Aim for Understanding: Identifying Members with Increased Risk for Severe COVID-19 Complications

How our core values drove us to build an actionable rules-based model for care teams

Alina Schnake-Mahl, ScD, MPH, Research Scientist — Data Science
Lon Binder, Chief Technology Officer
Marcy Carty, MD, MPH, Associate Chief Health Officer

Back in January, pre-COVID-19, we launched a new set of company values: Put Members First, Be All In, Bring Your Whole Self, Aim for Understanding, and Lean into Discomfort.

In a post from Iyah and Toyin on April 16th, we discussed what the COVID-19 crisis means for Cityblock and the work we’re doing, anchored within one of our core values: “Put Members First.”

Over the coming weeks, we’ll share more on our response to COVID-19 and how our values are driving our work to provide care for marginalized communities.

Today, we’ll explain how “Aim for Understanding” drove us to build an actionable rules-based model to help our care teams better understand COVID-19 risk within our member population.

For full access to the model, please visit New England Journal of Medicine Catalyst.

When COVID-19 began to hit New York City, we knew our member population was going to be disproportionately challenged.

Cityblock cares for our partners’ most medically complex members, in neighborhoods and communities typically left behind by health care innovation.

Chronic diseases, such as diabetes, hypertension, COPD, and coronary artery disease, are more prevalent in low-income and communities of color, in part because often those communities have experienced decades of disinvestment and limited economic opportunity, experience chronic stress due to racial discrimination, lack access to quality primary care and education, and have high rates of uninsurance and underinsurance. These same diagnoses seem to elevate the risk of hospitalization, intensive care, and death if a person contracts COVID-19.

We knew we needed to advance our capabilities to rapidly identify those at the highest risk of death and serious illness from COVID-19 so that our care teams could better escalate, track, and coordinate care across our member population.

We built an algorithm to help identify which of our members were at the greatest risk of serious illness and death from COVID-19.

Our work is detailed and the full open-source model is available in the New England Journal of Medicine Catalyst. We are grateful to the team at Catalyst for the opportunity to amplify this work.

Our model can help organizations proactively outreach to their highest risk members for COVID-19 screening and management, and equitably distribute care according to likely need.

Developing the algorithm took deep collaboration across multiple teams.

On Wednesday, March 11th, still early in our understanding of what COVID-19 would mean for our communities and our members, one of our founders, Iyah, convened a group of folks from our actuary, clinical, and data science teams to discuss what it would take to build an open-source risk model based on available factors and literature.

The team pulled together quickly. Immediately, on a train ride home from New York to Boston, our associate chief medical officer, Dr. Marcy Carty, began a literature review to begin identifying known risk factors. And by that Friday night, already working remotely because of the impending stay-at-home order in NYC, our three teams met on video to map out a plan to put a list of the highest-risk members in our clinical leaders’ hands by Monday morning.

Cityblock was well-positioned for this type of collaboration. Teams from the clinical, growth, and technology areas of the organization already had strong relationships because of collaborative work around metric development and analyses.

We understand each other’s skills, allowing us to efficiently divide up tasks, and trust that each team member would deliver on their responsibilities. For example, Alina, a research scientist with a doctorate in public health, reviewed Marcy’s initial literature review and conducted a more systematic review to ensure we’d identified all relevant articles. And while Gerardo, our principal data scientist, began building and running the code sets, Lon, our CTO, worked to help coordinate the effort and ensure prioritization of the effort against other top organizational priorities.

By the evening of Saturday, March 14th, we had a first draft of the diagnoses and factors to include in our first iteration of the model. By Sunday night, we were running the output through quality control and aggregating the list to make it actionable for care teams.

Our teams were able to channel chaos, break down silos, and because we were driven by a deep desire to help our care teams immediately outreach to those at highest risk and put our members first.

Applying the algorithm took flexibility and innovation.

Given the unique and urgent environment, our teams had to be flexible — which sometimes meant being comfortable adopting or tweaking systems. Our goal is always to prioritize members’ well-being. Hence that meant adopting temporary systems.

For example, although most of our work is documented and coordinated through our custom-built care management platform, Commons, we recognized that deploying engineering resources to build static interfaces for care teams would not allow us to quickly rollout COVID tracking, or allow us to iterate as we learned.

So in order to leverage and apply the algorithm to our work, we built and iterated quickly, working outside of normal workflows and databases.

Some of the processes and products we built and adopted included:

• Real-time COVID-19 dashboards to communicate metrics around at-risk members and outreach efforts
• New assessments to identify and track members with known COVID-19 exposure and symptoms
• Processes to utilize data from the health information exchange (HIE ) that alerted us when one of our members had COVID testing or went to the emergency department or hospital with a COVID-like syndrome
• Daily cross-market huddles to facilitate ongoing collaboration and information sharing
• Process flows for high and medium risk members to better equip our clinical teams doing outreach and management
• Proactive scripting to engage members and their loved ones with COVID-19 education and public health practices
• Training for all team members who were interacting with the new data and with our members

We’ll continue to innovate, learn, and share.

We are already translating the lessons we learned from identifying and treating COVID-19 high-risk members in New York to our members in Connecticut and Massachusetts, where the peaks of infection are a few weeks behind.

We know that other communities across the country and around the world will continue to battle COVID-19. We are humbled to be a part of a global collaborative effort to keep our communities healthy. If you or someone you know would find this algorithm beneficial, please access the full work at NEJM Catalyst, and give us a shout if we can help.