Jonas Botz1*
Danqi Wang1,2
Nicolas Lambert3
Nicolas Wagner3
Marie Génin3Edward Thommes4
Sumit Madan1,5Laurent Coudeville4Holger Fröhlich1,2*- 1Department of Bioinformatics, Fraunhofer Institute for Algorithms and Scientific Computing (SCAI), Sankt Augustin, Germany
- 2Bonn-Aachen International Center for Information Technology (B-IT), University of Bonn, Bonn, Germany
- 3Quinten Health, Paris, France
- 4Sanofi, Paris, France
- 5Department of Computer Science, University of Bonn, Bonn, Germany
The COVID-19 pandemic has highlighted the lack of preparedness of many healthcare systems against pandemic situations. In response, many population-level computational modeling approaches have been proposed for predicting outbreaks, spatiotemporally forecasting disease spread, and assessing as well as predicting the effectiveness of (non-) pharmaceutical interventions. However, in several countries, these modeling efforts have only limited impact on governmental decision-making so far. In light of this situation, the review aims to provide a critical review of existing modeling approaches and to discuss the potential for future developments.
Introduction
In December 2019, a new virus (SARS-CoV-2), causing a respiratory disease - later named COVID-191, was discovered. At the time of the outbreak, many healthcare systems around the world were not well prepared for the pandemic that later emerged. While the virus was initially detected in China, measures to prevent its spread to other regions of the world were often hesitant and taken too late. Whereas compartmental spatio-temporal models of disease spread in epidemiology have been known in principle for a long time (1), many countries initially lacked robust and systematically collected surveillance data to which these models could be fitted. In general, it has been difficult to translate insights from modeling into actionable decision support for the government.
Based on these considerations, the French-German collaborative project AIOLOS (Artificial Intelligence Tools for Outbreak Detection and Response) has recently started with the aim to strengthen the resilience of national healthcare systems against future outbreaks of respiratory infections2. More specifically, AIOLOS identifies three areas, where population-level computational modeling, including techniques from Artificial Intelligence (AI) and machine learning (ML), could potentially impact the preparedness against future pandemics based on various data sources (Figure 1):
1. early warning of a new outbreak,
2. monitoring the spatio-temporal spread of a disease,
3. predicting the impact and effectiveness of different interventions to support decision-making at scientific and policy levels.
Figure 1. Overview of potential impact areas of population-level computational modeling for increased preparedness against pandemic situations, including relevant data sources.
This paper aims to review existing population-level computational modeling work in each of these areas. Our ambition is thus significantly different from published reviews, which solely focused on mathematical models of COVID-19 disease spread (2) or AI/ML algorithms for patient-level disease diagnosis and prognosis (3).
Early warning
Surveillance data
Health surveillance data is the traditional source of information for detecting a pandemic outbreak. The goal of respective computational approaches is to detect anomalies in a data stream consisting of discrete events, i.e., cases reported by doctors. For this purpose, several statistical tests have been suggested in the literature, including methods proposed by the Robert Koch Institute in Germany (4) and the Center for Diseases Control and Prevention in the USA (5), the Farrington method and its variants (6, 7) and Bayesian methods (4). Altogether, the R-package “surveillance” lists almost 20 algorithms for the early detection of pandemic outbreaks using surveillance data (8), covering three different scenarios:
1. spatio-temporal data of individual infectious events,
2. temporal event history of a defined set of individual units (e.g., specified households),
3. events aggregated over regions and time periods.
Due to data privacy concerns, typically only data of the last category are made publicly available and considered for governmental decision-making. A comparative simulation study pointed out elevated false positive rates for many algorithms with sensitivities ranging between 20 and 67% (9). Furthermore, a principal challenge is that traditional surveillance data in many countries are not systematically recorded in a fully automated and digitalized manner. Moreover, surveillance data in several countries do not cover several relevant aspects, such as hospitalization and ICU admission rates. Hence, this data could come too late for an early warning system. In response to this situation, several authors have thus proposed to systematically monitor wastewater for virus particles rather than waiting for reports by doctors (10, 11), and according measures are currently being implemented in the USA, Europe, and Israel. Noteworthy, Israeli researchers already used such an approach a few years ago to detect a silent polio outbreak (12, 13).
Social media
Given the shortcomings of traditional surveillance data, several authors have more recently explored the potential of social media. Jain and Kumar (14) proposed a keyword extraction approach, in which they first used the term frequency-inverse document frequency (TF-IDF) technique, identifying relevant keywords from tweets, and secondly, used a linear discriminant analysis (LDA)-based classifier to find relevant keywords in newspaper really simple syndication (RSS) feeds. Subsequently, the relevant keywords were used to analyze tweets from the respective period, and machine learning classifiers were developed to filter out irrelevant tweets. They found that Support Vector Machines (SVMs) and a Naive Bayes classifier most accurately classified tweets (F1 = 0.77).
Lopreite et al. (15) performed statistical tests (Kolmogorov-Smirnov and Anderson-Darling) to compare the cumulative frequencies of pneumonia-related tweets from the winter seasons of 2018/2019 and 2019/2020 in selected European countries. They found an exceeding number of pneumonia-related postings in the winter season of 2019/2020 before the outbreak of COVID-19. In a similar direction, Mavragani (16) retrieved Google Trends data for the topic of “Coronavirus” and calculated Pearson correlation coefficients between Google Trends data and the respective categories of cumulative/daily cases/deaths. The results showed strong correlations of Google Trends data with COVID-19 cases and deaths in the examined European countries. The authors conclude that information epidemiology is a viable instrument to monitor the disease spread and identify regions in which cases have not yet peaked, hence contributing to an early warning system.
Going methodologically one step further, Yousefinaghani et al. (17) used a real-time anomaly detection approach utilizing the Seasonal-Hybrid Extreme Studentized Deviate algorithm (18) to identify the onset and peak of COVID-19 waves in Google Trends and Twitter data from the US and Canada. This study also evaluated the correlation between tweets and Google trends data with official COVID-19 case numbers. Pearson correlation analysis demonstrated a strong correlation between officially reported infected cases and the relevant posts and searches. Unlike other studies, the authors quantitatively prioritized COVID-19 symptoms in detecting disease trends. For example, “cough” and “fever” were better trend indicators compared to “tiredness” and “loss of smell.”
Broniatowski et al. (19) identified health-related, influenza-related, and case-reporting tweets with logistic regression, which were used with Google Flu Trends to predict influenza outbreaks at municipal and regional levels.
Further, Kogan et al. (20) used a Bayesian probabilistic model to develop an early warning algorithm for COVID-19 based on social media (Google Trends, Twitter, UpToDate), fever incidence rates, and predictions made by the global epidemic and mobility model (21), resulting in a time-to-event prediction. The algorithm was validated on COVID-19 surveillance data as well as incidence rates of influenza-like illness, demonstrating that an uptrend in COVID-19 infections could be predicted up to 7 days in advance with an accuracy of ~75%. Table 1 summarizes the techniques employed by the discussed papers.
Table 1. Included early warning studies.
Disease monitoring
Spatio-temporal modeling of disease spread
There are different approaches for modeling the spatio-temporal spread of an epidemic situation described in the literature (see Tables 2–5):
• mechanistic compartmental models formulated as differential equation systems, which have been classically used in epidemiology (22, 26–33, 35–38, 40, 64),
• machine learning approaches, including Bayesian learning techniques (41–49),
• agent-based modeling approaches (50–55),
• hybrid modeling approaches combining several of the aforementioned techniques (39, 56, 58–63, 65).
