Using race and ethnicity to estimate disease risk improves prediction accuracy but may yield limited clinical net benefit

Using race and ethnicity to estimate disease risk improves prediction accuracy but may yield limited clinical net benefit

Abstract: https://www.acpjournals.org/doi/10.7326/M23-3166 

URL goes live when the embargo lifts          

A cross-sectional study analyzed survey data of U.S. adults to compare the benefits of race-aware versus race-unaware predictions for disease risk. The analysis found that the clinical net benefit of race-aware models over race-unaware models was smaller than expected. The researchers provide a widely adaptable framework for deciding whether to include or exclude race from disease risk predictions. The findings are published in Annals of Internal Medicine.

Researchers from Harvard University studied data from National Health and Nutrition Examination Survey and National Lung Screening Trial between 2011 to 2018 and 2002 to 2004, respectively. The researchers aimed to present a decision-analytic framework for considering both the statistical and clinical utility of race and ethnicity in disease risk estimation, using cardiovascular disease, breast cancer, and lung cancer as case studies. For each disease, the researchers obtained risk estimates for a sample of individuals using a clinically recommended race-aware risk model. They converted these estimates to race-unaware estimates using statistical marginalization of race and ethnicity. The researchers then compared how clinical decisions would change using race-aware versus race-unaware risk estimates, and quantified how these changed decisions translate into utility gains or losses for different racial and ethnic groups under a shared decision-making context and a rationing context. 

The researchers found that, assuming the race-aware models yield accurate estimates, the race-unaware models underestimate risk of cardiovascular disease and lung cancer for Black individuals, but overestimate risk of breast cancer in Asian individuals and lung cancer in both Asian and Hispanic individuals. While including race in disease risk models shifts predictions for many patients, the researchers found that most individuals end up receiving the same clinical recommendation under a race-aware model as they would under a race-unaware model. The researchers also found that the overall clinical benefits of race-aware risk predictions were not as large as expected given the miscalibration of the race-unaware predictions. However, when used to inform rationing, race-aware models may have a more substantial net benefit. Overall, the results suggest that race-aware risk models yield smaller gains in net benefit over race-unaware models than the improvement in predictions might suggest. 

Media contacts: For an embargoed PDF, please contact Angela Collom at [email protected]. To speak with corresponding author Madison Coots, MS, please email her directly at [email protected].