At a panel discussion at the recent HIMSS AI Forum 2025, "The Role of AI in Advancing Digital Health Transformation," participants discussed how health system leaders assess where and how to apply artificial intelligence within the healthcare sector and the practical use cases of generative AI, analytics and machine learning for long-term value.
Cole Zanetti, professor and director of digital health at Rocky Vista University College of Osteopathic Medicine, sat down with MobiHealthNews to elaborate on a few key topics addressed at the HIMSS AI Forum, including the practical use of genAI analytics and machine learning, and where AI should be applied.
MobiHealthNews: What is your advice regarding applying artificial intelligence within the healthcare sector and the practical use of genAI analytics and machine learning?
Cole Zanetti: I always work with the end in mind, and to me, our biggest priority is addressing patient safety and medical errors. That is a huge problem that we have known about.
It is a major issue for systems change – knowing that human beings are valuable and systems are valuable and how we can leverage technology to mitigate these risks.
As a physician, there is nothing more important for us to do if we clearly identify that these tools can reduce the harm that is currently occurring, even though these tools are not perfect. The perfection that we are trying to reach, the real comparator, should be to what is our current state of human perfection, right? So, how we approach implementing this, no matter what health system or organization you are in, you should start there.
Where are we struggling in terms of medical errors, patient safety and patient harm, and what tools have been shown effectively to be able to address this, and are there outcomes better than our current state?
And if you approach it from that angle, this is not just about numbers, this is not a fantastical theorization of what AI can do and how it can transform medicine. This actually could save lives. So, I think that is where we need to start the real conversation on how this impacts real patient lives today and patient care.
MHN: Please explain the difference between a low-risk and high-risk AI solution.
Zanetti: I'll give you an example of what would possibly be considered high risk. So, think about biomedical software for teleretinal evaluation. So, when a patient gets their teleretinal exam, a picture is taken of the retina and sent to an ophthalmologist to be read. And there is a software solution that exists that FDA cleared that can actually diagnose diabetic retinopathy using artificial intelligence without a physician. It has been cleared to do so. That would be considered high risk because you don't even have a physician involved, but it went through robust testing and was cleared by the FDA for that to be used.
So, if it is FDA-cleared, there are a few things to keep in mind, though. FDA-cleared does not necessarily mean it is foolproof, but it has been cleared for use because it is like with AI solutions that the robustness of evaluation is getting better but that would be high risk.
Low risk would be something like robotic process automation. So, you need to get medical records for your patients, and they had an event where they were seen outside your health system. So, you can set up a bot that would take the patient's information as a trigger, would automate an electronic fax to that other health system, if we have to send a fax or log into a health information exchange, find the record, download it, and push it into your patient's medical record within your system.
So, I would consider that lower risk because you are moving a document with clear identifiers from one location that is secure to another secure location so that it is accessible to the care team to adjudicate care; it is not diagnosing. It is not changing treatment. It is just moving data from one place to the next that we need. That would be considered lower risk.
MHN: You noted at the HIMSS AI in Healthcare Forum that there is so much information in the healthcare realm to digest that sometimes it is hard to keep up with what is available. Can you elaborate on that? With so much information out there, where should AI be applied?
Zanetti: I think a clear example is in clinical decision support and personalized medicine. We want the patient in front of us, the person, the human being in front of us to be treated as the individual and unique person that they are. So, your genetic background, pharmacogenetic insight to know what medications would work for you or not work for you.
Potentially genetic sequencing, if we have that data, if we do not have those things, what is your insurance information so we know what is going to be potentially covered and what is not?
Do you have support, a caregiver at home? Do you have transportation to come to and from a facility if you need to get specific levels of care? What is your education level for how we provide information to you to make sure that it is interpretable?
What is the level of comprehension you have as it relates to your current medications and whether they interact and how to take them? So, all of this stuff that is unique to each individual combined with as of today, what is the most evidence-based approach for care for you, given all of your medical conditions, all of your medications, all of your recent lab results, all of your vitals and the contextual information about your social vulnerability? Given that insight, what does the literature say we should do?
And I did not have to prompt AI to investigate that. Once I open up your chart, that should automatically happen. And then two things would occur. It would provide me with: These are the current gaps in care that exist between the current treatment and what the current evidence shows. Show me the references to justify that argument that I can click on and review myself, and if I agree with closing that gap in care, it auto-creates the order set that I need to sign off on to execute the care plan.
So, it is basically seamlessly integrating the data that we have and connecting it with up-to-date evidence that could be digested and connected to an EMR through artificial intelligence.
