Gregg Masters 00:09
This episode of PopHealth Week on HealthcareNOW Radio is brought to you by Health Innovation Media. I’m Gregg Masters Managing Director of Health Innovation Media and the producer co-host of the show. Connect with us via www.popupstudio.productions are following direct message me on Twitter by at@GreggMastersmph and that’s Gregg with two G’s. Joining me in the virtual studio is co-founder and principal co-host Fred Goldstein, president of Accountable Health LLC. And if you’re new to the channel PopHealth Week engages top industry talent spanning health systems health plans, physician enterprises, joint ventures employer purchasing coalitions or alliances, and the regulatory community writ large and population health best practices and strategies. On today’s show, our guest is Prashant Natarajan vice president at H20.ai, a market leader in the AI and machine learning space. Prashant is tasked with strategy and product-led growth. He is a best-selling author with his most recent release of Demystifying AI for the Enterprise and also serves as Co-faculty at Stanford Medicine. We discussed the growing body of AI and machine learning applications and use cases both administrative and clinical integrating health data via analytics driving value in the insurance informatics and clinical operational space by AI. Do follow up Prashant’s work on Twitter via @Natarprn @H20ai H respectively and on the web via www.H2Oai. And with that introduction, Fred, over to you.
Fred Goldstein 02:05
Thanks so much, Gregg and, Prashant. Welcome to PopHealth Week. It’s great to have you back.
Prashant Natarajan 02:08
Great to be here, Fred. And Greg. Hope you’re all well.
Fred Goldstein 02:12
We are thank you so much. It’s absolutely a pleasure to get you on. I know we’ve had you on before. But why don’t you give us a quick background?
Prashant Natarajan 02:18
I am Vice President of Strategy and products at H2o.ai, specifically responsible for healthcare, life sciences, and health insurance. And also do a lot of work on the product side of the house also with unstructured data, documents, and medical images. H2o.ai is the world’s largest oldest open-source AI machine learning company. And prior to coming to H2o, I had various stints across the globe at Unum Deloitte, Oracle Health Sciences, McKesson, and Siemens. Great to be back here, Fred, also to talk about Demystifying AI for the enterprise. Our new book.
Fred Goldstein 03:14
Absolutely, you’re a publisher, an author, I should say. And you’ve published a couple books on AI, correct?
Prashant Natarajan 03:20
That’s right, a couple of books.
Fred Goldstein 03:22
And so let’s start out. Can you tell us what’s the difference between AI and some of the normal analytics and things you see, people oftentimes are claiming they’ve got an AI system. But when you look behind the curtain, it really isn’t. So what is it?
Prashant Natarajan 03:38
There is a lot that has been claimed to be an AI system. And if we go down the list of everything that has gone by AI, it reminds me of a joke that my father used to tell me about counterfeiting and Scotch whiskey in India he used to say that there is more Scotch whiskey drunk in India than this made in all of Scotland. So to use that same analogy over here with AI think we see much more AI, or AI washing, as I call it in the book, taking something that is fundamentally not AI is something that is fundamentally the same as it was before. But changing your marketing material to now call it AI-powered, AI-enabled, et cetera, et cetera, et cetera. It’s gone to a ridiculous extent, enough across the board. So if you take at least the conversation between analytics and AI, there are some similarities over there. You could argue, even though there are significant differences, but if you take a look at the robotic process automation, which is really not learning, right, it’s templating. It’s drawing pictures and boxes around very well-defined structured workflows, there’s a lot of AI washing that is happening there. There are CRM systems, electronic health record systems, everybody who’s claiming to do some form of AI, but very few who are actually doing it and even fewer that are doing it right, and doing it transparently and responsibly.
Fred Goldstein 05:25
So for those of us who are less sophisticated in the say programming or AI space, what are some questions people could ask to ferret out whether the company that’s presenting their product really has AI built into it or not?
