Many Breast Cancer Risk Models: Which One Is Accurate?

“All models are wrong, but some are useful” - George Edward Pelham Box 1919 –2013
British mathematician/Professor of Statistics at the University of Wisconsin

We are moving into the era of risk-based screening and prevention for breast cancer and increased patient expectations that we can find out that risk. To do so, we need to have accepted methods for estimating the risk of developing breast cancer over time and the risk of having a mutation in a cancer-causing gene. Breast cancer risk models, which have been developed over the last two to three decades, are the best tools for this work, but they aren’t perfect.

Each type of risk helps decide next steps for the patient:

  • We use risk of breast cancer over time to decide the need for chemoprevention, MRI, earlier (or later) mammography and risk-based screening.
  • We use the risk of mutation to decide the need for genetic testing.

The major risk models used in the US are BRCAPRO, Tyrer Cuzick, Gail, Claus, and Myriad. Each model uses different groups of presumed breast cancer risk factors to make an estimate. The models assign a weight to each factor and then use an algorithm to combine these weighted risks to all the factors studied. The risk factors fall in three basic categories: hereditary, hormonal, and pathologic. The graphic below shows which risk factors fall into which categories:


So far, this seems fairly straightforward. But now, take a look at how each model uses a different combination of factors:


Add to this how each model can be used to decide different management plans:


You can see how models can give very different results for the same patient. Below is a graph, by model, of breast cancer risk over time. (Screenshot from CRA Health Risk Clinic Module).


How can they be so different? The models take in and leave out different factors.

Gail misses the hereditary piece because it only takes into account breast cancer in first-degree relatives. Claus also misses the hereditary piece because as it only looks at breast cancer and only once per relative. It misses the ovarian cancer history and the bilateral breast cancer. BRCAPRO hones in solely on the hereditary piece and is likely the most accurate in this category. Tyrer Cuzick (TC) takes into account hereditary, hormonal, and pathologic factors. Adding these together makes the risk estimates quite high. In addition, the latest version, TC 7, gives a much higher lifetime risk because it bases it on age 85 as a lifetime, while TC 6 uses age 80.

With so much complexity, it’s easy to get frustrated and lose faith in all models. However, as George Box noted above, the complexity of the models can be useful in the right situation.

We can help you make the models more useful. You can quickly run the various models and compare their results at CRA Health's Risk Express or go to to learn more about how our software can make identifying high-risk individuals easy and efficient. We can help you save lives.

In future blogs, we’ll discuss the models in more detail and how you can better use them to care for your patients.

About the Author: Kevin S. Hughes, MD, FACS


Kevin S. Hughes, MD, FACS is a co-founder and medical advisor to CRA Health. Dr. Hughes is the Massachusetts General Hospital’s Surgical Director of the Breast Screening Program, Surgical Director of the Breast and Ovarian Cancer Genetics and Risk Assessment Program, and Co-Director of the Avon Comprehensive Breast Evaluation Center, and serves as the Medical Director of the Bermuda Cancer Genetics and Risk Assessment Clinic. He is an Associate Professor of Surgery at Harvard Medical School. Dr. Hughes is actively involved in the establishment of standards and in research regarding the genetics, screening, diagnosis, and treatment of breast cancer. He is the author of numerous papers and book chapters on the subjects of breast cancer, screening, diagnosis and treatment, and risk assessment. More information can be found at: