Q&A: A UNC Health Rex Leader Shares a Clinician’s Perspective on AI

HEALTHTECH: How has healthcare’s AI outlook evolved in the past few years? What capabilities do organizations expect now that weren’t available a year or two ago?

UDAYAKUMAR: Initially, AI was limited to very simple tasks, such as automating workflows, but with recent advancements in algorithms, AI has become more sophisticated. In healthcare, AI is valued for its potential to analyze diagnostic data and to allow us to make faster and more accurate decisions. Overall, the evolution of our outlook on AI has opened up possibilities for improving patient outcomes and efficiency.

HEALTHTECH: What are some key concerns that clinicians have about growing AI adoption in healthcare?

UDAYAKUMAR: Our clinicians are asking important ethical questions. Ensuring data privacy, maintaining transparency around the algorithms and addressing potential bias are all very valid concerns. There is some mistrust and concern related to control of medical decision-making. This is where collaborative AI is very important. Across industries, organizations achieve the greatest benefit when humans leverage technology correctly so that they complement each other. We need to understand how clinicians and AI can amplify each other in healthcare. Each brings different strengths. No clinician is ever going to be able to compete with the speed or accuracy of AI, while AI cannot replicate the creativity that a clinician might have. We need to identify the best roles for the clinician and the best tasks for AI. If we can do that, it should mitigate the fear of competition.

EXPLORE: Learn how AI is making healthcare smarter.

HEALTHTECH: How do you think the workforce shortage is influencing AI deployment? What other considerations are driving interest in AI?

UDAYAKUMAR: AI can help reduce workloads by automating routine tasks and providing clinical decision support so that clinicians can focus on the most cognitively demanding or high-risk tasks. AI can also be used to optimize resource allocation. For example, staffing schedules and supplies are two challenging topics nowadays; perhaps we can use AI to optimize those schedules or identify locations that will have high resource or supply needs so that we can respond more proactively. AI can also address the shortage of healthcare professionals in specific specialties and geographical areas, which is where telemedicine and remote patient monitoring powered by AI come in.

HEALTHTECH: What are some lessons from your organization’s approach to AI that you can share with other healthcare systems?

UDAYAKUMAR: There’s a lot of excitement within UNC Health around the possibilities that AI will bring to the healthcare industry. Early on, UNC Health developed a responsible AI framework, which is based on four components: fairness, which means we ensure that we’re training our models on the correct population and identifying potential bias; accountability, meaning we track performance and establish an evaluation process; transparency, meaning we’re very clear on the usage and the limitations of AI; and trustworthiness, which means we need to educate others about AI. It was very important to us to have that framework established early on.

HEALTHTECH: What are some priorities related to AI for 2024? What do you think the next few years will bring?

UDAYAKUMAR: One of the most important priorities will be the ethical and responsible use of AI, with a focus on transparency and fairness. That means having regulatory frameworks and standards to govern the use of AI in healthcare, as well as measures to protect patient data and ensure confidentiality in complying with privacy regulations, such as HIPAA.

As far as what the next few years will bring, I’m very excited about AI-powered virtual assistants, which will become more widespread. We have an internal virtual assistant at UNC Health built in partnership with Microsoft. I have enjoyed piloting this to draft emails, create presentation content, summarize text and draft clinical education materials. From personal experience, this has decreased administrative work.

UNC Health is also working with Epic to pilot auto-drafted patient messages using ChatGPT. Multiple physicians at UNC Health are involved in that pilot. It is important to note that this model only creates a draft reply as an optional starting point for the clinician. The clinician is in full control of what gets written and sent to the patient. This has been very well received by our physicians, and they have said that the technology is improving patient engagement, allowing for more timely communication and decreasing documentation time.

Finally, we will likely see more progress in diagnostic interpretation, such as pathology, radiology and further development of tools that can help with earlier detection of disease.

RELATED: Understand how investing in IT infrastructure supports AI adoption in healthcare.

HEALTHTECH: What can organizations do to balance patient and provider expectations for AI?

UDAYAKUMAR: One of the most important steps is clear communication on the benefits and limitations of the technology to both physicians and patients. Patients should understand that AI is a tool to assist healthcare professionals, not a replacement.

Education is also critical, and that means that physicians need to receive training as part of their medical education on how to use AI systems. They need to be able to explain enough of the technology to a nonexpert. They need to be trained on how to address questions about whether an AI algorithm was fair or correct.

Finally, continuous improvement is essential. Regular evaluation of AI systems that incorporates feedback from patients and clinicians can help us identify areas for improvement. There needs to be an iterative process to refine the technology so that in time, the AI better aligns with expectations.