Preventing the Next Pandemic: A Critical Analysis of the Utilization of Artificial Intelligence in the U.S. Healthcare System

doi:10.21203/rs.3.rs-1721483/v1

Download PDF

Research Article

Preventing the Next Pandemic: A Critical Analysis of the Utilization of Artificial Intelligence in the U.S. Healthcare System

https://doi.org/10.21203/rs.3.rs-1721483/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this older preprint version

Read the latest preprint version →

The outbreak of 2019-nCoV (“COVID-19”) raised major concerns as it spread rapidly around the globe. The impacts are not fully known and cannot be overlooked given the emergence of COVID-19 variants. In response to this ever-changing virus, artificial intelligence (“AI”) has been explored for its capability to combat the pandemic. This critical analysis examines the utility of AI to improve pandemic preparedness and discusses the benefits of AI-based technologies, such as detecting disease outbreaks, mitigating spread, and speeding drug and vaccine development. It also questions whether AI applications could prevent the next pandemic.

A literature search was performed using PubMed, MEDLINE, and Scopus, and articles published after January 2020 that discussed AI applications in the U.S. healthcare system were ranked in terms of relevance on a scale of 1-5 according to their degrees of focus. Articles scoring three or higher were included in the final literature review. From a total of forty-one identified articles, twenty-four met the inclusion criteria and twelve were ranked of highest relevance by including discussions on AI models and methods, the strengths and limitations, and promising practice or recommendations.

This critical analysis concludes that AI-based technology is successful at improving pandemic preparedness, but challenges and limitations will arise as we continue to invest in AI-based technology; such as over- or under-performance due to imbalanced or incomplete data. Overall, the utilization of AI should aim to improve emergency preparedness strategies, mitigation efforts, and vaccine development, while alleviating the burdens experienced by medical professionals to achieve the best results.

artificial intelligence

machine learning

deep learning

healthcare

The Centers for Disease Control and Prevention (CDC) has taken the lead on COVID-19 surveillance, case management, and vaccine coverage and delivery in the U.S., but it has experienced challenges in reporting complete, timely, and accurate data—resulting in inadvertent spread of misinformation on social media (commonly known as an infodemic), as well as significant pushback from those (i.e., anti-vaxxers) questioning the overall impacts of COVID-19 or the effectiveness of the vaccine (Purnat et al. 2021). For the public to accept the message that public health authorities and scientists are trying to communicate, we need a clear picture of the current and future impacts of COVID-19. This highlights the need for accurate and efficient AI-based predictive modeling and techniques to detect disease outbreaks, mitigate its spread and speed drug and vaccine research and development.

A couple notable benefits of AI is that it can handle and compare large amounts of clinical and medical data and scientific literature to better inform epidemiological studies and vaccine discovery efforts; and it is crucial for decreasing physician workload and improving case management (Chen & See, 2020; Mohanty et al. 2020). Many understand the potential benefits of exploring AI as well after seeing its uses in the U.S. healthcare system. The AI-based drug discovery market alone is expected to reach over $2.1 billion by 2027 (Reports and Data 2020). Some applications have even led to rapid advancements in infectious disease contact tracing and in targeting medical supply chain issues (Vaishya et al. 2020; Tsyganov 2021). The greatest advantage of AI-based technology overall is that it can assist in refining pandemic prevention, mitigation and response strategies set out in emergency preparedness guidelines. Thus, investments in AI are on the rise as it is being explored for its potential to improve health outcomes and prevent the next pandemic.

In May 2020, the U.S. Department of Health & Human Services provided $6.6 billion to pharmaceutical companies as part of the Operation Warp Speed strategy to accelerate the development, manufacturing, and distribution of COVID-19 resources for both testing and vaccination (Welch et al. 2020). Yet, the following year we still saw spikes in new cases despite increased vaccination coverage and delivery. In response, the CDC announced its $2.1 billion investment towards improving prevention and control strategies (CDC, 2021). Most of the funding is being allocated towards supporting health department staff, training, and education, and strengthening workforce capacity. It is uncertain how much of that funding will go towards AI-based technologies that can improve mitigation, prevention, and response activities.

The CDC’s Data Tracker¹ reported a total of 78,518,045 confirmed cases and 936,162 case-related deaths as of February 23, 2022. Since vaccines became available, 215.1 million people were fully vaccinated, and 93.4 million people have received the booster dose as of February 23, 2022. However, community transmission in the U.S. continued to be high into February of 2022, with a 7-day case rate of 167.7 per 100,000. The emergence of the Omicron variant caused a significant increase in transmission, cases, and hospitalizations from November to December 2021 despite mitigation efforts. The emergence of other variants since Omicron emphasizes the importance of a more efficient way to collect, analyze, and disseminate data to improve pandemic preparedness processes. Without accurate and complete data, fighting the spread of COVID-19 has become an uphill battle.

The CDC relies heavily on three key data sources to compile their online COVID-19 reports: 1) aggregate count data; 2) patient-level data; and 3) death counts.² However, the inconsistency in COVID-19 symptoms has posed challenges in obtaining accurate and complete data at the state and local levels, resulting in differences in how cases are being tracked and reported across the country. This makes it difficult for public health officials to determine the actual risk of transmission, mortality, and other risk metrics needed for forecasting and prevention strategies.

The CDC created the COVID-19 Information Management Repository in response to the need for easily accessible, comprehensive, and up-to-date information at the local and national levels. However, the CDC also states on their website that:

While CDC strives to provide complete and accurate data . . . there might be discrepancies between numbers reported by CDC versus by health departments. When this occurs, data reported by health departments should be considered the most accurate.