Table 2. Included studies covering spatio-temporal monitoring of disease spread with compartmental models and their key aspects.
Table 3. Included studies covering spatio-temporal monitoring of disease spread with machine learning and Bayesian models and their key aspects.
Table 4. Included studies covering spatio-temporal monitoring of disease spread with agent-based modeling approaches and their key aspects.
Table 5. Included studies covering spatio-temporal monitoring of disease spread with hybrid models and their key aspects.
Compartmental models
General principle
To model and understand the evolution of an epidemic, compartmental models are often used. The underlying idea is to distribute the population into several interconnected compartments. The relationship between these compartments is given by a system of differential equations. With given or estimated initial conditions this mathematical system can be solved at any point in time The foundation of today's compartmental models was formulated nearly a century ago (1). In their study, Kermack and McKendrick examined the evolution of various pandemics and established the commonly used susceptible-infected-removed (SIR) model which is based on three compartments:
• S(t) - The susceptible population, i.e., the part of the population that can become infected,
• I(t) - The infected population, i.e., the part of the population that has the disease and can transmit the disease to the susceptibles,
• R(t) - The removed or recovered population, i.e., the part of the population that has recovered from the disease and that is considered immune. (With N = S(t) + I(t) + R(t) being the total population.)
The dynamics of the SIR model get described by a set of ordinary differential equations (ODEs), which include two free parameters, β - the transmission rate and γ - the recovery rate:
Due to its simple nature, there are also some limitations and assumptions with this model. Here we will mention some of them. First, the population size is assumed to be constant, the birth nor the death rates are incorporated, and the model does not allow for people to become reinfected. Second, both the transmission and the recovery rates are constant. Third, the model assumes that the infected person becomes infectious immediately after getting infected, whereas in reality there is a latency period. Another assumption is that there is homogeneous mixing of the population, and no social networks and mobility are considered.
To account for some of its limitations, the archetypical SIR model can be extended to include an age structure or additional compartments, e.g., compartment E for the - by the virus-exposed - population (susceptible-exposed-infected-removed: SEIR), compartment D for the disease-deceased population or compartment H for the hospitalized population.
Applications to epidemic disease monitoring
There is a vast literature on compartmental disease models over the last 50 years (66). Examples include the successful modeling of several epidemic outbreaks, such as SARS (22) and influenza (23–25). However, the highly dynamic development of the COVID-19 pandemic with corresponding public intervention measures required extensions and modifications (26, 27, 30, 32, 36, 37, 40). For example, Götz and Heidrich (32) used the number of registered deaths by COVID-19 rather than the registered cases, with the idea to evade the dark figure of undetected cases, including a delay term to account for the time between infection and death. Bahri (27) split between a young population (age < 60 years) and an older population (age ≥60 years) stating that the younger population has more infections, while the older population is at higher risk, with a much higher death rate. Similarly, Coudeville et al. (30) introduced an age-stratified SEIR model to estimate how different scenarios affect industry decisions on different time scales. In another study, the authors further used this model to derive the potential effects of various immunization programs based on vaccination (36).
Aravindakshan et al. (26) used a compartmental model including social distancing and mobility as parameters. The authors further estimated the impact of different non-pharmaceutical interventions (NPIs) on social distancing including other covariates (e.g., weather, day of the week) in a linear regression model and used its coefficients for simulating different scenarios. Schüler et al. (40) included NPIs by using a piecewise constant transmission rate depending on the corresponding NPI and analyzed effects on the district level. Similarly, Humphrey et al. (37) estimated the effect of testing and tracing in combination with social distancing measures by introducing an isolation compartment, resulting in a modified transmission rate. Prague et al. (35) estimated several parameters of an extended SEIR model from data about the incident and hospitalized cases in France at a regional level via a non-linear mixed effects model while considering NPIs. Moreover, the model by Prague et al. considers the fact that only a fraction of the actually infected patients is counted in surveillance data.
Bertozzi et al. (28) studied the disease spread in several European countries, first looking at the exponential growth and the self-exciting branching process and then using a compartmental model, focusing on the impacts of social distancing, enabling them to model and understand different stages of the pandemic.
Khan et al. (33) modeled an NPI-dependent transmission rate. Chang et al. (29) modeled the disease spread in the ten largest US metropolitan areas using bipartite networks with time-varying edges for mapping the hourly movement of census block groups (CBGs) to specific points of interest (POIs). Then, each mobility network gets paired with an extended SEIR model with a corresponding transmission rate. To illustrate the spatial dynamic coupling across locations, Pei et al. (34) used a metapopulation SEIR model including daily work commuting and random movement among 3,142 US counties. Using inference, they studied the effect of asynchronous interventions across these locations in the US and performed counterfactual simulations to estimate the evolution of the disease spread by implementing NPIs at different times. To account for the fact that COVID-19 is a pandemic with several waves, Khedher et al. (38) introduced multiple discrete states into their model.
Sartorius et al. (39) developed a discrete-time SEIR model, which incorporated information about population density and mobility using a hierarchical Bayesian model. They estimated their model via full Bayesian inference (Markov Chain Monte Carlo sampling).
Machine learning models
In addition to compartmental models, machine learning techniques, including neural networks, have become popular approaches for modeling and predicting disease spread. Examples include models for the disease spread in China (43) and worldwide (47). Fong et al. (43) tried to overcome the problem of a small dataset by using a polynomial neural network with corrective feedback, while Ibrahim et al. incorporated urban characteristics and NPIs via a variational Long Short-Term Memory (LSTM) encoder.
In addition to neural networks, other machine learning techniques have been proposed as well: for example, Al-qaness et al. (42) combined an Adaptive Neuro-Fuzzy Inference System (ANFIS) with a flower pollination algorithm (FPA) using the salp swarm algorithm (SSA), creating the FPASSA-ANFIS model. Nader et al. (48) developed a Random Forest algorithm; other studies employed extreme stochastic gradient boosting (XGBoost) (44, 46). Yeung et al. (49) compared different classical machine learning regression methods (ridge, decision tree, Random Forests, AdaBoost, and Support Vector Machines) and found Random Forests and AdaBoost to perform best. In general, classical, non-time series machine learning models could predict future pandemic development rather accurately.
Pavlyshenko (45) used a Bayesian machine learning approach for modeling the global spread of COVID-19 and its effect on the stock market, while (41) additionally included the spatial aspect via a spatio-temporal kernel function.
Agent based models
Agent-based modeling (ABM) is a sub-field of Artificial Intelligence (AI). The idea in ABM is to simulate a set of software agents, which can interact with each other according to a defined set of rules. ABM approaches can implement many characteristics such as social contacts of individuals or sub-populations, disease characteristics (e.g., virus transmission rates, virus variants), patient characteristics (e.g., age, sex, comorbidities, and risk factors), mobility and contact networks (e.g., household, workplace, school, community, tourism), healthcare services (e.g., hospitalization, bed occupancy) and governmental regulations or NPIs.
