Some health systems across the country interviewing for their first chief artificial intelligence officer — the hottest recent CEO in health care — are looking for executives with more clinical knowledge than artificial intelligence expertise. Conversely, some are looking for managers with more knowledge of artificial intelligence than they know about the clinical side of IT. But both types of expertise are significant in this position. The balance depends on the healthcare system.
Here’s one observation from an insider: Dr. Karandeep Singh, chief health AI officer and associate CMIO for hospital care at University of California, San Diego Health, who previously served as deputy chief medical information officer for AI intelligence at Michigan Medicine. Chief AI officers are not yet common in healthcare – so having a two-time chief artificial intelligence officer is quite a find.
sat down with Singh for a wide-ranging discussion about what UC San Diego Health expects in a chief artificial intelligence officer, what skills anyone who wants to become a health care AI executive should have, what is expected of Singh, and much more more.
Q. How did UC San Diego Health approach you about taking on the position of Director of Health AI? What were they looking for and who would you report it to?
AND. It was a role that we came up with and found together. I was interviewing for a position to lend a hand lead the Jacobs Center for Health Innovation at UC San Diego Health, which is our primary innovation center located within the health care system.
One of the things I wanted to make sure is that if we have an innovation arm of the healthcare system, we also play a role in the healthcare system thinking about the governance of AI and thinking about the responsible utilize of AI. When I interviewed for this role, we had many conversations about what it might look like. This eventually evolved into the position of Director of Health AI.
Question: This is not your first AI leadership role. You served as deputy chief medical information officer for artificial intelligence at Michigan Medicine.
AND. I am a researcher involved in research on applied artificial intelligence. I think there’s a lot of captivating work going on if you look at the computer science and statistics space in the artificial intelligence space. But what I’ve always been most interested in is how well do these tools perform when we actually utilize them at the bedside or in the healthcare system?
Can we scientifically determine what we can do to ensure that these tools, when used in the health care system, actually lend a hand our patients and doctors? At Michigan, I was an academic with an operational interest in artificial intelligence.
While there, I had the opportunity to join our nascent Clinical Intelligence Committee, which was the health AI management committee at Michigan Medicine. This role sort of had some elements of what a health AI director might do, but in my opinion it was still very much focused on managing health AI.
What I really wanted for the next stage of my career was to continue to have an academic interest in figuring out what works, what doesn’t work, and how we make things work.
But at the same time, I wanted to ensure a level of coordination and awareness across our leadership, so that as we design our healthcare system and develop it, we’re thinking about the opportunities available to us in artificial intelligence in a way that allows us to implement solutions in a way that we can explore and validate them whether they work.
I wanted to take a top-down approach to be able to identify and prioritize what things we should be doing in the AI space. But even to this day, I have an academic lab at the university that focuses on understanding what works, what doesn’t, and how we can provide better care.
Q: What skills do you think anyone who wants to become a healthcare AI executive should have?
AND. If you look around the country at places hiring AI directors, the first question asked is, “Why do we even need an AI health director?” Then, depending on why you think you need it, that really indicates what skills you might be looking for in someone to fill that role. So let’s look at why you might need it.
Many healthcare systems are already using artificial intelligence. At this point, they are not so much considering the utilize of artificial intelligence. Almost every healthcare system, if you look in their electronic health record, check what they’re doing, already has a lot of AI-based clinical workflows. The question is then asked: “Who is responsible for the use of artificial intelligence in the healthcare system? Who understands what is actually happening in the AI space and the workflows that AI is implementing in the healthcare system?”
When you get down to it, you start to realize that while there are people in the healthcare system who understand the various elements of AI-powered workflows, there isn’t a single person who really stands out as being in charge.
I think a lot of the reason why health systems should even consider this role is that if you’re using AI, when we know that using it inappropriately or irresponsibly could harm people, you need someone who is ultimately responsible for this and makes every effort to ensure that we have processes in place to prevent such events from occurring.
From this perspective, a health AI director is someone who should understand the basic principles of AI, potentially have some experience in informatics, but still understand the ultimate clinical utilize case, and have some level of understanding of informatics where they are able to it is to understand how to examine something to find out whether it works or not when it is integrated into the electronic health record and integrated into the clinical workflow.
As such, it’s a very diverse skill set. There are places across the country that will hire people with more clinical knowledge, or maybe less knowledge of artificial intelligence, and that’s fine. There are also places that will look for positions where the person has a lot of technical experience and maybe doesn’t understand the clinical workflow as much because they’re involved in electronic health record integration.
I don’t think either of these solutions are completely wrong because ultimately you need someone who understands and is responsible. The ideal person to fill this role is someone who understands a little of both worlds and can lend a hand translate the AI side to the clinician and the clinical side to the AI experts.
Q. Please describe the AI portion of your work at UC San Diego Health. Broadly speaking, what is expected of you? More specifically, what is your typical day like?
AND. To describe the AI part of my job, I would probably look at two different elements. The first piece is that we actually do a lot of work at UC San Diego Health and we actually build internally the AI that we utilize in healthcare operations.
One example is that we have patients in the ER who have been admitted to the hospital, and we often call them ER patients or boarding patients. The number of patients admitted to the emergency department is often determined by many other factors related to the health care system and issues such as wait times to the emergency department.
This is something we care about very much and we try to influence it to improve it. One way we do this is by forecasting the number of patients hospitalized in the emergency department over the next 10 days. This is a tool through which, while companies can propose specific solutions to achieve this opportunity, it is a tool that is quite specific to our institution in terms of some of the factors that we consider and what we want.
Even before I joined UC San Diego Health, the company decided this was a skill they wanted to develop internally. When I joined, we made some improvements to the model and tool. We have also engaged mission control leadership, and mission control is our nucleus where we actually utilize this information to reduce wait times in emergency departments.
We worked with this management to improve the model and be able to make better predictions. This is a tool completely built in-house. This is something I checked the source code for. I worked with our data scientists to identify some data issues and try to correct them.
We looked at other methods of doing the same forecasting to see if any of the recently introduced methods are better than some of the ones we utilize. This is the part of my role that I think requires knowing some of the algorithms, knowing some of the code to understand the key issues in the data pipeline, and being able to effectively guide our data scientists so they can go about building these tools to work accurately.
The second part is that we work with multiple vendors to implement solutions. Sometimes that partner is our electronic health record vendor. Sometimes these are other partners, such as start-up companies or other groups implementing solutions within our healthcare system. How do we know that the solutions implemented by these companies actually work?
So we engage with them to some extent and say, “Can you help us find out how well this is working in our health care system?” However, to some extent, at some point you need to independently check and make sure of some of these things, especially those that may have a immense impact on your health care system.
This is another place where we have this opportunity where we can actually go and do some of the inspection independently, without having to completely rely on our vendors to do it. This gives us another direct insight into what we expect to see as a benefit, or rather what we observe as a benefit.
The ability to create solutions ourselves and evaluate solutions developed by vendors are two capabilities that I believe healthcare systems should have. Otherwise you will be implementing, implementing, implementing. You’re not sure what works.
And then it becomes a really tough discussion about what things you will continue and what things you will turn off if you don’t go ahead and take that time to evaluate.