モデル性能とカスタム閾値を比較して大腸がんリスク評価を調整する：主にヒスパニック系コホートにおける研究

Colorectal cancer (CRC) remains a significant public health concern, particularly among underserved populations. This study evaluated CRC risk factors and compared risk assessment models in a predominantly Hispanic cohort residing along the U.S.-Mexico border. We analyzed data from 4,202 CRC screening participants, mostly women (78.3%), with a mean age of 57.3 years (SD = 5.54). Most were born in Mexico (86.2%) and had resided in the U.S. for an average of 25 years (SD = 16.5). Risk was assessed using four models: Freedman, Wells, Kaminski, and Yeoh. We examined model concordance, risk distributions, and gender-specific trends. Freedman's model showed that 41.1% of participants had a lifetime CRC risk above the national average. Men had a higher lifetime risk (51.3%) than women (38.3%), while more women had elevated 10-year risk (26.3% vs. 21.1%). Freedman and Wells models demonstrated a strong concordance correlation coefficient (CCC = 0.76), with an optimal Freedman cutoff of 0.97 (AUC = 87%). Kaminski and Yeoh identified more participants at above-average risk (27.8% and 35%) compared to Freedman (20.6%) and Wells (14.9%). Key contributors to elevated risk included smoking pack-years, obesity, and family history, with gender-specific variations. Freedman's model, offering both lifetime and 10-year risk estimates, emerged as the most practical tool in this cohort. Custom thresholds (0.89 for women, 1.41 for men) improved risk stratification, aligning with validated polyp-detection rates. These findings support the use of tailored CRC risk models for Hispanic populations to enhance targeted screening and prevention efforts.