UC San Diego Health seeks to better manage AI integration
Karandeep Singh, MD, spearheads a strategic approach to implementing AI for enhancing patient care and clinician well-being.
The movement toward using artificial intelligence is well underway at UC San Diego Health. Karandeep Singh, MD, now is tasked with bringing some order to its expanded use within the organization.
That’s a bit of a challenge for Singh, a nephrologist who completed his internal medicine residency at UCLA Medical Center. That’s because the AI “cat” is already out of the bag at UC San Diego Health, which has several AI-based initiatives in progress.
He steps into a new role of chief health AI officer at the three-hospital academic health system of the University of California San Diego. He comes to the position after playing lead roles at the University of Michigan Health, from where he was recruited.
Singh is beginning an investigatory process to plot a course to use AI to meet the future demands that healthcare systems will face. “We’re trying to measure things and see how well they’re working,” he says. “What’s the current state and then, honestly thinking a little bit further ahead from a strategy standpoint, how is healthcare going to be delivered in 2025 and 2030? And what are the actual problems we need to solve for us to get there?”
The questions Singh are facing represent those that the industry at large will need to resolve to gain the potential benefits that advanced information technologies can provide to help it address staffing shortages, physician and patient experience, and improvements in efficiency and quality.
Practical applications
Singh says his long-time interest in information technology had buttressed his training in medicine. “Even when I was in medical school, I was writing software, and I was writing software in my internal medicine residency,” he notes. He solved problems by digitizing residency educational materials and enabling paging systems to send bulk messages to colleagues.
His master’s degree thesis involved using natural language processing to better understand clinical notes, enabling technology to tease out risk factors for disease. “This has been kind of a natural progression – I spent the last eight years at the University of Michigan, really around trying to understand the impact of AI.
“What I really want to do is take the state of the art right now and see how far that actually gets us toward solving real problems,” he adds. “And that’s what ultimately led me to this role at UC San Diego Health.”
He’s now working with Christopher Longhurst, MD, chief medical officer and chief digital officer for the system. The emphasis for his new role is taking theoretical and possible uses of AI and making it real.
“I don’t really want to build the next state-of-the-art model – what I want to do is take the state of the art right now and see how far that gets toward solving real problems. That’s what ultimately led me to this role.”
Several efforts are now underway at the system. One uses AI in a sepsis prediction application, which has resulted in improvements in detecting and preventing the onset of the condition, resulting in cost savings and better outcomes for patients. Another AI application enables the system to predict estimated censuses at its facilities and better manage staffing and patient distribution.
Generative AI also is being used by UC San Diego Health, Singh says, noting that the system “was one of the first health centers using generative AI and large language models to communicate with patients … I think we have something like 15 generative AI pilots currently in progress.”
Managing the initiatives
Key to achieving that is ensuring that the governance process surrounding the use of AI is sound.
“What I’m hoping to do is to figure out what are the common themes (of current efforts) that we can kind of enshrine in our processes, enshrine in our governance and also enshrine in our use of technology, so that we can make the next handful of projects much easier,” Singh says. “I’m trying to understand existing processes and get a handle on AI governance … to try to make it more reproducible, so we can do the next 10 pilots with a lot less work than the first 10 pilots required.”
His organization – and healthcare overall – needs to focus on all aspects of positive change that can improve care delivery. It’s not just about using technology by itself without adapting other aspects of the adoption curve.
“One thing that learning health systems do really well is figure out the process and the people and the technology part – that’s what makes it different from just any ordinary health system,” he says.
Gathering evidence that supports the benefits of AI models will be crucial in rolling out the technology, he contends. “There has been a vast increase in people trying to measure how well these AI tools actually predict the outcome that they’re designed to predict, in a variety of health system environments,” Singh says. “The place where we’re still lacking a little bit is (to find out if) the model is predicting what it’s supposed to predict – is that actually making anything better?”
He differentiates between predictive models that try to ascertain which patients being discharged from a hospital are likely to be readmitted; that’s not as helpful as providing insight into how to actually fix the problems that result in readmissions.
Future potential
Large language models hold the promise of powering generative models and providing visibility into care delivery. Singh remains optimistic about the potential. “In the short term, we’re going to see a shift from folks putting their emphasis on predicting clinical outcomes … towards just improving the everyday experience of receiving and delivering care.”
And he’s hopeful that AI can provide the workload relief that clinicians have been clamoring for, particularly in relieving documentation burdens and providing “keyboard liberation.”
“And summarizing information is a really critical thing – historically, it’s been very difficult to summarize a mix of structured and unstructured information,” he says. “We spend a lot of time handing off care from one clinician to another, and a lot of that handoff involves summary. So it’s important to try and figure out where can we reduce cognitive burden and improve the experience of being a clinician. More longer term, I think there is going to be a focus on figuring out where can we automate aspects of care? Even if we don’t automate care, can we make it so that we don’t have to hold up care?”
While there will be disruption over time, AI and other advanced technologies have the potential to reorganize roles and expand access over time, he believes. “I do think that we have to rethink care – if you look at American healthcare, there are folks who have access to great care, and there are folks who have access to no care. What I would try to put more emphasis on is, what’s better, no care or some care delivered by AI. If we’re going to use it as a tool to expand access to care, I think I would look at it with that as an initial priority.
“But in the short term, I think (AI) is going to hopefully improve our ability to deliver better care and faster care.”