We already know that mammography is an imperfect screening tool. Authors of a study in the Annals of Internal Medicine recently re-analyzed data from the Breast Cancer Surveillance Consortium (BCSC) to try and tease out the actual rate of overdiagnosis from screening mammograms. Overdiagnosis was defined as the percentage of breast cancers detected by screening that, through mathematical modeling, would not have been predicted to cause clinical disease — either because the cancer would never grow or not grow fast enough before statistics predicted the person would pass away from another cause (for example, a heart attack or lung cancer, both more common causes of death in women than breast cancer).

The BCSC is a large and diverse dataset that included 35,986 women aged 50–74 who received at least one screening mammogram between 2000 and 2018. A total of 718 breast cancers were detected, of which 645 were detected through screening and about 10 percent diagnosed between screening periods. In this NIH funded study, the authors primarily examined the phenomenon of lead time bias. To do this, they used a Bayesian model to estimate how many individuals would have died of something else (theoretically, based on death tables from a 1971 birth cohort) before their breast cancer ever came to clinical notice. Importantly, the authors considered the impact both of cancers that would have never progressed and of cancers that may have progressed but not to the point where they were clinically apparent. False positive mammograms and/or biopsies were not considered as part of the definition of overdiagnosis.

Analysis suggested that having more mammograms was associated with a higher risk of finding a cancer that was not clinically important. This makes sense on several levels, but especially from a population-based perspective where those who are older have both a higher risk of cancer and a higher risk of death from other causes. Between the first mammogram and the last one, the rate of overdiagnosis increased from 3.1 percent (95% prediction interval [PI] of 1.6–5.1%) to 18.1 percent (95% PI 11.9–24.5%). Overall, the rate of overdiagnosis based on these calculations for biennial mammograms was 15.4 percent (uncertainty interval of 9.4–26.5%), translating to one cancer out of seven detected by screening statistically “destined” to never become clinically apparent.

It’s important to distinguish between overdiagnosis and false positives. With screening mammography, false positives are abnormalities on the mammogram that turn out not to be a malignancy after further evaluation. False positives are costly in terms of patient anxiety and need for subsequent imaging and biopsy procedures. Decreasing the frequency of screening limits false positives while preserving the benefits (see the USPSTF’s recommendation for biennial mammography). With overdiagnosis, however, the person genuinely has breast cancer, but identifying and subsequently treating it does not improve overall health or survival. Moreover, overdiagnosis artificially inflates survival rates. Finally, it’s important to consider that cancer treatments themselves can cause significant harms, even death, and should be accounted for in all-cause mortality. The challenge here, of course, is that overdiagnosis can only be identified in a population, and we can never tell whether an individual patient has been “overdiagnosed” or is in fact someone for whom screening has identified a cancer early when it is most treatable. We need to appreciate that overdiagnosis is an adverse effect of screening and cannot be separated from it. As part of shared decision making, we should be comfortable explaining to our patients that one in seven women diagnosed with breast cancer through screening will be unnecessarily treated as a result of overdiagnosis.

For more information, see the topic Mammography for Breast Cancer Screening in DynaMed.