A study by Kaiser Permanente researchers has shown that integrating mammographic AI with polygenic risk scores and clinical risk models can improve breast cancer risk stratification, guiding both personalized breast cancer screening and chemoprevention.
The study, published in the Journal of the National Cancer Institute, is one of the largest and most diverse to evaluate the ability of the three approaches to predict breast cancer risk.
“Our goal is to improve our ability to assess a woman’s breast cancer risk so we can personalize breast cancer screening recommendations,” said lead author Vignesh Arasu, MD, PhD, a radiologist and research scientist at the Kaiser Permanente Division of Research. “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.”
The study included 82,957 women (75% non-Hispanic White, 9% Asian, 7% Latina, and 4% Black) who enrolled in the Kaiser Permanente Research Bank between 2003 to 2020. All the women had a recent negative mammogram, and none had previously been diagnosed with breast cancer or had a genetic mutation known to increase breast cancer risk.
The researchers calculated each woman’s breast cancer risk using three different approaches: The Mirai mammography AI risk score, which looks for risk-related imaging biomarkers; the Breast Cancer Surveillance Consortium version 3 clinical risk score, which considers factors such as age, race or ethnicity, family history of breast cancer, breast density, and body mass index; and the 313-SNP polygenic risk score (PRS) that assesses risk based on the presence or absence of 313 breast cancer-associated single nucleotide polymorphisms.
During 10 years of follow-up, 2471 (3%) women were diagnosed with invasive breast cancer or ductal carcinoma in situ.
Arasu and team report that the C-index for breast cancer prediction when all three risk scores were combined was 0.70.
“This means that if you take any women who actually will get breast cancer in the future and pass her information to a risk model, that model will say she has a higher risk about 70% of the time relative to women who won’t get cancer,” Arasu explained.
The C-index for the combined model was significantly higher than that for individual models with only the clinical risk score (0.62), which is used by most clinical practices, the PRS (0.61), or the Mirai score (0.66).
The C-index for the triple model was also significantly higher than that for a model that combined the clinical and the polygenic risk scores (0.66).
The increase in the C-index of 0.04 “represents a moderate but meaningful improvement in discrimination when incorporating all three risk domains compared with the clinical plus polygenic model alone,” Arasu told Inside Precision Medicine.
He added: “The C-index reflects overall model performance across all possible risk thresholds and is commonly used as an initial, global assessment when integrating new predictors, such as AI-derived measures, to gauge their added value. However, future work will focus on defining clinically actionable risk thresholds to better characterize how these improvements translate into real-world tradeoffs between benefit and harm.”
Importantly, the improvements in risk prediction were consistent over time and across four common self-reported racial/ethnic subgroups of Asian, Black, Latina, and White women, which the authors say indicates that incorporating mammographic AI and ancestry-adjusted PRS into clinical risk prediction models may benefit all women.
The study also found that among the women at highest risk for developing breast cancer, the clinical risk score alone identified 19% of the women who went on to develop breast cancer over a 10-year period while the combined model identified 26% of these women.
Although all three model types are becoming increasingly accessible and part of care, Arasu cautioned that “more research is needed to see the benefit of using all three scores before a program that uses all three should be implemented.”
He added: “We already are planning the next research step, which will be to study the combined model in the on-going national WISDOM clinical trial in the next 1-2 years. The WISDOM trial already uses a PRS and clinical risk score to assess risk. Now, we will be adding mammographic AI.”
“By identifying more accurately which women are truly at high-risk, we hope to find more breast cancers as early as possible, when they are most easily treatable,” said Arasu. “For women who are very high risk, there is also the potential to discuss, in addition to annual mammography, risk reduction with a medication, such as tamoxifen.”
The post Mammographic AI Adds Predictive Power to Breast Cancer Risk Models appeared first on Inside Precision Medicine.