Health departments have been severely burdened by the impacts of the pandemic and the strain on maintaining health data records. Accurate data and timely reporting require the implementation of a unified case definition and standard nomenclature, an efficient codification of disease-related reporting, and electronic records data using exchangeable and interoperable information. (Garcia et al 2020). These will ultimately improve the efficiency and accuracy of data collection and reporting, while alleviating the burden experienced by health providers and data analysts. AI has the potential to take on these challenges and remove these discrepancies.

In a recent study aimed at developing a precise COVID-19 spread model via deep machine learning, researchers found that the model was most useful in preparing for new short-term outbreaks, because the data consisted of confirmed, recovered, and deceased COVID-19 cases, and the parameters of the model were time-dependent (Jung et al. 2020). Such a model approximates the infection and recovery rates and provides a depiction of predicted virus spread to better inform of emergency preparedness, mitigation, and response interventions. Moreover, it can allow emergency managers and public health officials to close the gaps in the development, manufacturing, and distribution of medical equipment and vaccines before the stockpile runs out. These findings are valuable in determining whether the application of AI-based technology can be used to prevent the next pandemic and ameliorate related U.S. health expenditure.

A critical analysis of the utilization of AI-based technology in the U.S. health care system can help to understand how AI models and tools are being used, which applications should be further explored or refined, and whether AI-based technology is more cost-effective than current methods being used. Given increased research, discourse, and publications on the utilization of AI in the medicinal industry, it is beneficial to identify the key applications of AI, the strengths and limitations of utilizing AI in the U.S. health care system, as well as its ability to highlight the gaps in knowledge still needing to be filled.

Specific Aims

The goal of the critical analysis is to determine the strengths and limitations of AI to improve emergency preparedness strategies and vaccine development, as well as answer the question of whether a particular AI model or tool has the potential to be cost-effective. Recommendations for future AI research and health policy are also necessary and will be discussed. The utilization of AI is vast but, for the purposes of this critical analysis, will be limited to applications that aim to improve pandemic mitigation, prevention, and response processes. The specific aims of this paper will be to answer the following questions:

What AI-based methods or models are being used to improve pandemic preparedness, such as prevention, mitigation, and/or response strategies?
1. Does the use of these technologies have the potential to be cost-effective?
What are the strengths and limitations of applying AI-based technology to support disease spread predictions and surveillance, and address vaccine development and medical supply issues?
What are promising practices and recommendations for the use of AI-based technology for pandemic preparedness in the U.S.?

¹ The above statistics are provided by the CDC’s Data Tracker and can be found at: https://covid.cdc.gov/covid-data-tracker/#datatracker-home

²Information regarding the CDC’s COVID-19 data sources can be found at: https://www.cdc.gov/coronavirus/2019-ncov/covid-data/about-COVID-data.html

To conduct the literature search, the following terms were primarily used: “artificial intelligence” and “pandemic.” This combination of search terms alone generated over 5,000 articles, systematic reviews, and/or case studies across the PubMed, Scopus, and MEDLINE search engines. These databases included evidence-based journal articles and reviews in the areas of health, social and life science, biomedicine and technology, and public health. Along with the above-mentioned search terms, other search terms were included such as “emergency preparedness,” and “machine learning.” See Appendix for a complete list of search terms. After the search was conducted, the selection process entailed a review of the article’s publication date, abstract and citations to determine relevance to the U.S. healthcare system, as well as whether the articles provide applicable and feasible recommendations on how AI was being or could be utilized to prevent the next pandemic. Articles published before January 2020 were excluded from the literature search. Irrelevant topics and articles that failed to thoroughly discuss the applications or impacts of AI-based technology during or since the pandemic were not included in the final literature review.

To address the first specific aim, articles focusing on the relationship between phases of emergencies and pandemic and the role AI might play in improving phase activities were selected for analysis. The general phases of disaster consist of mitigation, preparedness, response, and recovery (Lindsay 2012). For this analysis, the recovery phase was not assessed, and so articles focused mainly on COVID recovery were removed from the literature review. Emergency management involves many actors and frameworks to determine what activities are needed to prevent and respond to disasters and emergencies, as well as who is authorized to carry out certain activities.

The Department of Health and Human Services (HHS) is authorized to handle public health emergencies and developed the COVID-19 Response Strategic Plan to protect citizens, allocate HHS funding, and protect confidential data (2020). The plan includes definitions of the four phases and examples of activities for each phase. Mitigation refers to identifying the risk to reduce the impact of an event. State and local officials enforcing mask mandates and quarantine to prevent the spread of COVID-19 are examples of mitigation. Preparedness also aims to reduce the impact of risk but focuses mainly on training, resources, and surveillance. Response to an event includes the immediate activities required to save lives, such as the distribution of medical equipment and vaccines. According to WHO (2009), the pandemic phases focus on, but are not limited to, “only loosely correspond[ing] to pandemic risk; identify[ing] sustained human-to-human transmission as a key event; [and] better distinguish[ing] between time for preparedness and response.” Therefore, it was important to include articles that focused on how AI can identify and measure pandemic risk and spread to better distinguish between the preparedness and response phases.

Specific aim two of this analysis was focused on the strengths and limitations of certain AI-based technologies, and how these tools were being used either to prevent the next pandemic or improve emergency preparedness. Understanding public perception or concerns also helped to develop this analysis and reach this aim. Although the review of current news and events to determine public perception either in support or opposition to AI-based technology applied to healthcare was not used as a method in this critical analysis, it merely helped to understand whether the public believes AI will result in more harm than good. According to the Chair of the United Nation’s Law Track for AI for Good, public perception can serve as either a strength or limitation to the utilization and investment of AI moving forward (AI for Good Law Track, personal communication, August 18, 2020). A major point of contention often raised is that AI will replace the role of healthcare providers and public health officials. This misconception may come from the lack of understanding between artificial intelligence and augmented intelligence, and the role that each play in health care. For this reason, “augmented intelligence” was also used in the search terms of the literature review. Another concern with using AI in healthcare is how data privacy and security will be preserved (AI for Good Law Track, personal communication, August 18, 2020). Thus, the search term “data privacy” was also added. Understanding the pros and cons of the applications of AI achieves this aim and helped in the development of appropriate recommendations.

