Race Corrections in Clinical Algorithms Can Help Correct for Racial Disparities in Data Quality

Despite ethical and historical arguments for removing race corrections from clinical algorithms, the consequences of removal remain unclear. An important and underdiscussed consideration in this debate is the fact that medical data quality frequently varies across race groups. For example, family history of cancer is an essential predictor in cancer risk prediction algorithms but is less reliably documented for Black patients and may therefore be less predictive of cancer outcomes. We assessed whether race corrections could allow risk prediction models to capture varying data quality by race, focusing on colorectal cancer risk prediction. Using data from the Southern Community Cohort Study, we analyzed 77,836 adults with no history of colorectal cancer at baseline. We assessed whether the predictive relationship between self-reported family history of colorectal cancer and 10-year colorectal cancer risk differed by race. We then compared two cancer risk prediction algorithms -- a race-blind algorithm which included standard colorectal cancer risk factors but not race, and a race-corrected algorithm which additionally included race. Family history predicted 10-year colorectal cancer risk among White patients (OR: 1.74, 95% CI 1.25-2.38), but not Black patients (OR: 0.98, 95% CI 0.72-1.29). Relative to the race-blind algorithm, the race-corrected algorithm improved predictive performance, as measured by goodness of fit in a likelihood ratio test (p-value <0.001) and AUROC among Black patients (0.611 versus 0.608, p-value 0.006). Because the race-blind algorithm underpredicted risk for Black participants, the race-corrected algorithm increased the fraction of Black participants among the predicted high-risk group, potentially increasing access to screening. Race corrections can allow risk prediction algorithms to model varying data quality by race group, which frequently occurs in clinical settings.