More Race Data Needed on COVID Vaccination and Testing, Researchers Say

The U.S. is likely not getting the whole truth about the spread of COVID — but why? Because the data lack equity, Ziad Obermeyer, MD, said at an event sponsored by the Alliance for Health Policy.

“At the national level, we collect data from lots of poor and Black counties and, that might look great, but there’s this other problem, which is, what do we mean by COVID?” said Obermeyer, who is a professor of health policy and management at the University of California Berkeley School of Public Health, at the event on Friday. “How do we track COVID cases and COVID deaths? Well, we track it because people get tested for COVID, and we count that as COVID. But we also know that lots of people don’t have access to testing for COVID.” Those patients, who are often members of racial or ethnic minorities, “don’t show up in our measures.”

“We see the tip of the iceberg of the COVID cases that are diagnosed, and it’s very tempting to treat that as the ground truth,” he added. “But that’s not the ground truth, and in fact the reason it’s been so hard to predict COVID is because we’re missing that whole bottom part of the iceberg of undiagnosed COVID that is spreading without us tracking it, but it’s still killing people, and it’s still doing so much damage.”

Another problem is that current data on COVID are often lacking when it comes to numbers on race and ethnicity, said Samantha Artiga, director of the Racial Equity and Health Policy Program at the Kaiser Family Foundation. For example, the CDC reports race and ethnicity data at the federal level, but not at the state level, and race and ethnicity data are missing for 40% of vaccine recipients. Such data also are not available for any recipients of vaccine boosters, she said.

Smaller ethnic groups are even more shortchanged, especially at the state level. A total of 25 states plus the District of Columbia, for instance, don’t report racial or ethnic data for American Indians/Alaskan Natives, or for Native Hawaiians/other Pacific Islanders, Artiga said. States also differ in the way they use these classifications; 23 states don’t include Hispanic as a race category.

“These ongoing data gaps and limitations really limit our ability to get a complete and nuanced picture of who is and who is not getting vaccinated,” she said. “And this incomplete understanding and the limits on our ability to interpret and analyze the data hinder efforts to effectively direct resources and implement efforts to address disparities.”

Areas to focus on in the future include prioritizing data collection and reporting, increasing availability and accessibility of health data, and expanding the data available for smaller population groups. Researchers also should conduct outreach and education for clinicians and others who are collecting data “to help them understand the importance of reporting data by race and ethnicity, and why reporting that data is actually helpful,” Artiga said.

Regarding inequity in healthcare data more generally, Obermeyer discussed research he and his colleagues have done on artificial intelligence algorithms that are used in health screening to predict who will get sick. In particular, they studied one piece of software sold to health systems that contains algorithms “to make an important decision about who gets help … The market estimates of these algorithms is 150 to 200 million patients per year,” he said. “So, effectively, the majority of the U.S. population is being screened through one of these algorithms. And that algorithm score is determining whether you get access to extra help with your health today to prevent these chronic illnesses.”

“We were interested in whether those algorithms were biased,” Obermeyer continued. “We found that the particular product we studied was extremely racially biased; what it was doing was it was prioritizing healthier white patients ahead of sicker Black patients for extra help with their care.”

However, the good news is that this problem is fixable, he said. Obermeyer’s team was able to work with the company whose algorithm they studied and created a revised version that more accurately predicted patients’ healthcare needs, “and in doing so we reduced the bias in that algorithm by 84%, and we found similar results in lots and lots of different settings, with different partners in the healthcare sector.”

Healthcare payers and institutions need to have this kind of algorithmic evaluation on their radar, said Kadija Ferryman, PhD, a cultural anthropologist at Johns Hopkins University’s Berman Institute of Bioethics. “Maybe healthcare institutions might be rewarded for having unbiased algorithms or showing that their algorithms have decreased health disparities between groups,” she said. “There could be incentives for you to make sure that your algorithms are really doing what you say they’re doing and are proactively addressing some of the long-standing issues.”

One challenge with trying to analyze racial bias in healthcare is that health insurers often don’t have much demographic information about people, Obermeyer said. “These insurers get a ton of data from hospitals and providers, and there is no reason that they cannot also ask for demographic information along with all of the other data that the hospital is sending. The reason that they often don’t is because they would, in some cases, rather not know about disparities,” and that’s especially true of hospital legal departments. However, he added, according to state and federal regulators he has spoken with, “it is a bad defense, when someone comes asking about bias, to say, ‘We’ve never looked into it’ — that’s actually not protective, and if anything, it looks worse.”