An AI-based machine learning model predicted advanced-stage disease (AUC 0.81 vs 0.67) and 5-year survival (AUC 0.84 vs 0.71) better than traditional multivariable logistic regression.
Cohort (n=97,462)
Yes
Does a machine learning model improve prediction of advanced-stage disease and 5-year survival compared to multivariable logistic regression in women <40 years with invasive breast cancer?
An AI-based model outperformed traditional logistic regression in predicting advanced-stage disease and 5-year survival in young women with breast cancer, identifying high-risk patients missed by conventional criteria.
Absolute Event Rate: 0.81% vs 0.67%
1029 Background: Young-onset breast cancer (YOBC), defined as breast cancer diagnosed in women <40 years, is associated with aggressive tumor biology and affects patients differently depending on race, ethnicity, and socioeconomic status (SES). Traditional risk models may insufficiently predict disease severity and survival in this population. Artificial intelligence (AI) can integrate multidimensional clinical and demographic data to improve outcome prediction and identify high-risk subgroups. We evaluated AI-based prediction of advanced-stage disease and 5-year survival in a large U.S. cohort over two decades. Methods: We conducted a retrospective study using the SEER database, identifying women aged <40 years diagnosed with invasive breast cancer from 2000–2021. Variables included age, race, ethnicity, marital status, tumor stage, grade, receptor status, treatment, and county-level SES. A machine learning model was trained to predict advanced-stage disease (stage III–IV) and 5-year overall survival. Model performance was compared with multivariable logistic regression using area under the receiver operating characteristic curve (AUC). We also explored differences by race, ethnicity, and SES to understand disparities. Results: Our cohort included 97,462 women--38.9 percent non-Hispanic White, 26.7 percent non-Hispanic Black, 22.3 percent Hispanic, 11.1 percent Asian/Pacific Islander, and 1 percent American Indian/Alaska Native. Overall, 33 percent presented with advanced-stage disease, with higher rates among non-Hispanic Black (42 percent) and Hispanic (37 percent) patients compared with non-Hispanic White patients (26 percent, p<0.001). Women living in the lowest SES areas were 30 percent more likely to have advanced-stage disease than those in the highest SES areas. The AI model predicted advanced-stage disease (AUC 0.81) and 5-year survival (AUC 0.84) better than traditional regression (AUC 0.67 and 0.71, respectively). Importantly, 23 percent of high-risk patients identified by AI would have been missed by conventional criteria. Survival disparities persisted: 5-year survival was 77 percent for non-Hispanic Black women versus 89 percent for non-Hispanic White women, and 75 percent for women in the lowest SES areas versus 91 percent in the highest. The AI model’s predictions were consistent across all subgroups. Conclusions: As per the study AI-based models improved prediction of advanced-stage disease and survival while highlighting persistent racial, ethnic and socioeconomic disparities. These findings suggest AI-driven risk assessment can help identify high-risk young women and support equity-focused interventions to improve outcomes in YOBC.
Ethakota et al. (Wed,) conducted a cohort in Young-onset breast cancer (n=97,462). Artificial intelligence (AI) machine learning model vs. Multivariable logistic regression was evaluated on Prediction of advanced-stage disease (stage III–IV) and 5-year overall survival (AUC). An AI-based machine learning model predicted advanced-stage disease (AUC 0.81 vs 0.67) and 5-year survival (AUC 0.84 vs 0.71) better than traditional multivariable logistic regression.