Vignesh Arasu still remembers the moment a 42-year-old patient with no family history of breast cancer was diagnosed with stage II disease—despite having a recent mammogram. It was a wake-up call: even modern screening tools were missing critical clues. Now, as lead author of a landmark Kaiser Permanente study, Arasu is helping change that. His team’s research shows that combining artificial intelligence, genetic data, and traditional clinical factors can significantly improve how doctors predict a woman’s risk of developing breast cancer—offering a more precise, personalized path to prevention.
Breast cancer affects one in eight women in the U.S., and early detection remains one of the most powerful tools in reducing mortality. Yet current risk models, often based on age, family history, and breast density, only tell part of the story. The new study, published in the Journal of the National Cancer Institute, analyzed data from 82,957 women enrolled in the Kaiser Permanente Research Bank between 2003 and 2020—all of whom had mammograms showing no signs of cancer and no known high-risk genetic mutations. Over the next decade, 2,471 (3%) were diagnosed with invasive breast cancer or ductal carcinoma in situ (DCIS). The goal? To see which risk assessment tools could best predict those outcomes.
The results were clear: integration wins. The model combining mammography AI, polygenic risk scores, and clinical data achieved a C-index of 0.70—significantly outperforming clinical risk scores alone (0.62) or polygenic scores alone (0.61). Even combining just clinical and genetic scores reached only 0.66. The AI component analyzed imaging biomarkers on mammograms, while the polygenic score evaluated 313 single nucleotide polymorphisms linked to breast cancer. The clinical model factored in age, race or ethnicity, family history, breast density, and BMI.
Most strikingly, among women at highest risk, the combined model identified 26% of those who would go on to develop cancer within 10 years—compared to just 19% flagged by clinical scores alone. “Our study shows that each of the approaches identifies a distinct group of women, and that when all three risk tests are used we increase our ability to differentiate high-risk and low-risk women and provide more personalized screening recommendations,” said Arasu.
For Stacey E. Alexeeff, Ph.D., co-author and biostatistician at Kaiser Permanente’s Division of Research, the consistency of the results across timeframes was key: “From a modeling perspective, the key result is the consistent improvement in prediction accuracy when these risk scores are combined.” As AI and genomics continue to evolve, the potential for even greater precision looms. For millions of women, the future of breast cancer prevention may not lie in a single test—but in the intelligent fusion of many.