**2. Public health emergency**

#### **2.1 Definition of public health emergency**

Public health emergencies are a subset of public emergencies that are related to health incidents and have an inclusion and exclusion arrangement with public emergencies. The Emergency of Public Health (Emergency of Public Health) is described as "mainly including infectious diseases, mass diseases of unknown origin, food safety and occupational hazards, animal epidemics, and other events that seriously affect public health and life safety" in the "China National Overall Emergency Plan for Public Emergencies" promulgated on January 8, 2006.

#### **2.2 Overview of AI application in public health**

## *2.2.1 The stage of crisis recovery is important in the management of public health crises*

The recovery phase in public health crisis management refers to the stage during which the crisis is gradually alleviated and eliminated. The flow of factors that trigger disasters has slowed, and public health emergencies have been

*Artificial Intelligence (AI) in Evidence-Based Approaches to Effectively Respond to Public… DOI: http://dx.doi.org/10.5772/intechopen.97499*

effectively addressed. The government's goal at this time is to reduce the impact of public health crises, contribute to social and economic recovery, summarize crisis management flaws, and improve the experience of managing public health emergencies in crisis.

It is critical to use AI and other technologies to promote the resumption of work and development in order to ensure sustainable and stable economic growth. Intelligent network systems have been used in China [8] to carry out online workplace, online teaching, and other activities. Companies must "not close" during the epidemic, and schools must "suspend classes without suspension." During the nationwide "war epidemic," AI networks such as WeLink, DingTalk, and Tencent Conference were widely popularized in order to minimize the losses incurred by shutdowns and output shutdowns, which played a positive role in reducing crowd gathering and reducing the risk of cross-infection while going out [9]. On the other hand, using a big data platform to analyze the migration and traffic situation in each region, as well as AI technology to prevent the epidemic from resuming, and to genuinely achieve safe resumption of work and development. Manually processing these data makes it difficult to ensure the data's validity and timeliness. Experts use AI to assist them, collect crisis-related data, and determine the type of crisis [10]. Moreover, AI calculates the severity of the crisis' effects and analyzes the causes of the crisis. Furthermore, AI allows for early detection of a problem, allowing for more time to deal with it.

#### *2.2.2 The use of AI to predict the results of public health emergencies*

The period in which the crisis will break out in the crisis management of public health crises is referred to as the preparatory stage of crisis management for public health emergencies. Since the onset of public health emergencies is uncertain and unpredictable, it is important to track and alert them. For this stage, improving the ability to monitor and respond to public health emergencies is the main focus of the government's work. The preparatory stage of crisis management for public health emergencies consists of two parts: crisis early warning, crisis training and exercises [10].

Early notice of public health crises is a vital task in the planning stage of disaster prevention. When a crisis occurs, successful early warning will significantly speed up the organization's response time. To develop an infectious disease outbreak early warning system, governments of various countries currently depend primarily on conventional surveillance methods (collaboration of medical institutions at all levels, disease prevention and control centers, and influenza-like case monitoring sentinel hospitals, and medical institutions diagnose and record clinically diagnosed and confirmed cases of influenza). However, there are some disadvantages of this monitoring system: the data collected is from a single source, and there is no comparison or correction of data from other sources; the data acquisition process of daily sampling and weekly summary, the data results are comparatively lagging; the monitoring consumes a lot of manpower and material resources, and the monitoring covers the entire country; the monitoring consumes a lot of manpower and material resources, and the monitoring covers the entire country. The accuracy of the data would be affected by an error in any node in the network [10]. The use of AI to perform infectious disease forecasting and early warning work, as well as monitoring social media, online news posts, and government reports for signs of infectious disease outbreaks, can significantly assist relevant government agencies in keeping track of the epidemic, rationally allocating medical resources, and improving advances. The cost of national disease prediction and infection prevention and control is reduced by the success rate of prevention.

By scanning foreign language news stories, animal and plant disease reports, and various official statements, the AI system provided alerts to its customers, recognizing the first foreign alert of the epidemic at an early stage. Machine learning has been used to track, locate and report on infectious spread. It provides alerts to a wide range of clients, including health care, government, industry, and public health organizations. It also serves as an alert about the existence of a new coronavirus [11]. In several aspects of the global battle against the epidemic, AI has already played a valuable but fragmented role. Screening, contact tracing, contact alerts, diagnosis, automatic deliveries, and laboratory drug discovery are only a few of the applications. AI has already played a useful but fragmented role in many aspects of the global fight against the epidemic. It has been widely used in screening, contact tracing, contact alerts, diagnosis, automated deliveries, and laboratory drug discovery [11]. It also predicts whether or not a person is infectious in advance, as well as the seriousness of the infection. By doing some general data analysis, one can significantly reduce waiting time, determine whether or not one has come into contact with virus carriers, and prevent the virus from spreading. Systematic planning and drills are an important way to enhance emergency response in the event of a disaster. Knowledge map technology can be used in public health emergency training to combine and link all of the information in the knowledge base and create a bottom map that covers all knowledge and records the connections between knowledge and knowledge, significantly increasing the scope and depth of training. The use of AI technology to perform public health emergency simulation exercises, build public health emergency simulation scenarios, deduce public health emergency handling protocols, and summarize the effects of public health emergency crisis exercises. It will help the government assess the epidemic's condition, develop decision-making and deployment capabilities for epidemic prevention and control, and test a range of mature response plans in simulation, setting the groundwork for potential rapid response and precise policy implementation in actual combat in the future [18].

