Tales from the Healthcare AI Front: Applications and Evaluations
Discussions about artificial intelligence are everywhere in healthcare circles today, though the levels of adoption and comfort with this technology vary widely among digital and...
By Wendy Kerschner and Scott Dethloff
Discussions about artificial intelligence are everywhere in healthcare circles today, though the levels of adoption and comfort with this technology vary widely among digital and technology leaders, healthcare CEOs and executives, boards, clinicians, the vendor community, and other interested parties. Despite the diversity in understanding and utilization, it is undeniable that AI’s influence is here to stay. Across the healthcare spectrum, organizations are actively determining the most effective ways to integrate AI into their operations. This was particularly evident at the recent AMIA 2023 Annual Symposium in New Orleans, where shared use cases were abundant and spanned clinical, operational, and research applications.
From an operational perspective, there were examples on how organizations are using AI to prioritize patient portal messages:
- The in-basket burden is an issue contributing to clinician burnout. A team from New York University shared their success in implementing an AI model to identify the most critical emails for doctors and staff to respond to, after determining that only 6 to 10% of messages describe an urgent medical need.
- Another session focused on many examples of how AI is being leveraged to review conversations between medical professionals and patients captured by ambient listening devices and recommend diagnoses based on the conversations.
- A research team from Ohio State University shared their work in developing an AI-based chatbot to screen for social determinants of health in patients. They cited studies suggesting that up to 80% of health outcomes are influenced by social determinants of health yet current methods to obtain this information are limited and costly.
- At Memorial Sloan Kettering, clinical teams have established a new AI-based workflow based on AI driven data integration, clinical event monitoring, and a patient’s long-term history to develop and schedule radiation therapies for cancer treatments.
The innovative applications of AI in healthcare operations showcased in these sessions highlight a transformative shift towards more efficient and targeted patient care.
For those who are hesitant to embrace the science behind AI, there’s reassurance in the rigorous methodologies leaders are using to evaluate its applications. Conversely, concerns about the reliability of AI-driven queries were the subject of several AMIA presentations which highlighted the healthy skepticism of the clinical and informatics community.
- These concerns include the “black box” effect in that clinicians will never fully trust AI algorithms because they are unsure of the data inputs and how decisions are made from them. Examples given included how the use of different medical terminology in AI can significantly affect answers. An example would be the imprecise use of terms like “chronic fatigue syndrome” versus “myalgic encephalomyelitis” in queries that can affect the quality of data that algorithms produce.
- Other concerns expressed included the opposite, that clinicians may place too much trust in AI, or that health equity disparities could amplify inequalities in care delivery, for example in prosperous urban areas versus resource constrained rural areas.
- The risk of malpractice by relying on AI was also highlighted as a concern, in addition to potential for cybersecurity attacks that could manipulate data.
- The ability (or inability) of regulatory agencies to keep up with guidelines around the use of AI was also an area of discussion.
In short, the medical informatics community continues to grapple with the potential ramifications of AI technology.
Who’s in Charge?
A couple of years ago at a conference we attended, a speaker mentioned that our society was in the Middle Ages in terms of where AI was in its development. Today, there is no doubt that AI’s evolution will persist. From a leadership perspective, the primary question is how organizations will manage these new tools. We see many forward-thinking health systems now have a Chief AI Officer. This leader oversees enterprise AI/ML strategy, implementation, management, governance and potential ethical issues. They focus on leveraging AI technologies to improve patient care, operational efficiencies, and drive innovation in delivery. Alternatively, some have AI teams resembling the structure of those that manage internet and web-based applications today, with an infrastructure group overseeing it and front developers building and optimizing it for different departments that might use it independently in everything they do. There is little doubt that AI will redefine traditional roles and create new ones, paving the way for an important new segment of the healthcare workforce – and, optimally, ushering in a new era of innovation and improvement in healthcare itself.
