Turning Data into Equity: How Advanced Analytics and SDOH Integration Are Redefining Value-Based Care
Key Takeaways:
- Integration of social determinants of health with advanced analytics enables more precise risk stratification, expanding care management capacity and supporting value-based care models.
- A Midwest health system achieved a 6-fold increase in care management capacity, multimillion-dollar savings, and a ~40% reduction in avoidable hospitalizations using fine-tuned risk models and targeted interventions.
- Emerging AI tools—including ambient listening, automated outreach, and agentic AI—are improving data capture, reducing administrative burden, and enhancing population health management efficiency.
Michael Hoxter: My name is Mike Hoxter. I'm the chief technology officer of Lightbeam Health Solutions.
I've kind of always been an IT guy and nerd since I was getting my first job at like 16. I studied industrial engineering, which is not software, but it's a little bit more about how do you make processes better, and then got into health care from there, where I realized—as far as a background in trying to analyze processes and make them better—there is an enormous amount of low-hanging fruit across the health care system.
It seems like this is probably the sector I’ll be in for my whole career because there’s a virtually unlimited amount of stuff to do.
How are organizations successfully integrating population health risk stratification and SDOH data sources to better identify and support these at-risk populations?
Hoxter: The more data you can get on your patients, the better. I think the first thing that it shows when you have organizations starting to take advantage of SDOH data specifically—and really the best thing that that shows—is a cultural willingness to start using these sorts of technologies and things that are a little bit outside of the box of where the industry has been for a long time.
We're still a very heavily fee-for-service focused industry as far as how providers and health systems end up getting paid. You’ve got to keep the lights on, that has to be a very important part of what our health care providers do.
SDOH don’t necessarily help you with a billing event—with a wellness visit or when you’re seeing the patient—but it’s really good movement that these sorts of things are starting to be used. Because of the prevalence of value-based care plans and the fact that you're financially incentivizing systems to take better care of the patient when they're outside of the office, when they're not there, when there is no billing code that’s going to happen from that activity—there’s still value being driven.
We’ve seen time and time again, people tend to do what they're incentivized to do. The fact that there are incentives on the table to take care of those patients, to use these data, to do a better job—it’s just great movement in the industry as a whole.
Can you share any specific examples or case studies that illustrate how advanced analytics have directly led to measurable outcomes like reduced admissions or significant cost savings?
Hoxter: One of the studies that we've done somewhat recently was from one of our midwest health systems, utilizing a few different technologies. The stats from the study are effectively that we were able to about 6x the care management capacity through a few different tools—but risk stratification is a huge one because you just can’t expect clinicians to look through the charts of that many patients in a day. There are only so many hours in a day.
When you have a risk score, like a hepatocellular carcinoma (HCC) risk score based on claims—say it’s based on diseases and the patients are 3 times the average—that’s good information to have. But it’s pretty broad, right? It’s like, “the patient’s going to cost a lot.” That’s not really actionable information. [It’s] something you should do, but you don’t know what to do based on that.
That’s where, when you start to have much more fine-tuned types of risk stratification, it moves beyond “at a high level, the patient’s going to cost a lot” to “the patient has this specific risk for this specific reason—and here’s what you need to do about it.”
That’s the output of the models. The cognitive load of the clinician—taking it from “I need to do something for this patient” to “here’s what I do for this patient”—drastically decreases. You provide so much more very specific information about the patient when you go past just the International Classification of Diseases (ICD) codes.
It not only makes it easier for the care manager to take care of a patient, but it makes it much easier—in this particular case, about 6 times easier—for a care management team to take care of a population of patients. You can take care of many more because each individually takes up a lot less of your time.
On that one, we saved a couple million dollars and had a pretty substantial reduction in inpatient and emergency room (ER) visit admissions. It was about a 40% reduction in avoidable hospitalizations for the population that we were managing.
If you’re checking up on your patients and making sure that they have everything they need—historically, the only way you would do this is assigning a person to be checking up on them regularly. That can have a very sizable effect on a patient population. But when health systems try to do this, we’re limited by the number of people we can put on the problem.
You have a small number of nurse care managers assigned to care management. They’ll do a good job with that small high-risk population, but there’s only so much you can do. Improving the technology improves the throughput of that team of care managers and increases the scope of what we can cover—and it does so in a very efficient manner.
Since value-based models really emphasize proactive, equitable care, how can organizations use technology and analytics to better help hospitals, accountable care organizations (ACOs), and health plans close these care gaps and address health disparities?
