Scientists and clinicians increasingly question the wisdom of using race as a variable in algorithms that guide health and treatment decisions. A team of University of Washington School of Public Health researchers recently examined one such case: the use of race in the Multi-Ethnic Study of Atherosclerosis (MESA) algorithm used to calculate the 10-year risk of coronary heart disease. This score is widely used by physicians and patients to assess heart disease risk when coronary artery calcium (a measure of atherosclerosis) is known.
“Our work is part of a growing effort to assess the implications of including race and ethnicity in clinical risk prediction models,” said Quinn White with the Collaborative Health Studies Coordinating Center (CHSCC), a UW biostatistics doctoral student who is the first author for the study.
The team compared the original MESA risk score to an updated version that did not include race/ethnicity.
“We used the same model development process as that for the original MESA Risk Score. This included the use of LASSO and Ridge regression in two stages. The only difference in our approach was to exclude race/ethnicity, along with any interactions with race/ethnicity, from the set of candidate predictors considered in the variable selection step,” said White.
Findings revealed that the version of the MESA risk score calculation that did not include race predicted heart disease as well as the original calculation that did.
“In the case of the MESA risk score we have shown that if coronary artery calcium — a potent marker of subclinical cardiovascular disease — is known then there is no loss of predictive ability in omitting race/ethnicity from the risk score development. Race is a social construct, and when it can be replaced by other risk factors, or simply left out entirely, it is desirable to do so to avoid inadvertently perpetuating disparities,” said Robyn McClelland, CHSCC director and the primary investigator for the MESA Coordinating Center who is a co-author of the study.
“This is an important advance in that many people do not identify with one of the race/ethnicities in MESA, may be multi-racial, or may not wish to factor this into their clinical care. This opens the door for our score to be useful for more people in a way that does not highlight race/ethnicity in clinical decision making.”
White emphasized that developing equitable risk score models is more complex than simply including or excluding race/ethnicity.
“On one hand, there have been various examples of how the inclusion of race/ethnicity can cause harm to marginalized groups, such as race-based equations to calculate eGFR that assigned Black patients to have higher kidney function, resulting in restricted access to treatment. However, there is no guarantee that simply removing race/ethnicity from a risk score model will prevent harm to marginalized groups, and the consequences of this modeling choice must be thoroughly investigated for each risk score.”
White presented study results at the American Heart Association’s (AHA) Scientific Sessions in November 2024. In addition to White and McClelland, study authors included CHSCC research scientists Craig Johnson, MS and Spencer Hansen, PhD as well as Brittany Saldivar-Murphy, MD and Andrew DeFilippis, MD, MSc with Vanderbilt University, and Wendy Post, MD at Johns Hopkins University.
The updated MESA risk score will be released as an easily accessible online calculator in early 2025. White’s team is currently working on a manuscript to submit for publication in early 2025.
White’s work was supported by her MESA Research Assistantship and an AHA grant she was awarded as part of the Debiasing Clinical Care Algorithms AHA Data Grant Program.
