There is a very significant opportunity to leverage sociodemographic and socioeconomic data in new ways in order to more effectively care for patients in the context of accountable care organization (ACO) and population health work; and much of that opportunity has yet to be fully plumbed. That was the core message of the presentation “Effectively Using Sociodemographic Data in Healthcare Analytics,” presented by Justin Pestrue, administrative director of quality analytics at Michigan Medicine (the new name for the University of Michigan Health System), Ann Arbor, Mich., on Friday, March 24, during day two of the Health IT Summit in Cleveland, sponsored by Healthcare Informatics, and held at the Hilton Cleveland Downtown, in Cleveland, Ohio.
Pestrue began by noting something that numerous industry leaders have pointed out and are aware of: only a small proportion of patients’ health status can directly be impacted by care delivery in patient care organizations (some say that proportion accounts for perhaps 20 percent of the overall impacts on the health of individuals, with personal lifestyle and behavioral choices, personal environmental influences, and other influences being far stronger overall).
“One of the biggest opportunities we have in healthcare analytics,” Pestrue told his audience, “is how we use sociodemographic and socioeconomic data to deliver great care for patients. This is important data to consider.” One obstacle? The fact that it remains challenging to gather, analyze, and use sociodemographic and socioeconomic data. Pestrue showed his audience a slide of a photo showing a man bending down searching for something in the grass under a streetlamp. The parable he told was this one: “So there was this man who was walking down the street one night, and he saw a man searching for something in the grass under a street lamp. The first man said, “Are you looking for something?” The second man said, “Yes, my car keys.” The first man said, “Do you think they’re very nearby here?” The second man said, “No, not at all, I lost them far from here.” “Well, why are you looking for them here?” “Well, because this is where the light is.” The point of the parable, Pestrue told his audience, is that so often, leaders of patient care organizations work with data that is relatively easy to access and work with, rather than data that might be very important but is harder to access and work with.
“So we have a real tendency to look at the data that’s easier to get—and that’s often the elements that are in our EHRs [electronic health records] and in our billing systems. But sociodemographic and socioeconomic data can do a really good job of helping us to better understand our patients,” he noted, with sociodemographic data include gender, race, ethnicity, age, and place of residence, and socioeconomic data including education level, income, etc., and both types of data being corralled together under the rubric “social determinants of health.”
“Among the conversational themes around these elements,” Pestrue told his audience, “are equity and justice; issues around policies and payments; and the elements around quality of care and population health management.” Speaking of the federal Centers for Medicare and Medicaid Services, he noted that “CMS is trying to identify how we appropriately risk-stratify populations. There’s a lot of work going into how demographics play into that,” he noted, adding that, “When I worked for a faith-based healthcare organization, they made it a part of their mission to ask, are we serving our population as equitably as possible, and fulfilling our mission?”
Meanwhile, at Michigan Medicine, Pestrue told his audience, “We’re looking much more closely at our quality outcomes and population health initiatives.” And, in that context, he noted, “The National Quality Forum has an ongoing Disparities Project. And one of their great recommendations is that while you want to account for risk, there are some challenges in adjusting for demographics. In addition, a lot of APMs [alternative payment models] have specific written requirements about using socioeconomic data as well.” Importantly, he said, “We were all introduced to these ideas when a lot of the public health data came out suggesting that very little of the longevity of people is determined by their engagement in the healthcare system. Healthcare plays about a 10-to-20-percent influence—the remaining factors are behaviors, social impacts, and genetics. So, if the healthcare system is only accounting for a small proportion, how do we account for all the other elements? That is one of the biggest opportunities we have—not only impacting our population health and public health challenges, but how are these factors being thought of in an ED visit or when a patient is on a chronic disease registry?”
At Michigan Medicine, Pestrue said, he and his colleagues have been drilling down in particular on looking at the underlying conditions that may help to explain variations in patient utilization. And, he said, drilling down into their data, he and his colleagues found that “There are racial groups using emergency services at more than twice the rate of other groups. So it’s clear that sociodemographics has an impact on how we use healthcare.” In that context, he and his colleagues have gone to work and drilled down into the data to try to uncover patterns. They took their covered-life population of 200,000, and divided that population into groups, depending on how many chronic disease registries those patients are on—zero to five. Then, they layered on top of that categorization work further segmentation using the LACE tool—an acronym that stands for “length of stay, acuity, comorbidity, and emergency department visits”—and looked at LACE-related data from the previous five years. They then risk-stratified which patients might be readmitted.
