Under resource constraints, race-aware models demonstrated greater potential by better prioritizing high-risk individuals.

Image Credit: peshkova-stock.adobe.com

New research provides a framework for assessing when and how race should be included in clinical risk models while factoring in accuracy, equity, and utility.

In the study published in Annals of Internal Medicine, researchers explored the implications of including race and ethnicity in clinical risk prediction models. Using data from the National Health and Nutrition Examination Survey (NHANES) and the National Lung Screening Trial (NLST), the study examined models predicting risks for cardiovascular disease, breast cancer, and lung cancer.

Race-aware models were compared, incorporating race and ethnicity as variables, with race-unaware models generated through statistical marginalization. While race-aware models showed superior statistical calibration across racial and ethnic groups, their benefits in clinical decision-making were less pronounced than expected.

For 95% or more of patients, predictions from both model types led to the same clinical decisions. For patients whose care decisions differed, the benefits of screening or treatment were modest, as these patients often had risks near decision thresholds.

Significant miscalibration was found in race-unaware risk prediction models, underestimating cardiovascular and lung cancer risks for Black individuals while overestimating breast and lung cancer risks for Asian individuals and lung cancer risks for Hispanic individuals. For White individuals, predictions were largely consistent between race-aware and race-unaware models.

Despite these inaccuracies, the clinical utility gains of race-aware models were modest, improving net benefit by up to 2.0 per 10,000 individuals for cardiovascular disease and less for other diseases. These improvements were most pronounced in populations where race-unaware miscalibration was worst, such as Black individuals for cardiovascular disease, Asian individuals for breast cancer, and Hispanic individuals for lung cancer.

Under resource constraints, race-aware models demonstrated greater potential by better prioritizing high-risk individuals, particularly among underrepresented groups. This emphasizes their role in equitable health care resource allocation, though their broader impact remains dependent on clinical contexts and decision thresholds, the researchers noted.

While race-aware models improve statistical predictions, their impact on clinical outcomes is influenced by how decisions are made. The findings suggest that race-aware models could be valuable in addressing disparities under specific circumstances, such as rationing, but their use should be carefully contextualized to avoid reinforcing biases.

The study acknowledged several limitations, including assumptions about the accuracy of existing race-aware models and the simplified methodology for creating race-unaware predictions. Future research should investigate how these models perform in real-world settings and their potential to mitigate health disparities.

“Decision making surrounding the use of race in the use of prediction models is not always straight forward,” the researchers wrote. “There is unlikely to be any single path to including or excluding race that is appropriate for all models and conditions. Therefore, among those responsible for the use of prediction models, an explicit discussion of the role of race within each model is warranted.”

Reference

Coots M, Saghafian S, Kent DM, Goel S. A framework for considering the value of race and ethnicity in estimating disease risk. Ann Intern Med. Published online December 3, 2024. doi:10.7326/M23-3166