The first phase of the vaccine rollout, which is supposed to deliver shots to roughly 24 million health care workers and residents of long-term care facilities, has been stymied by poorly conceived distribution plans based on judgement calls. Without better use of sound science and data, vaccine plans for the next two phases of the rollout, which aim to inoculate nearly 180 million Americans, could descend into complete chaos.
Most of the missteps so far stem from the same problem: prioritization decisions that ignore the science of risk assessment and leave too much to chance. From Stanford Medical Center’s reliance on a simplistic and arbitrary distribution plan that kicked front-line health care workers to the back of the line, to Massachusetts General Hospital’s use of an app that relied on workers to self-report their level of risk, to an Arizona county’s misplaced trust in a survey issued to health care workers, we’ve seen that the people who need the vaccine the most tend to get left behind when allocation decisions are made by faulty risk models or are up to the discretion of a handful of individuals, despite their best intentions.
For all of the logistical complexities of distributing the vaccine, the solution to determining the order of the queue for shots is straightforward. In any target group — whether it’s health care workers or long-term care residents or individuals over 75 — an individual’s risk for adverse outcomes when exposed to Covid-19 should determine his or her place in line.
Many states seem to know that overlooking science and data when making vaccine plans is no longer an option. This is especially true when it comes to the fast-approaching task of reaching people with underlying medical conditions that increase their risk for severe Covid-19, since identifying and ultimately reaching these people is not as simple as knowing their occupation — a nurse, say — or, in the case of long-term care residents, their place of residence. Stratifying those between the ages of 16 and 64 with underlying health conditions is a far more complex undertaking that requires combing through individual health records and understanding the local social determinants of health at play. And then reaching these people — getting shots in arms — requires understanding their level of connection to the health system, such as whether they regularly see a doctor or go to a hospital, for example, or live near a pharmacy.
But here, too, states seem poised to repeat mistakes, as most are signaling that they will turn to a combination of two ingredients to identify high-risk individuals: the number of people who live in a given county and the Social Vulnerability Index (SVI), a tool often used in the aftermath of natural disasters such as hurricanes that combines 15 factors, ranging from the percent of people living in poverty to lack of access to a vehicle and crowded housing, to determine the vulnerability of each U.S. census tract. And in many cases, pharmacies will be the primary vehicle for distribution during upcoming phases of the rollout — a decision that falsely assumes all Americans have immediate access to a pharmacy.
That’s a recipe for disaster. Using population as a factor means that dense communities, merely because they are populous, will get far more doses than they need and more sparsely populated communities will get far fewer doses than they need. For example, Los Angeles County, the largest county in the country, would get more doses than it needs because it is home to 10 million people, not because it is home to the greatest share of high-risk people.
What’s more, the SVI fails to account for individual clinical history or key social determinants of health at the county level, such as air quality and proximity to grocery stores based on ZIP code. And relying on pharmacies for distribution leaves out people who live in pharmacy deserts, which are more common in low-income neighborhoods, communities of color, and rural areas.
Instead of the population size plus SVI model to determine vaccine allocation, states should turn to a three-pronged data set.
First, the SVI must be supplemented with available data on social determinants of health. Second, the model should be based on peer-reviewed medical literature about SARS-CoV-2, the virus that causes Covid-19, and other coronaviruses. Third, the model should include individual clinical history whenever possible. This, of course, would require states to work with health insurers, which is exactly the kind of public-private coordination needed to successfully roll out vaccines to all Americans.
Health insurers have a longitudinal view of their members’ clinical histories, meaning they have an exhaustive record of individuals’ care and treatment. Hospitals, on the other hand, have more of a transactional view, such as when a patient experiences an emergency health event. And to ensure that those without insurance are not missed, the model needs to flag communities with limited access to key points of vaccine delivery, such as pharmacies, hospitals, county health clinics, and federally qualified health centers.
I have seen firsthand how drawing on insights from more precise data sources can help tackle these urgent challenges for making vaccine plans. In December, my company, Cogitativo, conducted a pilot project at the request of the Department of Health and Human Services, which was exploring ways that states could optimize their prioritization decisions by identifying communities at greatest risk and determining the best pathways to reach high-risk individuals. We analyzed demographic, clinical, and social determinants of health data, along with peer-reviewed medical literature on Covid-19 and other coronaviruses, to create risk scores for more than 20 million Californians and precisely identify the communities with the most pressing needs, down to the ZIP code.
We identified a number of key risk indicators that have gotten almost no attention in the national conversation. We found, for example, that significant drivers of elevated risk of developing serious Covid-19 include not only conditions like high blood pressure and diabetes, but also conditions like lupus, sickle cell disease, and severe mental health issues.
Regarding social determinants of health, we also found that individuals who lived far from a grocery store, to take one example, were at an increased risk of ending up in the hospital with Covid-19. And in terms of vaccination pathways, we found that 20% of at-risk Californians aren’t engaged with the health care system. These individuals are among those who will be missed unless states launch targeted outreach programs communicating the critical importance of the vaccine.
This HHS pilot project shows the kind of granular insights that can be uncovered when we turn to science and data. Our nation — and our economy — cannot afford more vaccine missteps, and budget-strapped states cannot afford an endless series of costly do-overs.
Of course, reaching this comprehensive view of risk requires sifting through massive amounts of data that would be impossible for any one state or public health agency to do alone — but that’s no reason to take an easier path. To complete the vital task of making effective vaccine plans, states must partner with major provider and payer networks as well as county health clinics and federally qualified health centers that typically deliver care to underserved populations, all of which have the best line of sight into the needs of their patients and beneficiaries.
The health of our country and economy depend on it.
Gary Velasquez is the co-founder and CEO of Cogitativo, a data science company.