Equal treatment for equal risk: Should race be included in allocation algorithms for Covid-19 therapies?

When the Omicron surge threatened to overwhelm hospitals and the number of infections greatly exceeded the quantities of anti-Covid therapies that might help keep people out of the hospital, the need to prioritize individuals at the highest risk was clear. Yet controversy quickly arose around what attributes — including race — might be used to evaluate risk and prioritize access to scare resources.

“The left is now rationing life-saving therapeutics based on race, discriminating against and denigrating, just denigrating, white people to determine who lives and who dies,” former President Trump declared at a rally in Arizona on Jan. 15. “If you’re white you have to go to the back of the line to get medical help!”

Various conservative political commentators amplified the message that “race-based medicine” — as Trump termed it — was being used to prioritize scarce Covid-19 therapies. Utah and Minnesota subsequently removed race from their Covid-19 algorithms, reportedly fearing legal challenges.

This outrage at so-called race-based medicine from the right has obvious parallels with outrage on the left about the use of race in risk prediction for clinical decision-making. This includes recent efforts to remove race from the eGFR equation used to estimate kidney function.

Yet the Covid-19 example demonstrates how including race in risk prediction models used to allocate health care resources can — perhaps counterintuitively — support treating patients with similar health needs equally. We think this principle is important to understand, alongside other considerations and concerns.

While the question of whether to include race as a factor in predictive algorithms is contentious, there is broad agreement that individuals with similar risks for an outcome, such as needing to be hospitalized for severe Covid-19, should be treated similarly regardless of race. We call this egalitarian principle “equal treatment for equal risk.”

When race has no prognostic information independent of relevant clinical characteristics, there is no controversy, since only characteristics that contribute to prognosis are included in risk models and race does not enter the picture. Controversy arises only when race is predictive of differences in risk for an outcome, despite clinical characteristics that appear to be similar — which is often the case when there are health disparities.

Health disparities, unfortunately, are not rare and are often not subtle. Among people infected with SARS-CoV-2, the virus that causes Covid-19, some studies indicate that Black people have about twice the risk of hospitalization as white people, even after accounting for age, other medical conditions, socioeconomic factors, and insurance type.

Understanding hospitalization risk with Covid-19 is important when making decisions about who to prioritize for treatments such as Paxlovid and monoclonal antibodies, which are specifically used to keep infected people from needing to be hospitalized, since there are insufficient quantities to treat everyone and — this is important — not everyone is equally likely to benefit from these treatments. Including race in algorithms makes it possible to include the excess risk in Black people not accounted for by the other characteristics. Thus, this supports both the egalitarian principle of equal treatment for equal risk, and also supports the utilitarian principle of achieving the greatest good for the greatest number of people — regardless of race.

What is known about the higher risk for Covid-related hospitalization that Black people experience that is unaccounted for by other observed variables? It is real and measurable. It is large. It has causes (even if these causes are difficult to measure or not always precisely known), and those causes are disproportionately experienced by Black people. Leaving race out of risk calculations does not treat Black and white people equally — it systematically ignores those (unknown or unmeasured) causes of greater Covid risk that are more common in Black people than white people. The risk associated with these unmeasured causes is no less important than the risk associated with known and measured characteristics.

Acknowledging that race is predictive of disparate risk does not in the least imply that race is a direct causal or biological factor. As a social construct, race can cause health outcomes only indirectly — through racism. It also can act as a proxy for causes that might be unknown and unmeasured, including socioeconomic factors, cultural factors, and genetic factors. Regardless of the causes, the use of race within prognostic models to inform allocation decisions is based on measurable risk — not solely on politically more contested grounds of restorative justice.

To be sure, there are good reasons to leave race out of clinical algorithms. Some have contended that using race in prediction models for medical decisions implies, falsely, that race is a biological characteristic, that it gives credence to the notion that differences are biologically based as opposed to social. It has also been argued that social determinants of health should be used instead. In addition, the use of race in medicine has proven singularly inflammatory across the political spectrum in a way that may exacerbate already frayed social divisions or undermine trust in the health care system.

We believe that including race in risk models should be done as a last resort, when there remain large risk differences across racial groups despite accounting for all observable clinical characteristics — particularly when its inclusion may reduce disparities. Disparaging all race-aware risk prediction models is misleading when including race can sometimes improve both health outcomes and fairness under principles that most Americans would find eminently reasonable.

Fairness should be a central goal of every decent society, although different notions of fairness may conflict. We suggest that equal treatment for equal risk is an important fairness criterion, though not the only one. In the presence of disparities in health care outcomes, using race-unaware algorithms to prioritize resources might paradoxically be discriminatory as doing so systematically ignores important causes of risk that disproportionately affect one group — usually the group that is already disadvantaged.

David M. Kent is a general internist, director of the Predictive Analytics and Comparative Effectiveness Center at Tufts Medical Center, and professor of medicine at Tufts University School of Medicine. Keren Ladin is director of the Research on Ethics, Aging, and Community Health Lab and associate professor of occupational therapy and community health at Tufts University. O. Kenrik Duru is a general internist and professor of medicine at the David Geffen School of Medicine at the University of California, Los Angeles.