In the literature, ABM approaches have been used on different scales. Staffini et al. (53) used socio-economic and disease-related information to study the spread of the SARS-CoV-2 virus and the influence of NPIs in Italy, Germany, Sweden, and Brazil. Shattock et al. (55) included risk groups and seasonal patterns in the transmission model and estimated the effect of various NPIs as well as vaccination campaigns on the pandemic evolution, hospitalization, and deaths in Switzerland. Colosi et al. (54) used an ABM approach to estimate school-specific reproduction numbers depending on the COVID-19 variants.
Various authors further extended these models by including demographic features as well as more profound contact networks - through deeper population mobility simulations - to simulate synthetic populations, the disease spread in this population, and the effect of a large set of NPIs (50–52). Hoertel et al. (50) focused on possible post-lockdown measures to reduce epidemic rebounds and therewith estimated the effect of protecting/shielding persons at risk; while Hinch et al. (51) and Kerr et al. (52) both developed a simulation platform, OpenABM, and Covasim, which enables to simulate the disease spread depending on various settings, including different NPIs.
Hybrid models
One of the main limitations of machine learning is the assumption of test data being drawn from the same statistical distribution as training data. This results in a major challenge if there is a covariate shift of test data relative to the original training data, e.g., due to NPIs, seasonal effects, new virus variants, or further unknown factors. Hence, the utility of conventional machine learning models in a highly dynamic situation such as the COVID-19 pandemic must be questioned. In this regard hybrid modeling approaches combining compartmental models and machine learning, or compartmental models and ABM approaches could provide an interesting alternative.
Several authors have explored hybrid models of the spatio-temporal disease spread in this regard: For example, Dandekar and Barbastathis (56) used a neural network to model the influence of NPIs on the compartment of infected patients. For model training, they employed the universal ODE approach, which combines neural networks with ODEs in a joint framework (67). Menda et al. (65) introduced a neural network to relax the assumption of a constant transmission rate. Their model is formulated as a non-Gaussian state-space system, which is estimated via Certainty-Equivalent Expectation-Maximization (57).
Wang et al. (60) combined their extended SIR model with spatial cellular automata (CA) and then introduced a Convolution Neural Network (CNN) paired with an LSTM recurrent neural network to learn the dynamical parameters of a compartmental model, which also includes the population of undetected or asymptomatic individuals.
Watson et al. (61) first used a probabilistic graphical model estimated via Bayesian inference to predict the velocity of cumulative cases. Moreover, they developed a Random Forest model to give daily projections and interval estimates for cases and deaths in different US states. Both models were then combined into a compartmental model to make forecasts of incidence rates.
A different type of hybrid model is presented by Fritz et al. (62). They combined a statistical spatial regression with a Graph Neural Network (GNN) incorporating social connectedness and co-location maps.
Several authors combined ABM approaches with SEIR models (58, 59, 63). Hadley et al. (63) derived the transmission and hospitalization rates depending on the agent's age, comorbidity status, and testing status to forecast the ICU bed demand. Silva et al. simulated a society (i.e., persons, houses, businesses, government, and healthcare systems) including a large set of social and demographic parameters and estimated different scenarios based on social distancing measures. Capobianco et al. introduced the PandemicSimulator including – besides a SEIR model – a moving and interacting society, a government that makes policy decisions, and optional testing and contract tracing strategies. They also suggested adding a hidden Markov model to adapt infection rates over time and a reinforcement learning (RL) layer to find the optimal policy to minimize the public health impact.
Social media and internet searches
The epidemic spread has been shown to be correlated with search engine usage on the web in the past (68). Nowadays people also share their opinion on social media networking sites such as Twitter, Reddit, and Facebook. These opinions can also be utilized to track epidemic disease spread. Masri et al. (69) studied using tweets' time and geolocation data to improve the monitoring of the Zika virus (ZIKV) epidemic. The collected tweets were counted and compared with weekly data of the U.S. ZIKV cases, revealing a high Pearson correlation coefficient value of 0.67 by applying a 1-week lag on tweets. Adding this 1-week-lag tweet data to the case counts in an auto-regression prediction model improved the coefficient of determination (R2) from 0.61 to 0.74, which showed that tweet metadata is a significant predictor of future ZIKV cases.
Various authors have also used social media to support the surveillance and monitoring of an epidemic (70–72). Missier et al. (70) identified tweets related to dengue epidemics by classifying them into mosquito, sickness, and news-related classes. Chen et al. (73) created an ongoing collection of so far 123 million COVID-19-related tweets identified using various keywords and shared it with the research community for further analysis.
To better understand and model the trajectory of COVID-19 in the US, Klein et al. (74) manually annotated 10,000 pre-filtered tweets into three COVID-19 associated classes (probable, possible, and other cases) and used Bidirectional Encoder Representations from Transformers (BERT) to automatically classify tweets. The classifier achieved an F1 score of 0.64 for differentiating three classes. Given that “probable” or “possible” tweets were primarily distributed in the states reporting COVID-19 cases and posted before the first confirmed case, the model could successfully identify candidate COVID-19 cases and high-risk regions.
Similarly, Liu et al. (75) collected COVID-19-related Reddit posts from North Carolina, which showed a similar trend of observed confirmed cases and deaths as to the government data. They further classified these posts while performing NER to obtain mitigation types (such as distancing, disinfection, personal protective equipment) and detection types (such as symptoms, testing) and analyzed for a certain time period the change of people's sentiments toward masks in these posts. For disease monitoring, Magge et al. (76) built a system to collect symptoms and disease mentions from social media platforms and normalized them to unified medical language system (UMLS) terminology. Using deep learning methods (such as BERT and RoBERTa) that were trained on multiple available corpora (such as TwiMed, MedNorm, DS-NER), they achieved an F1-score of 0.86 and 0.75 on DailyStrength and Twitter datasets, respectively. They also applied their system on Twitter posts to collect COVID-19 symptoms.
Users also share their opinions on COVID-19 measures on Twitter by supporting, refuting, or just commenting on them (77). These opinions from German-speaking countries were manually labeled, and Beck et al. utilized predictions by transformer-based models. Jalil et al. (71) performed sentiment analysis on tweets' text to classify them into positive, negative, and neutral. For the analysis, they used the COVIDSenti dataset (78) and reached the highest accuracy of 96.66% with the proposed Multi-depth DistilBERT method. Table 6 provides an overview of the use of social media and internet searches for disease monitoring.
Table 6. Included studies focusing on disease monitoring via mining of social media and internet searches.
Pathogen sequences
Pathogens are, like any organism, under evolutionary pressure and will thus mutate to optimize their adaptation to the human host. Accordingly, different pathogenic variants will occur over time. Deep learning approaches have recently been introduced to identify such variants during sequencing (79). In addition, phylogenetic tree inference, a classical approach from computational biology based on a sequence alignment followed by a statistical tree inference (either maximum likelihood or Markov Chain Monte Carlo) with a dedicated likelihood function (80), is often used. Incorporation of spatio-temporal information into the construction of phylogenies could potentially provide important information on the spread of virus variants. Still, phylogenies are not only informed by pathogen sequences, but also by external factors, such as the sampling process, the proportion of the pathogen genome sequenced in each sample, the quality of the sequence data, and the mutation rate of the pathogen itself (81).
Several authors have suggested approaches to construct temporal phylogenies (82–84) and applied this strategy to SARS-CoV-2 (85–87). More recently, Didelot et al. (88) showed that transmission events between hosts could be estimated by coloring different hosts in a phylogenetic tree reconstruction. Müller et al. (89) extended phylogenies to networks by incorporating recombination events and applied this strategy to influenza.
