Machine learning could help public-health authorities make sure that the most vulnerable people move to the head of the COVID vaccination queue.
With limited supplies of COVID-19 vaccines now available, states are scrambling to decide who should be first in line to receive the shots. But in many cases, that question is being settled based on loosely labeled groupings of at-risk populations, such as “the elderly” or “essential workers.” While it may sound simple, this approach is short-sighted, because it ignores the incredibly complex mix of factors that put people at greater health risk and have been shown to drive increased incidences of hospitalization and mortality due to COVID-19.
A precise determination of health risks must be the foundation of a successful vaccination effort. Otherwise, scarce resources will be squandered on those less in need while blocking vaccine access for the most vulnerable.
How should states identify those most at risk with precision? They must have a way not only of analyzing basic demographic data but also of taking into account underlying health conditions, social and environmental determinants of health, and the latest research on the mechanisms of the coronavirus.
For that difficult but critically important task, states must heed the advice of frontline doctors and researchers who have been working around the clock to solve this public health challenge. Many of them say that the best tool at our disposal is machine learning.
Machine learning leverages all the data available and can detect important hidden patterns that will otherwise go unnoticed—and alert us to the heightened needs of people who might otherwise slip through the cracks. I saw this firsthand in a recent project involving my data science company, Cogitativo, and the insurer Blue Shield of California. In the early days of the pandemic, Blue Shield of California stepped-up and sought to provide personalized health counseling and support services to its most COVID-vulnerable members. The insurer engaged us to identify those vulnerable members.
Cogitativo built a machine learning platform that brings together factors about an individual’s health history (in particular, whether the individual has a specific high-risk underlying condition) combined with social, environmental conditions and the most up-to-date medical literature on COVID-19 and other infectious diseases. These factors predict the risk of having adverse health outcomes from COVID-19 infection.
Many of the findings were surprising. For example, the risk-scoring tool found that individuals who did not live within a close-enough vicinity to a grocery store were at an increased risk of ending up in the hospital, on a ventilator, or even dying from COVID-19. Another finding was that individuals who had experienced severe mental health issues were at greater risk. Based on these and other insights, Blue Shield of California provided various services to its members, including free meal delivery, medication delivery, telemedicine, and in-home clinical visits.
The lesson from our project in California was clear: Relying solely on age or making judgment calls about whether employees in a particular industry are at greater risk will result in vulnerable populations being overlooked and put at unnecessary risk or worse.
This flawed approach for allocating constrained vaccine supplies could exacerbate inequities that have already devastated communities of color during the pandemic. Latinx and African Americans in the U.S. have been three timesmore likely to contract the virus and twice as likely to die from the virus than corresponding white populations. We cannot, for example, fail to account for one important risk factor such as housing density, which is much greater in urban communities. The vaccine distribution effort must play a role in reducing these disparities, not widening them.
A failure to prioritize the most vulnerable for vaccination will cause avoidable deaths. As someone who has experienced the pain of losing family to avoidable health events, I ask that states act now to solidify their vaccine distribution strategies. The virus is surging again. In recent weeks, hospitalizations in the U.S. have soared to record highs, daily reported cases have blown past previous records, and there were three consecutive days of more than 2,500 deaths for the first time. Furthermore, public health officials are warning that some of the darkest days of the pandemic lie ahead.
States must use the most advanced technological tools at their disposal. Machine learning provides an equitable, precise and expedient capability to allocate our precious vaccine supplies. The use of science and of data-driven decision-making can help ensure that states reach the right people at the right time—and that no American gets left behind.
Gary Velasquez is the co-founder and CEO of Cogitativo, a data science company.
