In a recently published study in PLOS Digital HealthThe researchers reviewed the existing literature on the use of artificial intelligence (AI) in healthcare to characterize the AI applications used in the clinical applications during the 2019 coronavirus disease (COVID-19) pandemic and examined the location the timing and extent of the use of AI in healthcare and examine the regulatory approval processes in the United States (USA).
background
Despite the large number of US Food and Drug Administration (FDA) approvals for AI applications in healthcare over the past six years, the uptake of AI applications in various healthcare settings has been limited. Additionally, in contrast to the significant and rapid growth of telemedicine and vaccine technologies, there is limited information on the development and use of AI applications during the COVID-19 pandemic.
While previous reviews have reviewed the potential uses, challenges, and implications of AI applications for clinical response to COVID-19, many of the reviews found methodological flaws and potential biases in the use of AI applications in clinical practice. A Lack of Reviews provides a comprehensive account of the development, testing, and application of AI in clinical responses to COVID-19.
About the study
In the present scoping review, researchers searched the academic and gray literature for studies that would provide answers to the following four questions: 1) What AI applications are used in clinical responses to COVID-19? 2) What are the locations, schedules and scope of use of these applications; 3) How are these uses different from pre-pandemic health technology and what are the US FDA’s stringent approval criteria for these uses? and 4) What publicly available evidence recommends the use of AI applications in healthcare?
The study began by consulting healthcare stakeholders such as physicians, patient advocates, insurers and healthcare system representatives, researchers, policy makers, industry representatives and public health officials for recommendations on study design and documents to include in the review.
Various databases were then searched for literature available after January 2020 on the use of AI in clinical responses to COVID-19, and the AI applications identified from the literature review were examined in detail for additional developer and usage information.
Applications were included in the assessment if they met three criteria. First, the application had a patient health-related function as part of the patient development, diagnosis, decision-making, or treatment process. Applications that were part of drug development or biomedical research were not considered.
Second, the AI application has been directly involved in the clinical response to COVID-19. Applications available through government health websites and medical testing and symptom assessment websites for even limited periods qualified for the review. Finally, the application used artificial intelligence, machine learning or deep learning algorithms.
The extent of the use of the AI application was determined by the number of patients treated with the use of the AI application in the clinical response to COVID-19. The Organization for Economic Co-operation and Development’s classification into high-, middle-, and low-income countries was used to determine the location of AI application usage.
Results
Results reported the use of 66 AI applications in clinical response to COVID-19, grouped into six functional categories. The Lung Assessment AI applications evaluated computed tomography (CT) scans or X-rays, or both, for the presence of pneumonia, pneumothorax, or other lung abnormalities due to COVID-19. Symptom checker AI applications used patient-supplied demographics, risk factors, and symptoms to calculate COVID-19 risks and make health recommendations.
Patient deterioration AI applications monitored the vital signs and health status of COVID-19 patients and provided information to make healthcare decisions. These were used by doctors in hospitals, assisted living facilities and to monitor patients who were isolated in their homes.
Several applications predicted the likelihood of COVID-19 infections from various sources of information, such as B. breath volatile organic compounds, geographically grouped data, patient demographics, blood test results, and luminescent signals from antigen test strips.
Some applications used demographic and medical record data to predict the risk of serious consequences from COVID-19 and have been used by doctors in hospitals, telemedicine facilities and outpatient clinics. Other applications performed various functions such as B. Image acquisition, treatment response prediction, immune response detection, etc.
These applications used neural networks, advanced tree-based methods, and supervised and unsupervised machine learning methods. A large number of the applications were deployed between January and June 2020, during the initial stages of the pandemic, and were mainly used in high-income countries such as the US and China, while very few were used in middle- and low-income countries.
No clinical studies evaluating the use of AI applications in responding to COVID-19 have been found, and the few publications supporting the use of some of the applications have been independent assessments.
Conclusions
In summary, the scoping review reported the location, function, potential benefit, scope, nature, and input data of AI applications used in clinical practice during the COVID-19 pandemic. While there is an abundance of scholarly literature on AI models, there is a dearth of scholarly reports on the use of these AI models in clinical practice.
The limited evidence in the literature makes it difficult to determine the benefits of using AI in pandemic response efforts. More studies on the real-world applications of AI in healthcare are needed.