Understanding the purpose that AI could serve in preventing the next pandemic and improving related U.S. healthcare expenditures is key. For aim three, research on promising practices and recommendations on the use of AI-based technology for emergency preparedness and mitigation processes were selected for review. Many of the articles included for review discussed best practices on implementing AI-based technologies for the purposes of early detection of outbreaks, ongoing surveillance to mitigate the spread, and ensuring that equipment and supplies for tests and vaccines. are produced and available.

Literature Search Results

The databases used for the literature search produced a gross total of 5,871 articles, including systematic reviews, randomized control trials, and cohort & case control studies. The total numbers of articles found in each database are depicted in Table 1.

Table 1 (n = 5871)

Search Results #1 – Total Number of Search Results

SCOPUS	PubMed	MEDLINE
2,432	2,097	1,342
Note: Duplicates were not yet removed.

Selection Process

In conducting the selection process, the databases were filtered by publication date to only include articles published after January 2020. Duplicates were then removed between databases, reducing the literature search results substantially (n = 2390). Table 2 presents the scoring system used to eliminate irrelevant articles, and those that scored below a three were not included in the final review. (*Note: U.S.-based application of AI was necessary for inclusion) Twelve (n = 12) of the articles included in the final literature review received the highest-ranking score of five by including discussions on U.S.-based applications, recommendations for preventing the next pandemic, and addressed all specific aims of this analysis. Although AI recommendations were highly considered for inclusion, non-AI specific recommendations were also considered. The ranking system was used to determine which articles would be most insightful to the findings and discussion sections of this critical analysis.

Table 2

SCORING CATEGORIES	Point Value
U.S. Based*	1
Non-AI recommendations for preventing the next pandemic	1
Addresses specific aim 1	1
Addresses specific aim 2	1
Addresses specific aim 3	1
Total Possible Points	5

Due to the vast application of AI, most of the articles were excluded based on not meeting most of the criteria related to relevance, applicability, or feasibility. To determine the relevance of each article, the title, abstract and citations were reviewed—eliminating most articles from the literature search. Determining the applicability and feasibility of non-AI or AI recommendations made in the articles with respect to preventing the next pandemic required a further review of the discussion or conclusion sections. A review of the methods and findings determined the inclusion or exclusion of the articles in the literature review, and the adjusted totals are depicted in Table 3.

Table 3 (n = 41)

Search Results #2 – Adjusted Total Number of Search Results

SCOPUS	PubMed	MEDLINE
6	26	9

Final Literature Review

Of the forty-one articles selected from the secondary literature search, twenty-four met all inclusion criteria. The articles selected for final review relate to how AI is applied in different pandemic phases or how effective certain AI models or methods are in improving phase activities. Based on the content of the articles, they were sorted into four major categories: 1) mitigation 2) emergency preparedness; 3) vaccine response; and 4) reviews of AI models and methods.

The first category included articles that referred to the use of AI to identify the transmission risk of COVID-19 to reduce its impact and spread; and includes the following topics: Risk Modelling; Big Data; and Forecasting. The second category included articles that refer to how AI can improve training, resources, and surveillance; and are based on the following topics: Preparedness / Prevention Training & Strategies; Supply Chain Management; and Surveillance. The third category included articles that referred to the use of AI in vaccine research & development (“R&D”) and are based on the following topics: Drug Identification & Repurposing; and Reverse Vaccinology. The fourth category included articles that provided an in-depth review of AI-based models and methods and their effectiveness in preventing the next pandemic. Overall, all articles included in the literature review covered many of the above-mentioned categories and/or topics. The articles included in the final literature review are described in Table 4 below. The Literature Review Matrix provides a hyperlink to the articles, the author and year it was published, a summary of the article, the AI model(s) or method(s) discussed (if any), and the above theme categories that were covered. A full description of AI models and methods in Table 4 can be found in the Appendix.

Table 4 (n = 24)