“Our team did a little study looking at how many readmissions the LACE score predicts or explains,” Pestrue said. “And we ran a model and compared it to strictly demographic experiences of patients.” And, pointing to a slide with data on it, he told his audience, “You can see that the R squared is 0.057; but looking at a model that strictly involves sociodemographic factors, it’s 0.047. What that means is that 80 percent of what we get from LACE, we can get from a pure sociodemographic model. So that points to how important these factors are in the outcomes of our patients.”
In any case, Pestrue told his audience, “We created a risk-adjusted readmission rate adjusted by the LACE score itself. And we factored in the clinical conditions and then broke out the rates of readmission. Not surprisingly, males are readmitted at higher rates. In the LACE-Plus model, if you’re a male, you get more points, they recognize that that risk is higher. Different payer groups account for differences, and also gender. Being male. If you’re a single male who’s on Medicare, your likelihood is going to be tremendously higher.”
Drilling down further into the data, Pestrue told his audience that he and his colleagues went deep on location-related demographic data, scanning the service area around Ann Arbor by zip code. What they discovered was interesting, because zip code alone turned out to be a rather crude indicator of health status. Instead, drilling down to the sub-zip code level, they found a variety of “hot spots” of neighborhoods of people with challenging sociodemographic and socioeconomic characteristics. Zeroing in on a map showing different sizes and colors of circles and explaining what the colors and sizes meant to the audience, Pestrue explained the map as a map of readmissions in their service area by zip code, with circles showing areas within zip codes of notably higher or lower rates of readmissions.
“One of the things I wanted to illustrate is that, often when we look at zip codes, we’re missing the market,” Pestrue said. “The biggest white circle in the middle of the map—a zip code on the western edge of Ann Arbor—is in the zip code I live in.” As it turned out, the “most insightful circle,” was a mobile home park on the edge of a mostly-affluent area. What does that circle mean? It means that the residents of that trailer park are living with less-favorable sociodemographic and socioeconomic characteristics than are their neighbors in that zip code overall. “We have a tremendous amount of readmissions coming from that one mobile home park. We haven’t done anything about that yet, but we’ll be taking a closer look at what’s happening in that area. And I didn’t pick that zip code to analyze for that reason; it just happened. But I think if we do this across most of our collective catchment areas, we’ll see more dots like this. In the public housing sphere, they’re doing a lot of research into depreciation areas and deprivation indexes. They’re taking census data and figuring out where we’re under-resourced, and where residents have a lot of challenges in paying for transportation or medications, and this is a project in Utah where they’re studying that. A lot of studies are finding that the tighter the neighborhood studied, the sharper the data will be. So we’re starting to look for deprivation indexes in our area as well.”
Pestrue told his audience that much work remains to be done in all these areas, but that some of the initial work that he and his colleagues have done at Michigan Medicine points to the tremendous potential inherent in drilling down into sociodemographic and socioeconomic data, and cross-hatching such social determinants of health-related data with clinical and claims data available to the leaders of patient care organizations. And, in that context, Pestrue cautioned that “There’s a risk-adjustment trap. When we as an organization or industry risk-adjust out for these demographics, we have to be very careful to not then do analysis on those risk-adjusted models around the factors put into the models in the first place. Let’s say there’s been a historic disparity around a particular racial group,” he said. “If we say, that group has historically had negative outcomes, and we take a risk-adjusted model and stratify it by race, it washes out the actual factors involved. I’ve seen that before.”
So, what is possible around demographic data? Pestrue recommended to his audience that they “understand the importance of SDS and SES on your organization’s results.” In that, he said, “Healthcare analytics teams should have core competencies in SDS and SES factors and geo-analytics. There are tremendous insights we can gain from all of that data.”
To do so effectively, he said, “We need to actively work to capture accurate actors and supporting data. We need to be able to geocode patient addresses, and capture from multiple channels. And how much are we willing and able to ask the tough questions?” he asked. It’s important, he said, to engage models that utilize this information, while understanding the risks. And he said, while many remain concerned about the idea of purchasing data—after all, there are entities who purchase credit scores from credit agencies and use that data in ways that can harm consumers—in the context of trying to improve the management of services to patients and consumers at risk, it might well be worth purchasing some forms of data in order to enrich the ability to better organize the management of services to one’s community.
In the end, Pestrue said, there is a tremendous opportunity to leverage sociodemographic and socioeconomic data to enrich the data analytics work that can support robust population health management and care management initiatives in patient care organizations health system-wide. And, he said, we are at the beginning of a very long journey around learning how to optimally use that data to improve population health management and service optimization going forward.