Breast cancer risk models can misclassify many women

Breast cancer risk estimates for individual women vary significantly depending on which risk assessment model is used, and women are likely to receive very different recommendations depending on the model used and the cut-off used to define “high risk,” according to a new study from UCLA . The study is published online in Journal of General Internal Medicine.

Current incidence rates indicate that approximately one in eight women born in the United States today will develop breast cancer at some point in their lives. The risk increases with age.

As precision medicine advances in healthcare, breast cancer risk models are increasingly being used to identify women who would benefit from breast cancer risk reduction drugs and complementary MRI screenings. Easy-to-use risk models are available online, and women often receive a risk estimate in their mammography screening reports. An important question is: How accurate are these models?

In 2019, the U.S. Preventive Services Task Force recommended that physicians offer risk-reducing drugs, such as tamoxifen, raloxifene, or aromatase inhibitors, to women who have a high risk of breast cancer and a low risk of adverse drug effects over the next 5 years.

While a 5-year risk limit had previously been set at 1.67%, the task force recommended a new, higher 5-year risk limit of 3%. And while the current tools for assessing breast cancer risk at the population level work well, little attention has been paid to their performance at the individual level or to the variation in risk estimates for the ≥ 3.0% 5-year threshold at the individual level.

The current study involved more than 31,115 women who were part of the Athena Breast Health Network, a nationwide quality improvement initiative at University of California medical and cancer centers. It focused on three commonly used risk assessment models: the Breast Cancer Risk Assessment Tool (BCRAT, also called the Gail model), the Breast Cancer Surveillance Consortium (BCSC), and the International Breast Intervention Study (IBIS, also called the Tyrer-Cuzick). Model).

The researchers found that using a threshold of ≥ 1.67%, more than 21% of women were classified as high risk of developing breast cancer in the next 5 years by one model, but as intermediate risk by another model.

Using a threshold of ≥ 3.0%, more than 5% of women had disagreement about risk severity between models. When all three models were used, almost half of the women (46.6%) were classified as high risk by at least one model. Because most women are not diagnosed with breast cancer within 5 years, the authors would say many women would be misclassified as high risk.

“This study highlights the risk of a blanket approach to using risk prediction models to inform medical screening and treatment decisions at the individual level,” said Dr. Joann Elmore, the paper’s senior author and professor of medicine in the Division of General Internal Medicine and Health Care Research at the David Geffen School of Medicine at UCLA. “All three models we examined had similar accuracy at the population level, but there was significant disagreement in our analyzes as to which of all three models was identified as ‘high risk’.”

The authors say their results highlight the trade-off between sensitivity and imprecise classification as “high risk” when using the two different thresholds currently recommended. For example, using the ≥ 1.67% cut-off for chemoprevention consideration, about half of the women who will be diagnosed with breast cancer in the future could be correctly identified as high-risk, but many more women would be incorrectly identified as high-risk. women classified. While using the more conservative cut-off of ≥ 3.0% would result in far fewer women being misclassified as high risk, most women with a future breast cancer diagnosis would be missed.

The study has some limitations. For example, the cohort was drawn from women who had participated in a longitudinal screening study. And although the authors had extensive risk factor data on many participants, some family histories and data on polygenetic risk scores were missing.

The authors note that newer risk models are being developed that include information about breast cancer susceptibility genes and genetic susceptibility variants, which could improve predictability. Meanwhile, several recent studies suggest that quantitative imaging biomarkers and artificial intelligence algorithms could complement or replace current, subjective clinical risk assessment tools.

Additional authors included Jeremy S. Paige MD, Ph.D., Christoph I. Lee MD, MS, MBA, Pin-Chieh Wang Ph.D., William Hsu Ph.D., Adam R. Brentnall Ph.D., Anne C. Hoyt MD and Arash Naeim MD, Ph.D.