In a recent commentary published on DeepAI, “Artificial Intelligence-based Clinical Decision Support for COVID-19 – Where Art Thou?,” faculty members from the Malone Center for Engineering in Healthcare discuss the notable absence of AI-based clinical decision support (CDS) in the early phases of the COVID-19 pandemic, and identify opportunities to improve “AI readiness” for future healthcare challenges.
In the article, the team explains that prior to the COVID-19 outbreak, artificial intelligence and machine learning (AI/ML) for healthcare had finally reached a high level of maturity, performance, and reliability. Now, in a global pandemic that is putting an unprecedented strain on our healthcare system, AI-assisted CDS is arguably more important than ever. Such tools could certainly assist clinicians with diagnosing, triaging, treating, and monitoring COVID-19 patients.
Why, then, are AI-assisted CDS tools seemingly limited in contributing to the fight against COVID-19?
According to the commentary, COVID-19 has revealed some of the major barriers that impede the rapid deployment of new AI solutions. For example, using AI to screen and treat COVID-19 is hampered due to a lack of historical training data.
The team writes, “While the application of AI-based CDS for COVID-19 has been limited to date, this is somewhat understandable. AI is highly dependent on sufficiently diverse, representative data; AI/ML algorithms cannot be trained and adequately validated without creating large data sets that reflect the clinical use case, and such data will always be limited in the early phases of an emerging infectious disease.”
Yet, despite the lack of data, there are opportunities to improve data collection for AI-based CDS development. Because they require large amounts of medical record data, these studies must be reviewed by the respective Institutional Review Board (IRB), which are established to protect the welfare, rights, and privacy of human subjects. In light of COVID-19, many new studies are being submitted for review to IRBs, leading to delays that prevent AI work from moving forward. To promote “AI readiness,” the team calls for the implementation of IRB sub-committees dedicated to AI algorithm development. This will allow for rapid turnaround review of data science-related projects, while still complying with all ethical standards.
The authors also suggest that healthcare organizations should invest in a robust data infrastructure, if they hope to benefit from AI innovations. AI tools are only empowered by a reliable data infrastructure that is accessible to researchers in a timely manner. This access should be significantly expedited by establishing strong pathways for data sharing and usage, including the standardization of data formats.
The authors say that as the pandemic progresses and more data is acquired, becomes available for research, and is shared, it’s likely that AI’s contribution to combating the COVID-19 crisis will grow.
“Once the dust of this first wave of the pandemic has settled, we, as a community, should spend some time to identify and analyze the organizational, institutional, or regulatory hurdles that this healthcare crisis has highlighted, as well as the solution paths that emerged to bring AI-based CDS systems for COVID-19 to the bedside,” write the authors.
Faculty who contributed to the commentary include:
Mathias Unberath*, Kimia Ghobadi, Scott Levin, Jeremiah Hinson, Gregory D Hager