One of the biggest “buzz terms” in healthcare and health IT is undoubtedly artificial intelligence, or AI, but there are still plenty of questions and debate on how AI can best be leveraged to lower costs, industry-wide.
Leonard D'Avolio, Ph.D., an assistant professor at Harvard Medical School and CEO and co-founder of healthcare technology company Cyft, based in Cambridge, Mass., has a unique perspective when it comes to the best applications for AI in healthcare. Indeed, while artificial intelligence is often labeled as a “solution” that people buy, D'Avolio sees it differently, calling it a “capability”— particularly in the realm of healthcare’s value-based care evolution.
To supplement our Top Ten Tech Trend piece on AI—Healthcare Informatics spoke to D'Avolio about his viewpoint on AI, and where the biggest opportunities lie as healthcare leaders continue to look for innovative ways to improve quality and lower costs. Below are excerpts of that discussion.
As things stand in this space today, what do you see as the best use cases and greatest areas of benefit for AI in healthcare?
AI, as a capability, allows us to optimize, analyze, customize, and process things in ways we haven’t been able to do in the past. For health plans, especially those that have intervention capabilities, like Medicare and Medicaid Advantage plans, they [take on] all the financial risk for keeping people healthy. There is an obvious opportunity there for getting right care to the right people, optimizing operations, and using the dollar as efficiently as possible.
But hospitals are a totally different game; they get paid on the number of people they see and the complexity of the care they deliver. So for them, the most obvious opportunities are to see more people more quickly and to improving billing, to capture the full complexity of what they do. These are extremely different applications of what is in effect technology that lets us learn from data in different ways.
As we move to value, if any organization, whether a hospital, health plan, clinic, or anything in between, is given a fixed amount of money with which to operate as effectively as possible, the idea that they would not employ the best possible software for discovering how to operate as efficiently as possible is laughable. If true value-based care occurs, claiming to have a solution with AI involved will be as dramatic as claiming you have software that employs a database. If your job is to operate efficiently, and there are technologies to help you discover the best ways to do so, you are behaving irresponsibly as a steward of that organization if you are not using that technology. The challenge in healthcare is not whether or not this math lets us discover operational efficiencies, but rather what are the sub-industries and problems where we could get paid to do a better job?
The most obvious application of AI is not at the bedside; that’s the bait. There are probably two dozen better applications that we are doing today, or will be doing, that do not involve affecting a clinician’s decision making at the bedside. Probably the biggest mistake that IBM made, and I do get it from a marketing standpoint, was choosing to be able to “revolutionize cancer in two years,” versus doing something that is a bit more operationally focused where the stakes of the decisions aren’t so life-and-death.
Behind any one clinician making a decision are thousands of operational decisions that move supplies or paper from point A to B, and streamline phone calls. So there are just so many areas to improve the efficiency of healthcare. If 70 percent of the cost of care is administrative, as some estimates say, shouldn’t that be the focus? It is dramatic to say that life-changing medical decision-making is being passed off to a robot, but that’s not where this stuff is making a difference today.
The whole idea is using these technologies to identify people most likely to benefit from specific care management, intervention and attention. So, the patient is not in front of you and the AI is not saying “give them this drug,” but rather you have a care management team and on any given day they have to decide how to best allocate their time. This is a good opportunity for AI to use all of the data at our disposal, be it EHR (electronic health record) data, claims data, or device data—to use all of that to say if you are going to try to prevent falls, this is the list of people, in order, who you should approach with a falls prevention program.
Leonard D'Avolio, Ph.D.
Do you believe that hospitals and health systems fully understand what the best use cases are for AI, or is there a disconnect there?
We have had to learn how to help teams understand how AI works and how to evaluate it with their own data and patients, so that they gain faith in it. You have to ease them into it and you can’t do it in chunks. In time, you can win confidence. Machine learning is 10 percent of the job and workflows, while process and education is the other 90 percent. That’s the greatest challenge—in healthcare we are used to buying solutions, handing it to IT and having them do the installs. But this works differently; it’s about partnerships, learning and adaption. Amazon doesn’t install a third-party population bookselling algorithm and Google doesn’t consider data to be the exclusive property of IT. They succeed because they have re-thought how they approach learning and improvement. Data is the grease between the gears, but it’s a means to an end. It’s really about change management.
The biggest thing that’s missing is that AI is a capability; no one buys it. This is no different than what happened when the Internet became popular in the 1990s. People started spending billions of dollars on Internet-enabled stuff and they had no idea what they were buying, but there was a fury of believing they needed to get involved. When it settled down, it was clear that the constant connectivity would change how you do business. But now, what’s happening is that everyone in healthcare is saying that AI will be important, so they’re talking about it as a solution. But it’s not.
The real challenge now, and what you have to bridge as you get to value, is that people have to understand what AI is and isn’t. And if it’s a capability, you have to understand what problems that capability can help solve. The hard work of solving problems is no different today than it has been forever. But anyone who is thinking about AI as a thing that you purchase and install has put themselves in a position to fail expensively.
When you hear concerns about AI someday replacing certain professions in healthcare, does that strike you as valid?
It’s not a valid fear. It’s just something that sells stories because talking about replacing humans is something that’s super interesting. If you are responsible for leading a healthcare organization and moving toward value-based care in any way—and this isn’t an IT decision— and it’s a CEO, CFO, CMO, and CIO decision—you need to ask yourself, would my organization benefit from a different and better approach to learning from our data? Your team should be creating a list of operational challenges that would benefit from a better understanding of what’s really happening, what is likely happening, and as a result, what’s the best choice to make in light of all of the data we have.
If you are in the value-based care realm, and you cannot come up with at least 10 operational and clinical opportunities, you are probably not paying enough attention to your organization. In a value-based environment, the idea that in the next three to five years, machine learning will not be a core competency, is incorrect. If you are truly at risk and you are not employing the best approaches possible to learn from data, you will be in [trouble].
Can you offer a timeline for when AI in healthcare will be much more rampant?
None of what I am talking about is applicable to fee-for-service healthcare. Incentives are totally different; only scheduling or billing matters in that environment. But then you have the 13 percent of healthcare delivery that is truly value-based. Right now, 40 percent of Americans have some form of coverage via CMS (the Centers for Medicare & Medicaid Services), and by all indications, reimbursement from those models of care will become increasingly value-based. You will need to adopt the tools that help us learn from our data and operate more efficiently.