When an individual patient asks a physician, "What is my risk of getting breast cancer," the physician has little to offer.
When an individual patient asks a physician, “What is my risk of getting breast cancer,” the physician has little to offer. “For a woman sitting in my office, her risk is either 0 [she won’t get breast cancer] or 100% [she will get breast cancer,]” explained JoAnn Elmore, MD, University of Washington, Seattle, WA. “The ideal risk model needs to work well for the individual woman, so we can counsel her, but we have lots of work to do before we have an ideal risk prediction model,” she said.
The Gail model, developed in 1989, is widely used to estimate expected cancer incidence among populations, but not individuals. This model is based on agreed-upon risk factors that include age, age at first period, age at first birth/or nulliparity, number of first-degree relatives with breast cancer, and history of benign atypical hyperplasia of the breast.
“Websites, such as the American Cancer Society, get hundreds of thousands of hits by women attempting to use this model to estimate their risk of developing breast cancer,” Elmore explained.
A study by Rockhill et al. attempted to validate the Gail model in 82,109 women enrolled in the Nurses’ Health Study. Over the first five years of the study, about 1000 women developed breast cancer (1.65%). The expected incidence, according to the Gail model, was 1.55%, “so the risk model worked well at a population level,” Elmore said.
The validation study also found, however, that many of the women diagnosed with breast cancer did not turn out to have high risk factors. In fact, 42% had a lower risk level than cancer-free counterparts. “Among 47 women with a high Gail risk score followed for five years, only one will be diagnosed with breast cancer,” she stated. “The Gail model is equivalent to a coin flip for an individual patient,” she said.
Why is it so difficult to develop an ideal risk prediction model? Elmore said that many risk factors are not highly sensitive or highly specific. Moreover, among most women, the risk is too small to be predicted.
Communicating With Patients
Breast cancer is a major fear of most women, and they routinely overestimate their risk, Elmore continued. A group of women were presented with this example: a hypothetical patient with two family members with breast cancer, a history of atypical hyperplasia of the breast, and age over 40 years at first birth. The women’s risk estimates were all over the map, from 4% to 50%. Physicians don’t do much better at estimating an individual’s risk. When the same case scenario was presented to radiologists, 96% overestimated the patient’s risk.
Patients, and even physicians, have a hard time interpreting the risks/benefits of a particular treatment in published studies. The same paper gave the following results for a specific treatment: reduced death rate by 34% compared with controls; produced an absolute reduction of 0.06%, and prevented one death among every 1588 women during 10 years.
“We have a problem in this country with numeric literacy. Health information on the Web is breeding a generation of cyberchondriacs,” she said. Use risk models cautiously, and communicate carefully with patients, she added.