Prashant Natarajan 05:40
That’s a fantastic question, Fred. And I would say that, the first thing to do is request your readers to buy our new book Demystifying AI for the Enterprise price. Not because it’s a glib answer, because we go into a lot of detail about what is AI and what is not AI. Right. And so an AI system should do certain things at the very minimum, it should be able to leverage data at its fullest, data of all its kind structured data, unstructured data, documents, images, etcetera should be able to put that data to work. How it does that we talked about in chapter one and two. it does that by recognizing patterns. It does that by making predictions. It does that by extrapolating. Right. And essentially, artificial intelligence is very simplistically the process of putting our data to work. Using linear algebra, concepts of statistics, probability, pattern recognition, computer vision, natural language processing, in order to look into the future. And when you fundamentally look at what humans want to do. As human beings, we have never been in favor of looking at today’s news and being happy enough with that we want to look forward, the oldest texts that exist to us today, are surprise, surprise, are ???, where people were trying to predict the weather people were trying to predict the humidity or predict infestation, right? Same thing with respect to predictions of trade predictions of monitoring weights and measures, standardization. So the I would say that human existence, and trade and commerce and education have all relied on a large extent of prediction, with prediction not being certainty. Today, we are at that perfect storm, Fred, where computing power, data technology, human resources, have all come together in a way to make our dreams possible.
Fred Goldstein 08:18
And if we think about the use of AI and machine learning and those tools today, particularly within healthcare, where are we seeing progress being made? You know, everybody talks about it, it’s out there in the clinical setting, it’s in the imaging setting, it’s in the operational areas. Where are we seeing true value today?
Prashant Natarajan 08:39
It’s a interesting question, because I think it depending on who you are in the healthcare continuum, organizations, and people, more importantly, are seeing value across board. So let’s take a provider organization health system, for example, either an integrated delivery network or an academic medical center or a community health center. So what I’m seeing a lot of is just to stick to health systems and health care providers, many uses of AI in the following areas. Number one patient engagement, for example, the recommendation engines, identifying the best educational resources for a patient given that point in time that patient’s needs. That’s not just personalized, but that is also context sensative, using recommender systems, right, that’s one huge area that we are focusing on looking at the supply chain, which has taken a huge hit during this pandemic, and taking a look at not just ?? room and inventory management, but looking into orders looking at anomalies, looking at invoices and purchase orders in order to speed up the supply chain process to make it more efficient to provide relief to the humans of healthcare, looking at precision medicine to combine genomic data and phenotypic data together to be able to create multimodal AI models that not just leverage computer vision, but also natural language processing, and episodic or time series data, across all the way to operational efficiencies in addressing some real challenges with employee burnout, employee churn what is happening with physicians and nurses in our country today, looking at keeping the lights on with respect to cash flow projection with respect to determining the best investments, so that health systems and other organizations are solvent and are able to put that dollar back to patient care where it belongs. There are so many success stories out there, Fred, many of which that we are actively involved in, that we truly are at the cusp of a million streams of value that can come out of AI over the next few years. It’s already happening. But the potential is so much more.
Fred Goldstein 11:23
I know I was on a panel at HIMSS, you know, a few months back, and there was a discussion of some of the work that I think was done at UC San Francisco using AI in terms of the process of getting faxes and stuff. Could you talk about that some?
Prashant Natarajan 11:37
That’s a very interesting use case, huge value proposition for the University of California, San Francisco, specifically the Center for Digital Health Innovation. led by Dr. Aaron Neinstein. Now we work closely with Dr. Michael Blum and Bob Rogers and Marybeth talk and others over there. In order to look at the various PDFs and scans that come into UCSF and UCSF Veritas intake and referral automation system that looks at the millions of documents, they get to separate the referrals from those from, for example, DME requests or school forms and lab results, etc. And then be able to rapidly process those multi-page complex referrals in order to do a few things. The first was to improve the experience of the people involved in the intake specialists to help them focus on specific things that needed to be caught. Rather than doing the mundane, and being able to help them focus on other things that only humans can do. And that machines can’t do. What it also led to, as Bob Rogers and others talk about frequently, is also a dramatic improvement in the overall efficiency of the workflow itself from soup to nuts, not just within the health system, but also with all the other players, most importantly, the patient and their caregivers. And also, as it has been stated a few times now in various forums that you may have also heard, read is testimonials from patients in terms of the improved speed of how they are able to interact with the health system. And also, especially when you take think about rare diseases and terminal diseases, where days actually matter. Such a thing not only brings operational efficiency to the table, which is something that we all expect. It also brings an improved experience for the humans of the head of healthcare. Everybody who’s involved in that episode, all the way from start to finish and most importantly, also has an immediate impact on the patient experience.
Fred Goldstein 14:10
Fantastic. And one of the issues, Prashant I know that it’s come up a few times over the last year or two years is these uses of AI within clinical settings and ultimately discovering that the system isn’t producing the right answer for various groups or subgroups in that population and the bias in the data. What what is causing that and what’s being done about that.