Hoxter: Getting a hold of the data and making use of it—doing anything is far more valuable than doing nothing, which is sadly where a lot of groups are. They think it’s a good idea, but they’re busy.
If you look at an example of a diabetic patient, a lot of times that’s the scope of information you get: “this patient is diabetic.” That’s not worthless as a data point, but a diabetic patient who lives in a food desert in a neighborhood with bad walkability is very different from a diabetic patient who lives a couple of blocks from a Whole Foods and can easily get nutritious food.
These are very different problems to solve, even though on the surface, as a care manager, what you may be given in many health systems today is just “patient is diabetic.”
It takes a lot to dig into that—to really look beyond diabetes as an ICD code and ask: what does this patient really need? Have they filled their meds? Do they need to get out on a walk? Can they not access nutritious food?
The actual root causes for many patients’ conditions aren’t in the electronic medical records (EMR). They’re not in a diagnosis code. There have been a ton of studies showing that one-half or more of the overall health care experience is related to social determinants of health, not ICD codes.
It’s these factors—where you live, your lifestyle, your home life. We built algorithms quite a while ago and found that, especially within a Medicare population, even marital status matters. Those who are widowed have drastically higher health care variability. There’s no ICD code for that—you won’t get that from billing records—but it’s a key data element that helps you understand who the patient is as a person. That really helps clinicians treat the patient more efficiently and effectively.
What innovations or trends do you see shaping the next phase of population health management and risk stratification?
Hoxter: Well, I think we’re getting a lot better tools. I use AI every day. The rate of improvement of these base models and the technology being built around them is moving at a rate I’ve never seen in my career, I’m really hopeful for what’s coming.
Some of the technologies that I think will be really impactful—there’s quite a lot happening in the ambient listening space. First and foremost, doctors are the focus, but I see it expanding to nurses doing less intensive care visits and care management conversations.
The conversation you have with a person contains a lot of information that may or may not actually be captured and used. AI is very good at taking those conversations and turning them into discrete data elements that can then be used for stratification. I imagine we’ll see improvement in the quality of data collection, which provides better source data for risk stratification.
Once you have good data, you can make good patient lists. But we can give you far more to do than you can handle. We deal with this all the time—there are 10 times as many patients where, if you could, you’d have someone talking to them and reaching out, but we’re all being asked to do more with less.
How do we expand reach without linearly scaling our nurse workforce? That’s not in the budget for most health care systems. What we see is more adoption of automated outreach and communication. I’ve gotten AI voice agents calling me for follow-ups on doctor visits. A year ago, they were really bad—patients didn’t like them. I’m pretty tech-forward, but most seniors would reject them outright.
More recently, though, with how fast these are improving, I’ve gotten a couple and thought, “That was actually pretty good.” What we’re seeing in our stats as we roll out these programs is that post-conversation satisfaction surveys are reaching parity with humans.
We’re getting to where a voice agent can have a conversation with a person, and patient satisfaction is nearly the same as with a human. Some days it’s even higher. That’s another area where we’re hitting a bottleneck—when you have good data, you hit a limit on how many humans can do the work. That’s where AI fills in gaps.
Agentic AI can perform many of the tasks care managers do. The “cold call” or “warm call” to check on a patient, collect a data element—you spend 15 minutes getting them on the phone, you get your data point, everything’s fine, and you move on. It’s necessary but inefficient. We can have an agent do that.
Then your care manager staff can focus on the patients who actually need them—spend time with patients who need you, rather than collecting data. The more AI and automation fill those gaps, the less time clinicians spend on administrative work and the more patients they can manage.
At the end of the day, that’s what it’s all about. One of our big pushes is that we want the patient, when they’re not in the office, to be treated as just as valuable as when they are. That’s where the tools are going.
Culture eats strategy for breakfast. The cultural change is such an important part of this technology. We’re seeing more health care organizations not just adopt a tool, but embrace the concept: we need to use these tools to better care for our population as a whole.
It’s taken a while—we’ve been doing value-based care management for over a decade—and it’s now part of the DNA and culture of these organizations. That cultural adoption—being curious, wanting to see what’s next with this technology—takes you far, because there are so many great tools out there to improve population health.
Be curious. Check them out. The rate of change is crazy. If you saw something 6 months ago that you didn’t like, it may be really good now. If you had a bad experience with AI, check again—it’s improving at a crazy rate. I’m excited to see where things go.
