Race in CDS Tools
By Lisa A. Eramo, MA
For The Record
Vol. 35 No. 1 P. 10
Does it contribute to ongoing health care disparities? Growing evidence says ‘yes.’
Health care organizations have collected data about race for decades—since the dawn of EHRs and even before that. However, what do organizations do with that data? More specifically, should they use it for clinical decision support (CDS) to address health care disparities? Experts agree that the answer is anything but straightforward, and in some cases, data about race can actually exacerbate disparities rather than help solve them.
Identifying Health Care Disparities
It’s clear that race plays an important role in addressing health care disparities. Nothing made this more apparent than the COVID-19 pandemic.
“Historically, there have been a number of health inequities that have existed for generations. They’ve continued to exist specifically in communities of color especially when it comes to preventive care and access to care,” says Sherri D. Onyiego, MD, PhD, FAAFP, a family medicine physician and medical director for Texas with Equality Health.
Khadijah Breathett, MD, MS, FACC, FAHA, FHFSA, a tenured associate professor of medicine and an advanced heart failure and transplant cardiologist at Tools Indiana University, agrees. “During the COVID pandemic, we approached an era of social justice that’s receiving an increasing amount of publicity and media attention. People can no longer ignore these longstanding issues,” Breathett says. “These are real problems that are devastating to minoritized populations. Maintaining the status quo is discriminatory and unsustainable for US health care and society.”
Understanding CDS Algorithms
The question is this: Should race be a part of CDS algorithms, including the tools, risk calculators, and decision aids that clinicians use to make medical decisions?
In many cases, the answer is no, says Breathett, who provides the example of estimated glomerular filtration rate (eGFR) that’s used to monitor kidney function. Until recently, many health care providers and laboratories have included race in the calculation. However, more recently, research has shown that using race may be inappropriate. In September 2021, the National Kidney Foundation and American Society of Nephrology task force released a report suggesting that current GFR calculators overestimated eGFR in Black populations and, to a lesser extent, non-Black people.
“This means Black people were incorrectly classified as having less severe kidney disease,” Onyiego says. “This led to inaccurate diagnoses and kidney disease staging as well as treatment delays resulting in Black patients not being considered for kidney transplants.”
As a result, many health care organizations are starting to remove race from the CDS tools, says Andrew Resnick, MD, director and chief medical and quality officer at The Chartis Group. “There’s a movement to identify racialized medicine and get rid of it,” he says. “With the eGFR calculation, it makes renal function of Black patients look artificially higher, reducing access to treatments including kidney transplants.”
UC Davis Health, for instance, stated on its blog that it would stop using race as a parameter for eGFR on May 4, 2021. The blog states: “A single eGFR calculated without the race parameter will be reported as a discrete value in the chart. Using this calculation is more sensitive for the detection of CKD [chronic kidney disease]. This supports the goal of narrowing some of the disparities in CKD, including the higher prevalence of CKD in African Americans compared to white patients, the higher rate of progression to ESRD [end-stage renal disease], and transplant eligibility.”
Onyiego provides a similar example that uses race and ethnicity in clinical decision-making: the vaginal birth after cesarean (VBAC) calculator. “The VBAC calculator is a simple tool that predicts the likelihood that a woman will have a successful vaginal birth after having a previous cesarean section, and obstetricians have been using it for years,” she says. “However, recent literature has shown that the inclusion of race as a risk factor for this calculation is primarily built upon a social/structural basis rather than a biological one. This can potentially result in women with a less than favorable score not being offered an option to undergo a trial of labor and vaginal birth.”
For example, research has shown that Black, American Indian, and Alaska Native women are two to three times more likely to die from pregnancy-related causes compared with white women. Cesarean deliveries founded upon race-based VBAC calculations/predictions could be one of the drivers of these health inequities, Onyiego says. “There have been efforts to remove race from this VBAC calculator,” she adds.
A recent article published in the New England Journal of Medicine highlights countless other examples where race data are used inappropriately in multiple specialties—all to the potential detriment of minority patients.
