One early adopter's tips for AI deployment success
Next week at the HIMSS AI in Healthcare Forum in San Diego, an analytics leader will discuss the artificial intelligence use cases with which he has experienced the most success – and offer tips and perspective on best practices for AI implementations in clinical settings.
Dr. Luis Ahumada, director of health data science and analytics at John Hopkins All Children's Hospital – speaks on a panel with Sumit Nagpal, CEO of Cherish Health – about his experiences so far as an early adopter of AI and machine learning models.
He will provide some real-world examples of how Johns Hopkins is using AI tools to improve its healthcare delivery processes – automating and accelerating clinical diagnostics and freeing physicians to focus more on patient care. He describes overcoming barriers to AI adoption, best practices for integrating it into existing workflows, and much more.
“AI has helped,” says Ahumada, but it is “an ongoing project.”
At John Hopkins, “we're trying to better understand where the key benefits will lie,” he said, because “resources are not available to everyone and AI is expensive.”
Ahumada sees value in two types of AI, with different characteristics in terms of size, shape and scope.
“One of them is LLMs, but they won't solve everything,” he said. “The other is what we traditionally call machine learning: creating models for prediction and high-risk calculators, things like that. There will be a hybrid between the two at some point. But over the last ten to fifteen years we have focused on the second.”
Both are important for one use case that John Hopkins All Children's Hospital has focused heavily on: clinical documentation.
“We have risk calculators for everything under the sun: readmissions or surgical risks, complications, things like that. But at the same time, a lot of those problems are caused by inefficiencies within the system, in the documentation process. And as far as I'm concerned, we need to solve that first. And that is probably what the current tools we have at our disposal are aimed at.”
Data integrity is the “cornerstone and foundation for everything we do with LLMs and ML.”
Dr. Luis Ahumada, Johns Hopkins All Children's Hospital
To achieve that goal, the health care system is focused on “getting better data,” he said. “We can use machine learning, we can use LLMs, we can use a lot of different things to better collect this data.”
Yes, there are also major challenges around missing data, validation and data integrity, says Ahumada. “Data is not perfect. It should be, but it isn't. That's a problem because we use the data we collect every day, every second, to create models.”
Another fundamental obstacle has to do with “collecting and putting all that data together,” he said. “Traditional machine learning loves huge amounts of data.” But it can often be challenging to create “even a small registry for hundreds of thousands of patients,” he explained.
But despite these challenges, Johns Hopkins is moving forward with a range of different use cases for generative AI and machine learning, which Ahumada in San Diego will describe in more detail.
Data integrity is the “cornerstone and foundation for everything we do with LLMs and ML,” says Ahumada.
But he's also concerned about costs, and about helping ensure that health care systems of all sizes — not just those with the resources of Johns Hopkins — have access to the kinds of AI tools that can help them improve their clinical and improve administrative processes.
Just this week, a report from KLAS was published showing that large healthcare systems are already seeing success with generative AI, but smaller systems are more limited in their adoption.
“AI should be accessible to everyone, but that is not true because it is expensive,” he said. “You can use advanced LLMs, but even with that you can set up an LLM in your shop (but) people have to understand that in order to do that, you have to have people who know how to do it, and they're not cheap. So yes, there are a lot of different benefits. But it's going to cost money.”
Attend this session at the HIMSS AI in Healthcare Forum taking place December 14-15, 2023 in San Diego, California. Read more and register here.