How AI and interoperability affect the spread and treatment of COVID-19.
Artificial intelligence (AI) has an opportunity to help patients and providers alike during the coronavirus disease 2019 (COVID-19) outbreak. AI has been known to speed up workflows and administrative processes, along with improve patient outcomes. If used correctly, this technology has a chance to help mitigate the spread of the virus.
Darren Schulte, MD, chief executive officer of Apixio, discussed in an interview with HCPLive® how AI is being leveraged during the COVID-19 outbreak, how a lack of interoperability between electronic health records (EHRs) could lead to worse patient outcomes, and the role of digital data during a pandemic.
Editor's note: The following interview transcription has been lightly edited for style and clarity.
HCPLive: Thank you so much Dr. Darren Schulte for joining me today. How are you?
Schulte: I'm doing well, and you?
HCPLive: I'm doing great. I'm hoping you can start off by letting us know how AI is being leveraged to improve processes during COVID-19?
Schulte: Probably in 2 ways. I think the first is to try and get a handle—and certain countries have tried this—on the spread of COVID-19. So, trying to determine based on testing which positive patients have interacted with other individuals that then would need to quarantine or take additional self-isolation measures. Certain countries have done so; a lot less so in the United States given the relative unavailability of testing and testing kits.
I think the second main way in which I've seen it being used is to determine characteristics of individuals who are COVID-positive, especially those who ended up being more severe in terms of their illness. Based on those characteristics, can we get ahead of those who might fall into those high-risk categories for additional quarantine or social distancing.
However, that said, most of the country and a lot of other countries have already gone into shelter-in-place. And so, we're already doing quite a lot to keep people from interacting with others, with the exception of doing things like trying to get groceries and the like. But for those in the highest risk categories, even going to get groceries or other essential items should be something that they have others help them with.
By getting characteristics of those highest risk individuals, in addition to just those who are older, among us, I think is how AI and leveraging that across medical records can be useful today.
HCPLive: Are you seeing that AI is being used more in particular areas of the country or in particular medical centers? Are there any trends related to that?
Schulte: Hard to tell. I think medical centers are probably faced more with real issues of protective personal equipment shortages—how do I get ventilators, how do I set up intensive care units, how do I set up hospitals in terms of additional beds. So, they're kind of more on the frontlines of immediate care.
I think what I'm talking about is more public health measures that are being taken into account by public health departments, researchers, health systems. So, what would be the best way then to use AI and technology during this time? I think that at this point, like I've mentioned, taking into account those characteristics of people who are going to suffer more from COVID. We're starting to learn that your body's immune reaction to the disease is potentially what's causing some folks to have particularly worse outcomes. But we'll learn that through determining from thousands of medical records and lots of data, how that's in fact, the case.
And as I mentioned, also trying to determine spread, communicability, transmissibility from smartphone data and other data when other countries are using that. Probably in this country. I'm sure privacy concerns would be invoked, but I think that's where we could see it.
I think you're going to see a lot of things after this pandemic that we’ll learn to prevent more of this than that. Future more intensive testing, determination, how this spread is evolving across this community.
For now, I think we're learning how AI could be useful, and probably more so for the next pandemic than this 1, given that it's well spread across most countries of the world, unfortunately.
HCPLive: Is there are a particular level of training and understanding that physicians need to have to use AI?
Schulte: I think if built correctly, no, except for the data scientists who are considering how best to use the techniques. But if done well, it should be something that should inform with insights, physicians, public health officials—those that are helping to advise cities and localities how to react to a particular infection. What's the best social distancing required? How do we balance between way of life versus isolation? Hopefully it doesn't take a PhD to understand it. It might take certainly PhDs to create the algorithms that derive insights.
Another key area that AI can be helpful with is rapid discovery of drugs that can be useful in treating the disease and looking at the drugs being used in individuals, its properties, and how 1 can use that to fight the infection. Secondly, go through particular ways in which vaccines can be developed. I think AI techniques can be leveraged to create the types of vaccines that can be most effective, because you want to get a particular candidate in testing as quick as possible so that we have a chance to develop a vaccine within a reasonable period of time. And I think new machine-learning and AI techniques can be used to create vaccine candidates, as well as determining particular drugs that can be used to fight the infection in individuals with COVID, for example.
HCPLive: Right, and then that would obviously speed up the whole testing and diagnosis process.
Schulte: Correct. So, definitely treatment. So, on the 1 hand, there's: how can we limit spread? How can we test effectively? How do we determine that? Who should be quarantine? And then when you're ill: what drugs are effective and ultimately, how do we get a vaccine into the marketplace so that we can effectively create the immunity we need across the population. And that's the best way and the ultimate way to solve this pandemic.
HCPLive: Along with AI helping with drug discovery, it's also helping with the rapid dissemination of information. Can you dig a little deeper into that?