Literature Review Matrix

Title (hyperlink embedded)	Author, Year	Summary	AI Model(s) or Method(s) Discussed	Theme Category
How Big Data and Artificial Intelligence Can Help Better Manage the COVID-19 Pandemic [Review]	Bragazzi et al., 2020	The review is aimed at providing an overview of potential applications of AI and Big Data to manage the pandemic, such as i) a quick and effective pandemic alert; ii) tracking and diagnosis; 3) identifying pharmacological treatment; 4) facilitating implementation of PH interventions; 5) building smart, health, resilient cities.	Susceptible Exposed Infectious Removed (SEIR)	1,2,3,4
Application of machine learning and artificial intelligence for Covid-19 (SARS-CoV-2) pandemic: A Review	Lalmuanawma et al., 2020	The review is aimed at providing an overview of AI and ML as methods for screening, predicting, forecasting, contact tracing, and drug development in relation to COVID-19. The accuracy and specificity of these methods were also assessed.	MYCIN	1,2,3,4
Artificial Intelligence for COVID-19: Rapid Review	Chen & See, 2020	The review is aimed at discussing AI applications being deployed at various levels of the health care system in response to COVID-19.	COVNet	1,2,3,4
A Survey on Applications of Artificial Intelligence in Fighting against COVID-19	Chen et al., 2022	The survey investigates the main uses of AI to combat COVID-19. In addition, a summary of uses for AI-based COVID-19 research as well as the main challenges and potential directions of AI is discussed.	Various	1,2,3,4
Artificial intelligence for the novel corona virus (COVID-19) pandemic: Opportunities, challenges, and future directions	Ahmed et al., 2022	The article discusses the uses of specific subsets of AI techniques being used to battle COVID-19, such as machine learning (ML), deep learning (DL), and natural language processing (NLP).	Machine Learning (ML), Deep Learning (DL), Natural Language Processing (NLP)	1,2,3,4
Deep Learning applications for COVID-19	Shorten et al., 2021	The article provides applications of deep learning to combat COVID-19.	NLP, Recurrent Neural Networks (RNNs), AlexNet	1,2,3,4
The Role of Augmented Intelligence (AI) in Detecting and Preventing the Spread of Novel Coronavirus	Long & Ehrenfeld, 2020	The article provides applications of augmented intelligence to combat COVID-19, such as prediction, prevention, and detection of future global health risks.	Google Flu, HealthMap	1,2,3,4
The role of artificial intelligence in tackling COVID-19	Arora et al., 2020	The article concluded that AI plays a vital role in the fight against COVID-19, and provides specific methods being used.	HealthMap, AI4COVID-19	1,2,3,4
Artificial intelligence vs COVID-19: limitations, constraints, and pitfalls.	Naudé, 2020	The article concluded that AI has not yet been impactful against COVID-19. It further purported that the use of AI is inhibited either by a cumbersome or limited amount of data, and that it will require a balance between data privacy and human-AI interaction.	HealthMap	1,2,3,4
Artificial Intelligence (AI) applications for COVID-19 pandemic [Review]	Vaishya et al., 2020	The review focuses on the applications of AI in healthcare delivery to combat COVID-19 and other pandemics.	N/A	1,2,3
Machine Learning and Image Analysis Application in the Fight against COVID-19 Pandemic: Datasets, Research Directions, Challenges and Opportunities	Somasekar et al., 2020	The articles discusses how machine learning and image analysis research can provide support to subsequent medical diagnosis.	ML	1,2,4
Infodemic Signal Detection During the COVID-19 Pandemic: Development of a Methodology for Identifying Potential Information Voids in Online Conversations	Purnat et al., July 2021	The article explains how an infodemic results in risk-taking behaviors and mistrust in health authorities and public health responses. It also discusses an AI-based signal detection tool developed by WHO that will enable prioritization of health response activities in relation to COVID-19.	EARS	1,2,4
EARS - A WHO Platform for AI-Supported Real-Time Online Social Listening of COVID-19 Conversations	Purnat et al., May 2021	The article explains how an infodemic results in risk-taking behaviors and mistrust in health authorities and public health responses. It also discusses an AI-based social listening tool developed by WHO that will enable prioritization of health response activities in relation to COVID-19.	EARS	1,2,4
Application of Artificial Intelligence in COVID-19 drug repurposing [Review]	Mohanty et al., 2020	The article discusses the application of AI in COVID-19 drug repurposing and the benefits.	Various	1,3,4
Predicting commercially available antiviral drugs that may act on the novel coronavirus (SARS-CoV-2) through a drug-target interaction deep learning model	Beck et al., 2020	The article discusses the use of a deep learning model called Molecule Transformer-Drug Target Interaction (MT-DTI) to identify commercially available drugs that could act on viral proteins of COVID-19.	Molecule Transformer-Drug Target Interaction (MT-DTI)	1,3,4
How artificial intelligence may help the Covid-19 pandemic: Pitfalls and lessons for the future	Malik et al., 2021	The article provides an overview of the prospective applications of AI-based model systems in healthcare settings during the pandemic.	N/A	1,2
AI for COVID-19: Battling the pandemic with computational intelligence. [Review]	Xu et al., 2021	The study aims to review existing studies on applications of AI techniques to combat the COVID-19 pandemic.	N/A	1,2
How Might Artificial Intelligence Applications Impact Risk Management?	Banja, 2020	The article discusses the substantial risks with AI applications, such as malfunctions, privacy concerns, and consent to data.	N/A	1,2
Utilization of machine-learning models to accurately predict risk for critical COVID-19	Assaf et al., 2020	The article discusses three different machine learning models used to predict patient risk for COVID-19 to improve hospital management and resources.	Neural Networks (NN), Classification and Regression Decision Tree (CRT)	2,4
Artificial intelligence, public control, and supply of a vital commodity like COVID-19 vaccine.	Tsyganov, 2021	The article discusses AI applications in the context of a vaccine shortage to improve and public control.	Various	3,4
Artificial intelligence and machine learning approaches for drug design: challenges and opportunities for the pharmaceutical industries. [Review]	Selvaraj et al., 2021	The article discusses AI and ML methods used to support drug design as well as how these methods present challenges and opportunities for the pharmaceutical industry.	Computer-Aided Drug Designing (CADD), NN	3,4
Regulatory responses to medical machine learning.	Minssen et al., 2020	The article examines implications for medical machine learning (MML) as it relates to data privacy and other regulatory related responses.	AI/ML-software as a medical device (AI/ML-SaMD)	3,4
A framework to support the integration of priority setting in the preparedness, alert, control and evaluation stages of a disease pandemic	Kapiriri et al., 2021	The article aims to discuss how priority setting informs the COVID-19 pandemic planning and preparedness frameworks.	N/A	2
Principles of supply chain management in the time of crisis	Okeagu et al., 2021	The article discusses the principles of supply chain management in light of COVID-19 and how increased transparency and improved technology can prevent shortages.	N/A	3

Applications of AI Methods or Models

Seven predominant applications were identified across the articles that were being used to improve pandemic preparedness: i) detection and diagnosis of COVID-19; ii) monitoring treatment of patients; iii) contact tracing; iv) forecasting; v) development of drugs and vaccines; vi) alleviating the burden of healthcare workers; and vii) prevention, mitigation, and response strategies. Many of these applications have the potential to be cost-effective when deployed. One thing to note when discussing the applications of AI is the difference between supervised and unsupervised machine learning. Supervised machine learning aims to develop a predictive algorithm using regression (whether linear or multiple) or classification methods (i.e., decision trees or neural networks), which is often used in forecasting. Unsupervised machine learning allows computers to explore large amounts of unclassified data to discover novel diseases or treatment patterns (Chen & See 2020).

