Generative AI and analytics in Healthcare with Prashant Sarpamale

Generative AI and analytics in healthcare represent a powerful synergy, where generative AI’s ability to create data and insights aligns with analytics to drive data-driven healthcare improvements, from personalized treatment plans to operational efficiency enhancements.

In this week’s CDO insights we have with us Prashant Sarpamale, the CEO at Althea.ai. He is an executive with extensive experience in conceptualizing and leading Data & AI initiatives at scale within the Healthcare Payer and Provider segment. A results-focused leader with a strong focus on innovation and harnessing the full potential of teams. 

He talks about how healthcare’s transformation through data-driven technologies and Generative AI, inquiring about the origins of these discussions and seeking examples of successful use cases. He also unwinds the differentiation between AI and Generative AI’s productivity enhancements, considers a hypothetical impact on past Covid-19 conversations, and talks about recent observations and future adoption predictions in healthcare.

AIM: How has the evolution of data-driven technologies in healthcare transformed the way data is structured and utilized, from unstructured data a decade ago to its widespread application in day-to-day healthcare practices, including by doctors and other healthcare professionals, and how does this relate to the emergence of generative AI in healthcare?

Prashanth Sarpamale: Back in 2012 when I joined the healthcare industry armed with a lot of knowledge and data and analytics but with absolutely no knowledge in healthcare, it was a short introduction when looking at the state of maturity, from a healthcare perspective for analytics. It was primarily a B2B industry and not so much a B2C and that has repercussions in terms of how data is employed. It was a data rich industry and there were a lot of quantitative methods employed, but primarily in a very structured fashion. You can think about a lot of data sets in the healthcare industry and they’ve traditionally been extremely structured. You can think of HCC, CPT codes. You can think about how we go about billing, diagnosing diseases etc, and traditionally, all those applications, how you go about setting rates, and about measuring things etc, have been very structured and that was the way how things were done for a long time. And that was what was driving the state of analytics at that time. But somewhere around the time when Obama care got passed the entire focus started shifting from B2B to B2C and sort of placing the consumer first. So the entire concept of the Triple Aim came into focus, which was on number one around quality, number two patient outcomes and number three around cost. So to drive that and to put consumer experience first, you’ve got to put yourself in the consumer shoes. Understand what the consumer is thinking and the consumer is a very amorphous being. In terms of understanding perception, behavior, longitudinal consumer data, etc,  all these are very retail, analytical concepts, which are slightly more probabilistic rather than being very deterministic in terms of how the industry and the data sets. And everything was sort of structured that sort of created the shift to analytics. And predictive and prescriptive analytics, sort of drove a lot of the transformation from a technology perspective, etc. Because now you have to understand digital, surveys, and unstructured data because a lot of these data sets are coming in ways and shapes which you’re not used to, which are not structured. So you have to move from relational databases to non-relational databases. So that sort of drove a lot of those changes and started making the move towards more and more sophisticated ways of doing analytics. And then finally now we are at a point wherein we are saying “We can drive diagnosis, we can improve drug development, etc.” So, I don’t know whether we’ll get there because the industry sometimes gets its own way. But it’s exciting to see that journey, that sort of gives you a sense for how you sort of evolved as a healthcare industry.

AIM: How did the conversation around Generative AI in healthcare begin? Were problem statements in the healthcare industry the starting point for considering how Generative AI could be applied, or was it initially driven by speculative interest and hype? Please share insights on the origins of these discussions.

Prashanth Sarpamale: I think a lot of us in the industry sort of started off in the same way that a lot of people externally have started off. A lot of us have been observing the buzz and seeing the movements in technology. I think this all started out as a huge improvement from a large language model perspective. As a technologist, you start taking a look at what these models do, how they are different from previous iterations and generations and how they’ve sort of improved the state of the art. But what is critical is, how you apply it into the healthcare industry and that’s where the rubber meets the road. And I think just as a practitioner of healthcare analytics and somebody who’s been in the industry for some time, you start applying it to the unique context of healthcare which is quite different in a lot of other industries, they might go for the most sophisticated applications right away. Because that is where it has the largest impact. In healthcare it is quite different. You probably go in for applications which have the least disruption first. Because this is a bit counterintuitive, but think of the way that the new drug is introduced to market, or a new treatment is introduced to market, it goes through a lot of trials and, you need to ensure that the efficacy and effectiveness is 100% or very close to, very high or rather improves upon the current state of the art. So only then it’s sort of released into the market after a lot of value discussions. 

