"We must utilize AI to change the way healthcare is delivered and how patients can be more engaged in their care"

Q. Can you brief us about Apixio and your interesting journey to how you got here?

Sachin: At Apixio, our mantra has always been to achieve better healthcare through data insights. Apixio has created a proprietary artificial intelligence platform that’s able to render computable data from clinical, administrative and other notes. The text in these documents, the unstructured data, contain 70 to 80 percent of the information about an individual’s healthcare, most of which is not captured in claims or other administrative data. Certainly, you need both pieces, the structured and the unstructured data. We pride ourselves on being able to tackle both of those. What we can do with that data is assemble patient phenotypes from the smart aggregation of various insights that are generated from the two data types. Our artificial intelligence platform can then provide these insights for a variety of different use cases which we can get into.

Prior to Apixio, I was with a healthcare services company and the value-based care space, and prior to that I was in investment banking, a way back when I started my career as an engineer.

Q. Apixio was acquired by Centene late last year. Can you brief us about that?

Sachin: Centene had been a recent customer of ours. They acquired WellCare, who was one of our key customers and they had seen the direction that we were taking our platform and the potential that we had – to use the artificial intelligence capabilities, to improve value-based care, and other activities that are important to a health plan. That was a perfect fit for our next chapter. And importantly, we have had the benefit of having seen 40 plus other customers data inform the quality of our insights that are gathered. Now we have access through Centene to twenty-five million patients’ worth of additional data from which to train our algorithms and develop new capabilities.

Q. Centene is one of the largest health insurance companies in the country with a specific focus on the Medicaid population. Is that right?

Sachin: That’s right. They’re largely focused on Medicaid, but with the acquisition of WellCare they have a pretty significant footprint in terms of number of lives covered in the Medicare space as well.

Q. Where do you see AI in healthcare today and what are the big opportunity areas.

Sachin: In the last handful of months and the last few years, a tremendous amount of buildup in different organizations, plans, providers, analytics firms are utilizing AI to change the way in which healthcare is delivered and how patients can be more engaged in their care. That’s really where the sizzle or the interest lies in any company pursuing their activities within the healthcare realm. So, getting closer to that point of care, getting closer to the patient, that’s where you can really drive some of the changes that are being looked for. But you can’t leave behind the administrative or the plan administration side of it as well that sometimes doesn’t get talked about much. The ability to have a technology platform that works across all those areas and be effective in terms of the access to the data, the analysis that you conduct on it, and mining all of the different pieces of information to form a holistic view of what that care journey looks like for both the payer and the provider. And to the patient as well, as that is how you unlock all of the value. The big opportunity lies is bringing that all together. So, you have got certainly a continuum of folks that operate in different parts, but you want to be able to bring that all together and then have that bear out in the type of care that’s delivered.

Q. The money that’s being made today from applying AI directly seems to be a lot more in the administrative functions where you can see a very direct correlation between what you put in and what you get out of it. Is that right?

Sachin: I think that is spot on. Certainly, with appropriate focus on where we all would like to see healthcare go as it relates to the provider and the patient side of it. No doubt that is where we all want to see improvement. Because if you think about what is caused a lot of the abrasion within healthcare delivery in the US, it’s the burden of those administrative activities that prevent the providers from being able to provide the right type of care. So, we all have an eye towards that. But as it relates to where you also need to have important business focus, it is on that administrative side. And I would say you’re right, certainly in terms of being able to demonstrate a clear ROI and then importantly, as you think about value based care and how those contracts are structured and how you drive the action that is desired from all parties to that set of activities, that’s where you want to make sure that you’ve got the administrative software as well, with the benefit of efficiencies gained from artificial intelligence platforms or other technologies, especially when you look outside of Medicare Advantage and think of the other lines of business that typically don’t have as robust a margin profile.

Q. Tell us how Apixio is bringing about some of these improvements in administrative efficiencies by applying AI in the context of revenue and payment operations.

