Clinical concordance between the Gail and Tyrer-Cuzick breast cancer risk models in a Turkish population with high mammographic density prevalence: A cross-sectional study.

The Gail and Tyrer-Cuzick models draw on different input variables and can assign the same woman to different risk categories, yet whether this discordance alters clinical risk stratification in populations with high mammographic density prevalence has not been formally assessed. This study aimed to compare clinical concordance and density-driven risk reclassification between the 2 models in a Turkish screening cohort. In this prospective cross-sectional study, lifetime breast cancer risk was calculated using both models in 1365 women aged 35 to 75 undergoing routine screening at Ordu University Training and Research Hospital from June 2025 to January 2026. Mammographic density was assessed by independent radiologists blinded to clinical data. Concordance at the ≥20% lifetime risk threshold was evaluated using Cohen kappa and the prevalence-adjusted bias-adjusted kappa. The mean age was 50.6 ± 7.8 years; 77.2% had dense breasts (Breast Imaging Reporting and Data System Type C/D). The Gail model classified 40 women (2.9%) as high risk (≥20% lifetime risk); the Tyrer-Cuzick model classified 232 (17.0%; P < .001). Cohen κ was 0.22 (95% confidence interval: 0.15-0.29), deflated by marginal asymmetry; the prevalence-adjusted bias-adjusted κ was 0.70, corresponding to substantial agreement (Prevalence Index, 0.80; Bias Index, 0.14; observed agreement, 85.2%). A total of 182 women (13.3%; 95% confidence interval: 11.6-15.3%) classified as low or average risk by Gail were reclassified as high risk by the Tyrer-Cuzick model. The reclassification rates were 14.5% for dense breasts (Type C/D) and 24.1% for extremely dense breasts (Type D). The sensitivity analyses confirmed that mammographic density was the primary driver of inter-model discordance (κ = 0.19; upstaging rate = 10.8%). In this Turkish screening cohort, the Gail and Tyrer-Cuzick models showed only fair concordance at the ≥20% threshold for supplemental magnetic resonance imaging, with mammographic density as the principal driver of disagreement. Outcome-linked validation in populations with high breast density prevalence is needed before either model can be recommended as the reference.