How Kaiser Permanente is using gen AI to ‘paradoxically’ make care more human again

The implications for AI in healthcare are vast and significant — however, medical facilities and institutions shouldn’t be too hasty in implementing it, Daniel Yang, VP of AI and emerging technologies at healthcare organization Kaiser Permanente, said onstage at this year’s VentureBeat Transform. 

Yang, who is also a practicing medical doctor, outlined his organization’s steady, deliberate approach to adopting the groundbreaking technology. 

“There’s a tremendous amount of hype and opportunity in this space,” Yang told the audience. “We really need to see the evidence, see that it works, see that it’s safe, that it’s effective in real world settings.” 

Advanced alert monitoring saving hundreds of lives

The “antidote” to chasing after the bright, shiny object of AI is really doubling down on an organization’s core mission, Yang emphasized 

For Kaiser Permanente — which has 250,000 employees, 40 hospitals and 618 medical facilities — that core mission is “pretty simple”: Delivering safe, high-quality care that is also affordable. That in mind, the goal is to build directionality into their AI strategy. 

“We want to decrease disparities and inequities in the communities we serve, and we want to support and allow our employees to thrive,” said Yang. 

One successful AI application at Kaiser Permanente is its advanced alert monitoring program, he noted. Using predictive analytics, the app identifies hospitalized patients that are likely to clinically decompensate (get sicker) in the following 12 hours. For instance, if a patient is hospitalized for hip surgery, the programs can try to predict whether they may get infected or suffer a pulmonary embolism and unexpectedly land in the ICU. 

Algorithms assess risk every hour using real time data from electronic health records, Yang explained. This includes a patient’s vital signs, their medical conditions and laboratory results, among other data. The app can then alert if risk increases above a certain threshold or if other red flags flare up. 

“The idea is we can intervene earlier, thereby reducing or preventing that negative outcome,” said Yang. 

He explained that Kaiser Permanente performed a “very robust” clinical trial across 21 hospitals in Northern California, and published its findings in The New England Journal of Medicine. The results were “astounding”: The program was saving more than 500 lives per year. 

However, Yang noted that the algorithm itself didn’t save a single life — it was the workflow redesign. 

“We had to design the entire workflow around it,” he said. “That is 90% of the work.”

Notably, his team decided that they didn’t want the physicians to receive alerts because that would be distracting (and there are always risks of false positives that could waste important time). Instead, they trained critical care nurses to review patient charts. Then, if appropriate, they would activate a rapid response team of doctors, nurses and pharmacists to evaluate and decide on next steps. 

“I can’t emphasize enough that AI was not replacing the judgment of our clinicians,” said Yang. “It was augmenting it, it was supporting it.”

Gen AI making care more human again

When asked about generative AI, Yang noted dryly that “I’m probably the first person to be on this stage that hasn’t mentioned gen AI in the first 30 seconds.” 

However, Kaiser Permanente is beginning to implement the technology, notably with a clinical AI scribe that he called the first of its kind in the country, “if not in the entire world.”

He pointed to a “terrifying statistic” that burnout rates in healthcare range from 40 to 70%. 

“So with that context in mind, it’s no surprise that one of the most exciting areas of generative AI in healthcare is not where you think it is,” said Yang. “It’s not treatment. It’s not diagnosis. It’s not personalized medicine. It’s reducing the administrative tasks of our clinicians.”

As a practicing physician in internal medicine, he still works part time at Kaiser Permanente, in its urgent care clinic in San Francisco, he explained. During an eight hour shift, he’s expected to see 18 patients, which can be like a “hamster wheel.” Not surprisingly, doing that day in and day out can be extremely stressful. 

To help slow down that hamster wheel, Kaiser Permanente’s scribe technology generates a first draft of a clinical note based on a recording of a patient encounter that they consent to upon visiting the office, which the doctor can later review. They’ve found that some clinicians are saving up to an hour a day with the technology, Yang noted. 

“What really excites me about this use case is generative AI paradoxically making care more human again,” he said. 

When many people visit the doctor, they end up staring at their doctor’s back, he noted, because their eyes are focused on the computer writing their notes. 

“The scribe technology allows the provider to be liberated from the keyboard,” he said, and as a result, they’re able to spend more time talking with the patient. And, patients love the tool, too, he said, as it creates more transparency into their care because doctors are dictating findings out loud. 

“Gen AI takes the complexity out of healthcare,” said Yang. “It makes care more of a conversation between two humans.”

Integrating responsible AI from the start

While demand for this type of technology from clinicians was huge — they were “desperate” for it — there were challenges because the technology was so new, Yang explained. There wasn’t a tremendous amount of evidence on its effectiveness and robustness across diverse care. While his team had a good sense that it worked in the primary care environment, they didn’t know how well it would work in specialties such as urology or ophthalmology where clinical workflows were different. 

This put the healthcare institution at a crossroads: They could either implement the technology or sit on their hands and wait for evidence to be generated. Normally, that can take a few years due to the pace of science, Yang pointed out. So, Kaiser Permanente decided to generate that evidence themselves with “the most robust quality assurance testing.” 

This involved testing functionalities across a very diverse setting and soliciting feedback from thousands of providers. Not surprisingly, it wasn’t perfect; there were hallucinations and other issues. 

“We created this flywheel where we were learning by doing, we were getting feedback, we were taking that feedback and we were giving it back to the provider to educate them on the use of this technology,” said Yang. 

Kaiser Permanente’s responsible AI philosophy is to “measure twice and cut once,” or be strategic about where to deploy AI and be more thoughtful upfront about identifying and mitigating risks. 

Yang pointed out that when people hear the words “governance” and “guardrails,” their initial reaction is “bureaucracy, red tape, molasses, you’re gonna slow everything down.” 

However, he pointed out, “nothing slows innovation, particularly in healthcare, more than patients who are harmed by technology.”