Sachin: In simple terms, if you’re looking at a series of hundreds of pages of a patient’s chart and you’re a human, let’s make it this area that we’re talking about and particularly our primary use cases within risk adjustment activities. If you’re looking at hundreds of pages of charts for thousands of patients over the course of months, you’re likely to get tired, fatigued, very naturally. It might miss a detail or make a decision that may be as inconsistent with a decision you might have made around two weeks prior with the AI platform. AI type capabilities can allow you to, instead of looking at all 200 pages in this document, see eight pages that matter for what you are trying to do in this activity. It doesn’t have to be a risk assessment, could be anything. It could be a quality initiative and a variety of other activities. And if you’re only looking at those pages and you’re generally guided there and making either a confirmatory decision or you’re saying, ‘hey, actually, I don’t agree, because for our plan X, Y and Z matters a little bit more for whatever that uses.’ You can then make that change and being a lot more efficient with your time. And that’s really where I think you gain those efficiency of scale. Also, if you’re only looking at claims data, you may find 20 to 30 % of the information that’s really rich data. It is the record of truth as it relates to payment. When you think further down into other areas where you would want to expand those capabilities, as we were talking about, point of care, clinical discovery, things of that nature, that’s where you do want to look at the unstructured data. That unstructured data certainly has important details, but also has a richness of data and depth of data from the physician’s notes. So, the physician may code at a certain level and say that I have these two conditions, but they may also add in their notes because they don’t necessarily want a bill for that. But the patient also has these other symptoms that we may want to keep track of. That’s what you also want to know and so that’s where we think the entire profile is important, especially as we talk about things like value-based care.

Q. So, from what you have described, you are primarily talking about natural language processing, is that right?

Sachin: That’s part of it and then there are other techniques as well that can be used to combine for insights.

Q. Did you build the technology on your own? Can you brief us about the evolution of the technology and how you got it to where it is today?

Sachin: Yes, we did build everything. It was purpose-built and was in-house for risk adjustment initially. Certainly, we have used a variety of NLP and machine learning techniques. Think about our platform as it has a core capability of being able to find these insights. You can tune the algorithms to find what it is that you are looking for in a chart. It does not have to be this risk adjustment case. I can then tune those algorithms to find other information, whether it’s a quality initiative. I just want to maybe search in a simple way for all diabetes patients who have had an eye exam or something else. You can do all of those activities by upfront, tuning the platform to run those different use cases. That’s really the way in which we envisioned it. So, think of it as there is this base layer of capability and then on top of that, you build out different applications for different use cases. So, as it relates to risk adjustment, an important area for us to select, certainly because there’s a tangible benefit that folk see right up front in terms of being able to appropriately deliver care for what may be a more higher acuity patient population. It also gave us the richness of data over time. We noticed this after we crossed 10 million patient records from across the U.S. and now we’re worth of 20 million. This diversity of data in the risk adjustment function allows you to have confidence in a narrow confidence interval, in the insights that you’re delivering. That’s really important because you’re going to not only believe in the decision that you’re making as a health plan, but you also want to believe in those decisions being made as a provider to ultimately drive adoption of these technologies.

Q. What you’re really talking about is being compensated for the care that you provide and more specifically making sure you’re not leaving money on the table by missing something in the coding process that could be a legitimate claim for a payment. Is it a fair statement?

Sachin: That’s correct. On the other side of it, one of our important full solution capabilities from a compliance standpoint, you also want to look through and review those same charts and make sure you haven’t previously submitted something that shouldn’t have been. In that case you can proactively flag and note it so that payment is essentially recouped or taken out from what you may be finding for other more higher acuity populations. So, it’s important to do both activities.

Q. So, one of your clients, Centene, puts in a dollar of investment in this technology. What can they expect to get out of it in terms of order of magnitude of returns?