When it comes to the pandemic, this critical analysis first aimed to identify AI methods or models that were being used to improve prevention, mitigation, and response strategies generally; AI methods or models identified in the literature will be discussed below according to seven main applications. With respect to prevention, AI is currently being used to improve the training of emergency preparedness staff and officials, the management of the health supply chain, and the operation of surveillance systems being used to study COVID-19. When it comes to mitigation, AI is being used for risk modelling, big data, and forecasting. Lastly, in response to the current impacts of the pandemic, AI is being used for drug identification and repurposing and reverse vaccinology.

I. Detection and Diagnosis of COVID-19

AI is a power tool for detecting and diagnosing COVID-19 as it quickly recognizes patients’ symptoms and efficiently analyzes the results of medical imaging technologies (Vaishya et al. 2020). COVNet, for example, is an automatic 3D deep-learning framework developed for the detection of COVID-19 and can extract relevant information from computed tomography (CT) scans to generate a probability score that distinguishes between patients with COVID-19 or pneumonia. This model can process each CT scan in under five seconds on average (Chen & See 2020). The ability to diagnose patients quickly and accurately is very cost-effective because it allows for proper allocation of health resources.

II. Monitoring Treatment of Patients

As AI is being introduced into the healthcare field to aid clinical decision-making and monitor patients, it also has the capability to provide daily updates and pharmacological solutions to physicians about their patients’ treatments (Vaishya et al. 2020; Bragazzi et al. 2020). MYCIN was the first medical expert system, developed in 1976, to assist physicians with treating bacterial infections by incorporating 450 medical expert rules on suggesting certain antibiotics to patients (Lalmuanawma et al. 2020). The evidence suggested that using a machine learning approach with this system can support the treatment of communicable diseases, such as COVID-19, and is also cost-effective.

III. Contact Tracing

There are various AI-based tools currently being used to track the spread of COVID-19 and identify hot spots in real-time, such as HealthMap and AI4COVID-19. HealthMap is led by a team of epidemiologists and developers at Boston Children’s Hospital to monitor and collect publicly available data on COVID-19 to effectively track its spread (Long & Ehrenfeld 2020). The use of aggregated big data in AI-based technology is promising for epidemiological trending but can result in underperformance if poorly designed. An example of such failure was Google Flu, which reported substantial errors in trend estimates (Long & Ehrenfeld 2020). AI4COVID-19 a mobile health application that records three second audio samples of coughs and sends the recordings to an AI cloud engine to provide a preliminary diagnosis of COVID-19 within two minutes (Arora et al. 2020). Since it can be used in wearable devices, it can be employed for effective monitoring and contact tracing as well as diagnosis. Digital contact tracing methods that identify and manage people who are recently exposed helps to avoid further spread and use of health resources; thus, being cost-effective.

IV. Forecasting

AI-based methods have been helpful for epidemiological forecasting of COVID-19 spread by applying deep learning to socio-medical informatics and engineering (Shorten et al 2021). Big Data is often used for forecasting and defined by velocity, volume, and variety because it is comprised of large amounts of data from different sources and can be acquired, processed, and manipulated at incredible speed (Bragazzi et al. 2020). Evidence suggests that when combined with an AI approach and trained on COVID-19 data, like the Susceptible Exposed Infectious Removed (SEIR) model, epidemiologists can predict the pandemic curve, allowing for case detections six to ten days in advance.

V. Development and Vaccines

When it comes to the research and development of drugs and vaccines, AI has many applications and benefits. One main benefit is that it can identify already existing antiviral drugs that can combat COVID-19 using a deep learning model known as the Molecule Transformer-Drug Target Interaction (MT-DTI) model (Chen & See 2020). There are other uses of AI as well, such as drug repurposing/repositioning and discovering new drug-like compounds by constructing new molecular structures that inhibit viral proteases at the molecular level (Mohanty et al. 2020). Computational drug repurposing has shown great potential for treating COVID-19 infections in some studies, yet the main challenge is properly diagnosing COVID-19 patients and identifying the relationship between the vaccines being developed and the disease it intends to treat (Xu et al. 2021). However, AI-based models have the capability to address these challenges. AI has proven to be useful for the design of drug delivery and speeding the testing process for COVID-19 vaccines in real-time due to the fact that AI can diagnose and monitor treatment of patients much faster than standard diagnostic testing (Vaishya et al. 2020).

VI. Alleviating the Burden of Healthcare Workers

As previously mentioned, AI can help to improve the detection and diagnosis of patients while also monitoring their treatment in real-time. This both decreases physician workload and allows for a greater volume of patients to be evaluated at a given time; ultimately, facilitating more rapid case detection and treatment. Alleviating the burden of healthcare workers is arguably one of the main benefits of the utilization of AI in the U.S. healthcare system. Due to the everchanging impacts of COVID-19, healthcare workers have experienced burnout which affects the quality of patient care (Vaishya et al. 2020).

VII. Prevention, Mitigation and Response Strategies

AI has proven to be a valuable tool in public health efforts, helping to characterize the epidemiology of COVID-19 and model disease transmission, even in the early stages of the pandemic. The application of AI in this area would be particularly helpful for policy makers in addition to clinical decision making by individual clinicians. Aside from this, other potential applications of AI consist of a real-time alert system, successful implementation of public health interventions, and standardizing emergency protocols that will result in healthier and more resilient cities (Bragazzi et al. 2020). Comparing the utilization of AI versus non-AI based approaches to identify and treat COVID-19 patients could be used as a conceptual approach to improving WHO’s pandemic phases (Vaishya et al. 2020).