Why? Because the cost of failure is extremely high. Think of any false positive that could come about as a result of these models, and any treatment that occurs. The cost of those failures could be extremely high, therefore, you sort of start off, especially in healthcare with lower impactful but something that sort of has a bigger impact on your day to day work. Think of productivity related applications. That’s where my perspective is, it will start and it has from an actual application perspective. There has been some very, very interesting work which is sort of coming out of a lot of startups. Some of the large companies around things like drug discovery or image analysis, and they’re hitting the headlines. But, in terms of actual application, this is how we play out, and that’s my perspective.

AIM: Could you provide examples of near-success use cases for Generative AI in healthcare that initially began as research but are now transitioning into production within enterprises, considering the critical nature of healthcare applications?

Prashanth Sarpamale: Some of the actual applications that will sort of pan out initially in every industry, are productivity related applications. Even things like clinical research the way it’ll actually impact that entire industry is it will shorten the timeline of doing research or it will improve the quality of research. For example, in healthcare IT co- pilots in terms of coding or looking at coding quality etc. Right from that base layer to the next level in terms of how you go about the entire healthcare processes. In terms of how pre-authorizations are done, how claim processing is done, how engagements with customers are done in terms of chat bots, that entire process should get much easier because of these generative driven applications. When you sort of go through an appeal process for a claim, think of how difficult and how painful it gets if the entire process gets much simpler because it’s easier to sort of take a decision on that case within the insurance company and get that communicated out to you in a much simpler and in quick fashion. So all these productivity related aspects in my view will drive the initial value. Even things like image analysis, in a doctor’s office or in complex cases, you want to sort of utilize imagery to drive better decision making. This could be a pre-authorization decision, or this could be a diagnosis related decision, this could be medical treatment related, analysis, etc. Being able to sort of fuse a lot of this imagery and actually marry that with longitudinal clinical data and provide insights will sort of improve. But at the end of the day, what is it really doing? It is improving productivity across the board, improving accuracy, and improving quality. So that’s really the implication of generative AI in a lot of these processes.

AIM: How can you help us differentiate between the potential productivity improvements offered by AI and Generative AI within a specific context?

Prashanth Sarpamale:  I look at it as sharpening the sphere. If you look at the previous generation and where we are headed to with generative AI. So for example same medical imaging from a generative AI perspective, you are able to use something called generative adversarial networks (Gans) to drive better imagery. What does it do? It sort of brings together, significant amount of data, which is already there and simulates and gives you significant amount of synthetic data, which is similar to what is out there in the real world so that you can compare and contrast and be able to sort of pinpoint and be able to simulate pinpoint in a much better fashion. So the quality of your diagnosis will go up, your ability to predict and pinpoint much earlier in the process will go up. So the ability to prevent intervention and drive better outcomes will be much improved. And the benefit with some of these micro improvements is significant. For example, if you are able to sort of move to an earlier stage of detection in kidney disease that is a significant improvement because, keep in mind that 5-3% of all cases drive 55% of medical costs. That’s a very important statistic to keep in mind because it’s a very small sliver of people who are chronically ill or significantly ill, who drives significant medical cost in the system, and all the treatment and drugs everything goes to this small population and the entire focus is to ensure that more people don’t get added to their population or how to reduce that population. So that we can prevent people adding to that small sliver of population who drive the significant medical costs. Now, if you consider detect much earlier or slightly earlier, then that’s a significant win in terms of cost, experience and lifespan, for that person. So that way it is sharpening the sphere.

AIM: Hypothetically, how might conversations with leaders about Covid-19 have played out differently two years ago if we had the capabilities of Gen AI, especially in terms of accelerating research efforts?

Prashanth Sarpamale: Yes, I would agree with that contention. So improving the productivity of the research process is shortening the time to market is something that technology is getting better and better and generative AI is sort of pushing that envelope to that extent. Maybe we could have said reduce the time to market there.

From a covid-19 perspective, the other thing that comes to mind is, if you  step back from the entire drug discovery process, clinical research, trial and look at the social impacts in terms of how covid-19 was managed. If you look at how covid-19 spread across the world, it went from the east to the west, the initial wave and a lot of the west sort of reacted very late. Primarily because there wasn’t a lot of awareness in terms of what is happening,what is coming? 

