MERIDIAN – As part of its value-based care initiative, Blue Cross of Idaho is using artificial intelligence to notify primary care providers when patients are at risk of developing serious health issues.
The project has been under development for almost a year and has grown from two to 23 staffers.
“It’s not magic,” said Marc Roberts, director of analytics for the company. “We’re not trying to look for weird correlations you wouldn’t expect. A lot of them are obvious.”
The artificial intelligence (AI) or machine learning (ML) project has drawn national attention, including a mention in the Wall Street Journal.
“On one of its first data runs, the automated ML program made a startling discovery: Blue Cross patients who get regular teeth cleanings and dental checkups have markedly better cardiac health in the following year,” the paper noted. Other correlations bore this out, meaning Blue Cross will look at ways to encourage members to seek preventive dental care, which is cheaper than heart-related costs.
Blue Cross didn’t develop the AI program on its own. It’s been working with experts in the field, including DataRobot, software from a Boston-based specialized predictive modeling company, and IBM Watson Health, which uses the Watson AI system best known for winning the Jeopardy! TV show against humans in 2011. DataRobot took tasks taking two to three months and reduced them to a week, while Watson helps correlate disparate claims into a single medical episode, Roberts said.
But Blue Cross still needs to work with the software for insights. “It helps us do the grunt work of predictive modeling and automates manual processes,” Roberts said.
For example, diabetes and renal disease are often connected, so the system could track where patients are in the disease cycle, said Drew Hobby, senior vice president of health care economics.
What’s new is Blue Cross sharing information with primary care providers (PCPs), Roberts said.
“We think the next level of effectiveness is sharing with the PCPs,” he said. “When PCPs have that information, they can reach out and care for the members,” potentially before a serious health issue, he said.
The notification tells PCPs a member they’re caring for is predicted to be a high-risk member in the next 12 months, based on information such as recent diagnoses and prescriptions the PCP might not know about, combined with demographic information, Roberts said.
“The PCP would reach out to a patient and say, ‘Hey, I noticed you’ve been to the emergency room or started a new prescription, and I’d love to get you in for a checkup,’” he said.
On a monthly basis, such notifications will typically be fewer than 1% of a PCP’s patients, he said.
“As we’ve tested the model, 95% of the people do have a high-cost event in the next year,” he said. “We feel quite confident that not only is it a short list, but a very highly accurate list.”
PCPs started getting the AI data in November, but so far the program is promising, Hobby said.
“The initial feedback is that the PCPs love getting more data.”
Value-based care means instead of being reimbursed by how many services are provided, providers are reimbursed by outcomes, a method intended to reduce health care costs for businesses. At this point, 70% of Blue Cross of Idaho’s claims are in a value-based care arrangement, Hobby said.
In traditional primary care, case managers interact with people with major complications, but that’s often after-the-fact, Hobby said. With value-based care, case managers can interact more proactively and better identify patients in greatest need of help, he said.
While it will take six months before Blue Cross will have outcome data on AI, the value-based care program itself is starting to show results. In 2018, Blue Cross of Idaho paid out approximately $20 million in shared savings payments to its value-based care providers.