic network marker
The idea of dynamic network marker [12] (DNM) or dynamic network biomarker (DNB) is an elaboration of the critical slowing theory [27] of highdimensional systems. We can employ the discrete dynamic system to express the dynamic development process of the regional network, provided that the spread of an infectious disease is a complex dynamic process of a nonlinear system. When a complex system approaches a critical point or tipping point, the DNM theory states that there exists a dominant group, i.e., the DNM Group, which fulfills three basic properties:

i.
Within the DNM group, the Pearson correlation coefficient (PCC) between each pair of members rises significantly.

ii.
The Pearson correlation coefficient (PCC) between the DNM Group member and the nonDNM Group member drops rapidly.

iii.
For each member of the DNM group, the standard deviation (SD) increases dramatically.
The emergence of the DNM group with strong fluctuation and high correlation signifies the arrival of the critical transition, according to the properties given above. As a result, these traits can be utilized as three criteria to characterize a complex biological system’s critical state.
The algorithm of ARNNLNE
Based on ARNN and DNM methods, we propose a novel critical warning method for infectious diseases, namely, the landscape network entropy based on the autoreservoir neural network (ARNNLNE). The calculation process of this method is mainly divided into the following four steps, as shown in Fig. 2b.

[Step 1]: Constructing a regional network structure
In a country or region, the geographical location information is modeled to a network, where each node represents a region. There is an edge between two adjacent areas in this network, indicating their adjacency relationship. Taking Germany as an example, based on the geographic locations and traffic routes of these 16 provinces, a regional network can be constructed as shown in Fig. 2b, which has 16 nodes and 27 edges. This network can also be partitioned into numerous local networks, which are composed of central nodes with their firstorder neighbors. Therefore, a local network \(N^{k}\) has E + 1 members, that is, a central node k with its firstorder neighbor nodes \(k_{j} \;\;(j = 1,2,...,E)\).

[Step 2]: Predicting the daily new cases time series of COVID19 by ARNN
The daily new data for each region can be regarded as onedimensional data and then the original data of multiple regions can constitute highdimensional data. For each time point \(T = t\), choosing the appropriate training length m and prediction length L, we could use the highdimensional data \(I^{t}\) as the input and the future predictions \(O^{t}\) can be obtained by solving the ARNNSTI equation iteratively, as depicted in Fig. 2b.

[Step 3]: Calculate the ARNNLNE index
For any local network \(N^{k}\) with E + 1 members, its network entropy index \(H_{k}\) at the time point \(T = t\) can be calculated according to formulas (2), (3).
where \(C^{k} (t) = \left( {c^{k} (t  L + 1),c^{k} (t  L + 2), \cdots ,c^{k} (t)} \right)\) represents the sequence of daily new cases in the local network or region \(N^{k}\) at the time point \(T = t\), \(c^{k} (t)\) denotes the new confirmed cases of COVID19 at \(T = t\) and L1 is the predicting length. While \(C_{l}^{k} (t) = \left( {c^{k} (t  L + 2),...,c^{k} (t),\overline{c}^{k} (t + l)} \right)\) stands for the predicted sequence of daily new cases at \(T = t\). Calculated by the ARNN method in [Step 2], \(\overline{c}^{k} (t + l)\) is the predicted daily new cases in the region \(N^{k}\) at \(T = t + l\). In addition,\(C_{j}^{k} (t)\),\(C_{j,l}^{k} (t)\) in formulas (3) are the sequence and the predicted sequence of daily new cases in the firstorder neighbor node \(k_{j} \;\;(j = 1,2,...,E)\) of the central node k at \(T = t\), respectively. According to the local network entropy \(H_{k} (t)\), the average network entropy of the whole region \(H_{t} = \sum\limits_{k = 1}^{K} {H_{k} }\) can be calculated. Additionally, the number of local network members considered here must be at least more than 2, that is, the number of neighbors of the central node k is greater than 1. If a center node has no neighbor node or only has one neighbor, we let \(p_{j,l} = 1\) to guarantee the normal calculation of formulas (2), (3).