In addition, The Washington Post reported in 2019 that a widely used algorithm used to predict which patients will benefit from extra medical care dramatically underestimated the health needs of the sickest Black patients.
“We cannot allow these errors to occur in population health management algorithms at a broad level or at the individual level when it’s clinical decision support at the bedside,” says Jesse Ehrenfeld, MD, president-elect of the American Medical Association (AMA).
Efforts to remove race from CDS tools make sense given the fact that race is not biologically based, Onyiego says. In fact, in 2020, the AMA revised its policies to recognize that race is a social—not biological—construct.
“Race has been included in a number of clinical decision support tools over the years, but now there’s more awareness of the fact that race is not tied to any biological differences,” Onyiego says. “It’s the structural and societal implications on groups of people that put them at higher risk for worse outcomes. When we factor race into clinical tools and algorithms, it can lead to over- or undertreatment, or it can prevent people from receiving procedures to improve their outcomes.”
Ehrenfeld agrees. “There’s a false conflation of race with inherent biology of genetic traits. What I think is being recognized is that race is often a proxy for risk factors and structural racism. We don’t have any evidence where race is a primary biological determinant.”
There’s increased awareness at the federal level, too. In October 2021, the House Ways and Means Committee published a report describing how the inclusion of race in CDS tools negatively affects health equity and how stakeholders can address it. In addition, at the request of Congress, the US Agency for Healthcare Research and Quality is examining how clinical algorithms may introduce bias into clinical decision-making.
Embracing CDS 2.0
Still, CDS tools of the future may be sufficiently sophisticated to account for the nuances related to race and health outcomes. “We need to replace race-based algorithms with race-conscious algorithms,” Ehrenfeld says. “The latter incorporates race to promote equitable health outcomes despite persistent racial inequities in our institutions.”
Making a better model starts with clinical guidelines, Breathett says. “Many clinical guidelines are not based on diverse populations, and health care systems are starting to reevaluate this,” she adds. “Some of these guidelines have been published and used for a long time, and people are recognizing that the way we’re doing things is wrong, and it contributes to worsened care.”
Breathett adds, “Those making decisions about what factors go into these models must be a diverse group of people. The more diverse the team, the better the outcome. It’s also important that these models are tested and studied on a diverse population. If they’re only studied among white men, it’s not going to produce accurate results for other populations.” Health care systems, she says, “also need to be willing to look at who develops the algorithms and on whom those algorithms are tested.”
AHIMA recently echoed this sentiment, stating bias must be accounted for when developing artificial intelligence (AI) and other physician-assisted technologies. In its recommendations to further health equity in US federal policy, AHIMA states: “As health care continues to advance and implement new technologies, it is crucial for those who develop, train, and implement those technologies to appropriately account for bias during operational activities. If a diverse set of experts are not present at every step of the development process, then AI and other physician assistive technologies will be implemented with unintended biases. Ensuring there is diverse representation throughout all steps of the development and implementation process will assist in limiting the presence of unintended bias in next generation health care technologies.”
Transparency is also important so physicians are fully informed about the algorithm, how it works, and whether it would be appropriate to use for certain patients, Ehrenfeld says. “When you know the inputs, the data sets, and the variables that are in the model, it’s pretty easy to understand what drives the outputs,” he says. “Where things get tricky is when you start using machine learning. Some companies choose to make it very clear in terms of the underlying basis for the algorithm. The AMA supports this. Other companies have been much more reluctant to share the details. We think this is problematic.”
Breathett says clinicians should always ask themselves these three critical questions when using CDS tools:
1. Has the calculator been validated in a diverse population?
2. Does the calculator contribute to biased decision making or racism?
3. How can I tailor my care plan to address bias, structural racism, and social determinants of health?
Addressing Structural Racism
Perhaps even more important is recognizing the effects of structural racism—particularly in the absence of race data in CDS tools, Breathett says. “To completely remove race from CDS tools without thinking about the things that contribute to bias is wrong,” she says. “What we need to do is address the factors that contribute to structural racism and address those in our clinical decision making so we can provide equitable care.”