Sachin: So, I think typically from an efficiency of workflow standpoint, customers would typically look for is something in the 4-7 times return in terms of efficiency, of effort, of what’s being done by their folks. And from a dollar perspective, it’s a wide range and it depends on what the initiatives you’re doing. I am not speaking broader to some of the other things that we work on with customers beyond risk adjustment, that can vary a little bit more.

Q. Recently, several emerging data partnerships have been announced – Truveta, Mayo Clinic, Highmark-ChristianaCare. What are your thoughts on this trend?

Sachin: In the last handful of months, I think most of the health systems have come to the realization that the path for them is to have a partner that can help them get there faster rather than perhaps developing the capability in-house. The challenge through all of this is going to be how do you keep that data integrity at a high level? There’s certainly some compliance type of steps that need to be held there, especially as it relates to HIPAA. But if you can clear all of that, then you’ve got high integrity of data and then you need to very specifically define what is the success for this activity that we’re pursuing. I think that is generally alluded to in some of the partnerships that you’re referencing and grow into it over time so that you have confidence that the decisions that you’re making, using those technologies are ones that you can feel really good about. They are not going to either impact you from a financial viability standpoint, but more importantly, that are going to be good decisions for you in delivering care for patients. In one of the organizations that you mentioned, the Mayo Clinic in particular, they referenced that they’re going to be utilizing some of these wearables, technologies and other types of data. I think that’s really exciting and interesting.

Q. One of the things that we see when it comes to applying AI in the context of clinical outcomes, algorithms require a lot of retraining. All the variables need to be adjusted when you are moving from one population to another. So, if you have an NLP algorithm that can scroll through charts and surface opportunity areas, it’ll work just the same in any hospital, any health plan across the country, you don’t really have to do a lot of tweaking to it. Is that a fair statement?

Sachin: It depends. Certainly, there’s different guidelines for each type of organization that you’re working with, plan or provider group that might matter to how you approach each situation. So, there might be custom tuning, but as a general concept, your comment is fair.

Q. You’re a part of Centene now and I guess it is a whole different feel from being a native company. Does your relationship with Centene preclude you from doing any kind of business, especially with their competitors?

Sachin: No, it does not. That is the short answer. So, part of the focus of this transaction, and in particular one of Centene underlying thesis, was that we would continue to sell externally and focus our efforts equally there. The simplest way to do how Optum operates within United in serving both the parent company as well as the broader market. So, we continue to work in that regard to win and have won new contracts with other players in the market.

Q. What are the big challenges for AI that the healthcare industry needs to address before we can realize its full potential?

Sachin: I think there is the widespread adoption or the way in which you drive this fast or appropriately with what are the privacy requirements and what is covered under HIPAA and what other considerations do you need to be aware of? There are other government task forces around this that need to be kept in mind. So, it’s the appropriate attention of how fast technology firms would want to move to say – ‘yeah, give me all the data and I’ll run it through, and we’ll get you that much more high-quality insights and analytics.’ But on the other side, you have to move at the right speed. I would say that the ability to get there should be picking up pace as you start getting folks comfortable that you are able to maintain the integrity and security of the data. That happens with more and more players now. Sometimes a big situation that comes up at some point in the future and there’s a breach. Someone is exposed and that becomes a concern. So, with more firms being focused on that as a table stakes item to be successful in winning new engagements with plans and providers, I think it drives some of the discipline even more so around that. When you think about the different axes of how you propagate or become competitive in healthcare analytics with the use of AI, there are three different vectors: there’s the quality of your data science, which you have general control over; there’s the quality of data volume or the quantity of data volume. This is when you have enough diversity of patient data from which to feel comfortable, or can certainly say – ‘hey, for this case, I don’t want to have too much of a bias in this direction or that direction.’ And then there’s the data liquidity piece and it’s really the data liquidity that’s going to be a rate-limiting factor here when you think about those three vectors, because that is driven by not only decisions by health systems and providers, but also from a regulatory standpoint.