A recent study found that elimination of race and ethnicity as risk predictors for colorectal recurrence is associated with worsened fairness of the risk prediction model, resulting in inappropriate care for ethnic and racial minorities in the population.
The inclusion of racial and ethnic parameters in the risk prediction models is often inadequate, reflecting the inequalities in the provision of healthcare services. There is a substantial lack of studies in the literature that have explored the consequences of the omission of race and ethnicity from risk prediction models on treatment-related decision-making.
This retrospective study investigated the effect of the inclusion of race and ethnicity in predicting colorectal cancer risk on racial bias. The findings are published in JAMA Network Open.
A total of 4230 patients were enrolled in the study. The mean age of the participants was 65.3 ± 12.5 years. The cohort comprised Hispanic, non-Hispanic White, Black or African American, and Asian, Hawaiian, or Pacific Islander. Of the total participants, 42%, 31%, and 27.1% were diagnosed with stage I, stage II, and stage III colorectal cancer, respectively.
Colorectal Cancer Recurrence in Racial and Ethnic Groups
The study participants’ unadjusted 3-year recurrence rate of colorectal cancer was 11.2%. The 3-year incidence of colorectal cancer recurrence among non-Hispanic White, Black or African American, Hispanic, and Asian, Hawaiian, or Pacific Islancer patients was 9.5%, 13.8%, 13.0%, and 12.5%, respectively.
Calibration of Risk Prediction Algorithm for Colorectal Cancer Recurrence
The calibration of the race-neutral model was excellent, whereas the calibration of the race-sensitive model was good overall. The race-neutral model may overpredict the colorectal recurrence risk for low-risk individuals and underpredict the recurrence risk for high-risk individuals in Black or African-American patients.
Discriminative Ability of Risk Prediction Algorithm for Colorectal Cancer Recurrence
The discriminative ability of the race-neutral model was lowest in Asian, Hawaiian, or Pacific Islander individuals and highest in Black or African American individuals. The discriminative ability of the race-sensitive model was comparatively less variable across the racial groups, and the model has the lowest overall racial bias.
False-Positive and False-Negative Rates of Colorectal Cancer Recurrence
The false-positive rates (FPRs) and false-negative rates (FNRs) of colorectal cancer recurrence in the race-neutral model were 68.2% and 7.2%, respectively. The race-stratified group had the largest difference in the FPRs across the differential racial and ethnic groups.
Negative and Positive Predictive Values of Colorectal Cancer Recurrence
The negative predictive values (NPVs) and positive predictive values (PPVs) in the race-neutral model were variable across the racial and ethnic groups. The race-sensitive model had a lower PPV and improved NPV for individuals from the Asian, Hawaiian, or Pacific Islander group. Moreover, the standardized net benefits of the race-sensitive model were higher for Asian, Hawaiian, and Pacific Islander individuals, while the race-neutral model showed higher net benefits for Black or African American and non-Hispanic White individuals.
Khor, S., Haupt, E. C., Hahn, E. E., Lyons, L. J., Shankaran, V., & Bansal, A. (2023). Racial and ethnic bias in risk prediction models for colorectal cancer recurrence when race and ethnicity are omitted as predictors. JAMA Network Open, 6(6), e2318495. https://doi.org/10.1001/jamanetworkopen.2023.18495