Using AI-Driven Insights to Understand COVID-19 Decision-Making
By Matt Phillion
Researchers with Surgo Health have used an innovative approach to map out the interwoven drivers behind vaccine hesitancy in the U.S. during the pandemic, using AI to identify causal factors behind why individuals chose or declined to be vaccinated against COVID-19. The technique could help build understanding about patient decision-making beyond the pandemic, offering insights into behavioral drivers and guiding organizations toward choices in their health initiatives and interventions.
The study used a form of AI called a causal Bayesian network to map out these drivers and uncover a system-level understanding of barriers to help providers, healthcare organizations, and even government agencies create more effective strategies.
One of the focuses of the study was whether political affiliation had a causal impact on vaccine hesitancy. What the study uncovered, though, was that the two strongest contributions to whether a patient chose to get the vaccine were contributing to social responsibility and protecting others in their community.
“As a company, we have been obsessed with the question of why patients are doing what they’re doing. Answering the why is critical to improve outcomes,” says Dr. Sema Sgaier, CEO and co-founder of Surgo Health. “It’s also pivotal to making sure everyone gets the healthcare benefits they need. During the pandemic, so much was said about the reasons why people were not taking the vaccine, whether it was political affiliation, race, and we knew the answer was far more nuanced.”
So they turned their attention to answering the fundamental question: What was keeping people away from this life-saving prevention?
The study drilled down into the behaviors of the respondents: what explained the things they believed, whom they believed and trusted, what information sources they were using, and what was impacting their decision-making.
“There was a lot of discussion in the news about how Republicans weren’t vaccinating, and we did a deep dive into thinking about why this might be the case,” says Dr. Vincent Huang, director of data science and AI at Surgo Health and co-author of the study. “In some ways, the results were not surprising to us in that things like social responsibility, if they felt responsible for the well-being of those in their immediate family and community, if they would regret if those people got sick” loomed large in the minds of those who got the vaccine.
“We didn’t give enough credit to ourselves and how much we care about others,” he says.
The two causes make sense, Sgaier notes. “Those two things were strongly explanatory,” she says: “social responsibility and the anticipated regret if they didn’t get vaccinated, and what that would lead to.”
The Surgo Health study applied causal AI to tease out factors that were explanatory in those behaviors, not just associative, Sgaier explains. “If you look at other countries that did really well on the vaccination front, these two factors were really front and center,” says Sgaier. “The campaigns they used, for example, or even just in terms of the societal structures—there was a responsibility for each other.”
Getting the bigger picture
Given the complexity of healthcare, getting the whole story from a data set is key, says Huang. This is where the AI used in the study comes into play. “With a causal Bayesian network, it builds a spiderweb, an interconnected relationship between variables, and this allows us to figure out [what happens to the outcome] when you change something,” he says. “It’s a more system-level view.”
“We get this hierarchal view: what’s more important versus less, what influences what. It’s a very transparent narrative,” says Sgaier. “It offers an explanation of the problem we’re seeing.”
Recently, Surgo Health conducted a study in rural India using the same technique, looking at how to influence mothers to give birth in the hospital instead of at home for a safer experience.
“The government was thinking maybe we need to build more clinics closer to the villages, but over the course of the study we found that if you have a transportation system, you don’t need to build hospitals closer to where they live,” says Huang.
The study also found that how the benefits were communicated to expectant mothers was important. “There are community health workers to help mothers, but [the workers] didn’t know exactly how they should be coached to help,” he adds.
It was particularly important for mothers to preplan where to deliver their baby rather than leaving the decision to the last minute, according to the study findings. “We can tell workers how to access care, how to allocate resources, and all of these impact multiple fronts with their healthcare decision,” says Huang.
The right data can positively influence behavior, Sgaier says, but behavior change is still very difficult. Making it happen starts with understanding the why. For example: Whom does the patient trust as sources of information and engagement, and what should those sources be saying to enable beneficial behavior?
Returning to the question of COVID-19 vaccination, the sources of truth in the U.S. focused on being very informational, Sgaier notes. “ ‘Here is where the vaccine is available. It’s good for you.’ It was very simplistic,” she says. “But if social responsibility is the factor you’re addressing, how do you bring that to the forefront?”
The answer is to tell the patient they are a key member of the community, show them what they can do for the community, and shape the message around explaining their responsibility to other community members.
“People don’t want to feel regret,” says Sgaier. “So we need to make that real for people: ‘If you don’t get this vaccine, this is what will happen.’ We have to get message crafters to do this really well and shape those campaigns in ways that are specific and touch on the core thing that is influencing the person.”
And, of course, the campaign must be communicated by someone the patient trusts. “People across the board trusted healthcare providers,” she says.
The future of causal AI and improving outcomes
Surgo Health’s ethos, Sgaier says, is to understand behavior and bring novel tech on board to get to the core of an issue. Starting off by studying COVID-19 vaccination made sense—an ever-present and immediate issue faced by the entire world—but Huang says the technology could be used anywhere for a clearer look at healthcare data.
“For example, how do we get people with chronic diabetes to take their medications? Or in maternal health, how do we close the equity gap, and what are the drivers of that gap?” says Huang. “It’s not issue-specific: We went big into COVID because it was a crisis, but as we work with other issues, it’s a question of do we have the right data to start with, investing in data generation itself, and going out to collect that data.”
The company has also noted that behavioral health and pharmaceutical studies could benefit from this type of research, the latter to ensure that trials are equitable and represent all facets of the community. “We can better understand what stands in the way of people participating in trials,” says Sgaier.
Surgo Health is animated by refusing to settle for the status quo of data analysis. “There are technologies and tools that could move us ahead, so let’s be open-minded about these tools,” says Sgaier. “The other part is understanding the value of data. It has to be high quality, equitable, representative. We talk a lot about AI, but it starts with data. Responsible use. Spending the time to understand where we’re getting it and why and how, and then spending the effort to translate that into action.”
Addressing what happens beyond the walls of a healthcare facility can have a huge impact on quality of care, Huang points out.
“In healthcare, so much of it is biology and mechanics, but here we need to understand the minds of people,” says Sgaier. “We need to shift from a body approach to a mind-and-body approach.”
Matt Phillion is a freelance writer covering healthcare, cybersecurity, and more. He can be reached at matthew.phillion@gmail.com.