A statistical model commonly used to predict the risk of breast cancer in women was not accurate when used to evaluate women with atypical hyperplasia, according to a new Mayo Clinic study. Atypical hyperplasia (atypia) describes breast tissue with an increased growth of abnormal cells that might become cancerous.

The Gail model calculates probabilities that a woman will develop invasive breast cancer during the next five years, and by age 90. The model has come to be called by the name of its developer, National Cancer Institute researcher Mitchell Gail, M.D., Ph.D. Its original purpose was to identify groups of high-risk women for participation in breast cancer chemoprevention studies. But the model has since been used in clinical settings to counsel individuals about their risk of developing breast cancer. Predicting risk at an individual level is much more challenging than when risks can be averaged across groups.

"The assumption has always been that this model works well in women with atypia, but this had never been validated," says Shane Pankratz, Ph.D., Mayo Clinic statistician and a lead investigator in the study. "We found that, for the group of women with atypia, the model predicted significantly fewer invasive breast cancers than were actually observed, and we also observed that the model was not able to reliably identify the women who were actually at higher risk of developing breast cancer."

The Gail model considers the woman's family history of breast cancer, her age, and her ages at the onset of menstruation and at first live birth, as well as the number of breast biopsies undergone and presence of atypical hyperplasia found in biopsies. About 5 percent of women who undergo biopsies for suspicious lumps or other breast concerns have atypia. About 25 percent of those with atypia will develop cancer within 25 years.

Mayo Clinic researchers tested the Gail model in 331 women with atypia who had benign breast biopsies at Mayo Clinic between 1967 and 1991. Of these women, 58 developed cancer during an average of 13.7 years of follow-up. In contrast, the model predicted that 34.9 women would develop breast cancer in that period.

Using these and other data, researchers also calculated the model's performance for individuals using the concordance statistic (c-statistic), which reflects how closely the actual timing of breast cancer events aligned with model predictions. A c-statistic of 0.5 is observed if the predictions are no better than random chance; a c-statistic of 1.0 is observed if the predictions are perfectly concordant with the actual outcomes. In this study, the c-statistic was 0.5, reflecting that the Gail model worked no better than a coin flip in predicting which of the women with atypia would develop invasive breast cancer.

When assessed across other groups of women without respect to the presence of atypia, the Gail model typically performs better. In that setting, it has been shown to predict approximately the same number of breast cancers that later occur.

Lynn Hartmann, M.D., Mayo Clinic oncologist and co-investigator on the study, says that there is strong interest in predicting breast cancer risk. For example, the Gail model, posted on the National Cancer Institute's Web site, attracts 25,000 viewers each month.

"Doctors counsel women at high risk to have more frequent or intensive surveillance or to consider chemoprevention strategies such as tamoxifen or raloxifene," says Dr. Hartmann. "When making such decisions, women and their physicians must have highly reliable risk estimates."

Researchers are pursuing other avenues to better predict individual risk. Previously, Mayo Clinic researchers found that women with multiple sites of cellular atypia in a breast biopsy have significantly increased risk of developing breast cancer. In a study published earlier this year, Mayo researchers found that women whose atypia tissues express COX-2 enzymes were more likely to develop breast cancer, and the higher the COX-2 levels, the higher the risk.

Other Mayo Clinic researchers involved in this study were Judy Boughey, M.D.; Amy Degnim, M.D.; Robert Vierkant; Karthik Ghosh, M.D.; Celine Vachon, Ph.D.: Marlene Frost, Ph.D.; Shaun Maloney and Carol Reynolds, M.D.

Copyright 2008- American Society of Registered Nurses (ASRN.ORG)-All Rights Reserved