A case in point is breast cancer. In this issue of the
Journal, Wacholder et al.
3 report that when 10 recently identified SNPs
that have been associated with breast cancer were added to the
Gail model, a commonly used tool to assess the risk of breast
cancer, discrimination and prediction in estimating risk were
only moderately improved. The Gail model is based on self-reported
risk factors (age at menarche, number of breast biopsies, age
at birth of the first child, and family history of breast cancer).
The authors conclude that the clinical usefulness of common
genetic variants for identifying women at increased risk for
breast cancer is, at present, still negligible. Intriguingly,
however, Wacholder et al. also found that a model with the 10
SNPs alone performed just as well as the Gail model in predicting
the development of breast cancer. The addition of the SNP data
to the Gail model almost doubled its discriminatory power, which
was expressed as the area under the receiver-operating-characteristic
curve (AUC); the AUC was calculated to be 61.8% in the final
combined model. The authors correctly conclude that such an
AUC is still far too low for application in clinical settings,
but in admitting so, they also implicitly dismiss the pure Gail
model for this purpose.