Calibration of the US breast cancer risk assessment tool in 114,533 Mexican women: Evidence from the Mexican Teachers' Cohort

Background: Breast cancer risk prediction models are increasingly used to identify women for risk-based screening and chemoprevention. The US Breast Cancer Risk Assessment Tool (NCI-BCRAT/Gail) is a widely used model, yet its performance in Mexican women is unknown. The Mexican Teachers' Cohort (MTC), a prospective cancer cohort in Mexico, is a unique resource that can be leveraged to assess the usefulness of breast cancer risk models in this understudied population. Thus, we evaluated calibration and discrimination of this model in a large, diverse cohort of Mexican women. Methods: Absolute invasive breast cancer risk was calculated from enrollment to December 31, 2019, in 114,533 cancer-free women aged 25-80 years. Missing model predictors were assigned to the lowest-risk category. We compared expected and observed cases overall and by 5-year age groups and Indigenous ethnicity using expected-to-observed (E/O) ratios with 95% confidence intervals (95% CI). Discrimination was assessed with the area under the ROC curve (AUC). We quantified the proportion of women with a predicted risk above the 5-year high-risk threshold among those who developed and those who did not develop breast cancer. Results: Over a mean follow-up period of 11.2 years, we identified 1,490 women with invasive breast cancer among cohort participants. The NCI-BCRAT/Gail model predicted that 2,115 women would develop breast cancer, leading to an expected-to-observed (E/O) ratio of 1.40 (95% CI 1.35-1.49). Overestimation was most pronounced in the 50-54 age group (E/O = 1.62; 95% CI 1.41-1.87). Among Indigenous women, 79 developed breast cancer compared to 135 women who were predicted to develop the disease (E/O = 1.71;95% CI 1.35-1.49), and for age 50-54 E/O ratio was 2.71 (95% CI 1.29-5.86). The model's discriminatory accuracy was 63% (95% CI, 62%-65%). Yet, 82% of women who developed breast cancer did not reach the 5-year high-risk threshold. Also, 7% of non-cases had a predicted risk above the 5-year high-risk threshold. Further, results using different breast cancer incidence estimates for Mexico and 5-year high-risk thresholds will also be presented. Conclusions. The NCI-BCRAT/Gail model overestimated invasive breast cancer risk in Mexican women from the MTC. Overestimation was particularly salient in older women and Indigenous women. Using this model “as is”, most women who developed breast cancer would have been classified as average risk. These findings suggest that the NCI-BCRAT/Gail should be recalibrated and validated before clinical use in this population.