Efficacy Endpoints: Mortality, breast cancer mortality
Harm Endpoints: Unnecessary surgical procedure, false positives, overdiagnosis leading to treatment
Narrative: Screening mammography (mammography in women without any signs or symptoms of possible breast cancer) has been studied in large randomized trials of nearly a half million women. The theoretic basis for the intervention is sound. It is presumed that therapeutic intervention at a point when cancer is visible on a mammogram but not yet palpable in the breast will, for a small number, result in earlier, ultimately life-saving, therapy. Overall mortality rate, however, was not improved in the groups in these studies assigned to receive regular mammograms. When aggregating data from those trials in which randomization was appropriate (resulting in balanced groups), there was also no identifiable reduction in deaths due to breast cancer.
The statistical result is slightly different when one accepts all trial data rather than restricting data to appropriately randomized studies. While overall mortality remains unchanged, in this analysis breast cancer mortality is reduced by approximately 15%. If accurate, this would represent a NNT of approximately 2000. However, breast cancer mortality is not as important as overall mortality, because individuals deciding whether to undergo screening mammography will typically want to avoid death, rather than simply avoiding death from one possible cause. The lack of overall mortality benefit with a potentially small benefit noted in breast cancer mortality suggests that mammography is likely to be taking as many lives as it is saving.
This is an important, though uncommonly discussed, issue in the translation of evidence from cancer screening trials.1 It is known that overdiagnosis (treatment of cancers that would have been no threat), and high false positive rates (misdiagnosis) lead to medical harms and unnecessary surgeries, chemotherapy, and radiation. It is presumed that a small number of these individuals also die due to aggressive therapies such as chemotherapy and major surgery, and given the small, though disputed, margin of benefit suggested by the analysis above it seems likely that if there is a benefit to screening mammography it is balanced out by mortal harms from overdiagnosis and false-positives.1 If screening mammography reduces the rate of one type of death, but the overall rate of death is not affected, it is mathematically implied that there is an increase in some cause of death that is offsetting the reduction in breast cancer deaths. Unfortunately, because most massive trials of mammography are simply too large to reliably track every cause of death, it is often difficult to tell with precisions where this increase occurs.
Harms due to breast cancer screening are discussed at length in the Cochrane review, and in more recent analyses.2,3 These include false positives that result in surgical procedures, and emotional distress due to false positives. In addition, despite the lack of identifiable benefit to screening mammography, women in groups assigned to receive mammograms were 20% more likely to undergo mastectomy and 30% more likely to undergo surgery. Finally, if it is true that breast cancer deaths are reduced it has been estimated that for every one patient who avoids death from breast cancer approximately 10 to 20 women are treated unnecessarily as cancer patients, typically receiving surgery, radiation, and chemotherapy.3,4 This additional estimate of harm is not represented in the NNH statistics listed above.
Caveats: This topic is complex and difficult to study even in randomized trials, where some patients assigned to mammography do not comply with plans for screening and others assigned not to receive mammograms undergo screening (i.e. 'contamination'). Moreover the studies are dated, and there have been significant improvements in breast cancer therapy since these trials. Therefore these data are highly imperfect. This has led many researchers to examine observational data, from which inferences regarding cause and effect are less valid, but which may be useful in explaining epidemiologic trends showing no significant reduction in breast cancer deaths since the advent of screening mammography. Most recently, an observational study of the effects of screening in the United States over the past thirty years suggests that 50% of screen-detected cancers represent overdiagnosis, and that while early stage cancer diagnoses have doubled, advanced stage breast cancer is about as common today as it was before mammography was in widespread use.5
Importantly, overall mortality may not be affected by mammography because breast cancer deaths are only a small fraction of overall deaths. This would make it very difficult to affect overall mortality by targeting an uncommon cause of death like breast cancer. If this is the reason for trial data demonstrating no overall mortality benefit then it means that it would take millions of women in trials before an overall mortality difference was apparent, a number far higher than the current number of women enrolled in such trials. If this is the proper explanation then any important impact on mortality exists, it is small enough that it would take millions of women in trials to identify it. This belies the public perception of mammography.
It is important to note that screening mammography saves some number of individual lives in the sense that there are almost certain to be women whose breast cancer is amenable to curative treatment at the time it is identified by mammography, but not amenable to cure at the time it is identified by physician examination or self-examination. However the number of cancers for which the window of curative potential exists exclusively during this time is likely to be small, and however many lives are saved in this way an equal number appears to be lost due to radiation from mammograms, or overdiagnosis, or false positives, or some combination of the three. The NNT that we calculate here is designed to be a tool for those attempting to make decisions about health care interventions, and in aggregate the best available evidence suggests there is no overall benefit to screening mammography. This should not be taken to mean that some individuals are not saved—it is likely that they are—however an equal number of individuals appear to lose their lives due to mammography, and there is no way to know which group one will eventually fall into.
Two recent data reviews deserve further mention. The United Kingdom commissioned an independent review after dissenting voices swelled, for the purpose of better informing shared decision making and educational materials about the harms and benefits of screening.6 Unfortunately the review concluded that screening mammography trials were inadequately powered to detect an impact on all-cause mortality, and therefore used breast cancer mortality as a primary outcome. They concluded a 20% reduction and used this in their discussion of harms and benefits. As noted above, cause-specific mortality is both scientifically unstable (conclusion reversal is common when all-cause mortality is considered)7 and disease-centered rather than patient centered (patients would prefer to avoid death altogether). Thus, either screening mammography does not save lives or else we have inadequate data to say whether it does or does not. In neither case can a benefit be scientifically claimed.
A circumspect, and patient-centered, response to the UK review provides a brief, smart data analysis suggesting that screening mammography is taking significantly more lives than it is saving.8 This latter review takes into account improvement in cancer treatments and likely overdiagnosis (using the US data, which were published after the UK-commissioned review), and finds that the short and long-term sequelae of radiation therapy in cancer treatment regimens is leading to two to three lives taken by mammography for every one life saved. We have nonetheless chosen to retain the 'Red' (data suggest no benefit) rating of screening mammography rather than change to 'Black' (data suggest harms>benefits), as this review presents a best-guess projection about modern day impact.
As a final note, it is important to be aware that the primary reason for the failure of screening mammography is the ongoing inability to determine which cancer cells in which humans represent a true threat to the host. If this could be determined successfully then any screening test like mammography would be a success. Should technology and science advance to the point where prognosticating based on clinical material (i.e. biopsy cells or mammogram characteristics or individual's biologic characteristics) was overwhelmingly accurate and reliable, early detection programs of all sorts would all become far more likely to result in life-saving benefits.
Author: David Newman, MD
Published/Updated: October 6, 2010
The title bar is color-coded with our overall recommendation.
If you have suggestions, requests, or questions about a particular NNT review, please send us a message and we’ll try to address it as soon as possible.