By Anita Cattrell and Katie Bobroske
Evolent is combining clinical data, social determinants of health and the latest evidence about COVID-19 to identify and outreach individuals at the greatest risk of severe outcomes from the disease.
In COVID-19 prevention triage, anyone can be a first responder—once they recognize the need. Neighbors are making masks for the nurse next door. Families are grocery shopping for Grandma. But if a population isn't as communal as a street of tight-knit neighbors or as connected as a family, can we still find and support those most vulnerable to the new coronavirus?
As individuals everywhere try to figure out how they can help in the pandemic, health plans and health care providers have tried to answer the same question on a larger scale: How do we identify high-risk individuals who might fall through this makeshift safety net? How do we marshal the resources to help them?
At Evolent, we've been helping our partners respond to the challenge, quickly adapting our predictive analytics to prevent the riskiest exposures by easing one hand off the door handle at a time.
Building a COVID-19 Risk Score
Every one of us reads in the news that COVID-19 is affecting men more than women, poorer areas more than affluent ones, and that smokers, pregnant women, and people with compromised lung function are at particular risk.
The list goes on and on, to the tune of 3,000 clinical and academic papers published on the virus already.
But how can health plans use this information to reliably locate those individuals at highest risk? For complex populations such as Medicaid recipients, simply flagging individuals based on clinical criteria—a diagnosis of a respiratory condition, for example—would result in call lists that overwhelm outreach operations and drain valuable resources. A member with asthma who has their symptoms under control, a well-stocked pantry and a month's supply of medications will likely be of lower risk for poor outcomes than someone who was recently hospitalized with chronic obstructive pulmonary disease (COPD) and doesn't know how they will afford next week's groceries or prescriptions.
Our multidimensional approach to this challenge allows us to prioritize members based on a comprehensive understanding of their health. We combine a rich set of clinical and social determinants data with the latest research on the virus to assign a COVID-19 Risk Score for each member.
We started by combing through the most recent research on COVID-19 from leading medical journals and institutions. This points us toward individual characteristics that are associated with an increased risk of contracting COVID-19 and, for those patients who develop the virus, associated with severe complications. Based on this evidence, we estimate relative weights for each patient characteristic. These weights are applied to each member's set of conditions and severity level, to calculate the clinical dimension of the overall COVID-19 risk score.
Another dimension of the score draws on prior Evolent research that went into the Social Needs Index, measuring the extent to which social and economic hurdles may keep a member from staying healthy. The index is computed using data on the member's income level, education level, neighborhood type and the local economic situation—all components which may indicate the level of preparedness for the virus. With this data, we identified members with housing insecurity or homelessness and those living in crowded spaces, as they may be more exposed to virus carriers. Likewise, members struggling with food scarcity may be less able to organize food delivery or self-isolate when needed.
Finally, we incorporated three additional measures for each member: risk of future hospitalization, risk of mortality and health care utilization score. These scores are drawn from the risk-stratification models, which Evolent has used and refined for years in our clinical programs. While not specific to the virus, leading research consistently suggests that people with multiple clinical conditions or at higher risk of death due to other factors are also at heightened risk for severe effects from COVID-19. We wanted to ensure that no potentially high-risk member goes unidentified as the medical community continues to uncover and understand the mechanisms of the virus.
The three components combine to create a risk score for each member, enabling us to make priority lists for outreach. And we've been helping health plans, accountable care organizations and physician groups call those individuals, one by one.
Scaling Up Proactive Prevention
Many payers with call centers in place already reach out to high-risk individuals for months-long care management programs. But an urgent community triage effort is helpful to reduce the spread of COVID-19.
These calls can differ from typical care management calls in that the goal is to prevent an exposure and help with resources. Callers can be enlisted from other areas of the organization and offered a simple script: Do you have what you need at home to get through the next 30 days? How many days of medication do you have left? Here are resources on the latest guidance on social distancing: Would anything prevent you from following them? How can we help you overcome those barriers?
One Midwest health plan that Evolent supports whittled a plan population of 300,000 members down to the 8,500 highest-risk and the next 20,000 moderate-risk individuals. Over the course of a week, a blitz of outreach by 85 team members pulled from across seven plan departments outreached to an average of 3,100 people each day to ask what they needed that would help them stay home and avoid exposure—and to help them get it, from prescription deliveries to social services to food resources.
Among many drafted to make those one-on-one calls was a population health manager. She reached a woman with lung cancer who needed food but was forbidden by her doctor to leave the home. With no internet to order food deliveries, no family nearby and few food options in her rural town of 4,000 people, she was stuck.
The population health manager reached out to several stores, finally connecting with the motivated owner of an independent grocery who was willing to be a first responder.
"He said that they do not offer delivery services, being a small, locally owned operation. However, due to the circumstances with COVID-19 and people not being able to leave their homes, he offered to personally deliver groceries to the member. I provided her his contact information, and she was very grateful. She stated she was going to reach out to place an order with him right away."
As new research emerges on the virus, we continue to update the COVID-19 risk scoring algorithm, refining its predictive powers. And every week, we reprioritize members to ensure our outreach efforts are targeting those most in need of support. It's a process any health plan can follow, and one we'll need to see implemented more broadly to minimize the human and financial costs of this and future pandemics.
Easing one hand off the door handle at a time. It's how we apply population health and data analytics to care for our members, supporting them in every stage of their health care journey.
For more information on our approach to risk stratification during the pandemic, see Playbook: Evolent's COVID-19 Risk Stratification Approach
About the Authors
As Evolent's Chief Innovation Officer, Anita is responsible for establishing and promoting a consistent approach to innovation that enables Evolent to focus on clear objectives, test for outcomes, assess the impact of rapid-cycle pilots and ultimately define how we scale proven innovations. In her previous role as Evolent Health's Senior Vice President of Research and Development, Anita led the teams responsible for advanced analytics, clinical program design, and evaluation of clinical initiatives.
A data scientist, Katie has collaborated with Evolent's clinical analytics and research and development teams since 2013. She is a PhD candidate in Operations & Technology Management at the University of Cambridge, where she develops data mining techniques to analyze medical and pharmaceutical claims data.