Siftwell Analytics is a prescriptive analytics company that provides health-related social needs data analytics to health plans. Trey Sutten, co-founder and CEO, recently spoke with Healthcare innovation about trends in this space and some of the ways health plans are using this data to improve patient outcomes.
Sutten has served as chief financial officer and chief executive officer of a managed care organization and as chief financial officer of the North Carolina Department of Health and Human Services, which is responsible for the state’s Medicaid program. Siftwell raised $5.8 million in the company’s first round of venture capital funding.
Healthcare Innovation: Could you give us an example of how your company works with health plans?
Sutten: We work in four areas. The first is emerging risk, meaning emerging conditions or rising costs, worsening conditions, things like that.
Next are the quality measures. How do we help plans understand their members and drive things like medication adherence, wellness visits, and cancer screenings?
The third area is member retention. Who will leave your plan, why will they leave your plan, and what can you do about it?
The last area is risk adjustment: making sure plans understand the acuity of their members and are paid for it. The way the technology works, and it’s pretty consistent across all of those use cases, is that we get their data set, we combine it with our own, and we have thousands of data points on people across the country. We match it to our data sets, and then we use machine learning to make predictions, and then we interrogate those predictions for related explainable factors or related causal factors.
HCI: Can you give an example of how it works?
Sutten: We had a client who said, “We would really like to improve our members’ compliance with cancer screening measures.” So we ran the predictions and told the plan that there are approximately 12,000 women who are unlikely to be screened for breast cancer. Of those 12,000, it is divided into a group of different cohorts with similar characteristics that are driving that non-compliance: whether they are obstacles or blockers of some kind. Of the 12,000, let’s say, there are 65 cohorts. Here is a cohort with an 80% chance of default. The reasons they are unlikely to go for cancer screening is that they live more than 20 miles from a screening center. They don’t have transportation. They need child care when they leave. Additionally, they are of a socioeconomic status so it will be important for you to talk to them about the fact that this is a covered service.
Then the customer starts making calls and collects the information from those calls. We take the structured data and build psychographic models for everyone else who hasn’t been called yet. Maybe 8% will say they don’t want to go for religious reasons. Another 11% will say, “Well, I heard that this can be dangerous and could actually increase my chances of developing cancer, so I don’t want to go.” We take that information and say, ‘Okay, when you start reaching out to the remaining 12,000 people that you haven’t contacted yet, think about using religious organizations to raise awareness about this group. Consider radio ads in these areas and billboards in these areas to raise awareness of its benefits. For the rest, the standard calling campaign is the right way to go.’ Now they’re taking all that context that we originally gave them and they’re using different channels, different modalities, to better engage those members, to drive compliance.
HCI: Do you have any thoughts on collecting race, ethnicity, and sexual orientation data? New York State had just announced that it will propose requiring insurers to collect that data. First, what big push is there to collect and report that type of data, and what are some of the issues health plans face in dealing with that data?
Sutten: I think there are a lot of existing processes, a lot of data collection points that can be leveraged. I think from a plan perspective, it’s completely doable and I think it’s the most responsible thing to do over time. On the topic of data collection, whether on race, sexuality, or gender identification, we run into some complicated issues. Some people may not want to identify themselves if they have questions about sexual identity. When you address some of the questions about race, I think there could also be issues with self-identification. If you come from the black community, there have been well-documented cases where people have been betrayed by the system. Think of North Carolina as an example and the eugenics project.
HCI: So there is a lack of confidence in reporting this?
Sutten: Absolutely. So, as an individual, what would be my motivation to self-identify, rather than preferring not to say so? In terms of health equity on the plan’s part, as soon as it knows it has a problem, it should fix it. I think with that information comes the responsibility to do something about it.
HCI: Before collecting that information, are health plans a little blind in knowing how deep the disparities are between different groups?
Sutten: I think a lot of plans go blind to a lot of different data. I think this is just an example. I was talking to a plan recently that said they only have 60% completion factors on some of the information we’re talking about. I don’t know of any problem that you can solve if you don’t fully understand it, so I think it’s a really important thing to build on. But I don’t think this is some kind of golden architecture that’s going to solve some of these problems. I think we all need to come together and commit that this is an issue in our country, and we all need to do our various parts that we have control over to move the ball down the field and promote health equity.
HCI: Are there reasons why health plans might have a hard time, even once they have this data, deciding what to do with it internally rather than turning to a company like yours to help them?
Sutten: What we do relative to what people have internally is very different, and I speak from firsthand experience. The technologists we have on our team are simply not available, and particularly not available in the healthcare space. Technologists like my co-founder will work for Google, Microsoft and OpenAI. They are incredibly difficult to recruit. When you look at small regional plans, they are hard to find and hard to afford.
Part of Siftwell’s strategy is how to bring together the best technologists, combine them with the best minds in managed care, and bring that kind of cohesive set of skills and experiences to managed care plans. What plans generally do is correlation and retrospective analysis which is completely different than prospective machine learning and real artificial intelligence. Everyone is talking about AI right now, right? But there are those of us who really do it and those of us who talk about it. What I know from talking to my peers in the field is that there isn’t much real data science within managed care plans right now.
HCI: As Medicaid managed care plans begin to get paid in a different way for addressing whole-person care issues, will we see more plans placing greater emphasis on this, simply because they are paid that way?
Sutten: Yes, when you really see the market moving quickly and broadly, that’s when the right financial incentives are in place. I was on the board of directors of the Association of Community Affiliated Plans. These are non-profit organizations. There is a difference between the motivations for your ACAP or nonprofit plans and your commercial plans. But in all cases, even when it’s a nonprofit plan, “no money, no mission,” so a lot of these things come in the form of unfunded mandates. In this case, I think the regulators are showing how serious they are by weighing certain measures and including funding for some of them as well. And we’re seeing that in California, certainly in New York, and there’s a big emphasis on North Carolina as well.