New variants may influence the transmission rate of a pathogen. Davies et al. (90) first retrospectively estimated the lineage-dependent growth rates of SARS-CoV-2. Based on that, they further calculated the expected competitive advantage of a new lineage and predicted the impact on the reproduction and the transmission rates via a discrete-time compartmental spatio-temporal disease model.
Decision support
Healthcare resource planning
Modeling can not only help to alert and monitor a pandemic situation, but forecasts generated by corresponding models can also give guidance on necessary actions. Therefore, there is no clear boundary between early warning, monitoring, and decision support.
One important aspect of decision support is the management and planning of available public healthcare resources. In this regard, Ivorra et al. (91) developed a compartmental model for China, in which they included the hospitalization rate. With the help of their model, they estimated and planned the demand for clinical beds. With a similar ambition in mind, Hadley et al. (63) proposed an agent-based modeling approach. Lorenzen et al. (92) developed a machine learning model (Random Forest) using electronic health records of more than 40,000 patients in Denmark, which predicted the number of ICU admissions and ventilator use. Kandula et al. (93) developed a compartmental model for predicting influenza hospitalization rates using Google search trends. Moa et al. (94) proposed a linear model to forecast the overall severity of an influenza season in Australia based on only five parameters.
Planning and evaluating NPIs
In addition to healthcare resource planning a further aspect of modeling is to support the planning and evaluation of NPIs. In this context three different types of studies have been conducted (see Table 7):
• those that retrospectively evaluate the effects of NPIs (26, 27, 31–34, 40, 48, 49, 56, 60, 97, 98),
• those that make forecasts on the effects of a specified NPI in the sense of scenario planning (26, 31, 35, 38, 50–55, 58, 96),
• and those that develop methods for optimal control policy identification (59, 100–104).
Table 7. Included studies covering decision support.
Retrospective evaluation of NPIs is generally challenged by the fact that NPIs are highly heterogeneous. Historically, often several NPIs have been applied at the same time, and there is neither a control group nor any kind of randomization. Systematic differences across countries in terms of demography, population density, climate, or cultural aspects complicates using of one country as a control for another one, even if typical statistical matching or weighting techniques known from observational studies are applied. Moreover, there is the question of the corresponding outcome to consider, given that observed incident cases will depend on the applied test strategy and thus underestimate the true number of infected people.
One type of approach has been to try to associate NPIs with the spatio-temporal modeling of disease spread, e.g., by introducing the NPI effect on the transmission rate and reproduction number in a compartmental model (26, 31, 35, 38, 40, 56). Correspondingly, authors have then used such models to make scenario forecasts, e.g., regarding the effect of social distancing (31, 35, 38, 96). Also, other types of spatio-temporal disease spreading models have been used for the same purpose, such as ABM approaches (50–55, 58), Bayesian hierarchical modeling (97), and machine learning (48, 49, 98, 99). The work of Yeung et al. specifically investigated the influence of socio-cultural aspects on the growth rate of COVID-19 incidences in 114 countries. The work by Barros et al. considered causal machine learning techniques.
Also, more traditional statistical analysis approaches have been applied recently, such as the synthetic control technique (95), which uses incident case numbers from the same country in the treatment and control group, depending on when an NPI has been put in place. Additionally, Mader and Rüttenauer analyzed the effect of vaccinations.
To find optimal control policies, offline RL strategies have been proposed by several authors. While Kwak et al. (100) solely relied on deep learning and only focused on health aspects, other studies (101–104) focused on a hybrid modeling strategy incorporating an extended SEIR compartmental model for predicting potential NPI effects. Moreover, the latter studies incorporated the economic costs of NPIs as well. Finally, Capobianco et al. (59) combined their hybrid ABM approach with offline RL to optimize the reopening policies.
Discussion
Statistical tests have been used traditionally to detect outbreaks based on surveillance data. Recent years have witnessed an increasing use of other data sources, such as social media and internet searches. Even though such data types are likely to contain relevant signals, these are most likely biased toward certain user communities. Hence, early warning signals detected via “digital traces” should be seen as a complement to traditional surveillance data, but not as a replacement.
Regarding the monitoring of pandemics, specifically, the existing modeling efforts for COVID-19 have highlighted numerous challenges, such as the unknown number of truly infected persons (due to limitations of tests and test strategies, or due to asymptomatic disease) and the dependency on the spatio-temporal spread on external factors, such as NPIs and the compliance to those measures, weather, population density, and socio-economic aspects. Hence, many authors have extended traditional epidemiological compartment models and combined them with statistical inference and machine learning techniques, partially resulting in hybrid neural network /compartmental modeling approaches. While these are clear advancements, it should be seen that the spatio-temporal spread of an infectious disease is generally determined by a complex interplay between a pathogen (e.g., its genetic adaptability), individual (e.g., genetic variants, disease history, lifestyle, socio-economic conditions), society (e.g., testing strategy, vaccination rate, NPIs and compliance to those, population density) and environment (e.g., climate, weather). NLP techniques could help at this point to mine social media and news articles to complement surveillance data and to gain an understanding of the sentiment of the population with respect to specific NPIs, while at the same time taking into consideration the biases of this type of data and the principally limited accuracy of text analytics as such. Altogether, further developments of modeling approaches are needed, which better combine data modalities across all relevant scales, i.e., ranging from the pathogen up to the environment level. This, however, will in turn require better availability, integration, and accessibility of necessary data, including electronic health records. The investment into such a data infrastructure is thus a prerequisite to making significant progress on the modeling side.
Models will only have an impact if they can support the human decision process. In recognition of this fact, several authors have tried to support scenario planning by associating NPIs with the predicted spatio-temporal development of the disease, or by forecasting healthcare resources and economic impact. While forecasts under the scenario of no further taken action might be improved by considering the aspects mentioned above for spatio-temporal modeling, predicting the effect of an NPI is principally challenged by several aspects: (i) The NPI could be new and thus there is no direct historical comparison, and (ii) there is always a lack of a proper control group, i.e., it is not possible to perform a study akin to a Randomized Clinical Trial. RL techniques are thus generally challenged by this inability to experiment with a new policy. It is thus unlikely that decision-makers would immediately trust the recommendation of an optimal NPI estimated by an RL algorithm. A better approach might hence be to offer a ranking of the predicted effectiveness of multiple NPIs together with the estimated economic costs, which should not be neglected.
Conclusion
In response to the ongoing COVID-19 pandemic, many countries currently review their strategies to be better prepared against future outbreaks. One important aspect in this context is to invest in data analytical capabilities, including modeling. Computational modeling approaches could help to earlier detect an outbreak, monitor the spatio-temporal spread, and to support the decision-making process by governmental authorities.
In this paper, we reviewed the diversity of existing modeling approaches for all three areas. Of course, each model is adjusted to a specific healthcare-related question by fitting it to particular data. In conclusion, models for early outbreak detection as well as spatio-temporal disease spread could be further improved by better combining and integrating data modalities across multiple scales. The ongoing COVID-19 pandemic in this context provides a “global laboratory” with the opportunity to retrospectively validate existing techniques as well as develop new ones. At the same, there is a need for funding bodies and governmental decision-makers to invest in corresponding data ecosystems. Models are likely to increase their impact on decision-making if they become more accurate and are at the same time explainable. Showing point estimates of a black-box model without highlighting epistemic uncertainties or providing further explanations of the most influential features is thus discouraged.
