How AI can help get COVID vaccines in order

The task of prioritizing vaccine distribution is complicated, and one health system has turned to AI for help getting their program in order.

As doses of the COVID-19 vaccine are distributed around the country, healthcare policymakers and providers are discussing how best to prioritize who should be toward the front of the line for inoculations.  

After all, even among eligible patients there are those who are more susceptible to the virus than others, so providers are increasingly putting AI to use to determine who they are.

As a recent article explains, for example, from the early days of the pandemic researchers at    South Dakota-based Sanford Health worked to determine priority patients, and they’ve developed an AI algorithm that helps identify which of Sanford’s patients who are 65, most at risk, and, thus, should be vaccinated first.

As Sanford chief physician Jeremy Cauwels explained, beginning in the early days of the virus in this country, a Sanford data analytics team started compiling available research on who was at greatest risk of having the most severe outcomes of the disease.  While compiling the research on COVID-19 risk factors, Sanford has also added data on the roughly 85,000 COVID-19 patients Sanford has treated across its system. 

"With that 85,000 people, what we can do is take a real-time picture that evolves over time using computer learning, to tell us what patients or what people in the Midwest get the sickest from COVID-19,” Cauwels said. “And which ones spend more time in the hospital, which ones have the highest risk of death.”

According to the article, to prioritize vaccine doses, Sanford first narrows its pool of possible recipients to members of the groups state health officials say are eligible to be vaccinated — currently people 65 and older – then apply the computer algorithm to that group to assign a patient’s level of risk for worse outcomes from COVID-19.

"If you were to just go with the three most most mathematically important for predicting whether you're going to do well or not [with a COVID-19 diagnoses], you could go just with age, obesity and kidney disease," said Cauwels.

Indeed, the article notes, most healthcare in Sanford’s region are using some form of random choice to decide who within priority groups is first for a vaccine, but Cauwels believes an AI-based approach is more equitable than random choice to administer the vaccine.

“I think the reason for that is the difference between what we've seen here and what we've seen in South Florida, where people are lined up for blocks next to a hospital because they found out that doses are available,” said Cauwels. “Or you have people that are waiting hours on the phone in the Minneapolis area to try to get into a finite number of vaccines slots.”

Still, Sanford stakeholders recognize there are many people in their area who are not regular patients with any provider and thus may have no health data in any system. The organization has tried to address that gap with a website where anyone can sign up to be part of the patient database from which the system prioritizes vaccine distribution. 

Officials said nearly 90,000 people have signed up, providing basic medical and personal information, since vaccines became available.

"We use as much as we know about their medical information to put them on the list, right in between the people that we've known about for 20 years, to make sure that we're doing the best thing we can to distribute equitably between all populations," said Cauwels.