Study evaluates statistical models for estimating breast cancer risk

Researchers from Trinity College Dublin, St James's Hospital, and collaborating institutions have carried out the most comprehensive review to date of tools used to estimate breast cancer risk in women with a family history of the disease.

The findings, presented at the American Society of Clinical Oncology (ASCO) Annual Meeting, suggest that while some commonly used risk models perform reasonably well, none are highly accurate at identifying which women will go on to develop breast cancer.

Women with a family history of breast cancer are often offered an assessment of their own risk of developing the disease. As part of this assessment, doctors use statistical models that predict a woman's likelihood of developing breast cancer in the future. This is important as it helps guide decisions about screening and prevention. These decisions may include when to begin mammograms or MRI scans, whether to offer medicines that reduce risk, or, in some cases, whether preventative surgery to remove the breasts, should be considered.

Until now, there has been little clarity on which models work best for this group.

The Cochrane review analyzed 45 studies examining breast cancer risk models in women with a family history of breast cancer. Researchers assessed how accurately the models predicted future breast cancer risk. Four models had been studied often enough to allow detailed analysis: Gail (BCRAT), Tyrer-Cuzick (IBIS), BOADICEA, and BRCAPRO.

Some models outperform others but room for improvement remains

When looking at future breast cancer risk, the review found that the BOADICEA model showed the most balanced overall performance in women with a family history of breast cancer. The Gail and BOADICEA models produced risk estimates that were generally close to the number of breast cancers that actually occurred in studies. However, the Tyrer-Cuzick model tended to overestimate risk, while BRCAPRO tended to underestimate risk.

When researchers examined how well the models could distinguish between women who did and did not develop breast cancer – known technically as discriminatory accuracy – all models showed only modest performance, but none came close to the accuracy needed to fully personalize care.

Lead author Dr Sarah McGarrigle explained that the findings highlight both the value and the limitations of these models in risk prediction.

These tools are already widely used in clinical practice, and we now have a clearer picture of their accuracy in women with a family history of breast cancer. Our findings suggest that these tools have value in supporting risk assessment and that is encouraging, but we still have a long way to go."

Dr. Sarah McGarrigle, Lead Author

According to the authors, the findings support continued efforts to improve personalized breast cancer risk assessment, particularly for women with a strong family history of the disease.

"For women who are at elevated risk based on their family history of breast cancer, understanding their personal risk can shape some of the most important decisions of their lives, such as whether to screen more frequently and whether to consider preventive treatment," explains Professor Elizabeth Connolly, senior author of the study. "It matters enormously that the tools we use to guide those conversations are as accurate as possible. We're not there yet, but we're making progress, and this review helps point the way forward."

Many of the studies evaluating these models were of poor or unclear quality, limiting confidence in some of the results.

"Better-quality studies and continued improvement of these models are needed so women and clinicians can make decisions based on the most accurate information possible," McGarrigle added.