Strengths & Limitations of AI-based Technology

As discussed, AI methods and models are successful at improving pandemic preparedness, but challenges and limitations will arise as we continue to invest in AI-based technology. This critical analysis also aimed to address the main challenges in applying predictive models to support disease forecasting and surveillance, and to address vaccine development and current medical supply issues. These challenges are i) the generalizability of AI-based technology, ii) collecting multiple types of data in datasets, iii) the lack of infrastructure, iv) imbalanced data in the datasets, and v) false predictions (Somasekar et al. 2020).

I. The Generalizability of AI-based Technology

The generalizability of AI-based technology relies on the extent to which it can produce accurate information that can be used for future decision making. However, certain subsets of AI are not as generalizable as others and should be considered for future COVID-19 research. The Deep Convolutional Neural Network, also known as ResNet-101, has the highest accuracy and specificity in COVID-19 screenings with rates over 99%; whereas the Support Vector Machine and DarkCOVIDNet, which does not apply a deep learning approach, in contrast to machine learning, only has 77.5% and 87–98% accuracy, respectively (Lalmuanawma et al. 2020).

II. Collecting Multiple Types of Data in Datasets

Although AI has the capacity to handle and compare large amounts of clinical and medical data, it must be trained on collecting multiple types of data in datasets in real-time to be the most accurate (Somasekar et al 2020). Collecting data on COVID-19 can be difficult if there are delays in updating the data, such as with the CDC’s surveillance system. If open datasets cannot be updated daily or if the data is poor, the results produced by AI models cannot be as accurate as desired.

III. The Lack of Infrastructure

Most of the time, the failure of AI projects is due to the cost of operation and the lack of infrastructure to support it (Dilmegani 2021). Therefore, more investment in AI is needed for it to be successful in carrying out the purpose intended. As it becomes mainstream practicce, researchers and innovators hope to see more successful AI projects.

IV. Imbalanced Data in the Datasets

Similar to poor or delayed data, missing or null values in a dataset can affect the accuracy of results produced by AI models. This is often seen in risk prediction datasets and fatality prediction datasets (Somasekar et al. 2020). Without complete and timely data, AI models cannot do their job.

V. False Predictions

When it comes to predictive models, a greater number of null samples images in medical imaging datasets results in false predictions while testing the model (Somasekar et al. 2020). Under- and over-performance is a big challenge when testing AI models and something to consider with COVID-19 detection and forecasting. A notorious example of underperformance due to poor design was Google Flu, which was a model that utilized search engine queries to improve the tracking of the flu epidemic (Long & Ehrenfeld 2020).

Recommendations

As a final aim, this critical analysis aimed to briefly discuss promising practices and recommendations for using AI-based technology for pandemic preparedness in the U.S. Overall, AI investments should go towards improving emergency preparedness strategies, mitigation efforts, and vaccine development, while alleviating the burdens experienced by healthcare workers to achieve the best results. Compared to other viruses, AI-based technology has been successful when applied to fight COVID-19, especially in detection and diagnosis, drug and vaccine development, and forecasting (Chen et al. 2022). Although this critical analysis was able to meet all aims by providing an overview of the current applications of AI-based technology and its strengths and limitation, it is limited in determining the actual success of AI to prevent the next pandemic outside of a theoretical perspective. First, a more thorough review of research opportunities and emerging trends is required. To determine whether the utilization of AI can in fact prevent the next pandemic after such review, however, will require successful implementation and evaluation of these AI-based technologies in the U.S. healthcare system.

To my best knowledge, I do not have any financial or non-financial interests that are directly or indirectly related to the work submitted for this publication. Additionally, I declare that the authors have no competing interests as defined by Springer, or other interests that might be perceived to influence the results and/or discussion reported in this paper.

J.H. wrote the manuscript text, reviewed the manuscript, and prepared all figures. Dr. Elizabeth Andrade, Professor at GWU, conducted a peer review of the manuscript before submission. None of the date or figures in this manuscript have not been published elsewhere, nor are they under consideration (from you or one of your Contributing Authors) by another publisher. Lastly, no funding was requested for the research conducted in preparation of this manuscript.