[Step 4]: Identify the preoutbreak stage
The landscape network entropy \(H_{t}\) can quantitatively detect the warning signal of critical transition from the normal stage to the outbreak stage. Through the hypothesis ttest, we can convert \(H_{t}\) into the reciprocal of the corresponding pvalue, which is called the ARNNLNE index in this paper. When \(p < 0.05\), we can see \(H_{t}\) to be significantly different from the mean value of the vector \((H_{1} ,H_{2} ,...,H_{t  1} )\), the time point \(T = t\) can be regarded as the tipping point of the epidemic. Hence, the threshold for the ARNNLNE index is set at 20, corresponding to the significance level \(p = 0.05\). If the ARNNLNE indicator is lower than the threshold, the state of the infectious disease is considered to be in the normal stage at the time point \(T = t\), and then the new calculation will continue at the next time point \(T = t + 1\). When the ARNNLNE indicator exceeds the threshold, it can be regarded as a formal early warning signal.
From the perspective of a complex system, the dynamic process of the spread of COVID19 can be described by the evolution process of a nonlinear dynamic system with bifurcation points, where the system undergoes drastic changes. The ARNNLNE method is designed to detect the preoutbreak stage before the catastrophic transition to the outbreak stage and is applied to six countries or regions. Specific experimental results in “Results” for analysis.
Data processing and the parameter in ARNNLNE
In this paper, the algorithm is applied to COVID19 epidemic datasets [33, 34] in six countries or regions, including Germany, Italy, Netherlands, Spain, parts of Europe, and Canada. Considering that data collection may generate noise, we perform moving average processing on the acquired original data to reduce the impact of noise. The moving average lengths applied to each dataset are shown in Table 1. In addition, if the raw data of COVID19 are less than or equal to 0, it would be replaced by the average data of the previous 3 days.
As shown in Fig. 2a, the ARNN framework directly converts the observed highdimensional dynamic information \(I^{t} = (i_{1}^{t} ,i_{2}^{t} ,...,i_{D}^{t} )^{\prime},\quad t = 1,2,...,m\) into the reservoir, and maps the highdimensional spatial data to a onedimensional delay time vector \(O^{t} = (o^{t} ,o^{t + 1} ,...,o^{t + L  1} )^{\prime}\), where m is the training length and L1 is the predicting length. The parameters including m and L applied to the six datasets are shown in Table 1.
Results
As a modelfree method of nonlinear event prediction, the ARNNLNE method has been applied to the datasets of COVID19 confirmed cases from six nations or regions, including Germany, Canada, Italy, Netherlands, Spain, and parts of Europe. The time ranges of datasets are exhibited in the Table 2 and the ARNNLNE method’s parameters for each dataset are listed in Table 1.
The application of ARNNLNE in several countries
In this work, we collected the daily new cases [33] of the COVID19 epidemic of 16 German provinces from June 28, 2021 to September 29, 2021. As depicted in Fig. 3a, we can construct a regional network with 16 nodes and 27 edges based on the adjacency information for German geographical location. The detailed information of each node can be seen from Fig. 3c. A yellow warning signal given on July 3, 2021, as can be seen from Fig. 3b, indicates that the COVID19 epidemic would enter the outbreak stage. As the highly transmissible variant (Delta) of SARSCoV2 swept all over the world, the number of confirmed cases began to increase in early July. Subsequently, the German disease control agency RKI assessed [28] that Germany entered the outbreak stage on August 20, 2021. Obviously, the date of the early warning signal provided by ARNNLNE is earlier than RKI's warning signal.
