Move over risk scores, what healthcare needs now is risk science

Risk science is an emerging requirement for value-based care and population health programs.
By John Supra and Shrujan Amin
10:53 AM
doctors discussing risk science at computer screen

“What’s their risk score?” That’s a natural question to ask when a discussion arises concerning patient risk. Unfortunately, it’s also how most conversations both start and finish.

The idea is that a single score sorted from highest to lowest will tell you which patients need your attention, regardless of the person’s specific clinical or social needs or the care programs that you may be able to offer them.

Indeed, historical models based primarily on claims data aim to capture a general health status of a patient and were designed around actuarial prediction of future costs.

But as we drive toward value-based programs where providers share in significant risk this approach is no longer adequate.

With so many provider organizations engaging in population health, we believe that they need to innovate and mature the level of sophistication and precision beyond that offered by the claims-based risk scores.

More specifically, providers often enter information into an EMR that doesn’t show up on a claim and as a result that information doesn’t make its way to the risk models or even quality for meeting clinical quality goals.

And what about the patient’s general lifestyle? Through population health we have learned that non-clinical care is also critical to a patient’s overall health and well-being including the potential success of a particular program for an individual patient. Bringing together previously disparate data sources, leveraging machine learning (ML) to create smarter and flexible models, and working with subject matter experts to understand programs is what we refer to as Risk Science.

A brief history of risk scoring

Let’s take a step back and consider the genesis of risk scores. Many of the historical models were developed by insurance companies to predict the expected healthcare cost of a patient. The primary source for this analysis is claims data because that was the information insurance companies had access to and it was digitized, and thus available for robust analysis, before most other sources. To quickly assess their population, they developed methods to quantify each member’s health relative to the rest of the population and determine expected variations from the population as well as what resources (costs and services or programs) need to be allocated for these patients.

That said, risk has come to mean the need to stratify and prioritize patients’ clinical needs for eligibility and enrollment into a clinical care program. For example, a hospital discharge nurse might focus on length of stay or readmission risk. In contrast, a member of a quality team that is responsible for clinical quality metric performance, may view risk as gaps in care. Making sure when a member presents for a visit with their primary care physician, they receive the proper screenings based on eligibility. Both programs are equally important but requiring different models to insure patients are correctly identified for each. In addition, providers and health systems are also taking on financial risk, like insurance companies, in terms of pay-for-performance and/or shared savings agreements and thus are interested in potential future healthcare expenses.

Indeed, in the early 2000s claims data was the primary option for analysis, as it was standardized and digitized before most other data sources. However, as EMR adoption increased dramatically, especially for office-based physicians, 87 percent in 2015 as compared to 42 percent in 2008 and providers began to participate in value-based programs.

Time to advance beyond risk scores

Our view of risk and risk scores needed to evolve and become both more specific and broader in order to align to various care programs while also inviting the inclusion of other data sources. For example, clinical data from EMRs provides additional information around the patient’s condition and severity that doesn’t always show up on a claim. Although correctly coded claims may determine which patients are diabetic, additional information that can be found in health records, such as HbA1c values, provide additional details to assist in discern severity. Along with clinical information, other data sources such as, social determinants of health, help fill in the gaps of a patient’s life that generally do not show up in claims or even clinical documentation. We envision a world where claims and clinical data gathered in the care process is triangulated with socioeconomic data that is both collected from patients and their families as well as evaluated based on where patients live and work.

Much of the social determinant work has explored the impact of environment at the ZIP code level as well as analysis of healthcare service and provider access and adequacy. We have found, however, that analysis at the Census block group level is required to achieve the granularity necessary to understand the impacts on health and well-being as well as assess issues related to transportation, education, income, healthy food options, access to exercise (including sidewalks and safety), or other environmental factors that may contribute to health.

With ongoing research on social determinants, newer elements continue to be identified, such as the positive effects of green space in easing depression. This approach creates a more precise understanding of the patient’s overall experiences and their impact on their health and well-being.

Another evolving source of additional data can be found in information that is collected at home by patients. This might include specific condition data such as home diabetes blood monitoring, scales (weight), home blood pressures cuffs, and fitness/exercise trackers. By using a broader set of data source, we are able to complement traditional clinical information to assist in both our risk work as well as provide insight into patient and family engagement. Data that connects physical activities and health outcomes can provide a deeper understanding of patient engagement, motivations, barriers, and connect them with the benefits and needs of the communities we serve. Identification of trends based on geographic locations such as people that live in closer proximity to community walking trails or parks accumulating more steps or utilize of public transportation, or community programs could help shape future policy decisions – an important data-driven approach to community planning that would be focused on the health and wellness of their citizens.

With the drive toward value-based population health there is an evolving need for innovative specialized data science work. One that leverages the breadth of data sources available, well beyond claims for health care services, that focuses on assisting providers, delivery systems, and the clinical networks taking on significant financial risk to better understand and model the risk they are taking while also providing insight into the development and performance of care programs, especially non-traditional care outside the delivery system walls.

We believe that these efforts make up an emerging requirement for high-performing health systems: The ability to excel in value-based programs demand expertise in Risk Science.  

John Supra is vice president of solutions and services at the Care Coordination Institute where his colleague and co-author Shrujan Amin is a data scientist.