Author contributions
Conceptualization, methodology, supervision, project administration, and funding acquisition: HF. Data curation, formal analysis, visualization, investigation, validation, and writing—original draft: JB, DW, NL, SM, and HF. Writing—review and editing: JB, DW, NL, MG, NW, ET, LC, SM, and HF. All authors contributed to the article and approved the submitted version.
Funding
This work has been supported by the AIOLOS (Artificial Intelligence Tools for Outbreak Detection and Response) project. The project was supported by the French State and the German Federal Ministry for Economic Affairs and Climate Action (grant number 01MJ22005A) and the French Ministry of Economy and Finance. Ce projet a été financé par le gouvernement dans le cadre de France 2030 in the context of the Franco-German call on Artificial Intelligence technologies for risk prevention, crisis management, and resilience.
Conflict of interest
Authors NL, NW, and MG are employees of the commercial company Quinten-Health. Authors ET and LC are employees of the commercial company Sanofi. None of the afore mentioned companies had any influence on the scientific content presented in this paper.
The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
Publisher's note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
Footnotes
1. ^https://www.who.int/emergencies/disease-outbreak-news/item/2020-DON229
2. ^https://www.digitale-technologien.de/DT/Redaktion/EN/Standardartikel/Internationale_Koop_Projekte/Frankreich/ki_innovationsprojekte_de_fr_projekt_aiolos.html
References
1. William OK, McKendrick AG. A contribution to the mathematical theory of epidemics. Proc R Soc Lond A Math Phys Character. (1927) 115:700–21. doi: 10.1098/rspa.1927.0118
2. Shankar S, Mohakuda SS, Kumar A, Nazneen PS, Yadav AK, Chatterjee K, et al. Systematic review of predictive mathematical models of COVID-19 epidemic. Med J Armed Forces India. (2021) 77:S385–92. doi: 10.1016/j.mjafi.2021.05.005
3. Dogan O, Tiwari S, Jabbar MA, Guggari S. A systematic review on AI/ML approaches against COVID-19 outbreak. Complex Intell Syst. (2021) 7:2655–78. doi: 10.1007/s40747-021-00424-8
4. Höhle M. Surveillance: an R package for the monitoring of infectious diseases. Comput Stat. (2007) 22:571–82. doi: 10.1007/s00180-007-0074-8
5. Stroup DF, Williamson GD, Herndon JL, Karon JM. Detection of aberrations in the occurrence of notifiable diseases surveillance data. Stat Med. (1989) 8:323–29. doi: 10.1002/sim.4780080312
6. Farrington CP, Andrews NJ, Beale AD, Catchpole MA. A statistical algorithm for the early detection of outbreaks of infectious disease. J R Stat Soc A. (1996) 159:547–63. doi: 10.2307/2983331
7. Noufaily A, Enki DG, Farrington P, Garthwaite P, Andrews N, Charlett A. An improved algorithm for outbreak detection in multiple surveillance systems. Stat Med. (2013) 32:1206–22. doi: 10.1002/sim.5595
8. Meyer S, Held L, Höhle M. Spatio-temporal analysis of epidemic phenomena using the R package surveillance. J Stat Softw. (2017) 77:1–55. doi: 10.18637/jss.v077.i11
9. Bédubourg G, Strat YL. Evaluation and comparison of statistical methods for early temporal detection of outbreaks: a simulation-based study. PLoS ONE. (2017) 12:e0181227. doi: 10.1371/journal.pone.0181227
10. Lastra A, Botello J, Pinilla A, Urrutia JI, Canora J, Sánchez J, et al. SARS-CoV-2 detection in wastewater as an early warning indicator for COVID-19 pandemic. Madrid Region Case Study. Environ Res. (2022) 203:111852. doi: 10.1016/j.envres.2021.111852
11. Maida CM, Amodio E, Mazzucco W, La Rosa G, Lucentini L, Suffredini E, et al. Wastewater-based epidemiology for early warning of SARS-CoV-2 circulation: a pilot study conducted in sicily, Italy. Int J Hyg Environ Health. (2022) 242:113948. doi: 10.1016/j.ijheh.2022.113948
12. Sharara N, Endo N, Duvallet C, Ghaeli N, Matus M, Heussner J, et al. Wastewater network infrastructure in public health: applications and learnings from the COVID-19 pandemic. PLOS Global Public Health. (2021) 1:e0000061. doi: 10.1371/journal.pgph.0000061
13. Brouwer AF, Eisenberg JNS, Pomeroy CD, Shulman LM, Hindiyeh M, Manor Y, et al. Epidemiology of the silent polio outbreak in rahat, israel, based on modeling of environmental surveillance data. Proc Nat Acad Sci. (2018) 115:E10625–33. doi: 10.1073/pnas.1808798115
14. Jain VK, Kumar S. Lev AI - 105 - an effective approach to track levels of influenza-A (H1N1) pandemic in India using twitter. Procedia Comput Sci. (2015) 70:801–7. doi: 10.1016/j.procs.2015.10.120
15. Lopreite M, Panzarasa P, Puliga M, Riccaboni M. Early warnings of COVID-19 outbreaks across Europe from Social Media. Sci Rep. (2021) 11:2147. doi: 10.1038/s41598-021-81333-1
16. Mavragani A. Tracking COVID-19 in Europe: infodemiology approach. JMIR Public Health Surveill. (2020) 6:e18941. doi: 10.2196/18941
17. Yousefinaghani S, Dara R, Mubareka S, Sharif S. Prediction of COVID-19 waves using social media and google search: a case study of the US and Canada. Front Public Health. (2021) 9:656635. doi: 10.3389/fpubh.2021.656635
18. Hochenbaum J, Vallis OS, Kejariwal A. Automatic anomaly detection in the cloud via statistical learning. arXiv. (2017). http://arxiv.org/abs/1704.07706 (accessed May 20, 2022).