Is it a Cough or a Covid? COVID-19 Detection Using Artificial Intelligence from Cough Sounds.

Increased disease screening and early warning capabilities can help to dramatically delay the spread and effect of a disease. Recent progress in developing deep learning AI models to classify cough sounds as a COVID-19 prescreening tool has shown early promise. Cough-based diagnosis is a non-invasive, cost-effective, and scalable method of diagnosing COVID-19 that, if approved, could be a gamechanger in the battle against the virus. Cough sounds have recently been tested as a preliminary diagnostic or a prescreening technique for Covid-19 identification in asymptomatic individuals by AI researchers [11]. This is advantageous because the virus can trigger subtle changes in the body that can be identified by complex algorithms combining audio signal processing and machine learning, even though no symptoms are present. This technology may also be more efficient than the standard strategy of prescreening for COVID-19 based on temperature, especially in asymptomatic patients.

#### *2.2.3 AI accelerating healthcare outcomes*

AI expands data access. AI's predictive ability is based on the volume and variety of data available; optimizing emerging tools requires extensive data access across the healthcare ecosystem. To prevent gaming of findings and prejudice, data scientists must commit to robust research over several parameters. AI allows for more concentrated collaboration. Thousands of inputs must be incorporated by scientists and technologists in collaboration with clinical specialist physicians,

*Artificial Intelligence (AI) in Evidence-Based Approaches to Effectively Respond to Public… DOI: http://dx.doi.org/10.5772/intechopen.97499*

including lab results, vital signs, drug administration, prescription doses and durations, length of stay in hospital, and patient and hospital demographics, to name a few. Clinicians can participate in the validation process and feature engineering for each organ- or condition-specific version of an AI surveillance system so that the solution can produce customized, actionable risk scores that clinicians will use. Moreover, transparency in clinical surveillance is aided by AI. These surveillance solutions can enable clinicians to apply their own clinical judgment to the performance by offering a visual representation of how and why AI made the predictions. Any AI-enabled tool should do the same thing to promote clinician buy-in and the requisite change management for widespread adoption.

#### **2.3 AI aids decision-making by simulating a real-life epidemic**

The government's approaches or policies to combat the outbreak are unquestionably important in effectively controlling the virus's spread. AI can be used to help them make the best decision possible.

Key parameters that define the characteristics of the spread, such as the transmission rate, incubation time, population density in the region, and so on, can be used to create a simulated model that mimics the actual environment of pandemics.

Following the development of the environment simulator, Reinforcement Learning can be used to determine the best strategy for achieving our aim of preventing virus spread while minimizing economic costs.

#### *2.3.1 The SIR models*

A simple compartmental model in epidemiology, known as the SIR model, is commonly used to simulate the spread of disease [12]. *S* represents the number Susceptible/Healthy individuals, *I* represents the number of Infectious individuals and *R* represents the number of Recovered individuals. It can be modeled by the below set of ordinary differential equations:

$$
\frac{dS}{dt} = -\frac{\beta IS}{N} \tag{1}
$$

$$
\frac{dI}{dt} = \frac{\beta IS}{N} - \gamma I
$$

$$
\frac{dR}{dt} = \gamma I,
$$

where *N* is the total population, *β* is the probability of disease transmission in a contact between a susceptible and an infectious subject. *γ* is the probability of an infectious individual being recovered in *dt*.

#### *2.3.1.1 The agent-based model*

The SIR model is a fundamental model for studying individual flow between compartments, assuming that all individuals within a compartment are homogeneous. An Agent-Based Model is developed to simulate the behavior of heterogeneous individuals, taking into account their characteristics [13, 14].

We may, for example, identify various types of agents, such as individuals, families, businesses, and governments, and then enable them to communicate with one another. Each type of agent may have different attributes, such as age, location of the person, location of the house/business, and wealth of the agents. Different activities, such as going to work, going home, or making business connections, can be simulated at different times.

Different social, epidemiological, and economic parameters, such as individual mobility, incubation, transmission, recovery time, income, and GDP, must be specifically defined by domain experts, using empirical evidence, or designed by the author to simulate the attributes and actions of agents. During simulation, the economic impact and pandemic statistics, such as individual wealth and the number of active cases, can be generated for evaluation.