Ahmed A, Boopathy P, Sudhagara Rajan S (2022) Artificial intelligence for the novel corona virus (COVID-19) pandemic: Opportunities, challenges, and future directions. Int J E-Health Med Commun 13(2). doi:10.4018/IJEHMC.20220701.oa5
Arora N, Banerjee AK, Narasu ML (2020) The role of artificial intelligence in tackling COVID-19. Future Virol. https://doi.org/10.2217/fvl-2020-0130. 10.2217/fvl-2020-0130
Assaf D, Gutman Y, Neuman Y, Segal G, Amit S, Gefen-Halevi S, Shilo N, Epstein A, Mor-Cohen R, Biber A, Rahav G, Levy I, Tirosh A (2020) Utilization of machine-learning models to accurately predict the risk for critical COVID-19. Intern Emerg Med 15(8):1435–1443. https://doi.org/10.1007/s11739-020-02475-0
Banja J (2020) How Might Artificial Intelligence Applications Impact Risk Management? AMA J Ethics 22(11):E945–E951. https://doi.org/10.1001/amajethics.2020.945
Beck BR, Shin B, Choi Y, Park S, Kang K (2020) Predicting commercially available antiviral drugs that may act on the novel coronavirus (SARS-CoV-2) through a drug-target interaction deep learning model. Comput Struct Biotechnol 18:784–790. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7118541/pdf/main.pdf
Bragazzi NL, Dai H, Damiani G, Behzadifar M, Martini M, Wu J (2020) How Big Data and Artificial Intelligence Can Help Better Manage the COVID-19 Pandemic. Int J Environ Res Public Health 17(9):3176. https://doi.org/10.3390/ijerph17093176
Center for Disease Control and Prevention (2021), September 17 CDC to Invest $2.1 Billion to Protect Patients and Healthcare Workers from COVID-19 and Future Infectious Diseases [Press Release]. Retrieved from https://www.cdc.gov/media/releases/2021/p0917-COVID-19-funding.html
Chen J, See KC (2020) Artificial Intelligence for COVID-19: Rapid Review. J Med Internet Res 22(10):e21476
Chen J, Li K, Zhang Z, Li K, Yu PS (2022) A survey on applications of artificial intelligence in fighting against COVID-19. ACM-CSUR 54(8). doi:10.1145/3465398
Dilmegani C(2021) 4 Reasons for Artificial Intelligence (AI) Project Failure in 2022.AI Multiple.https://research.aimultiple.com/ai-fail/
Equbal A, Masood S, Equbal I, Ahmad S, Khan NZ, Khan ZA (2021) Artificial intelligence against COVID-19 Pandemic: A Comprehensive Insight. Curr Med imaging. 10.2174/1573405617666211004115208
Garcia M, Lipskiy N, Tyson J, Watkins R, Esser ES, Kinley T (2020) Centers for Disease Control and Prevention 2019 novel coronavirus disease (COVID-19) information management: addressing national health-care and public health needs for standardized data definitions and codified vocabulary for data exchange. J Am Med Inf Association: JAMIA 27(9):1476–1487. https://doi.org/10.1093/jamia/ocaa141
Haas Q, Alvarez DV, Borissov N, Ferdowsi S, von Meyenn L, Trelle S, Teodoro D, Amini P (2021) Utilizing Artificial Intelligence to Manage COVID-19 Scientific Evidence Torrent with Risklick AI: A Critical Tool for Pharmacology and Therapy Development. Pharmacology 106(5–6):244–253. https://doi.org/10.1159/000515908
Jung SY, Jo H, Son H, Hwang HJ (2020) Real-World Implications of a Rapidly Responsive COVID-19 Spread Model with Time-Dependent Parameters via Deep Learning: Model Development and Validation. J Med Internet Res 22(9):e19907. https://doi.org/10.2196/19907
Kapiriri L, Essue B, Bwire G, Nouvet E, Kiwanuka S, Sengooba F, Reeleder D (2021) A framework to support the integration of priority setting in the preparedness, alert, control and evaluation stages of a disease pandemic. Glob Public Health 1–13 Advance online publication. https://doi.org/10.1080/17441692.2021.1931402
Lalmuanawma S, Hussain J, Chhakchhuak L (2020) Applications of machine learning and artificial intelligence for Covid-19 (SARS-CoV-2) pandemic: A review. Chaos Solitons Fractals 139:110059. https://doi.org/10.1016/j.chaos.2020.110059
Lindsay BR(2012) Federal Emergency Management: A Brief Introduction. Congressional Research Services. Available from: https://training.fema.gov/hiedu/highref/federal%20em-a%20brief%20introduction-r42845%20-%20lindsay.pdf
Long JB, Ehrenfeld JM (2020) The Role of Augmented Intelligence (AI) in Detecting and Preventing the Spread of Novel Coronavirus. J Med Syst 44(3):59. https://doi.org/10.1007/s10916-020-1536-6
Malik YS, Sircar S, Bhat S, Ansari MI, Pande T, Kumar P, Mathapati B, Balasubramanian G, Kaushik R, Natesan S, Ezzikouri S, El Zowalaty ME, Dhama K (2021) How artificial intelligence may help the Covid-19 pandemic: Pitfalls and lessons for the future. Rev Med Virol 31(5):1–11. https://doi.org/10.1002/rmv.2205
Minssen T, Gerke S, Aboy M, Price N, Cohen G(2020) Regulatory responses to medical machine learning, Journal of Law and the Biosciences, Volume 7, Issue 1, January-June 2020, lsaa002, https://doi.org/10.1093/jlb/lsaa002
Mohanty S, Harun Ai Rashid M, Mridul M, Mohanty C, Swayamsiddha S (2020) Application of Artificial Intelligence in COVID-19 drug repurposing. Diabetes & metabolic syndrome 14(5):1027–1031. https://doi.org/10.1016/j.dsx.2020.06.068
Naudé W (2020) Artificial intelligence vs COVID-19: limitations, constraints and pitfalls. AI Soc 1–5 Advance online publication. https://doi.org/10.1007/s00146-020-00978-0
OIG Strategic Plan: Oversight of COVID-19 Response and Recovery (2020) Department of Health and Human Services. Available from: https://oig.hhs.gov/documents/root/166/COVID-OIG-Strategic-Plan.pdf
Okeagu CN, Reed DS, Sun L, Colontonio MM, Rezayev A, Ghaffar YA, Kaye RJ, Liu H, Cornett EM, Fox CJ, Urman RD, Kaye AD (2021) Principles of supply chain management in the time of crisis. Best practice & research. Clin Anesthesiology 35(3):369–376. https://doi.org/10.1016/j.bpa.2020.11.007
Ong E, Wong MU, Huffman A, He Y (2020) COVID-19 Coronavirus Vaccine Design Using Reverse Vaccinology and Machine Learning. Front Immunol 11:1581. https://doi.org/10.3389/fimmu.2020.01581
Pandemic Influenza Preparedness and Response: A WHO Guidance Document. Geneva: World Health Organization, THE WHO PANDEMIC PHASES (2009) ; 4,. Available from: https://www.ncbi.nlm.nih.gov/books/NBK143061/
Purnat TD, Vacca P, Czerniak C, Ball S, Burzo S, Zecchin T, Wright A, Bezbaruah S, Tanggol F, Dubé È, Labbé F, Dionne M, Lamichhane J, Mahajan A, Briand S, Nguyen T (2021) Infodemic Signal Detection During the COVID-19 Pandemic: Development of a Methodology for Identifying Potential Information Voids in Online Conversations. JMIR Infodemiology 1(1):e30971. https://doi.org/10.2196/30971
Reports and Data (2020) Artificial Intelligence In Drug Discovery Market Analysis, By Drug Type (Small Molecule, Large Molecule), By Offerings (Software, Services), By Technology (Deep Learning, Machine Learning), By Applications, By End Use And By Region Forecasts To 2027. Available from: https://www.reportsanddata.com/report-detail/artificial-intelligence-in-drug-discovery-market
Selvaraj C, Chandra I, Singh SK(2021) Artificial intelligence and machine learning approaches for drug design: challenges and opportunities for the pharmaceutical industries. Molecular diversity, 1–21. Advance online publication. https://doi.org/10.1007/s11030-021-10326-z
Shorten C, Khoshgoftaar TM, Furht B (2021) Deep learning applications for COVID-19. J Big Data 8(1). doi:10.1186/s40537-020-00392-9
Somasekar J, Pavan Kumar Visulaization P, Sharma A, Ramesh G(2020) Machine Learning and Image Analysis Applications in the Fight against COVID-19 Pandemic: Datasets, Research Directions, Challenges and Opportunities. Materials today. Proceedings, 10.1016/j.matpr.2020.09.352. Advance online publication. https://doi.org/10.1016/j.matpr.2020.09.352
Tsyganov V(2021) Artificial intelligence, public control, and supply of a vital commodity like COVID-19 vaccine [Open Forum]. AI & Society (2021). https://doi.org/10.1007/s00146-021-01293-y
Vaishya R, Javaid M, Khan IH, Haleem A (2020) Artificial Intelligence (AI) applications for COVID-19 pandemic. Diabetes & metabolic syndrome 14(4):337–339. https://doi.org/10.1016/j.dsx.2020.04.012
Welch P, Porter K, Doyle M(2020) Congressional Letter to the U.S. Department of Health and Human Services regarding Operation Warp Speed (OWS). Available from: https://welch.house.gov/sites/welch.house.gov/files/OWS%20letter%207.23.2020.pdf
Xu Z, Su C, Xiao Y, Wang F(2021) AI for COVID-19: Battling the pandemic with computational intelligence. Intelligent medicine, 10.1016/j.imed.2021.09.001. Advance online publication. https://doi.org/10.1016/j.imed.2021.09.001