For Canada’s 10 provinces, we gathered daily data [34] of the COVID19 epidemic from June 16, 2021 to October 17, 2021. In that Canada has a very large area and interprovincial transportation is mainly aviation flight, we constructed a fullyconnected regional network with 10 nodes, as illustrated in Fig. 4a. In this way, any two nodes are connected by an edge, and the corresponding region of each node is listed in Fig. 4c. At the end of 2021, the number of new cases per day began to increase rapidly as the government eliminated the remaining public health measures. An early warning signal, as presented in Fig. 4b, was provided by ARNNLNE method on July 22, 2021, indicating that an outbreak of the COVID19 is imminent. Canada’s chief public health officer, Theresa Tam, issued a warning [29] at a press conference on August 12, 2021, declaring that an epidemic was emerging in Canada and cases were developed along a strong recovery trajectory. Evidently, the time point of the ARNNLNE warning signal is earlier than the warning date issued by the government.
In addition, the ARNNLNE method is also applied to the historical datasets [33] of Italy, Netherlands, and Spain. For Italy, an early warning signal occurred on June 13, 2021, as portrayed in Fig. 5a. The epidemic in Italy entered the outbreak stage in early July, with a rapid increase in new confirmed cases per day. For Netherlands and Spain, the signals provided by ARNNLNE all emerged before the dramatic rise of the new case series, as demonstrated in Fig. 5b, c. These signals are also supported by governmentissued emergency events. See the Additional file 1: Figures for details.
The application of ARNNLNE in parts of Europe
Not limited to the analysis of COVID19 in a single nation, we also acquired historical data [33] on daily new COVID19 cases from July 20, 2020 to October 12, 2021 in 35 European countries. As shown in Fig. 6a, a regional network can be constructed based on geographical location. This network has 35 nodes, which represent one country, and 69 edges. See the Additional file 1: Fig. S4 for details. An early warning signal, as can be seen from Fig. 6b, was received by the ARNNLNE method on September 1, 2020, indicating that the COVID19 epidemic in Europe will enter the outbreak stage. Between July and October, the number of COVID19 cases increased at an exponential rate, peaking in the first half of November. Europe became the epicenter of the pandemic at the end of 2020, despite the deployment of the COVID19 vaccine in numerous countries. As a result, every country in Europe has to take some tougher measures to prevent the spread of COVID19. On March 9, 2021, ARNNLNE gave the early warning signal, which indicates that the epidemic started to enter the outbreak stage again, as depicted in Fig. 6b. Due to the mutation of the new COVID19 and the stagnation of vaccination programs, the number of new confirmed cases each day has increased rapidly, reaching a local peak again in late April, and European countries such as Germany, Italy have also entered a new round of blockade. On May 29, 2021, an early warning signal was provided by ARNNLNE, although the real confirmed cases were still at a low level. However, the number of daily new confirmed cases of COVID19 throughout Europe had risen dramatically and the outbreak of the COVID19 epidemic had reappeared in the last two weeks. In general, the ARNNLNE method can provide early warning signals before infectious disease outbreaks.
Discussions
Although the vaccine for the COVID19 epidemic has been developed and is now available in all countries, the situation of this epidemic is unlikely to contain rapidly. To reduce the risk of COVID19 infection in humans and alleviate the shortage of medical supplies, we need to present scientific approaches for the relevant medical departments to execute appropriate control measures promptly. The ARNNLNE method, which has been proposed in this paper, is a novel way for early warning of infectious disease outbreaks. This method utilizes shorttimeseries samples to obtain early warning signals and has great potential for realtime surveillance of emerging COVID19 infectious illnesses, as evidenced by its successful implementation in six countries or regions.
In addition, ARNNLNE is a modelfree scientific calculation method, which is not directly related to the mechanism of infectious disease transmission. But the change of ARNNLNE’s warning signal should correspond to the change of the basic regeneration number R_{0} [30], which is an indicator of describing the likelihood of infectious organisms spreading in a population not previously immunized. Theoretically, \(R_{0} = 1\) corresponds to the bifurcation point of the nonlinear dynamical system [31] of COVID19. Table 3 lists certain countries’ early warning signals of COVID19, as well as their R_{0} information [32].When the ARNNLNE index exceed 20, it can be an early warning signal. Evidently, R_{0} was near to 1 at the time point when the ARNNLNE index is provided, indicating that the proposed ARNNLNE method can provide an early warning signal of a disease outbreak before a critical transition from a normal state to an outbreak state of infectious disease.
