Even in the midst of the global pandemic created by the COVID-19 virus, interest in artificial intelligence (AI) and machine learning continues to move forward. Indeed, hospitals, medical groups, health systems, and health plans are leveraging AI and machine learning to manage clinical, operational, and care management issues that the pandemic is creating across the U.S. healthcare system.
At the same time, the future of AI and machine learning remains an open chapter in U.S. healthcare in the coming years. One person who has done a good deal of thinking about the potential of AI across U.S. healthcare is Tushar Mehrotra, senior vice president at the Minneapolis-based Optum, a data company that, according to its website, uses “advanced analytics and emerging technology [to] help simplify industry processes and deliver actionable insights to support health leaders and professionals,” including through the use of AI and the Internet of things (IoT).
Recently, the Washington, D.C.-based Mehrotra spoke with Healthcare Innovation Editor-in-Chief Mark Hagland regarding the near-term future of AI and machine learning in U.S. healthcare, across the provider and payer sectors. Below are excerpts from that interview.
Looking at the landscape from a 40,000-feet-up standpoint, what is your current perspective on the near- and medium-term potential of AI and machine learning to help effect some of the transformations that need to take place in U.S. healthcare?
The reality is that we’re just scratching the surface of what AI can do in healthcare. Many industries—financial services, retailing—have been investigating in the models and technology for over a decade, and are already reaping the benefits. Healthcare is just starting out. And even within healthcare overall, biopharma, pharma, are much more advanced than other sectors of the industry. And health plans are willing to spend more to invest in areas like predicting claims denials. Providers are the least mature, and even within the provider space, it’s diverse. Intermountain Health and other integrated systems have invested in certain areas, such as using AI to predict disease prevalence in their market, and what to do about that, and in radiology. In the next two to four years, you’ll see a much higher ROI from their investment in AI.
Has there been a consistent approach to the harnessing of AI across the health plan sector?
It’s definitely the larger health plans—Humana, Cigna, Aetna—that have put into place large data science teams. They’re exploring partnerships with Epic and some EMR vendors. And they’re looking at the administrative/automation side of AI, to better understand populations, and to automate processes, and to predict which members will need help, and when. So those are the three core buckets: the automation/cost savings side, the clinical side around population health/consumerism, to better understand members, and design better understand digital solution.
Would you agree that there’s been a growing digital divide between big integrated health systems, and community hospitals, when it comes to the leveraging of AI?
Yes, absolutely. Margins are so slim as it is, and technology isn’t a core competency for the community hospitals. They can’t even get basic reporting done right. I was with a CEO at a hospital in Louisiana recently who was frustrated by that. So they believe that AI is a buzzword that they need to figure out at some point, but it’s just too low a priority. You won’t see an uptick in those markets unless it becomes CEO-driven, and unless you get some sort of support from vendor partners.
In that context, AI is not a magic wand that can achieve results simply by being waved around, correct?
That’s right. First and foremost, it has to be linked to some sort of problem you’re looking to solve, as well as some sort of clinical or business strategy. I was with the CEO of an academic medical center a few months ago, and he said, I’ve invested $5 million in this, and it’s become just a research tool. Number two is, this is a unique cross-cutting area; you can’t just buy a technology and think you’ve addressed your AI problem. You have to have data sets that are rich and clean enough; you need a good platform; and you need solution architects and data scientists. And do you have the ‘translator’ to be able to help you interpret the insights from the models, and build them into action. It’s tough, because you talk to people and they say, we want this technology. But you have to link it to some sort of clinical or business outcome. Just as with traditional analytics, you have to link it to some sort of use case.
Among those patient care organizations that are doing this right, what are they doing right?
First, the leaders of those organizations have in their minds a clear link to what they’re trying to solve. Second, they’re not just hiring 20 data scientists at once. They’re moving forward by building individual use cases. One organization is looking at kidney disease in their commercial market, for example. And they’re being thoughtful about who they hire and what they want to accomplish. You see all these things like Ascension/Google, and not everyone is comfortable with that. But you can’t do everything on your own, either. So some organizations are becoming very thoughtful about with whom they partner externally. S the organizations doing this well are being very thoughtful about it. And Intermountain has 200-300 informaticists involved in this. That’s at one extreme end. And when you talk to their chief analytics officer, he says, our CEO wants us to show ROI for every single analytics professional.
Where and at what pace, is this headed in the next few years?
I think the pace will still be fairly gradual, particularly in the health system space. I don’t think you’ll see 40 health systems with massive success. I think you’ll see pockets of success, particularly among academic medical centers and major integrated systems. Where it’s heading—I think you’re going to start to see a more robust set of areas in which organizations are going to get more involved, including disease prediction, imaging diagnosis in radiology, pharmaceutical adherence and treatment, site-of-service coordination and predicting when to discharge; those are all clinical use cases. On the administrative side—claims denials. There’s a huge cost involved in claims denials. So can you predict that in advance and address challenges upfront? So, prior authorization prediction. And even provider burnout. I think you’ll see more dabbling.
Will the COVID-19 pandemic slow progress or speed it?
If I had a crystal ball, I think it will be a mix of both. We’re spending a lot of time around COVID demand and infection prediction, and supply chain for ICU beds, ventilators, etc. And so many organizations are building pretty cool tools to help manage what’s happening in their markets, such as Penn Medicine. But at the same time, there will be slowdown in certain areas, as elective procedures are being diverted. It’s unclear to me about the depth and scale. It will depend on how quickly you can flatten the curve, to some extent. And this will be used for the financials, and for workforce management. In my eyes, that’s where the focus will be, on modeling those things out.
