How Blending AI and Behavioral Science Can Advance Health Equity

By Marten den Haring, Lirio CEO

A number of frameworks, technologies, datasets and initiatives from the U.S. Centers for Medicare and Medicaid Services (CMS) are converging to empower health plans and health systems to further increase their efforts to advance health equity — and the organizations are under increasing pressure to take action now.  

Medicare Advantage (MA) Star Ratings declined again in 2024 such that only 31 of the 545 rated plans earned 5 stars, 75 earned 4.5 stars,123 received 4 stars, and the remaining 316 came in at 3.5 stars or worse. There’s more projected turbulence on the Star Ratings horizon, too. In late November, for example, UnitedHealthcare won its lawsuit against CMS in a court ruling that requires CMS to recalculate UnitedHealthcare’s Star Ratings for the 2025 plan year. Additionally, among the significant changes to Star Ratings, CMS has modified its program with the Health Equity Index (HEI) to reward plans for improving care among member populations with social risk factors (SRFs). The explicit health equity focus likely requires many plans to adjust their strategy to better support members with SRFs.

CMS has also set the strategic goal of having all Medicare beneficiaries in value-based care (VBC) arrangements by 2030 with the stated vision of “achieving equitable outcomes through high quality, affordable, person-centered care.” The agency also introduced changes to the ACO Realizing Equity, Access, and Community Health (REACH) model to “advance health equity.”

Taken together, this alphabet soup of initiatives introduces new quandaries for health plans and healthcare providers alike. For health plans, the challenges stem from members with social risk factors tending to underperform in key quality measures as part of Star Ratings — so health plans will have to adjust accordingly, if only to maintain their current rating. Niakan, Russo, and Robb explain that health plans need to consider how to “enroll and better serve beneficiaries with SRFs starting immediately, as current performance will impact Star Ratings measured when the initiative takes effect.”

On the health system and hospital front, CMS is trying to reduce its expenses and that can be at odds with making the ACO REACH model profitable for providers. At the same time, value-based care models in both the public and private sector could benefit from better compensating the providers focused on serving disadvantaged groups. 

AI, Data, Design, and Frameworks for Health Equity 

Improving access to care and participating in VBC arrangements are among health care C-suites’ top priorities, according to research conducted by Sage Growth Partners. Nearly two-thirds (62%) of healthcare C-suites rank improving access to care among their organization’s most important challenges, 81% are continuing with current VBC levels or aggressively adding new value-based contracts, and 39% say the industry is making progress reducing health inequities, an increase from 28% last year.

CMS has clearly stated that AI will play an important role in leveraging data to advance health equity, expand coverage, and improve health outcomes.  

Cruikshank, Wade, and Bajwa note in Harvard Business Review that AI’s power leveraging multiple data sets to address the drivers of health inequities make a powerful case for using it to close gaps in care across the patient’s journey. 

Indeed, health plan leaders are increasingly recognizing the opportunities to leverage technology to advance health equity, and improve the member experience. Likewise, hospitals and health systems have similar opportunities to use AI and data to identify ways to advance health equity with the necessary oversight to ensure the tools do not widen existing inequities or create new ones altogether. 

“Data can support identifying individual and community needs, developing interventions to address those, and monitoring progress on narrowing disparities,” Rawal, Seyoum, and Fowler write in Health Affairs. 

For those initiatives to work well, health plans and health systems need to ensure data is high-quality so that AI and machine learning models do not widen disparities by operating with data sets that favor certain populations over others. In one example, a study uncovered evidence of racial bias in an algorithm that both predicts health risks and influences treatment decisions — with potentially deadly results. Obermeyer, Powers, Vogeli, and Mullainathan write in the journal Science that Black patients received less care than similarly ill white patients due to racial bias in the training data set, which reflected historical withholding of care from Black patients. Without understanding the origins of the training data, it was possible to falsely conclude that the white patients were healthier than their Black counterparts. The researchers explained that addressing this particular disparity would change the percentage of Black patients receiving additional help from 17.7% to 46.5%. 

Avoiding such miscalculations will require payers and health systems to design equity strategies well from the outset by focusing on the changes that can be most impactful. Healthcare organizations have options when designing equity programs, including the Framework for Digital Health Equity, PROGRESS-Plus, the Health Implementation Framework, and the e-Health Literacy Framework. The frameworks were purpose-built to help identify equity categories to include in the design. Additionally, CMS developed its Framework for Health Equity “to advance health equity among members of communities, providers, plans, and other organizations serving underserved or disadvantaged” populations. 

The Role of a Large Behavior Model in Advancing Equity: Digital Behavior Change Interventions  

For both health plans and health systems to strategically advance health equity, a new approach is needed. That new approach involves taking advantage of increasing volumes of data about medical and non-medical drivers of health and applying a new, novel Large Behavior Model (LBM). 

Lirio defines health equity as “providing individuals or groups the correct support so that they can have a comparable experience and outcomes to everyone else. Importantly, equity is not giving everyone the same opportunity, but rather tailoring support so that each person gets what they need.” To achieve that equity, health plans and providers need to drive health behavior change in underserved communities that have traditionally faced a range of social, economic, and environmental non-medical drivers of health that can negatively impact outcomes. 

That will require data and technologies for engaging members and patients with dynamic, personalized communications that enable individualized healthcare experiences. Behavioral science is critical to delivering personalization that can positively impact how humans living in underserved populations and communities can engage with the healthcare system and their own health. That is where an LBM designed specifically for healthcare can play a critical role in advancing health equity. 

An LBM is akin to a large language model (LLM). But while LLMs encompass language, LBMs unite AI and behavior science. Lirio’s LBM is trained on vast datasets of human interactions so it can learn patterns in human decision-making and communication. With the data and learnings, the model generates insights that inform digital behavior change interventions to nudge patients/members to adhere to medications, comply with preventive screening recommendations, take actions to manage chronic conditions or diseases, and more. 

The Power of Personalization in Improving Care and Experience, Reducing Costs

Health plans, hospitals, and health systems are already amassing larger volumes of data than any other industry, and that information can be used to dramatically improve healthcare. AI, behavioral science, and an LBM can deliver personalization to engage more members and patients. Adigozel and Wilson share research that shows how personalizing healthcare improves customer experience by 10%, increases care quality by up to 25%, and reduces costs by as much as 10%.  

Combining AI and behavioral science can ultimately pay off in more engaged and satisfied members and patients that translate to higher Star Ratings for health plans and greater success for health systems participating in VBC arrangements or the ACO Reach model. The effects are far-reaching and extend to the broader population as well as underserved individuals and communities. 

Lirio’s LBM and ability to drive personalization can effectively advance health equity by incorporating rich data that healthcare organizations can leverage to identify existing disparities and apply behavioral science techniques to address those inequities.