My belief is that, increasingly, we will be able to harness data and be able to see a lot of these Black Swan events or should be able to see Black Swan events sooner as these ripple sort of flow. You think of putting us on a stone in a pond and then the ripple sort of spreads in the pond. If you are at the bank of the power and the ripple is coming in the middle, then it takes a long time to sort of reach the bank, but you should be able to shorten that tail if you will significantly with, better analysis and better understanding of how these things are going to play out with the, power of technology. For example, if you were able to synthesize a lot of the conversations. Just think of a simple application.

Go to Bing or go to Google today and say  “What is Covid-19?” And immediately the copilot will give you a list of very smart curated answers saying this is what I should do about it. It will immediately tell you to do this. Based on the power and understanding of the web. Was this there in 2019? No. Could it have helped? Absolutely it would have helped. Because it would have immediately told you on your phone that or wherever you searched, this is what you’ve got to do about it. When I say shorten the tail a shorten or the ripple, this is how it would have shortened. It would have reduced the number of casualties significantly, and that is the power of technology. 

AIM: What’s your recent observation regarding the adoption of Gen AI-based solutions in healthcare among different stakeholders, and how do you anticipate this adoption evolving in the healthcare ecosystem over the coming years?

Prashanth Sarpamale:  The adoption cycle will move exactly like I said which will initially start off with the back office or a lot of the healthcare processes and improve the healthcare processes. Then it’ll sort of spread into more medical domains in terms of adoption. As and when more evidence is collected, in terms of efficacy and effectiveness, then it will move into the provider’s office.

Another thing to remember is that providers typically and I don’t mean this in any negative sense, but typically providers don’t have the time in order to engage in deep conversations on analytics or technology. A lot of this has to be matured by the time it reaches a provider’s office. A provider isn’t going to sit in front of his or her patient and actually prescribe something or utilize a new technology until he or she is comfortable with it himself or herself. Typically that takes a bit of time on that bridge. So the way it’ll reach the provider’s office is simple productivity related application. For example, today, providers spend a significant amount of time doing a lot of documentation after each visit. I see that getting cut down,

Things will be built right into the EHR (Electronic Health Record) in terms of AI driven applications that will improve. People will start seeing the impact of that. After that, we’ll move into diagnosis, medicine, and prescriptions. And that’s the real sort of move.

AIM: Any concluding thoughts?

Prashanth Sarpamale: While we talk about Gen AI, any AI and any technology in the healthcare industry, I think there are two concerns that we have to keep in mind. First as a chief data and analytic officer as somebody who drives budgets in addition to outcomes and services. One thing you’re always worried about is cost. The outcomes or the value that you generate needs to outweigh the costs involved. Doing increasing amounts of AI driven activity is very expensive in terms of server and processing time etc. The amount of horsepower that you need to throw at the thing is significant. And the amount of time spent in experimentation with regard to these data sets is significant before you start to see value. That is one concern that people will have and that is my point about maturing of the technology before it sort of reaches the provider office. 

The second thing is ethical. In the healthcare field, there will be ethical considerations, as it moves deeper into the clinical trial space, into the drug development space or even treatment space. One thing people will be concerned about is, because Gen AI will throw up a lot of possibilities. Think about utilizing Gen AI and a molecule development or a drug development process. The number of options and combinations that it might throw up is significantly larger.

As compared to the past, it might throw up combinations which are not even thought of. And some of those might have ethical considerations linked to that. But increasingly this will throw up and open windows and doors which have not been even considered in the past and it will throw up a lot of ethical debate.

This sort of relates back to what a lot of AI leaders pointed out to the US Congress sometime back. In terms of, “Hey, we’ve got to get more thought process around, how to regulate around how to govern the entire process of using AI.” And for example, if you’re going to put a chip inside someone’s brain and that becomes a conduit by driving both data as well as enabling technologies for that person. The intent is noble but tomorrow that could open up a lot of possibilities and once it sort of becomes widespread, those applications will drive a lot of questions. So the entire ethical field needs to sort of develop. It’s not just a personal thing but having answers to some of those and having some of the guardrails will be most important as we go into healthcare. In this field because it is so closely related to health and people’s lives, people will have ethical concerns and something that the industry needs to be very very aware of.