Dr. John Frownfelter
If we’re not careful, value-based care could worsen health disparities
September 07, 2021
By Dr. John Frownfelter
Decades in the making, the transition to value-based care may soon be complete. In June, Liz Fowler, the new deputy administrator of the Centers for Medicare & Medicaid Services (CMS) and director of its Center for Medicare & Medicaid Innovation (CMMI), suggested that CMS may soon require providers to be reimbursed based on patient outcomes. This stands in contrast to the status quo of paying providers directly for their services.
Whether this change comes to pass remains to be seen. But if it does, it will radically reshape healthcare delivery in the United States. Value-based care has tremendous potential to improve patient outcomes and lower costs. But if we’re not careful, it could have the unintended side effect of worsening existing inequities in health outcomes.
The risk of value-based care
Value-based care has a noble goal of incentivizing providers to improve both patient outcomes and the quality of their care. Depending on the model, reimbursements may be tied to metrics including readmission rates, avoidable ED visits, or rates of hospital-acquired conditions.
But for any adverse outcome that value-based providers are trying to prevent, there are underlying, root causes that are often invisible to the care team. Up to 80% of health outcomes are determined by external, non-medical factors — none of which are easily captured in the patient’s chart. These factors can be broadly categorized as social determinants of health (SDOH): the conditions in which people live, work, and play.
Social determinants of health include factors such as whether patients live in a food desert, whether they have access to public transportation or pharmacies, or whether they can afford their medication or other health services. They can also include environmental factors such as pollution in the air or water.
By and large, however, the most detrimental social determinants are often closely linked to poverty. The result is the wide disparities in health outcomes that we see among Black, Hispanic, rural, migrant or otherwise underserved communities.
If providers don’t have the visibility to address these underlying health risks, they could end up inadvertently focusing their limited time and energy on the patients least affected by SDOH. Worse, they could end up actively perpetuating healthcare inequalities.
Case in point, a 2019 study in Science found that a predictive analytics algorithm used to help manage care for about 200 million Americans was assigning lower risk scores to Black patients relative to White patients with the same level of illness.
The reason? The algorithm was trained on insurance claims data, and thus predicted which patients were likely to have higher care costs. Because systemic barriers to care — rooted largely in SDOH — have historically resulted in lower healthcare spending for Black patients, the algorithm ended up deprioritizing Black patients, perpetuating the existing racial bias in healthcare.
Risk scores like the ones in the study are pervasive in healthcare, and are commonly used by value-based care providers to determine which patients should be prioritized to prevent adverse outcomes. But as the study shows, failing to account for SDOH can leave these algorithms prone to bias against the most vulnerable and marginalized patients.
Bringing SDOH awareness to value-based care
Value-based care providers shouldn’t have to choose between improving their overall patient outcomes and taking action to reduce health disparities. They can do both. But it will take more than just traditional predictive analytics.
In 2020, I co-authored a study in the American Journal of Managed Care that showed that machine learning, trained only on SDOH data, could predict patients’ risk for healthcare utilization to a high degree of accuracy. This is the way forward for value-based care.
The difference between the machine learning approach and the predictive analytics approach is that it’s focused on the root causes of poor health outcomes (SDOH), rather than their end result (higher care costs, as reflected by the insurance claims data). This enables healthcare providers to proactively address the SDOH factors driving patients’ risk, and by extension, the disparities we see in outcomes among socioeconomically disadvantaged groups.
Using public data from the Census and other government agencies, machine learning can predict which communities are most vulnerable to poor health outcomes, as well as the SDOH risk factors driving their vulnerability. This enables value-based health systems to strategically invest in community benefit programs, such as food delivery services or ridesharing services, that address the barriers to health in a given community.
Machine learning can also identify the SDOH risk faced by individuals. Through a process known as similarity clustering, machine learning matches patients with thousands of similar patients whose outcomes are already known, which can predict their risk and identify what their leading SDOH risk factors are. When you add in clinical data from the provider’s EHR, machine learning can provide a holistic view of patient risk that goes far beyond what’s possible with predictive analytics.
To be fair, some predictive analytics tools have attempted to account for SDOH factors. But these attempts often fail to provide value for clinicians because they stop at providing a risk score — without any context of why the patient is at risk or what can be done to reduce their risk. To be useful, we need more than predictive analytics. We need prescriptive analytics.
Prescriptive analytics differs from predictive analytics in that it provides actionable recommendations for how care teams can intervene to address patients’ risk factors, whether they be clinical or SDOH-related. To be clear, these recommendations are not meant to make decisions for clinicians, but they can ease their cognitive burden and give them direction on how to approach a patient’s care differently.
Regardless of whether value-based care becomes the law of the land, the transition to performance-based payment models will no doubt continue — and for good reason. As a nation we spend more on healthcare than any other country while achieving worse outcomes. It’s about time we incentivize both reducing costs and improving outcomes.
But we must also ensure that outcomes are improved equitably. Prescriptive intelligence, combining machine learning and SDOH data, can empower care teams to address the root causes of health inequities while still achieving the goals of value-based care.
About the author: Dr. John Frownfelter is the chief medical officer at Jvion.