Artificial Intelligence Gets Real

How real is artificial intelligence getting now, in U.S. healthcare? Pretty real. So real that the folks who created Clover Health, a Jersey City, N.J.-based PPO/HMO (preferred provider organization/health maintenance organization) in 2014, in order to operate as a Medicare Advantage health plan, created their organization as a health plan with a built-in technology platform—and with artificial intelligence a core part of their strategic planning. And that says everything about the current state of AI in healthcare, which is getting more real by the day now.

And, in that regard, Clover Health has been developing and using data and machine learning approaches to proactively determine which patients are at higher risk for hospitalization or rehospitalization, and can then intervene. Clover Health’s leaders have also used AI and machine learning to identify their highest-risk plan members, and, said Kumar Dharmarajan, M.D., the health plan’s chief scientific officer, speaking at the AHIP Institute & Expo (the annual conference of America’s Health Insurance Plans) in Nashville in June, “We decided to deploy a program for those top 5 percent of plan members, and really take charge of care management for them. So we built out an in-home primary care program.”

What’s more, said Clover chief data scientist Ian Blumenfeld, “You don’t do things just because technologies are cool. I’ve been doing machine learning for a long time, but would never want to build a tool that didn’t have a practical use case,” he testified. “It’s all about the company mission.”

All that said, building machine-learning models is particularly challenging in healthcare, Blumenfeld said at AHIP, because human beings have very complex personal health trajectories, and so creating algorithms and models to predict personal health outcomes is inherently complex and difficult to do. Still, the fact that Blumenfeld, Dharmarajan, and their colleagues at Clover have built a very strong AI-driven tool called CHARM—the Clover Health Acute Risk Model—which has been helping them to accurately determine who the top 5 percent of their plan members are in terms of health risk, and who represent 40 percent of their medical costs—speaks strongly to AI’s overall potential.

“AI is definitely real now,” says Tushar Mehrotra, senior vice president, analytics, at the Eden Prairie, Minn.-based Optum, a health services innovation company. “There’s a difference, though, between real, and fully scaled,” says the Washington, D.C.-based Mehrotra. “We’re looking proactively at addressing AI for the detection of atrial fibrillation, in the cardiology area; meanwhile, imaging, in the radiology area, continues to see advances in the diagnostic use of AI. Both payers and providers are just now beginning to build capabilities in this area. Health plans are hiring the data scientists to help build the capabilities to predict chronic disease incidence and emergence; so they’re ahead of providers. And pharma is the furthest along.” One very practical use that health plan leaders are turning to, he notes, is using AI to predict which parts of which apps health plan members will be using. And, in terms of using AI to leverage social determinants of health data to predict the onset of chronic illness, health plans are “already between a 5 and a 7 on a scale of 10. They realize they need access to claims data and need to partner with their providers on the clinical data to move forward.”

Speaking of the opportunities in the patient care delivery space, “I think the opportunities are in better prediction and diagnosis, and there’s a lot going on in that area right now, because we have the data and reference sets, and AI is powering prediction,” says Tim Zoph, client executive and strategist at the Naperville, Ill.-based Impact Advisors consulting firm. “We’re finally capturing data now to be more precise around targeted treatments. We’re having better conversations about the effectiveness of treatments, and analyzing that effectiveness.”

Indeed, Zoph says, “The data is there, as are the processing power and the storage; the algorithms are evolving forward very quickly.” Still, he adds, “I think the ability to absorb this data, and to overcome individuals’ fears of being replaced, will be challenging.”