Confusion Over Utility of Common Genetic Variations in Breast Cancer Risk Prediction

A paper published in the New England Journal of Medicine (NEJM) this week, entitled “Performance of Common Genetic Variants in Breast-Cancer Risk Models,” has led several media outlets to declare that common genetic variants have nothing to add when it comes to predicting breast cancer risk.  Here we’ll explain how the results of this study have been misinterpreted.

Researchers from the National Cancer Institute and other institutions looked at data from 5,590 women with breast cancer and 5,998 women without the disease.  These women, all between 50 and 79 years old, had participated in one of five studies, four in the United States and one in Poland.  Since it’s known already which women have or have had breast cancer, and which have not, the researchers were able to use their data, both genetic and non-genetic, to test the predictive power of different types of risk calculations.

The study tested five different models.  The “demographic model” considered only age, year of entry into the study and which study the woman was originally part of.  The “nongenetic model” added in several variables that are part of the so-called “Gail model,” which is the standard model used in clinical practice today to counsel women about their risk.  These variables were the number of first-degree relatives with a diagnosis of breast cancer, age at menarche, age at first live birth, and number of previous breast biopsies.  Two models used the demographic information plus genetic information for 10 SNPs (they differed in details of how the genetic risk score was calculated).  Finally, the “inclusive model” combined demographics, the Gail model and the genetic information.

The genetic models and the nongenetic model performed about the same, with genetics doing just a little bit better.  Perhaps not surprisingly, the best model was the one that used the most information.  With the inclusive model, which is based on genetic and non-genetic information, there was a net 12% improvement in risk classification over the nongenetic model for women with breast cancer.

So, SNPs did add something.  The improvement in prediction seen when SNPs are added to the nongenetic model is about the same as the improvement seen when the Gail model information is added to the simplistic demographic model.  Discounting the benefit of adding SNP information to the Gail model implicitly discounts the utility of the Gail model itself.  Peter Devilee and Matti A. Rookus made this very point in a NEJM editorial that accompanied the study.

It’s also important to remember that the study of common variations and their effect on common diseases like breast cancer is a relatively young science.  Many more variants are bound to be discovered, and these will only help further refine risk predictions.

It’s true that there is a cost associated with collecting a person’s genetic information, while collecting the information needed for the Gail model is essentially free.  (In fact, a risk calculator based on the Gail model is available at the National Cancer Institute website.) But for those people who do have their genetic information in hand, the current study shows that it can improve the estimation of their risk for breast cancer.  It should be remembered, too, that not everyone is fortunate enough to have access to their family medical history. Adoptees, for example, cannot get an accurate picture of their breast cancer risk from the Gail model.  For such women, the ability to use their own genetics to help assess their risk for breast cancer is an option that should not be dismissed.

Ultimately, it’s not about genetics vs. non-genetics — it’s about getting accurate estimates of risk to help doctors catch cancers early on, when they’re easiest to beat. Researchers should welcome any tool that can help them move closer to keeping more women healthy.  Giving up on genetics this early on in the game would be a real disservice.

Note: A similar study can be found in Nature Precedings.  The results are consistent with the NEJM study regarding the improvement in risk prediction when SNP information is added to the Gail model.  This work also shows that the improvement is larger for women at intermediate risk who are more likely to be reclassified.  Senior author David Hinds was at Perlegen Sciences when this paper was written and is currently an employee of 23andMe, Inc.

  • Will FitzHugh

    Hi, I’m curious where you got the figure 12% in “there was a net 12% improvement in risk classification over the nongenetic model for women with breast cancer”. I can’t find that statistic in the paper.


Return to top