In late 2020, then-HHS Secretary Alex Azar tapped Berkeley, California-based Cogitativo to help inform the Trump Administration's vaccine distribution strategy.   

The company had developed a machine learning model that could predict the probability a patient would end up in an intensive care unit bed.  

Using 60 million records from HHS, Cogitativo created risk scores that flagged individuals who might be at particular risk from the effects of COVID-19.  

Now, explained CEO Gary Velasquez, the company has been tasked with a new project: Working with the Veterans Health Administration to identify beneficiaries who have deferred care during COVID-19.  

At this point, said Velasquez in an interview with Healthcare IT News, "You're worried about the veterans you haven't seen."  

In early January of this year, the Cogitativo team was given access to patient data to help identify 12 clinical conditions that may be exacerbated by delays in services.

Within a few weeks, says Velasquez, the team had developed an algorithm for three: diabetes, chronic kidney disease and arteriosclerosis. That model is currently in the validation phase. Velasquez says Cogitativo is aiming to deploy models for about three conditions per quarter. 

"What we're scoring is down to the individual beneficiary level," he said, in order to "rank the probability of an acute clinical event or disease progression within the next 12 months."

Velasquez notes that many people in the healthcare industry use the phrase "demand" to signify pulls on medical resources.  

Instead, he says, "We need to think about need."

As with all artificial intelligence, the possibility exists for bias in Cogitativo's algorithms.   

"Every data set is biased," says Velasquez.   

For instance, he notes, the data set Cogitativo is using isn't currently representative of populations in the Dakotas or Wyoming – although he says if it contracts with clients based there, it will expand its information base. 

However, it is quite robust: The team is using thousands of peer-reviewed research papers in addition to a training data set of 200 million claims and 100 million electronic health records, on top of psychographic data and social determinants of health data.

"What I try to do to reduce bias is to get as big of a data set as I can for the population of interest," he said.

He stressed the importance of "making the unseen seen, the unheard heard."  

"Healthcare has trillions of noisy signals," he said. "How we can pick out the signals in that noise is fundamental."  

For him, it's about actionable information: "If we can pick out inequities in healthcare, stakeholders at the federal, state and commercial level can act upon our insights."

He notes, for example, that addressing such inequities requires a multipronged approach.  

"The world wasn't perfect before COVID," he said.  "Our mission is trying to move the needle, a patient at a time."

Kat Jercich is senior editor of Healthcare IT News.
Twitter: @kjercich
Email: kjercich@himss.org
Healthcare IT News is a HIMSS Media publication.