Many studies have used a clinical trials approach to study the impact of cancer screening.  These studies were effective in that they demonstrated that mammography reduced the risk of dying due to breast cancer, but the ability to compare a range of different possible screening strategies is beyond the reach of individual trials.

“There are a number of questions that are left unanswered [by previous clinical trials], like what is the best screening schedule,” said Kathy Cronin, Ph.D, mathematical statistician for NCI.  “It’s unlikely that those types of questions would be addressed directly for a number of reasons” including the fact that breast cancer screening is already a wide-spread practice with a known benefit.  Randomizing people into groups that conflict with numerous guidelines raises ethical challenges.

To overcome this challenge, the researchers compared 20 screening strategies, using six statistical models.  “The six models are all independently designed models, in that they all have their own approach to modeling disease progression, and their own approach at understanding what the benefit of screening would be,” said Cronin.  “In order to allow for comparison, the models all used a common set of inputs, and common set of results.”

The outcome measures used in all of the models included the percent decline in mortality, the number of life-years gained (the difference between the age someone dies from breast cancer and their normal life expectancy without breast cancer), the number of false-positive results (mammograms read as abnormal, for women who do not actually have cancer) and unnecessary biopsies (biopsies done in women with false-positive results), and the rate of over-diagnosis (early detection and diagnosis of cancer that, in the absence of screening would not  have been diagnosed in the person’s lifetime).

When making breast cancer screening recommendations, many organizations rely on evidence reviews, which are valuable means of examining groups of studies looking at the same topic.  However, there are many different types of screening regimens that could be created – by varying the age when women start screening, when they stop screening, and the frequency of screening – and it is not possible to look at all of these strategies through the traditional approach.  The statistical modeling conducted by the CISNET group attempts to add a different perspective in attempting to resolve these issues.

The model can compare multiple screening strategies by adding “biological assumptions about the progression of disease, and assumptions about the screening technology,” said Eric Feuer, Ph.D., chief of the Statistical Research and Applications Branch of NCI, and senior author on the paper.  “The models can then compare these strategies across the entire life course, which in any individual study would be difficult to do. Comparing the results across multiple models lets us determine how robust the results are with respect to these assumptions.”

The models, much like most screening recommendations, are population based – the models cannot make determinations for an individual woman.  “An individual women, in collaboration with her health care provider, needs to consider two things, in order to tailor those recommendations for her individual situation,” said Feuer.  “First, what is the risk of breast cancer for that woman, compared to the general population…The second is to try to weight the harms vs. the benefits.” The harms, according to Feuer, include false-positives, unnecessary biopsies, and overdiagnosis.

“We can’t assess values for a woman,” said Feuer.  “One important addition that could be made to research is more tools that could help women assess and weigh these harms and benefits.”

This type of comparative modeling has been used in other cancers and additional disease types.  Modeling is useful in situations where we can use information on the natural history and/or epidemiology of the disease to compare competing strategies (as with breast cancer screening) or competing technologies, such as for colorectal cancer where there are many different types of screening tests available.  Finally, the models can be used to pinpoint areas in need of further study.

Additional Resources:

Additional still photos: mammography, breast cancer

Cancer Detection B-Roll Footage: full length preview, high-resolution download

NCI Statement on Breast Cancer Screening