Ehrenfeld agrees. “The AMA supports the consideration of race and ethnicity as a marker for increased risk of disease,” he says.
Ensuring Data Integrity
Collecting accurate and comprehensive data about race is more important now than ever before, says Steven Spencer, MD, MPH, MBA, deputy chief health officer at Cityblock Health. This starts with the EHR. For example, does the EHR provide sufficient options for race? In addition, frontline staff must feel comfortable asking patients questions about their race. The first step for any organization that wants to do this properly is to stop making assumptions,” he adds. “Staff need to ask the question. They can’t assume race or ethnicity.”
Breathett agrees. “You can’t guess race or ethnicity by how someone looks or by their name,” she says. “Patients must self-identify that information.”
However, even when staff ask the question, some patients may be hesitant to provide the information. That’s where patient education comes in.
“I think we need to be better about explaining the ‘why’ behind the question,” says Spencer, adding that even on intake forms, patients simply leave the question about race blank, or they select “other.” “There’s a pervasive thought that regardless of class or insurance status, minority populations will not get the same level of treatment as nonminority populations,” he adds.
Research confirms this. A 2020 survey conducted by the Kaiser Family Foundation found that Black and Hispanic adults are more likely than white adults to report they were personally treated unfairly because of their race and ethnicity while getting health care in the past year. Black adults also are more likely than white adults to report negative experiences with health care providers. This includes feeling a provider did not believe they were telling the truth, being refused a test or treatment they thought they needed, and being refused pain medication. In addition, Black and Hispanic adults are more likely than their white counterparts to say it is difficult to find a doctor who treats them with dignity and respect.
Combating these challenges starts with data collection, and frontline staff must be comfortable explaining how the organization is using data about race to enhance treatment—not serve as a barrier to it, says Spencer, who provides these three talking points staff can use to engage patients:
• The organization asks this question of every patient—not just minorities.
• The organization collects the information because it wants to treat everyone equally.
• The information may be important for your treatment and the reason you’re seeking care.
However, seeing the bigger picture of health equity is also important, Resnick says. “It’s definitely a process,” he adds. “I’ve worked with many organizations to try to work through this. It’s not just about one conversation—it’s about building trust with the community you serve.”
Leveraging Data About Race to Improve Health Equity, Value-Based Care
Organizations can leverage race and ethnicity data to promote equitable care across all patient populations—a goal of value-based care, Onyiego says. “I think we’re going to continue to see that having accurate information about race, ethnicity, and language helps providers make better decisions about taking care of their populations,” she says. “We can use value-based care as a lever for health equity.”
For example, organizations can—and should—examine all clinical outcomes by race and ethnicity, says Spencer, who did this previously when he worked as medical director of population health at Abington Hospital in Abington, Pennsylvania. There, he examined multiple clinical outcomes by race and ethnicity, including a comprehensive look at disparities in mammogram rates.
Resnick agrees, adding that organizations should stratify all clinical conditions, outcomes, and process measures by race, ethnicity, and language (REAL); disability; and sexual orientation and gender identity (SOGI) data. He provides the example of readmissions. “If you’re trying to solve for readmissions, and you’re just looking at all patient readmissions, it’s going to be awfully hard to close the gap,” he says. “But if you look at it for each population, then you start to identify differences in how those patients have access to care.
Organizations can also leverage REAL and SOGI data to promote culturally competent care and inform culturally competent training for providers. “Culturally competent care needs to occur regardless of the setting,” Onyiego says. “It also goes beyond providers themselves and extends to the front desk staff, C-suite, and all points in between. It also includes things like providing health information in a patient’s preferred language, providing access to an interpreter, helping the patient absorb the information in their culturally preferred way, and more.”
“The AMA strongly believes in the importance of race and ethnicity data collection to better understand the distribution of wellness and disease across our society and the legacies of structural racism in our systems to inform the design and implementation of solution to advance and drive equity,” Ehrenfeld says.
— Lisa A. Eramo, MA, is a freelance writer and editor in Cranston, Rhode Island, who specializes in HIM, medical coding, and health care regulatory topics.