#### *2.3.1.2 Reinforcement learning*

Reinforcement Learning (RL) is an area of machine learning that focuses on learning strategies or sequential decisions in order to optimize long-term reward in the defined environment.

A basic reinforcement learning model is shown in **Figure 1** [15]. It involves an Agent who interacts with the environment in each time steps by taking different actions. At time *t*, the agent will receive the current State *St* of the environment and the current Reward *Rt* . The agent will perform Action *At* based on the *policy* and *St* . Then, the environment will move from current State *St* to the next State *St*<sup>+</sup>1 and the associated "reward" *Rt*<sup>+</sup>1 will be output. The state *St*<sup>+</sup>1 and reward *Rt*<sup>+</sup>1 will be fed back to the agent. The process iterates until the terminal state is reached. The goal of the model is to learn the *policy* which optimizes the cumulative reward.

To train the RL model and learn the optimal policy, one way is to use Monte Carlo Tree Search, which is a searching algorithm to determine the best moves. It repeated the process of "Selecting ➔ Expanding ➔ Simulating ➔ Updating" to update the nodes in a tree (**Figure 2**) [16]. Each node in the tree represents the action we can take, with a node value which can be the probability of winning or the expected reward. At "Selecting" stage, we select the path by the value of the node until we reach the leaf node at the end of the branch. At the leaf node, we "Expand" by randomly choosing the action from the action space. Then, we "Simulate" the complete rollout, until the terminal state and obtain the final cumulative reward. The reward will then be backpropagated to update the values in each node along the path.

RL, in combination with the Agent-Based Model and the SIR model, will help the government make the best decision possible to combat the pandemic [17]. The state and the environment in RL can be simulated by the Agent-Based Model and SIR Model.

The reward function can be designed based on the pandemic statistics and economic statistics generated from the Agent-Based Model. The trained RL model will

**Figure 1.** *Basic reinforcement learning model.*

*Artificial Intelligence (AI) in Evidence-Based Approaches to Effectively Respond to Public… DOI: http://dx.doi.org/10.5772/intechopen.97499*

#### **Figure 2.**

*Phases of the Monte Carlo tree search algorithm. A search tree, rooted at the current state, is grown through repeated application of the above four phases.*

be able to advise the government on the best course of action to take at various stages of the pandemic and scenarios in order to contain the pandemic with the least amount of economic effect. The agent is to be trained with empirical demographic data, pandemic data and economic data in pandemic time to simulate the impacts of policies conditioned with the predicted pandemic data. An Action space can be defined. Social distancing, lockdown, company and school closures, wearing a face mask, doing nothing, public hygiene promotion and so on. The impact of policies can be highly dimensioned vectors and subject to execution error. However, all these attributes can be inputs of the simulations and the prediction error can be reduced by more data input.

With the numerous simulations, government can have a full profile of impacts of policies to be taken in different scenarios by making assumption on the parameters of the effectiveness of policy, say the shut-down of schools reduce the younger age group infection by {38.2%, 50%, 61.8%}, and the impact of this reduction can be propagated to other age groups through a trained Boltzmann Machine which depict the dynamics of infection rate between the age groups. The fidelity of the simulations is correlated with the complexity of model and number of data used to train the model.

The reward functions can deviate across regions; however the reward function can be designed based on the pandemic statistics and economic statistics generated from the Agent-Based model. The trained RL model would allow the government to select the optimal policy and evaluate the drawbacks before the policy is implemented.

#### **2.4 Public health emergencies accelerate the implementation of AI**

Many AI technologies, such as robots assisting in hospital transportation, have accelerated as a result of the epidemic's social isolation. One of the most contentious examples is contact tracing. Many countries around the world have successfully developed contact tracing systems, allowing them to efficiently monitor the epidemic's spread. This strategy, however, is seen as an infringement of privacy in the United States, Europe, and other countries. Although it has a bright future, there are challenges in the areas of privacy, data processing, ethics, and social issues. When it comes to medical information, these questions must be seriously addressed in the light of public health or personal health. During a public health crisis, the government must strike a balance between citizens' rights and the need for effective prevention and control measures to efficiently control disease transmission until the outbreak is over, and then return to normal.

No one wants to replicate the epidemic's mistakes. AI will be used in the future to prevent epidemics from arising and spreading. Hospitals will make effective use of sensors and wearable devices to collect outbreak data and report possible hazards in a timely manner, allowing them to properly respond to the crisis and avoid losing control again. Inevitably, privacy concerns would arise in the former. There are some notable inconsistencies between privacy rights and the requirements of machine learning. Although privacy security necessitates as little data sharing as possible, machine learning necessitates as much data as possible. Many countries are concerned that misuse of contact tracing could compromise privacy, so they have developed and implemented a variety of privacy security technologies. AI techniques, processes, and technology are being used to develop health care and programs. The good news is that they can coexist, and AI is a double-edged sword that can help to foster global governance and cultural change.