Schulte: I think that we're talking about how to get in place trials like the 1 that the World Health Organization is doing to try to assess for different types of drugs versus usual standard of care and determine rapidly across dozens of countries what works and what doesn't. Data and data insights can be then analyzed, and relatively quickly we get answers as to what works and what doesn't. Certainly, drugs like the anti-malarial drugs that people have discussed, chloroquine has been considered, HIV drugs, anti-TB drugs are among the candidates. So, it's important to try to get those studies done in an effective way, in a way in which it's going to determine truly if the treatments are effective.
HCPLive: Taking a step back, you've mentioned a lot about EHRs and data. How does a lack of interoperability play a role in this pandemic?
Schulte: It's actually 1 of our Achilles heel in a public health respect, which is to say, if we had a data pool of information that is continually being fed from individuals in terms of their health records, that then we can use in a secure and deidentified manner, we can get a handle on more quickly, those who have worse outcomes and why which drugs are effective and why and what characteristics are among those that tend to recover fast. If we had 1 pool of digital information to analyze, we'd be in much better shape than we are now where every hospital has their own dataset.
Every public health department's relying on only self-reported information for the most part. We are really seeing the negative effects of a lack of interoperability between EHR systems, which is truly a shame in the time in which we really need it the most and the time in which we can really use AI to rapidly determine with the digital data the answers to the questions that we have.
HCPLive: Is it possible then that this is leading to worse outcomes for patients?
Schulte: It might. Or at least it leads to a lack of understanding of drugs that can help future patients—those that will present for hospital. If we have more effective treatments and we have data on it, we'd be more comfortable in telling physicians what to do, how to do it, and disseminate guidelines of care that can help now, as opposed to months later when the dust settles and then we have a handle on what works and what doesn't. But then we've already run its course.
HCPLive: How are government organizations working right now to try to improve these processes and make interoperability a little bit easier?
Schulte: I'm really excited to see that ONC recently published rules that relate to the 21st Century Cures Act, which was passed several years ago. These rules relate to, in creating more interoperability between records systems, blocking what we see as a lack of information sharing, they call it information blocking, prohibiting that activity, and mandating EHR systems set up and providers use APIs, which are ways that electronic systems speak to one another. Set up these APIs based on universal standards accepted at the national level. This would allow for data sharing that would be seamless and certainly a lot easier to use. Patients can get access to the data as easily as sometimes health insurance plans are able to, or other hospitals can, although it's still today with a lot of effort.
So, I'm excited to see that. It's going to take 12, 24, 36 months, it's going to cost money, but at least we have focus from the national level. Congress is behind it. HHS is behind it. And I believe that post-COVID, we’ll have even more incentives to do this, given the types of things we discussed and the types of benefits we would get from a common data pool and the ability to analyze it and derive benefit from it.
HCPLive: How can AI be used to help patients who have comorbidities or who are at higher risk or at least to help flag down those patients who might need extra attention?
Schulte: If we know that certain characteristics like age, comorbidities, autoimmune disorders, are associated with worse outcomes, we could be more aggressive in treating those patients earlier in their course of their disease. We take action sooner when there's early indication of symptoms of the disease, then, certainly with intel and insights, we can then take action to avoid worse outcomes.
Today, unfortunately, we're trying different things, but for the most part, we are doing the best to provide supportive care measures, maybe some drugs being tried, but in some cases, unfortunately, it's too late by the time you're in the intensive care unit or on a ventilator, which hasn't shown for the majority of individuals to be effective once you're that advanced in your illness.
HCPLive: Are there any other challenges that are present that AI might be able to help with?
Schulte: I think it is surveillance of populations early on so that when we see that there are new and novel viruses and outbreaks and that we determine the communicability. How fast does it spread? What in the public has been described by epidemiologists as “are not,” which is if I'm infected, how many other people that I'm around get infected? Knowing that early and understanding therefore, how aggressive to do testing, isolation as a result of coming in contact with those positive, the sooner we get that data will inform public health departments to take action sooner.
We've seen in the Bay Area that shelter-in-place—I think we've done it earlier than any other municipality in United States has—I believe caused us to begin to truly flatten our exposure curve. That was an aggressive stance taken early. With data though, we don't have to have guesswork, we can have a natural response based on information that can help us make better decisions.
HCPLive: Do you have anything else that you would like to add about how AI can be used during this time?
Schulte: I don't, but I would say that the most important thing is availability of data. So, if we can all understand that it's not enough to have data that are electronic, rather than handwritten and data that way. When I trained, we used to record information about care. It's not enough if you don't have digital data widely accessible in a secure manner. But that becomes really critical for public health. Public health in the 21st century is based on data and data insights, without which we are just guessing on what works and what doesn't, which is unfortunate with a rapidly spreading virus like the 1 we're unfortunately having to deal with today.