Inside the Duke approach to build and buy AI tools

Almost all leaders in the health care industry have a shared understanding only if AI will never replace the experienced clinical judgment of workers, technology can provide them with an essential respite against the administrative burden by treating tasks not added to value, noted a framework of the health system.

Terry McDonnell, chief nurse of the Duke University Health System, stressed that by managing routine tasks, AI can release clinicians to devote more time and attention to patient care.

“I always said that I can teach a skill to a nurse, but I cannot teach what the patients look like on the phone. It is a lived experience, it is a learned experience, and that is what the clinician brings. And I think that these are the things we have to focus on and make people have time and the bandwidth to really engage,” said McDonnell.

She said that this need to extend the automation of AI tasks is more important than ever, since the growing demand for care of the country is collided with a clinical shrinking workforce.

A major but often neglected bottleneck is the shortage of teachers available to form new nurses, even if requests for nursing programs remain abundant, she added.

These labor challenges are one of the reasons why Duke has invested massively in AI products designed to rationalize tasks and improve results for patients. The health system builds internal AI tools while purchasing others from other suppliers, said McDonnell.

For example, Duke uses an internal AI model that monitors EPIC patient data to report the first signs of deterioration, giving the prior warning teams they need to intervene, she noted.

“We are upstream – we do not react in an emergency. We intervene proactively when we see that clinical condition can change, and this is motivated by AI algorithms, “said McDonnell.

Duke uses another internal AI tool that focuses on sepsis. He analyzes patient data to detect that could be at risk and triggers early processing beams before the condition goes to serious condition.

The health system also works with Artisight to integrate computer vision into its hospital rooms, said McDonnell. She said that Duke installs cameras in the room, which will work with AI algorithms to monitor fall risks and ultimately automate documentation – such as the recording of a patient’s liquid production without a nurse never has to write or dictate a note.

She also noted that Duke had recently led an AI driver with Microsoft’s Nuance, and he deployed the clinical documentation platform fed by Ai Abridge earlier this year. Although such tools have proven to be effective in reducing professional exhaustion for doctors working in ambulatory and ambulatory environments, they are not yet fully optimized for the complexities of care for hospitalized patients, said McDonnell. However, she noted that Duke is currently working on a documentation pilot for patients hospitalized with Abridge.

Regarding the question of whether to build or buy, McDonnell said that it depended on the problem that the health system is trying to solve.

All AI pilots in Duke start with a problem declaration, she noted. Then, the leaders see if there is a solution on the market which addresses this problem, or perhaps a tool that one of their partners already develops, such as Epic or Microsoft.

If there are no well-completed solutions on the market, Duke then plans to co-develop a solution with an external technological partner. And if it is not an option, Duke sees if he can build a tool on his own shot from his engineering school and IT capacities, said McDonnell.

“We have great internal strength in our own IT and development teams. We are really lucky in this regard-not all systems have this luxury, “she said.

In advance, McDonnell encouraged health systems to balance practicality with innovation when testing new AI models.

“You cannot excite yourself for each shiny and brilliant toy that goes through the front door. But I also think that we are starting to learn that we can manage things, try things and quickly learn what will work and what will not work,” she said.

For McDonnell, the success of AI depends not on the continuation of the most recent technology, but on the choice and refinement of tools that really facilitate the workload of clinicians and to improve results for patients.

Photo: Yuichiro Chino, Getty Images