AI to be tested in Kaiser Permanente's AIM

Sep 29, 2023

Dr. Vincent Liu is the principal investigator of the Augmented Intelligence in Medicine and Healthcare Initiative (AIM-HI) for the Kaiser Permanente Medical Group.

Photo: Courtesy Kaiser Permanente Medical Group

Excitement around artificial intelligence has grown in the last five months, but there's a big evidence gap between what AI is hyped to do and what it is actually doing to make a difference in impacting positive patient outcomes, according to Dr. Vincent Liu, principal investigator, Augmented Intelligence in Medicine and Healthcare Initiative (AIM-HI), the Kaiser Permanente Medical Group.

The lack of demonstrated AI case studies prompted Kaiser to announce on May 17 a $3 million grant program to award three to five U.S. healthcare systems up to $750,000 for their proposals on the use of AI and machine learning to improve diagnoses and patient outcomes.

The Gordon and Betty Moore Foundation is financially supporting AIM-HI and selected Kaiser Permanente as the coordinating center for these initiatives, Liu said.

The prospective evaluation of the selected AI proposals will be done over two-to-three years.

The request for proposals is open and letters of intent are due June 30. A national advisory committee and expert reviewers from Kaiser Permanente will assess proposals and work closely with the selected project teams during their two-year grant period.

A report on the first wave of applications is expected this fall and those selected will be announced in December."Our job is to evaluate the application, find the ones most innovative and feasible, to learn from these innovators and to share lessons with the broader audience," Liu said.

WHY THIS MATTERS

AI and ML have shown promise in early warning systems that analyze hospital patients' data to identify when patients are at risk of serious decline and may need intervention. Large language models similar to ChatGPT streamline medical record note taking. AI and ML also offer computer vision technology to analyze medical images for tumors, cancers and surgical guidance.

Algorithms are proving AI is improving patient outcomes, according to Liu.

Kaiser has implemented early warning systems that identify high-risk hospitalized patients. One of these programs has saved an estimated 500 lives a year, Liu said.

"We at Kaiser Permanente have developed an advanced alert monitor algorithm … it's an early warning system knitting together complex data in near-real time to evaluate patients at risk of deteriorating," Liu said. "We rolled it out over 21 hospitals in the system."

The results were published in 2021 in the New England Journal of the American Medical Association, he said.

Kaiser is also looking at AI for imaging, to improve prognostication of breast cancer and to look at the effect of diabetes on the eyes.

The buzz and excitement right now is around large language models and generative AI, but there is much coming in the next one to five years, Liu said.

The next stage is for AI to identify patterns in imaging and pathology, and to identify signals physicians are not able to pick up.

There is both excitement and concern about generative AI being able to learn after being fed a massive amount of data. The human element can't be taken out of the loop, he said.

"There are ways errors can propagate through," he said. "We have to be cautious, AI can have unintended consequences. We have to find ways to integrate it into the workflow."

The goal is to make AI that's fair, equitable, safe and sustainable.

"AI has the potential to make us more efficient. It can help us address some of the workforce challenges, such as sorting through charts or reading a 50-page summary of a hospitalization," Liu said. "With generative AI we can enhance communication between all parties in healthcare. I think with any new and exciting technology, everyone sees the potential."

What AIM-HI is doing, he said, will give confidence to decision-making.

"Our goal is to bring together experts," he said, "to get AI into routine use."

Twitter: @SusanJMorseEmail the writer: [email protected]