Skip to Main Content

As public health leaders double down on their efforts to get shots to the 30% of eligible people in the U.S. that remain unvaccinated, one health center has described a unique strategy to reach those who might have fallen through the cracks.

During the early days of vaccine rollout, “we realized that the vast majority of this communication was being done either by email or by MyChart messaging,” said Jacob Stein, an oncologist at the University of North Carolina’s N. C. Cancer Hospital. But relying on digital communication can leave some patients behind. Older people might be less likely to frequently check online patient portals. Rural residents could have limited internet access. And many patients don’t own a device to check up on providers’ messages.


So Stein and a team at the N.C. Cancer Hospital decided to use its records differently: to identify patients who might have been missed by the greater health system’s vaccine communications.

First they narrowed down a set of people who were likely to still be living in the community: those who had received cancer therapy in the last year and still had follow-ups scheduled. Within that group, they looked for patients who didn’t have active patient portal accounts or email addresses on file, which meant they were highly unlikely to have received the system’s digital messages. They also included patients who lived in a county with a poverty rate over 20%. Of adults with annual household incomes below $30,000, 43% don’t have broadband at home, and 24% don’t own a smartphone, according to the Pew Research Center.

Over the first three months of this year, four nurses at the hospital called more than 500 people to share information about vaccine safety and logistics and answer any questions. They reached nearly 400 patients directly or by voicemail — and by the beginning of April, more than 100 had received or were scheduled to get a vaccine, as described in a research letter published in JAMA Oncology.


One patient was particularly hesitant; she had concerns about everything from potential side effects and risk, to cost and logistics, recalled Stein. But when a nurse at the center gave her a call to discuss the vaccine, it turned into a long conversation.

“At the end, she said, ‘This is the push I needed,’” said Stein. She told the nurse she’d be getting her shot.

UNC is far from the only health care provider attempting to increase vaccine outreach in marginalized communities. Many of those important efforts target entire communities, and focus their outreach on locations, rather than individuals. States have allocated additional vaccine supply for underserved areas and deliberately set up clinics in low-income communities; CVS prioritized vaccination sites in areas that ranked high on the social vulnerability index established by the Centers for Disease Control and Prevention.

While that particular type of support is important, not every patient living in a certain community will require it. Often, demographic-driven outreach efforts are informed by county-level datasets like the American Community Survey, which tracks variables like commuting transportation and broadband internet.

“Everyone thinks they can use that community-level data as a substitute for patient-level data, and they can’t,” said Rachel Gold, a senior investigator at the Kaiser Permanente Center for Health Research and lead research scientist at OCHIN. “It’s like flipping a coin, if you use the community-level data to predict what an individual patient’s going to need.”

Providers might be able to deliver support more precisely if social risks were tracked within a patient’s electronic health record. But practices only started collecting that kind of information, known as social determinants of health, in the last few years. And adoption is still low.

“That has been the rate-limiting step in using social data,” said Laura Gottlieb, a professor of family and community medicine at University of California, San Francisco, who researches social and medical services integration. “We’ve provided these tools now with these records, but their rate of uptake has been pretty remarkably low.” In an analysis of UCSF’s own system, tools for social risk documentation in EHRs were used less than 1% of the time.

Gottlieb said that’s slowly changing with a push from the Centers for Medicare and Medicaid Services, which is working to encourage collection of social risk data. And in July, the Office of the National Coordinator for Health Information Technology announced that it would incorporate social determinants of health into its standardized list of data elements for interoperable records, including factors like food, housing, and transportation insecurity; employment and veteran status; and material hardship and stress.

In the UNC project, providers skipped over the problem of standardization, identifying patients for outreach not by searching for a defined data element in their records, but by capitalizing on missing data that suggested they might not see digital-only messages. “We already have collected very simple things like race, ethnicity, and language,” said Tung Nguyen, who practices at UCSF. “And we’re not even making that actionable.”

“We get stuck in health care trying to identify the perfect way of doing this, and perfect is the enemy of good,” said UNC’s Stein. “These are simple, readily available data that give us a glimpse at what folks are experiencing.”

If better patient-level data were available, said Gold, one natural application would be to help patients experiencing transportation insecurity. Many Covid-19 vaccination efforts have attempted to reduce barriers to transportation: Cities run ride voucher initiatives and mobile vaccination clinics bring shots straight to neighborhoods in need. The same concept could apply to a group identified through a method like UNC’s: “For patients who don’t have transportation, you reach out and say we’re going to provide you a Lyft voucher and a vaccine appointment that doesn’t need to be scheduled,” said Gottlieb.

But Gottlieb emphasized that simply targeting an intervention at the patient level doesn’t mean it’ll have an impact. “You’d have to have an intervention that met the needs of that individual person,” she said. In her previous research, for example, very few people have taken health systems up on transportation services even when they’re offered directly, because other factors like job insecurity or childcare requirements still get in the way.

Gottlieb hopes to see more trials that conclusively measure how much of an impact these individualized interventions can have. Because UNC’s cancer center simply reached out to all the patients its search turned up, there’s no comparison group to prove whether the effort increased vaccination rates.

“Our approach was not perfect, but it’s why I think this kind of work is really critical,” said Stein. “We need people to just start trying and sharing how it went, because when one project gets some success, then the next group can tailor that and improve it.”

Stein and his colleagues also saw another sort of success with the project.

“Many of the patients who met our criteria for potentially missing the outreach for Covid were also people that were really struggling in the pandemic,” said Stein. While the average phone call made by the cancer center’s nurses was just over four minutes, a handful lasted longer than half an hour. In many cases, the person on the other end of the line was able to connect those patients with other support. “People were so relieved and appreciative that someone was reaching out to check on them,” said Stein.

“As much as an analytics approach is great to identify the patient population, an algorithm can never do that human element of listening.”

Create a display name to comment

This name will appear with your comment

There was an error saving your display name. Please check and try again.