19. Broniatowski DA, Dredze M, Paul MJ, Dugas A. Using social media to perform local influenza surveillance in an inner-city hospital: a retrospective observational study. JMIR Public Health Surveill. (2015) 1:e4472. doi: 10.2196/publichealth.4472
20. Kogan NE, Clemente L, Liautaud P, Kaashoek J, Link NB, Nguyen AT, et al. An early warning approach to monitor COVID-19 activity with multiple digital traces in near real time. Sci Adv. (2021) 7: eabd6989. doi: 10.1126/sciadv.abd6989
21. Chinazzi M, Davis JT, Ajelli M, Gioannini C, Litvinova M, Merler S, et al. The effect of travel restrictions on the spread of the 2019 novel coronavirus (COVID-19) outbreak. Science. (2020) 368:395–400. doi: 10.1126/science.aba9757
22. Zhang Z. The outbreak pattern of SARS cases in China as revealed by a mathematical model. Ecol Modell. (2007) 204:420–6. doi: 10.1016/j.ecolmodel.2007.01.020
23. Shaman J, Karspeck A, Yang W, Tamerius J, Lipsitch M. Real-time influenza forecasts during the 2012–2013 season. Nat Commun. (2013) 4:2837. doi: 10.1038/ncomms3837
24. Leonenko VN, Ivanov SV. Fitting the SEIR model of seasonal influenza outbreak to the incidence data for russian cities. Russ J Numer Anal Math Modell. (2016) 31:267–79. doi: 10.1515/rnam-2016-0026
25. Osthus D, Hickmann KS, Caragea PC, Higdon D, Del Valle SY. Forecasting seasonal influenza with a state-space SIR model. Ann Appl Stat. (2017) 11:202–24. doi: 10.1214/16-AOAS1000
26. Aravindakshan A, Boehnke J, Gholami E, Nayak A. Preparing for a future COVID-19 wave: insights and limitations from a data-driven evaluation of non-pharmaceutical interventions in Germany. Sci Rep. (2020) 10:20084. doi: 10.1038/s41598-020-76244-6
27. Bahri MK. Modeling the flow of the COVID-19 in Germany: the efficacy of lockdowns and social behavior. medRxiv. (2020). doi: 10.1101/2020.12.21.20248605
28. Bertozzi AL, Franco E, Mohler G, Short MB, Sledge D. The challenges of modeling and forecasting the spread of COVID-19. Proc Nat Acad Sci. (2020) 117:16732–38. doi: 10.1073/pnas.2006520117
29. Chang S, Pierson E, Koh PW, Gerardin J, Redbird B, Grusky D, et al. Mobility network models of COVID-19 explain inequities and inform reopening. Nature. (2020) 589:82–7. doi: 10.1038/s41586-020-2923-3
30. Coudeville L, Gomez GB, Jollivet O, Harris RC, Thommes E, et al. Exploring uncertainty and risk in the accelerated response to a COVID-19 vaccine: perspective from the pharmaceutical industry. Vaccine. (2020) 38:7588–95. doi: 10.1016/j.vaccine.2020.10.034
31. Giordano G, Blanchini F, Bruno R, Colaneri P, Di Filippo A, Di Matteo A, et al. Modelling the COVID-19 epidemic and implementation of population-wide interventions in Italy. Nat Med. (2020) 26:855–60. doi: 10.1038/s41591-020-0883-7
32. Götz T, Heidrich P. Early stage COVID-19 disease dynamics in Germany: models and parameter identification. J Math Ind. (2020) 10:20. doi: 10.1186/s13362-020-00088-y
33. Khan ZS, Van Bussel F, Hussain F. A predictive model for Covid-19 spread – with application to eight US states and how to end the pandemic. Epidemiol Infect. (2020) 148:e249. doi: 10.1017/S0950268820002423
34. Pei S, Kandula S, Shaman J. Differential effects of intervention timing on COVID-19 spread in the United States. Sci Adv. (2020) 6:eabd6370. doi: 10.1126/sciadv.abd6370
35. Prague M, Wittkop L, Clairon Q, Dutartre D, Thiébaut R, Hejblum BP. Population modeling of early COVID-19 epidemic dynamics in French regions and estimation of the lockdown impact on infection rate. medRXiV. (2020). doi: 10.1101/2020.04.21.20073536
36. Coudeville OJ, Mahé C, Chaves S, Gomez GB. Potential impact of introducing vaccines against COVID-19 under supply and uptake constraints in france: a modelling study. PLoS ONE. (2021) 16:e0250797. doi: 10.1371/journal.pone.0250797
37. Humphrey L, Thommes EW, Fields R, Coudeville L, Hakim N, Chit A, et al. Large-scale frequent testing and tracing to supplement control of covid-19 and vaccination rollout constrained by supply. Infect Dis Modell. (2021) 6:955–74. doi: 10.1016/j.idm.2021.06.008
38. Khedher NB, Kolsi K, Alsaif A. A multi-stage SEIR model to predict the potential of a new COVID-19 wave in KSA after lifting all travel restrictions. Alex Eng J. (2021) 60:3965–74. doi: 10.1016/j.aej.2021.02.058
39. Sartorius B, Lawson AB, Pullan RL. Modelling and predicting the spatio-temporal spread of COVID-19, associated deaths and impact of key risk factors in England. Sci Rep. (2021) 11:5378. doi: 10.1038/s41598-021-83780-2
40. Schüler L, Calabrese JM, Attinger S. Data driven high resolution modeling and spatial analyses of the COVID-19 pandemic in Germany. PLoS ONE. (2021) 16:e0254660. doi: 10.1371/journal.pone.0254660
41. Stojanović O, Leugering J, Pipa G, Ghozzi S, Ullrich A. A Bayesian Monte Carlo approach for predicting the spread of infectious diseases. PLoS ONE. (2019) 14:e0225838. doi: 10.1371/journal.pone.0225838
42. Al-qaness MAA, Ewees AA, Fan H, Abd El Aziz M. Optimization method for forecasting confirmed cases of COVID-19 in China. J Clin Med. (2020) 9:674. doi: 10.3390/jcm9030674
43. Fong SJ, Li G, Dey N, Gonzalez-Crespo R, Herrera-Viedma E. Lev AI - 14 - finding an accurate early forecasting model from small dataset: a case of 2019-NCoV novel coronavirus outbreak. Int J Interact Multimed Artif Intell. (2020) 6:132. doi: 10.9781/ijimai.2020.02.002
44. Mehta M, Julaiti J, Griffin P, Kumara S. Lev AI - 18 - early stage machine learning–based prediction of US county vulnerability to the COVID-19 pandemic: machine learning approach. JMIR Public Health Surveill. (2020) 6:e19446. doi: 10.2196/19446
45. Pavlyshenko BM. Regression approach for modeling COVID-19 spread and its impact on stock market. ArXiv. (2020). http://arxiv.org/abs/2004.01489 (accessed May 02, 2022).
46. Suzuki Y, Suzuki A, Nakamura S, Ishikawa T, Kinoshita A. Lev AI - 16 - machine learning model estimating number of COVID-19 infection cases over coming 24 days in every Province of South Korea (XGBoost and MultiOutputRegressor). Preprint. Infectious Diseases (except HIV/AIDS). medRxiv. (2020). doi: 10.1101/2020.05.10.20097527
47. Ibrahim MR, Haworth J, Lipani A, Aslam N, Cheng T, Christie N. Variational-LSTM autoencoder to forecast the spread of coronavirus across the globe. PLoS ONE. (2021) 16:e0246120. doi: 10.1371/journal.pone.0246120
48. Nader IW, Zeilinger EL, Jomar D, Zauchner C. Onset of effects of non-pharmaceutical interventions on COVID-19 infection rates in 176 countries. BMC Public Health. (2021) 21:1472. doi: 10.1186/s12889-021-11530-0
49. Yeung AYS, Roewer-Despres F, Rosella L, Rudzicz F. Machine learning–based prediction of growth in confirmed COVID-19 infection cases in 114 countries using metrics of nonpharmaceutical interventions and cultural dimensions: model development and validation. J Med Internet Res. (2021) 23:e26628. doi: 10.2196/26628
50. Hoertel N, Blachier M, Blanco C, Olfson M, Massetti M, Rico M, et al. A stochastic agent-based model of the SARS-CoV-2 epidemic in France. Nat Med. (2020) 26:1417–21. doi: 10.1038/s41591-020-1001-6
51. Hinch R, Probert WJM, Nurtay A, Kendall M, Wymant C, Hall M, et al. OpenABM-Covid19—an agent-based model for non-pharmaceutical interventions against COVID-19 including contact tracing. PLoS Comput Biol. (2021) 17:e1009146. doi: 10.1371/journal.pcbi.1009146
52. Kerr CC, Stuart RM, Mistry D, Abeysuriya RG, Rosenfeld K, Hart GR, et al. Covasim: an agent-based model of COVID-19 dynamics and interventions. medRxiv. (2021). doi: 10.1101/2020.05.10.20097469
53. Staffini A, Svensson AK, Chung UI, Svensson T. An agent-based model of the local spread of SARS-CoV-2: modeling study. JMIR Med Inf. (2021) 9:e24192. doi: 10.2196/24192
54. Colosi E, Bassignana G, Contreras DA, Poirier C, Boëlle P-Y, Cauchemez S, et al. Screening and vaccination against COVID-19 to minimise school closure: a modelling study. Lancet Infect Dis. (2022) 22:977–89. doi: 10.1016/S1473-3099(22)00138-4
55. Shattock AJ, Le Rutte EA, Dünner RP, Sen S, Kelly SL, Chitnis N, et al. Impact of vaccination and non-pharmaceutical interventions on SARS-CoV-2 dynamics in Switzerland. Epidemics. (2022) 38:100535. doi: 10.1016/j.epidem.2021.100535
56. Dandekar R, Barbastathis G. Lev AI - 8 - quantifying the effect of quarantine control in Covid-19 infectious spread using machine learning. Epidemiology. (2020). doi: 10.1101/2020.04.03.20052084
57. Menda K, De Becdelievre J, Gupta J, Kroo I, Kochenderfer M, Manchester Z. Scalable identification of partially observed systems with certainty-equivalent EM. In: Proceedings of the 37th International Conference on Machine Learning. (2020), 6830–40.
