Print

Email story

Comment

Email editor

Breast cancer prediction tool has many variables

By DR. V. UPENDER RAO
Published January 1, 2007

What is my risk of cancer? That is a question that physicians often hear from patients. The obvious answer is that we cannot predict risk accurately. I often discuss the Gail Model with my patients and colleagues. The Gail Model is a well known breast cancer prediction tool that is used by oncologists all over the world.

Although it is the best known model available, it is far from perfect and has several shortcomings. In the ensuing paragraphs I will discuss its genesis, components, statistical method and its merits and demerits as a breast cancer prediction tool.

The original Gail Model was based on the Breast Cancer Detection Demonstration Project's data set. It consisted of several factors thought to be associated with breast cancer, such as age at first menstrual period, age at first live birth, number of previous biopsies, number of first degree relatives with breast cancer, and whether or not atypical ductal hyperplasia was present in the biopsy specimen.

A modified version of the Gail Model (Gail Model 2) was used by researchers who were evaluating the role of tamoxifen as a breast cancer preventing agent for women thought to be at high risk.

Using these parameters, a probability of risk of breast cancer was calculated for the first five years from the date of diagnosis and a lifetime risk (until age 90).

A risk level of 1.67 percent in the first five years was agreed upon as a point above which the benefits would off set the possible toxicities of tamoxifen. This level of risk was used as the minimal entry criteria for women participating in the breast cancer prevention trial with tamoxifen.

This study proved the efficacy of tamoxifen as a breast cancer prevention agent for high-risk women. This indirectly validated the Gail Model as an appropriate tool for risk prediction and patient selection.

While the Gail Model may have been useful in predicting risk for a group of women with similar risk characteristics, it was neither designed to estimate risk of an individual woman, nor was it ever proven to do the same.

An in-depth analysis of the Gail Model by Joann G. Elmore and Suzanne W. Fletcher of University of Washington School of Medicine and Harvard Medical School, respectively, was published in the Dec. 6 issue of the Journal of the National Cancer Institute. They calculate and assign an accuracy factor to the Gail Model. This factor is known as the "concordance factor."

A concordance factor of 1 would predict the risk accurately 100 percent of the times tested. A factor of 0.50 carries an accuracy of only 50 percent (equivalent to the toss of a coin).

The researchers arrived at a concordance factor of 0.59 for the Gail Model. This means that the ability of the Gail Model to predict risk of breast cancer for an individual woman was better than the "toss of a coin," but not by much.

Given this imprecise tool, how can oncologists select appropriate patients for breast cancer prevention with tamoxifen? In my own practice, I incorporate breast density (which was recently reported as a strong and independent risk factor) into risk estimation for individual patients.

Such incorporation improves the concordance factor to 0.66, which increases the predictive accuracy significantly. Additionally, I look for the absence of risks for tamoxifen toxicity. For instance, endometrial cancer is a rare but serious side effect of tamoxifen, and if the patient has had a previous hysterectomy, this will not be a concern in analyzing the risk benefit ratio.

The 2004 Institute of Medicine report on breast cancer screening identified individual risk assessment as essential to improving early detection of breast cancer. To further that effort, I use the Gail Model with incorporation of breast density and other strategies described above to select patients for cancer prevention with tamoxifen.

Of course, an open discussion and the patient's own participation in the final decision making are important. I think it would be unwise to forgo effective prevention of cancer just because the predictive model is not 100 percent accurate.

V. Upender Rao, MD, FACP, practices at the Cancer and Blood Disease Center in Lecanto.

Note to readers: The Dec. 25 column contained a error.

Over the years, the number of colonoscopy screenings increased in the United States when Medicare policy broadened. This has led to more cancer being detected in patients who are asymptomatic, because right-sided cancers tend to cause fewer symptoms than those of the left side.

The Dec. 25 column said "symptomatic" instead of "asymptomatic."

[Last modified December 31, 2006, 20:26:06]