Gregg Masters 14:36
And if you’re just tuning in, you’re listening to PopHealth Week on HealthcareNOW Radio our guest is Prashant Natarajan, Vice President at H2o.ai and best-selling Author of Demystifying AI for the Enterprise.
Prashant Natarajan 14:52
I think it’s fundamental. It’s a conversation that is being talked to a lot h2o ai has always been thinking about responsible AI encompassing things such as reducing, identifying, reducing bias, both known bias, and unknown bias, but also incorporating fairness and ethics into AI. Not as an afterthought, but by design. And upfront. In order to do that, we should be able to use the latest machine learning techniques interpretability, and explainability, which is understanding what the algorithms are doing and what the models are doing and giving the data scientist idea and transparency about what that’s happening to also then explaining the results intended, unintended, positive and negative to the other humans who are not data scientists, who are clinical users, who are patients, who are business users, who are administrative users, that becomes very important. We cannot think of responsible AI as an afterthought, or as a patchwork or as a governance program that can be put into place. Because often we find as several well-known examples have shown us in the recent past, looking at it as an afterthought means that sometimes the impacts and the unintended consequences are well in operation, and that are causing negative things to happen, or are causing unknown things to happen even worse, right, without any knowledge. So I believe strongly that in order for us to do it by design, you got to choose the platforms, the products, the framework, and also, importantly, build a culture of empowerment and responsibility that is baked into these from the get go.
Fred Goldstein 17:04
Are you seeing, I know, this happens, it’s sort of across the board because of various issues in terms of willingness to be involved within a study, or hesitancy to get into a research study, or not being able to get to research study or unwillingness to share data. So we ended up getting data from various groups. And I know of some studies where they look at it and say, well, golly, you know, this is getting a whole lot of white people enrolled in this study. But we’re definitely underrepresented in other other groups such as blacks, or Asians, or Hispanics, or things like that, are you seeing more of an effort to try to build those datasets or get those individuals who are less willing to share their data to understand the importance of that, as well as recognize, you know, potentially risks associated with that as well?
Prashant Natarajan 17:55
The answer to this thread is always, always is we bring more people into the tent under the tent, right, not just into the tent after it’s built. But in the process of making that tent in the process of erecting the tent right. So it’s not enough to say that it can’t be the tent, as in Field of Dreams, build it, and they will come or build it, and we will put diversity programs in place and make sure they come right, that’s not gonna quite work, inclusion and diversity, fairness, and bias has to involve people in deciding whether that is an A-frame tent, or whether that’s a lean-to, or whether it has some other kinds of things because we don’t know what environment we are in. We don’t know the tent that is being proposed and the material that has been proposed to us even right for that environment. So the tent that is being built for a night of outside camping in Watts in Southern California has to be very different from the tent that has been prepared to stay on the top of Fourteener in Colorado. Right. And that starts with some basic criteria. To me, the best way to address these needs is to bring the right individuals, the right communities, and right populations into the mix. If we cannot do that, because there is historical data, well, that wasn’t done in the past and but we can look at it in terms of moving forward and fix it forward. My colleagues at H2o are working on other data science techniques, which look at, for example, the impact of variables, for whom data, for which data may not exist in the past, but looking at the dependency of those variables on the present. So if a population has changed, right. And the data for that population is representative of the population as it existed 10 years ago or 25 years ago, but we know we’ve got a different population today, we’ve got different needs today, we can use AI to do AI, which is one of H2o’s ??, right. So we don’t have to limit ourselves to human constraints and human biases on how to address problems and data of the past, we can use AI to do that as well.
Fred Goldstein 20:31
Fantastic. And one of the other areas it’s been interested in of interest to me is some of the techniques used in this in this field are essentially black boxes. And it’s hard to say, what actually occurred to create that outcome. Any thoughts on that, or how that gets addressed?
Prashant Natarajan 20:50
Healthcare is not a black box environment, it cannot be a black box environment, or it could be a black box environment in probably 1% of the use cases, it has to be a very transparent white box in 99% of the use cases. And the 1% of the use cases where it can be a BlackBox are ones where there is no impact on the quality of care on the cost of care on the patient on the caregivers, their family members, physicians, nurses, etc. Right? It’s very important for it, this is a great question, because the question of black box or white box, the vast majority of AI players out there, put solutions out there I won’t name names for it, but shame on many of them, they put out these BlackBox solutions that it’s almost like well, either use it because we said so or use it or, you know, trust us but don’t verify. That’s another metaphor, and things like that, right. So for, for me, the ability to be able to use interpretability and explainability, but also to use things such as feature importance. And to be able to build feedback loops, which we did in the UCSF case, which we are doing with other health systems across the country, providing humans, not just the ability, but the control over how the machine learning should happen and the retraining should happen. These are all important components of what converts a system from a BlackBox to a WhiteBox.
