The inclusion of race and ethnicity in clinical risk prediction algorithms has been subject to criticism, with studies showing negative effects in diabetes, kidney disease and high-risk care management, for example. But removing these factors entirely may backfire, results from a new study of cancer care suggest.
Investigators developed and validated four prediction models of the risk of postoperative cancer recurrence for 4,230 patients with colorectal cancer. The data came from a large integrated healthcare system in Southern California. Treatment was received between 2008 and 2013 with follow-up through 2018.
Race-neutral model
When the models explicitly considered race and ethnicity as a predictor, results showed increased fairness across multiple performance measures for patients from racial and ethnic minority groups. The race-neutral model, however, had worse calibration and false-negative rates among racial and ethnic minority subgroups when compared to non-Hispanic white patients.
The results have important fairness implications for the clinical community who are “interested in algorithms that are both accurate and fair,” Sara Khor, MASc, of the University of Washington, and colleagues.
Inappropriate recommendations
“Simply omitting race and ethnicity may result in worse prediction accuracy in subgroups that may lead to inappropriate care recommendations that ultimately perpetuate or contribute to disparities in health outcomes,” they wrote. “Our results highlight the need to use clinically relevant fairness criteria to evaluate existing algorithms and understand the potential implications of removing the race and ethnicity variable on racial disparities in health outcomes.”
The study was published in JAMA Network Open.
Related articles:
Prediction tools miscalculate type 2 diabetes risk in racial groups, study finds
Adopt new test, and take race out of kidney disease diagnostics, clinicians urged
Switch to race-neutral equation would affect millions of U.S. kidney patients: study
New prognostic tool supports end-of-life discussions for dementia patients, families: study