58. Silva PCL, Batista PVC, Lima HS, Alves MA, Guimarães FG, Silva RCP. COVID-ABS: an agent-based model of COVID-19 epidemic to simulate health and economic effects of social distancing interventions. Chaos Solitons Fractals. (2020) 139:110088. doi: 10.1016/j.chaos.2020.110088
59. Capobianco R, Kompella V, Ault J, Sharon G, Jong S, Fox S, et al. Agent-based Markov modeling for improved COVID-19 mitigation policies. J Artif Intell Res. (2021) 71:953–92. doi: 10.1613/jair.1.12632
60. Wang TX, Stoecker T, Stoecker H, Jiang Y, Zhou K. Machine learning spatio-temporal epidemiological model to evaluate Germany-county-level COVID-19 risk. Mach Learn Sci Technol. (2021) 2:035031. doi: 10.1088/2632-2153/ac0314
61. Watson GL, Xiong D, Zhang L, Zoller JA, Shamshoian J, Sundin P, et al. Pandemic velocity: forecasting COVID-19 in the US with a machine learning and Bayesian time series compartmental model. PLoS Comput Biol. (2021) 17:e1008837. doi: 10.1371/journal.pcbi.1008837
62. Fritz C, Dorigatti E, Rügamer D. Combining graph neural networks and spatio-temporal disease models to improve the prediction of weekly COVID-19 cases in Germany. Sci Rep. (2022) 12:3930. doi: 10.1038/s41598-022-07757-5
63. Hadley E, Rhea S, Jones K, Li L, Stoner M, Bobashev G. Enhancing the prediction of hospitalization from a COVID-19 agent-based model: a Bayesian method for model parameter estimation. PLoS ONE. (2022) 17:e0264704. doi: 10.1371/journal.pone.0264704
64. Amaro JE, Dudouet J, Orce JN. Global analysis of the COVID-19 pandemic using simple epidemiological models. Appl Math Model. (2021) 90:995–1008. doi: 10.1016/j.apm.2020.10.019
65. Menda K, Laird L, Kochenderfer MJ, Caceres RS. Explaining COVID-19 outbreaks with reactive SEIRD models. Sci Rep. (2021) 11:17905. doi: 10.1038/s41598-021-97260-0
66. Heesterbeek H, Anderson RM, Andreasen V, Bansal S, DE Angelis D, Dye C, et al. Modeling infectious disease dynamics in the complex landscape of global health. Science. (2015) 347:aaa4339. doi: 10.1126/science.aaa4339
67. Rackauckas C, Ma Y, Martensen J, Warner C, Zubov K, Supekar R, et al. Universal differential equations for scientific machine learning. arXiv. (2021). http://arxiv.org/abs/2001.04385 (accessed July 04, 2022).
68. Ginsberg J, Mohebbi MH, Patel RS, Brammer L, Smolinski MS, Brilliant L. Detecting influenza epidemics using search engine query data. Nature. (2009) 457:1012–14. doi: 10.1038/nature07634
69. Masri S, Jia J, Li C, Zhou G, Lee M-C, Yan G, et al. Use of twitter data to improve Zika virus surveillance in the United States during the 2016 epidemic. BMC Public Health. (2019) 19:761. doi: 10.1186/s12889-019-7103-8
70. Missier P, Romanovsky A, Miu T, Pal A, Daniilakis M, Garcia A, et al. Tracking dengue epidemics using twitter content classification and topic modelling. In:Casteleyn S, Dolog P, Pautasso C, , editors. Current Trends in Web Engineering. Lecture Notes in Computer Science. Cham: Springer International Publishing (2016), p. 80–92. doi: 10.1007/978-3-319-46963-8_7
71. Jalil Z, Abbasi A, Javed AR, Badruddin Khan M, Abul Hasanat MH, Malik KM, et al. COVID-19 related sentiment analysis using state-of-the-art machine learning and deep learning techniques. Front Public Health. (2022) 9:812735. doi: 10.3389/fpubh.2021.812735
72. Jahanbin K, Rahmanian F, Rahmanian V, Jahromi AS. Application of twitter and web news mining in infectious disease surveillance systems and prospects for public health. GMS Hyg Infect Control. (2019) 14:Doc19. doi: 10.3205/dgkh000334
73. Chen E, Lerman K, Ferrara E. Tracking social media discourse about the Covid-19 pandemic: development of a public coronavirus twitter data set. JMIR Public Health Surveill. (2020) 6:e19273. doi: 10.2196/19273
74. Klein AZ, Magge A, O'Connor KMS, Cai H, Weissenbacher D, Gonzalez-Hernandez G. A chronological and geographical analysis of personal reports of COVID-19 on twitter. medRxiv. (2020). doi: 10.1101/2020.04.19.20069948
75. Liu Y, Whitfield C, Zhang T, Hauser A, Reynolds T, Anwar M. Monitoring COVID-19 Pandemic through the lens of social media using natural language processing and machine learning. Health Inf Sci Syst. (2021) 9:1–16. doi: 10.1007/s13755-021-00158-4
76. Magge A, O'Connor K, Scotch M, Gonzalez-Hernandez G. SEED: symptom extraction from English social media posts using deep learning and transfer learning. medRxiv. (2021). doi: 10.1101/2021.02.09.21251454
77. Beck T, Lee J-U, Viehmann C, Maurer M, Quiring O, Gurevych I. Investigating label suggestions for opinion mining in German Covid-19 social media. In: Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). (2021), p. 1−13. doi: 10.18653/v1/2021.acl-long.1
78. Naseem U, Razzak I, Khushi M, Eklund PW, Kim J. COVIDSenti: a large-scale benchmark twitter data set for COVID-19 sentiment analysis. IEEE Transact Comput Soc Syst. (2021) 8:1003–15. doi: 10.1109/TCSS.2021.3051189
79. Bartoszewicz JM, Genske U, Renard BY. Deep learning-based real-time detection of novel pathogens during sequencing. Brief Bioinform. (2021) 22:bbab269. doi: 10.1093/bib/bbab269
80. Felsenstein J. Evolutionary trees from DNA sequences: a maximum likelihood approach. J Mol Evol. (1981) 17:368–76. doi: 10.1007/BF01734359
81. Villabona-Arenas CJ, Hanage WP, Tully DC. Phylogenetic Interpretation during outbreaks requires caution. Nat Microbiol. (2020) 5:876–77. doi: 10.1038/s41564-020-0738-5