Fred Goldstein 22:51
Prashant, you’ve kind of been all over in terms of working AI machine learning for a long, long time, you know, and looked at probably not just healthcare, but things in other spaces. Well, what excites you the most What do you think is going to be the most transformative thing over the next five years? 10 years?
Prashant Natarajan 23:10
I’d say the most transformative things from a technology standpoint, from a data standpoint. And I’m hearing this a lot from other thought leaders in the space and also from my customers is an increased focus on unstructured data, we have always been talking in healthcare and other verticals such as banking and insurance, where I’ve been working over the years, as being filled with 80% of unstructured data, I think it’s a low number, it’s probably likely higher from what we have heard. But also, more importantly, AI and machine learning have solutions for us to be able to go after these documents, images, and things like that and actually get rapid value out of it. If we have spent the last 40 years of digital digital transformation data platforms, mostly in capturing the structured data, which is that 20%. So just think of it, Fred, by being able to now go after the 80% of the data that has mostly remained out of reach, I think we are not just going to be able to create value out of that unstructured data for the humans and businesses and companies. But we are also going to be able to generate new value and new sources of structured data as well. I am so excited about that from an enterprise standpoint and as a digital standpoint. I’m also seeing an increase move towards what have you done for me lately mentality, which I think is good because it’s going to force us to focus on success, rather than just activities.
Fred Goldstein 24:57
If you were to say This one thing is achievable within the next couple of years, you know, is it new drug discovery? Is it on an operational perspective? Or are there a bunch of these that are potentially going to break through over the next few years?
Prashant Natarajan 25:16
I think there’s a bunch Fred. Maybe three years or five years ago, if you’d asked me this question I’ve given you hesitating, two or three guesses. Right now, I see such a rich opportunity across the pharma side of the house, not just in sales and marketing, but also Clinical Trial Management, patient recruitment, site selection, protocol, validation, r&d, repurposing, just on the pharma side, right on the health insurance side of the house, seeing a lot of action with documents, again, conversational AI, customer experience, and also population health, disease management, we already touched upon providers, also seeing a lot of efforts happening on the public sector health, they are X and both you and I, Fred, are very involved in various efforts that we do in providing solutions to the countries and the global governments as well. And it just the amount of investments that are happening, and also the value that is being created, will will only increase and it’s just a great time to be in AI in healthcare and life sciences
Fred Goldstein 26:34
well really fantastic. Prashant, I think you covered a lot of different areas. And it’ll be interesting to see over the next year two or three, where we begin to see some of these things directly impacting it, you know, some of it, we see, as you’ve talked about the sort of on the back side, it doesn’t get people so excited, although it’s making huge differences like the UCSF system, but those that can get really relevant to where the patient but as you pointed out, you know, some patients are very pleased with that because of the flow-through, et cetera. But where we really start to see an impact, potentially clinical care, population health will be fascinating to watch.
Prashant Natarajan 27:08
I strongly urge your readers to contact you and us to learn more about these real solutions that we have and how we can work together to democratize AI for healthcare.
Fred Goldstein 27:22
Once again Prashant, it’s always a pleasure talking with you, and thanks so much for joining us.
Prashant Natarajan 27:27
Thank you Fred.
Fred Goldstein 27:28
And back to you, Gregg.
Gregg Masters 27:30
And thank you Fred. That is the last word on today’s broadcast. I want to thank Prashant Natarajan vice president that h2o.ai and best selling author of Demystifying AI for the Enterprise do follow his work on Twitter via @natarpr and@ H2oai respectively and on the web via www.h2o.ai . And finally, if you’re enjoying our work here at PopHealth Week, please like the show on the podcast platform of your choice. Share with your colleagues and do consider subscribing to keep up with new episodes as they’re posted. We stream live on HealthcareNOW Radio weekdays 5:30am, 1:30pm and 9:30pm. Eastern and for you left coasters 2:30 am ,10:30 am at 6:30 pm Pacific for PopHealth week my co-host Fred Goldstein. This is Gregg Masters saying please stay safe everyone Bye now.