To test the accuracy of ARNNLNE, we compared our method with the traditional model in two ways. On the one hand, we compared ARNN with the traditional machine learning method SVR in predicting the daily new COVID19 cases based on the six datasets mentioned in the paper. The results of the comparison are shown in Fig. 7 and Table 4, which indicate that ARNN outperforms SVR in prediction.
One the other hand, regarding the early warning of the COVID19 epidemic as a binary classification problem, that is to distinguish whether the current state of the dynamic system is in a critical state or a stable state, we compared the proposed ARNNLNE method with the traditional machine learning model, support vector machine (SVM). As can be seen clearly in Fig. 8, the performance of the ARNNLNE method is better than the SVMbased system. It’s easy to calculate that the AUC of ARNNLNE is 0.825 and the AUC of SVM is 0.77.
Compared with traditional machine learning algorithms, the ARNNLNE method has the following advantages. Firstly, ARNNLNE is a modelfree approach that does not require training or testing procedures and feature selection during computation. Further, because the method is datadriven and has no direct relationship with the mechanism of epidemic spread, it can also be applied to other infectious diseases besides COVID19 such as handfootandmouth disease. Secondly, ARNNLNE can rely on small samples rather than longterm timeseries data. Therefore, it is suitable for application in some developing countries which lack public health infrastructure. Thirdly, ARNNLNE is performed according to the predictive information. Thus, it would give warning signals earlier than conventional methods.
Our proposed method is a datadriven approach without modeling the dynamics of the transmission of infectious diseases. In fact, numerous studies [35,36,37,38,39,40,41,42] have shown that proactive measures taken by governments to deal with the outbreaks are beneficial to control the spread of infectious diseases. Timely measures quickly taken by the government before the outbreak could lead to changes in early warning signals, which we did not take into account. In future work, we will try to analyze the changes in early warning signals under the active control of the government, and attempt to make further improvements to our algorithm.
Conclusions
In this paper, we proposed a modelfree early warning of the epidemic method, i.e., ARNNLNE. Based on the published data of the daily new COVID19 cases, this approach can provide early warning signals for the outbreak of COVID19. Specifically, ARNNLNE can utilize the prediction information of time series to make an early warning, which performs better than some traditional machine learning models. To verify the effectiveness of the ARNNLNE algorithm, we selected six nations or regions for critical transition warnings. The results of these numerical experiments prove that the proposed algorithm is valid and flexible. It’s worth noting that ARNNLNE only relies on small sample data, rather than longterm data. Therefore, it has great application potential for monitoring outbreaks of infectious diseases. In the future, the COVID19 epidemic will still bring serious harm to human society. Thus, it is very crucial to detect realtime changes and send out accurate warning signals of the COVID19 outbreaks. We hope that our work can provide a reference for health institutions.
Availability of data and materials
The historical raw datasets on COVID19 in Germany, Italy, Netherlands, Spain, and parts of Europe are available in the [JHU CSSE COVID19 Data] repository, [https://github.com/cssegisanddata/COVID19]. The historical original data in Canada is available from the Dalla Lana School of Public Health, University of Toronto, [CovidTimelineCanada] repository, [https://artbd.shinyapps.io/covid19canada/].
Abbreviations
 COVID19:

Coronavirus Disease 2019
 LNE:

Landscape network entropy
 ARNN:

Autoreservoir neural network
 ARNNLNE:

The landscape network entropy based on AutoReservoir Neural Network
 DNB/DNM:

Dynamic network biomarker/maker
 ARIMA:

Autoregressive and autoregressive integrated moving average
 SVR:

Support vector regression
 SVM:

Support vector machine
 LSTM:

Long shortterm memory
 RNN:

Recurrent neural networks
 AUC:

Area under the curve
 EWS:

Early warning signals
References
Tang B, Wang X, Li Q, et al. Estimation of the transmission risk of the 2019nCoV and its implication for public health interventions. J Clin Med. 2020;9:462.
Gupta A, et al. Extrapulmonary manifestations of COVID19. Nat Med. 2020;26(7):1017–32.
AlAly Z, Xie Y, Bowe B. Highdimensional characterization of postacute sequelae of COVID19. Nature. 2021;594(7862):259–64.
Huang C, et al. 6month consequences of COVID19 in patients discharged from hospital: a cohort study. Lancet. 2021;397(10270):220–32.
Verma P, et al. A statistical analysis of impact of COVID19 on the global economy and stock index returns. SN Comput Sci. 2021;2(1):1–13.
Hellewell J, et al. Feasibility of controlling COVID19 outbreaks by isolation of cases and contacts. Lancet Global Health. 2020;8(4):e488–96.
Huang J, Zhang L, Liu X, et al. Global prediction system for COVID19 pandemic. Sci bull. 2020;65:1884–7.
Huang J, et al. Global prediction system for COVID19 pandemic. Sci Bull. 2020;65(22):1884.
Telenti A, et al. After the pandemic: perspectives on the future trajectory of COVID19. Nature. 2021;596(7873):495–504.
Arun SS, Iyer GN. On the analysis of COVID19Novel corona viral disease pandemic spread data using machine learning techniques. 2020 4th International Conference on Intelligent Computing and Control Systems (ICICCS). IEEE, 2020.
Benvenuto D, et al. Application of the ARIMA model on the COVID2019 epidemic dataset. Data Brief. 2020;29:105340.
Drake JM, et al. The statistics of epidemic transitions. PLoS Comput Biol. 2019;15(5):e1006917.
Chen P, et al. Detecting the tipping points in a threestate model of complex diseases by temporal differential networks. J Transl Med. 2017;15(1):1–15.
Chen P, et al. Detecting earlywarning signals of influenza outbreak based on dynamic network marker. J Cell Mol Med. 2019;23(1):395–404.
Chen L, et al. Detecting earlywarning signals for sudden deterioration of complex diseases by dynamical network biomarkers. Sci Rep. 2012;2(1):1–8.
Zhong J, et al. SGE: predicting cell fate commitment during early embryonic development by singlecell graph entropy. Genom Proteom Bioinf. 2022;19(3):461–74.
Liu R, et al. Hunt for the tipping point during endocrine resistance process in breast cancer by dynamic network biomarkers. J Mol Cell Biol. 2019;11(8):649–64.
Liu R, et al. Singlesample landscape entropy reveals the imminent phase transition during disease progression. Bioinformatics. 2020;36(5):1522–32.
Chen P, et al. Autoreservoir computing for multistep ahead prediction based on the spatiotemporal information transformation. Nat Commun. 2020;11(1):1–15.
Liu R, et al. Predicting local COVID19 outbreaks and infectious disease epidemics based on landscape network entropy. Sci Bull. 2021;66:2265.
Jiang J, Lai YC. Modelfree prediction of spatiotemporal dynamical systems with recurrent neural networks: role of network spectral radius. Phys Rev Res. 2019;1(3): 033056.
Hochreiter S, Schmidhuber J. Long shortterm memory. Neural Comput. 1997;9(8):1735–80.
Ma H, et al. Predicting time series from shortterm highdimensional data. Int J Bifurcation Chaos. 2014;24(12):1430033.
Jaeger H, Haas H. Harnessing nonlinearity: predicting chaotic systems and saving energy in wireless communication. Science. 2004;304(5667):78–80.
Ma H, et al. Randomly distributed embedding making shortterm highdimensional data predictable. Proc Natl Acad Sci. 2018;115(43):E9994–10002.