Appendix of Search Terms

"Artificial Intelligence" AND "COVID-19"

"Augmented Intelligence" AND "COVID-19"

"Artificial Intelligence" AND "data privacy"

"Artificial Intelligence" AND "pandemic"

"Artificial Intelligence" AND "emergency preparedness"

"COVID-19" AND "data privacy"

"machine learning" AND "pandemic"

"pandemic" AND "emergency preparedness"

Appendix of Acronyms & Definitions

AI/ML-SaMD (AI/ML-software as a medical device) = software used outside of a medical device, in contrast to software in a medical device.
AI4COVID-19 – a mobile health application that records three second audio samples of coughs and sends the recordings to an AI cloud engine to provide a preliminary diagnosis of COVID-19 within two minutes.
AlexNet = utilizes Convolutional Neural Network to boost performance of computer vision tasks.
CADD (Computer-Aided Drug Designing) = computational and theoretical algorithm for developing new therapeutics leads.
COVNet – a fully automatic 3D deep-learning framework developed for the detection of COVID-19 and can extract relevant information from CT scans to generate a probability score that distinguishes whether a patient has COVID-19 or pneumonia. This model can process each CT scan in under five seconds on average.
CRT (Classification and Regression Decision Tree) = algorithm used for classification and regression.
EARS = An AI-supported platform developed by WHO and used for real-time online social listening of COVID-19 conversations.
DL (deep learning) = a subset of machine learning that is based on AI neural networks and extracts high-level features from raw data.
Google Flu = a model that utilized search engine queries to improve the tracking of the flu epidemic; famously known for underperformance due to poor design.
HealthMap – collects publicly available data on COVID-19 to effectively track its spread.
ML (machine learning) = studies computer algorithms to improve automatically by the use of data and experience.
MT-DTI (Molecule Transformer-Drug Target Interaction) – identifies commercially available drugs that could act on viral proteins of COVID-19.
MYCIN – a system developed in 1976 and designed to use 450 rules collected from medical experts to treat infections by suggesting treatments to patients.
NLP (natural language processing) = a branch of AI or computer science that allows computers to understand text and spoken word.
NN (neural networks) = trained computing systems that consists of input and results.
RNNs (recurrent neural networks) = a type of NN that consists of temporal sequencing.
SEIR (Susceptible Exposed Infectious Removed) – a model combined with an AI approach and trained on the SARS data to predict the pandemic curve.

No competing interests reported.

Download PDF

Version 1

posted

You are reading this older preprint version

Read the latest preprint version →

Preventing the Next Pandemic: A Critical Analysis of the Utilization of Artificial Intelligence in the U.S. Healthcare System

Status:

Version 1

Abstract

Background

Methods

Results

Discussion

Applications of AI Methods or Models

I. Detection and Diagnosis of COVID-19

II. Monitoring Treatment of Patients

III. Contact Tracing

IV. Forecasting

V. Development and Vaccines

VI. Alleviating the Burden of Healthcare Workers

VII. Prevention, Mitigation and Response Strategies

Strengths & Limitations of AI-based Technology

I. The Generalizability of AI-based Technology

II. Collecting Multiple Types of Data in Datasets

III. The Lack of Infrastructure

IV. Imbalanced Data in the Datasets

V. False Predictions

Recommendations

Statements and Declarations

References

Appendix

Additional Declarations

Status:

Version 1