82. Bouckaert R, Vaughan TG, Barido-Sottani J, Duchêne S, Fourment M, Gavryushkina A, et al. BEAST 2.5: an advanced software platform for Bayesian evolutionary analysis. PLOS Comput Biol. (2019) 15:e1006650. doi: 10.1371/journal.pcbi.1006650
83. To T-H, Jung M, Lycett S, Gascuel O. Fast dating using least-squares criteria and algorithms. Syst Biol. (2016) 65:82–97. doi: 10.1093/sysbio/syv068
84. Sagulenko P, Puller V, Neher RA. TreeTime: maximum-likelihood phylodynamic analysis. Virus Evol. (2018) 4:vex042. doi: 10.1093/ve/vex042
85. Wolf JM, Kipper D, Borges GR, Streck AF, Lunge VR. Temporal spread and evolution of SARS-CoV-2 in the second pandemic wave in Brazil. J Med Virol. (2022) 94:926–36. doi: 10.1002/jmv.27371
86. Duchene S, Featherstone L, Haritopoulou-Sinanidou M, Rambaut A, Lemey P, Baele G. Temporal signal and the phylodynamic threshold of SARS-CoV-2. Virus Evol. (2020) 6:veaa061. doi: 10.1093/ve/veaa061
87. Hoffer AD, Vatani S, Cot C, Cacciapaglia G, Chiusano M, Cimarelli A, et al. Variant-driven multi-wave pattern of COVID-19 via a machine learning analysis of spike protein mutations. medRxiv. (2021). doi: 10.1101/2021.07.22.21260952
88. Didelot X, Kendall M, Xu Y, White PJ, McCarthy N. Genomic epidemiology analysis of infectious disease outbreaks using TransPhylo. Curr Protoc. (2021) 1:e60. doi: 10.1002/cpz1.60
89. Müller NF, Stolz U, Dudas G, Stadler T, Vaughan TG. Bayesian inference of reassortment networks reveals fitness benefits of reassortment in human influenza viruses. Proc Nat Acad Sci. (2020) 117:17104–11. doi: 10.1073/pnas.1918304117
90. Davies NG, Abbott S, Barnard RC, Jarvis CI, Kucharski AJ, Munday JD, et al. Estimated transmissibility and impact of SARS-CoV-2 lineage B117 in England. Science. (2021) 372:eabg3055. doi: 10.1126/science.abg3055
91. Ivorra B, Ferrández MR, Vela-Pérez M, Ramos AM. Mathematical modeling of the spread of the coronavirus disease 2019 (COVID-19) taking into account the undetected infections. The Case of China. Commun Nonlinear Sci Numer Simul. (2020) 88:105303. doi: 10.1016/j.cnsns.2020.105303
92. Lorenzen SS, Nielsen M, Jimenez-Solem E, Petersen TA, Perner A, Thorsen-Meyer H-C, et al. Using machine learning for predicting intensive care unit resource use during the COVID-19 pandemic in Denmark. Sci Rep. (2021) 11:18959. doi: 10.1038/s41598-021-98617-1
93. Kandula S, Pei S, Shaman J. Improved forecasts of influenza-associated hospitalization rates with google search trends. J R Soc Interface. (2019) 16:20190080. doi: 10.1098/rsif.2019.0080
94. Moa A, Muscatello D, Chughtai A, Chen X, MacIntyre CR. Flucast: a real-time tool to predict severity of an influenza season. JMIR Public Health Surveill. (2019) 5:e11780. doi: 10.2196/11780
95. Mader S, Rüttenauer T. The effects of non-pharmaceutical interventions on COVID-19 mortality: a generalized synthetic control approach across 169 countries. Front Public Health. (2022) 10:820642. doi: 10.3389/fpubh.2022.820642
96. Kissler SM, Tedijanto C, Goldstein E, Grad YH, Lipsitch M. Projecting the transmission dynamics of SARS-CoV-2 through the postpandemic period. Science. (2020) 368:860–68. doi: 10.1126/science.abb5793
97. Flaxman S, Mishra S, Gandy J, Unwin HJT, Mellan TA, Coupland H, et al. Estimating the effects of non-pharmaceutical interventions on COVID-19 in Europe. Nature. (2020) 584:257–61.
98. Barros V, Manes I, Akinwande V, Cintas C, Bar-Shira O, Ozery-Flato M, et al. A causal inference approach for estimating effects of non-pharmaceutical interventions during Covid-19 pandemic. medRxiv. (2022). doi: 10.1101/2022.02.28.22271671
99. Haug N, Geyrhofer L, Londei A, Dervic E, Desvars-Larrive A, Loreto V, et al. Ranking the effectiveness of worldwide COVID-19 government interventions. Nat Hum Behav. (2020) 4:1303–12. doi: 10.1038/s41562-020-01009-0
100. Kwak GH, Ling L, Hui P. Deep reinforcement learning approaches for global public health strategies for COVID-19 pandemic. PLoS ONE. (2021) 16:e0251550. doi: 10.1371/journal.pone.0251550
101. Colas C, Hejblum B, Rouillon S, Thiébaut R, Oudeyer P-Y, Moulin-Frier C, et al. EpidemiOptim: a toolbox for the optimization of control policies in epidemiological. ArXiv. (2020). doi: 10.1613/jair.1.12588
102. Khadilkar H, Ganu T, Seetharam DP. Optimising lockdown policies for epidemic control using reinforcement learning. Trans Indian Natl Acad Eng. (2020) 5:129–32. doi: 10.1007/s41403-020-00129-3
103. Padmanabhan R, Meskin N, Khattab T, Shraim M, Al-Hitmi M. Reinforcement learning-based decision support system for COVID-19. Biomed Signal Process Control. (2021) 68:102676. doi: 10.1016/j.bspc.2021.102676
Keywords: pandemic, machine learning, artificial intelligence, agent-based-modeling, compartmental models
Citation: Botz J, Wang D, Lambert N, Wagner N, Génin M, Thommes E, Madan S, Coudeville L and Fröhlich H (2022) Modeling approaches for early warning and monitoring of pandemic situations as well as decision support. Front. Public Health 10:994949. doi: 10.3389/fpubh.2022.994949
Received: 15 July 2022; Accepted: 21 October 2022;
Published: 14 November 2022.
Edited by:
Vaibhav Srivastava, University of Petroleum and Energy Studies, IndiaReviewed by:
Gisele Umviligihozo, Simon Fraser University, CanadaSudipti Arora, Dr. B. Lal Institute of Biotechnology, India
Copyright © 2022 Botz, Wang, Lambert, Wagner, Génin, Thommes, Madan, Coudeville and Fröhlich. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
*Correspondence: Jonas Botz, jonas.botz@scai.fraunhofer.de; Holger Fröhlich, holger.froehlich@scai.fraunhofer.de