Chen C, et al. Predicting future dynamics from shortterm time series using an Anticipated Learning Machine. Natl Sci Rev. 2020;7(6):1079–91.
van Nes EH, Scheffer M. Slow recovery from perturbations as a generic indicator of a nearby catastrophic shift. Am Nat. 2007;169(6):738–47.
Deutsche Welle. Germany enters 4th coronavirus wave. 2021. https://p.dw.com/p/3zCH3.
Theresa Tam. Statement from the Chief Public Health Officer of Canada on August 12, 2020. Public Health Agency of Canada. 2021. https://www.canada.ca/en/publichealth/news/2020/08/statementfromthechiefpublichealthofficerofcanadaonaugust122020.html.
Williams JR. Models for the study of infection in populations. Handbook of Models for Human Aging. 2006; 165–182.
Van den Driessche P, Watmough J. Reproduction numbers and subthreshold endemic equilibria for compartmental models of disease transmission. Math Biosci. 2002;180(1–2):29–48.
ArroyoMarioli F, et al. Tracking R of COVID19: a new realtime estimation using the Kalman filter. PLoS ONE. 2021;16(1):e0244474.
Dong E, Du H, Gardner L. An interactive webbased dashboard to track COVID19 in real time. Lancet Inf Dis. 2020;20(5):533–4. https://doi.org/10.1016/S14733099(20)301201.
Berry I, et al. Open access epidemiologic data and an interactive dashboard to monitor the COVID19 outbreak in Canada. CMAJ. 2020;192(15):E420–E420.
Vaccaro C, et al. The impact of COVID19 first wave national lockdowns on perinatal outcomes: a rapid review and metaanalysis. BMC Pregnancy Childbirth. 2021;21(1):1–14.
AlSalem W, et al. The emergence and transmission of COVID19 in European countries, 2019–2020: a comprehensive review of timelines, cases and containment. Int Health. 2021;13(5):383–98.
Organisation for Economic Cooperation and Development. COVID19 in Latin America and the Caribbean: an overview of government responses to the crisis. OECD Publishing, 2020.
Robinson, O. COVID19 lockdown policies: an interdisciplinary review. Available at SSRN 3782395. 2021.
MendezBrito A, El Bcheraoui C, PozoMartin F. Systematic review of empirical studies comparing the effectiveness of nonpharmaceutical interventions against COVID19. J Infect. 2021;83(3):281–93.
Misra M, et al. Exit strategies from lockdowns due to COVID19: a scoping review. BMC Public Health. 2022;22(1):1–16.
Perra N. Nonpharmaceutical interventions during the COVID19 pandemic: a review. Phys Rep. 2021;913:1–52.
Patel U et al. Early epidemiological indicators, outcomes, and interventions of COVID19 pandemic: a systematic review. J Global Health. 2020; 10(2).
Acknowledgements
We would like to thank Prof. Rui Liu and Dr. Jiayuan Zhong for productive discussions.
Funding
The work was supported by National Natural Science Foundation of China (No. 11971176).
Author information
Authors and Affiliations
Contributions
MZL designed the research and performed the experiments and created the figures; MZL and SM wrote and edited the manuscript. ZRL provided direction for the project and its goals. All authors read and approved the final manuscript.
Corresponding author
Ethics declarations
Ethics approval and consent to participate
Not applicable. JHU CSSE COVID19 Data.
Consent for publication
Not applicable.
Competing interests
The authors declare that they have no competing interests.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supplementary Information
Additional file 1: Figure S1.
The country network of Spain. Figure S2. The country network of Italy. Figure S3. The country network of Netherlands. Figure S4. The regional network of Parts of Europe. Figure S5. The early signals of COVID19 in Netherlands and Spain.
Rights and permissions
Open Access This 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.
About this article
Cite this article
Li, M., Ma, S. & Liu, Z. A novel method to detect the early warning signal of COVID19 transmission. BMC Infect Dis 22, 626 (2022). https://doi.org/10.1186/s1287902207603z
Received:
Accepted:
Published:
DOI: https://doi.org/10.1186/s1287902207603z