Doublex

Reconsider the Mammogram

Yes, they result in false positives. But that’s because we’re using them incorrectly. A new model fixes that.

A woman undergoes a free mammogram inside Peru's first mobile unit for breast cancer detection.
With mammograms, it is becoming increasingly apparent that a one-size-fits-all recommendation is not the ideal approach

Photo by Enrique Castro-Mendivil/Reuters

A recent cover story in the New York Times Magazine made a convincing case against the mammogram. The author’s main criticism was that mammograms result in many false-positives, which other research has confirmed. Women get treated for cancers they don’t have, or cancers that are noninvasive but which doctors at the moment can’t distinguish from the malignant ones. All of this leads to a lot of wasted money, stress, and distrust of the system as a whole.

While this is almost certainly true, many in the scientific community prefer to look at mammograms in a different way. Mammograms are a life-saving screening method, but they are not being utilized properly. But there is a way to fix that. A group of researchers at the University of Wisconsin–Madison, Georgia Institute of Technology, and Harvard Medical School have started to use risk factor models that can help eliminate some of the harms associated with mammography and use it to its full potential.

With mammograms, it is becoming increasingly apparent that a one-size-fits-all recommendation is not the ideal approach. Professor Oguzhan Alagoz from the University of Wisconsin–Madison, his former Ph.D. student Turgay Ayer from the Georgia Institute of Technology, and Natasha K. Stout from Harvard Medical School are working on a model that will determine the optimal time for a woman to get her next mammogram. (Disclosure: I am an undergraduate research assistant for Alagoz although I did not work directly on this project.)

Currently, screening recommendations are based almost exclusively on age. The American Cancer Society and many other groups recommend annual mammograms starting at 40 years old, while the United States Preventative Services Task Force currently recommends biennial tests starting at 50. Individual physicians will then discuss earlier screening based on other risk factors. But age is not the only relevant risk factor for breast cancer. Family history, alcohol use, number of lifetime menstrual cycles, and breast density are just a few of the myriad other factors. Ayer, Alagoz, and Stout’s model accounts for many of these risk factors and personal history information and using this information, calculates the best time for a woman’s next mammogram.

The model works by calculating the risk of getting either invasive or noninvasive breast cancer. While noninvasive cancers pose little threat to the woman, they can sometimes progress to invasive cancer. Therefore the model includes the likelihood of the woman getting noninvasive cancer simply because that is a risk factor for invasive cancer. The model then makes a recommendation based on the risk of invasive or noninvasive cancer. If the woman’s overall invasive cancer risk exceeds the desired threshold, a mammogram is recommended.

In a recent article in Operations Research, the authors show how the model would be used. Take a 40-year-old white woman who has no history of breast cancer in her family, who started menstruation at 14, and who had her first child at 23. Because her chances of getting in situ cancer or invasive cancer are low (0.1 percent and 0.2 percent), the model recommends waiting to get a mammogram until she turns 42. They then look at a 50-year-old woman who has the same risk factors as the first woman but did not have any mammograms during that time. This time, the model recommends getting a mammogram because her risk, at this later age, is high enough to justify the screening.

If the woman has another negative mammogram, the intervals continue to increase. Conversely, an unusual mammogram result, such as a benign cyst in the breast, can prompt a woman to shorten the time between her next mammogram.

In the end, the model would create a single statistic that would account for the individual’s breast cancer risk factors and her previous screening decisions and results. Ideally, this statistic would be a starting point for discussion among the radiologist, physician, and patient. It would help with mammography decisions of course, but it would also be useful in discussing other breast cancer prevention treatments, such as the drugs tamoxifen or raloxifene.

By personalizing mammography decisions, we can improve the quality and length of life, yet reduce the overall number of mammograms.  The model could potentially help save numerous high-risk women while preventing undue harm to the rest of the public. In addition, the model could help reduce the $100 million that we overspend in mammography. These savings come from a simultaneous decrease in the number of mammograms and the number of false-positive mammograms. Using the screening guidelines set in this model, the researchers estimate that an average woman will need 14 fewer mammograms and at least halve the number of false-positive mammograms.

Of course, implementation of such a model may be controversial. When the United States Preventative Services Task Force changed its mammography recommendations, there was a major backlash. Many claimed that the task force did not truly have women’s interests in mind, and instead they were in the pockets of the insurance companies.

We still do not fully understand all of the risk factors for breast cancer, nor do we fully know why breast cancer affects some women and not others. We still do not know the optimal treatment policy for breast cancer, and whether some cancers would be better left untreated. And as we learn more about this disease, the model will need to constantly evolve to match the incoming research on these issues.

Despite these challenges, the model furthers an incredibly exciting idea that can be applied to other diseases: We can better tailor our health care decisions to our personal history as opposed to broad population based statistics. Many have claimed that personalized medicine is the future of health care. Innovations such as this model can help us adjust our health care recommendations to each person’s situation.