Single-nucleotide polymorphisms do not substantially improve risk prediction for breast cancer

Recently identified genetic markers, called single nucleotide polymorphisms (SNPs), that are associated with a small but statistically significant increase in the risk of breast cancer do not appear to substantially improve the accuracy of existing models that use clinical factors to predict an individual's risk, according to a study in the July 8 online issue of the Journal of the National Cancer Institute. Risk prediction models provide estimates of the risk of disease to assist in clinical management of individual patients. A woman at high risk of breast cancer might benefit from a preventative intervention, whereas the side effects of the intervention might outweigh the benefits in a woman at lower risk of breast cancer. An ideal risk model would provide much higher risk estimates for women who eventually develop breast cancer than for women who do not, a feature called discriminatory accuracy. Researchers have traditionally used clinical information, such as age and family history, to build risk models, but other characteristics can be incorporated into models. Scientists recently identified seven SNPs that are associated with a moderate increase in the risk of breast cancer.

In the current study, Mitchell Gail, M.D., Ph.D., of the National Cancer Institute in Bethesda, Md., compared the discriminatory accuracy of these seven SNPs to an established risk model, called the Breast Cancer Risk Assessment Tool (BCRAT, frequently called the Gail model). BCRAT uses a woman's age, age at menarche and at birth of first child, family history of breast cancer, and breast biopsy results to predict her risk of disease. He also tested whether the addition of the SNPs to the existing model could improve its accuracy.

Gail found that the seven SNPs had less discriminatory accuracy than the BCRAT model. In fact, when he doubled the number of SNPs in the model, by adding seven hypothetical SNPs with similar characteristics, he found that the SNP-based model was still less able to accurately discriminate risk than BCRAT. When he added the seven characterized SNPs to BCRAT, the discriminatory accuracy of the model improved modestly. However, the increase was less than would be gained by adding information on breast density, which is also associated with breast cancer risk.

"Experience to date and quantitative arguments indicate that a huge increase in the numbers of case patients with breast cancer and control subjects would be required in genome-wide association studies to find enough SNPs to achieve high discriminatory accuracy," Gail concludes.

In an accompanying editorial, Margaret Pepe, Ph.D., and Holly Janes, Ph.D., of the Fred Hutchinson Cancer Research Center in Seattle note that although Gail's calculations are based on several assumptions, the paper provides a starting point for important discussions about risk models.

The editorialists point out, however, that Gail did not actually calculate the proportion of women that are classified as high risk with the BCRAT model or how many more would be identified by adding SNP information. They also suggest that even if SNPS are not informative in the population as a whole, they may have value in subpopulations.

Source: Journal of the National Cancer Institute