Mammography's Effect on Breast Cancer Mortality: The Emperor's New Clothes

by Dr. Rebecca Gelman

For the past two years, the advisability of screening asymptomatic women under 50 years old for breast cancer using mammography has been debated in the medical literature, in the public press, at 3 international workshops and in the US Congress. The issues include: is a new, large (+ - 2 million women followed 5-7 years) randomized trial needed, should government agencies publish guidelines on mammography use, and should a national health insurance system in the US pay for mammography in this age group (at an estimated cost of $2-3 billion per year). There is good agreement by all the debaters on: what type of study provides useful data (randomized), which outcome is most relevant (breast cancer mortality), and the conclusions to be drawn from the world's medical literature (no significant decrease in breast cancer mortality in women under 50 in the face of a significant decrease in breast cancer mortality in women 50 and over). The debaters seem to differ on only three topics: is it appropriate to analyze mammography results separately by age at all, are the available studies fatally flawed, and should women under 50 be denied the presumed benefits of mammographic screening only because the studies have been too flawed to show significant mortality results in this age group.

In this debate, I am beginning to feel like the child in the fairy tale about the emperor's new clothes. The emperor purchased some marvelous new clothes that were made of such fine cloth that only intelligent people could see them. Actually, no one could see the clothes, but everyone praised them, since no one wanted to be thought "stupid". Only an unsophisticated child burst out with the comment that the emperor was naked. Not only do I fail to see the good of mammography in young women, I think that it actually might have had some harmful effects.

I have been shocked to find that no one in the current debate is focusing on what appears to be a very clear qualitative interaction between age and mammography's effect on breast cancer mortality. (If not tantamount to overlooking a completely naked emperor, this at least qualifies as slighting a very decollete empress.) Only one trial (the HIP trial, started in 1963) has shown a benefit of screening in women under 50 years old. However, the benefit in the HIP could not have been due to mammography (only 6 tumors found by mammogram in women under 50 compared to hundreds found by physical exams and patient self-exams). Six published trials (started 1976-1982) have found a detriment of screening in women under 50 for at least the first 7 years of follow-up (i.e., more breast cancer deaths in the screen group than the control group). In 4 trials followed more than 7 years, there is some indication that deaths in the control group are catching up or surpassing deaths in the screen group, perhaps because these trials started mammograms in the control group at 3-5 years. In contrast, only one trial has shown a detriment of screening in women over 50 during the first 7 years of follow-up; the other 6 trials have all shown a benefit. Even doing a simple Fisher exact test on the results of these trials (1 beneficial and 6 detrimental in younger women vs 6 beneficial and 1 detrimental in older women) produces a significance level of 0.03. Doing a two-sided logrank test stratified by trial, one finds that the detriment by 5 years in younger patients has a p-value of 0.05 and the benefit in older patients has a p-value of 0.006. The mortality effects are of roughly the same order of magnitude (in opposite directions): a difference of 5-6 lives/10,000 women screened.

I attribute the lack of attention to this qualitative interaction to four reasons. First, mortality in these trials is almost always summarized as cumulative mortality at the most recent follow-up, and the dates when data are reported separately by age group are not usually pre-planned and may be related to the results (informative stopping). Second, the statistical test used is almost always the Poisson test on cumulative mortality at the most recent follow-up; this is no more appropriate than using a binomial test in a treatment trial at the time when the better survival curve finally falls to meet the worse survival curve. Third, in the absence of other proven prevention methods, there is an understandable tendency to lapse into wishful thinking. Fourth, there is no immediately obvious explanation of how mammography can be harmful in the younger group. Excess breast tumors caused by the mammograms themselves makes no sense, since solid tumor radiation carcinogenesis has a 10-15 year lag time to diagnosis (and an additional 5-10 year lag for deaths), but the detriment from mammography is apparent 3 years after the trials start. Inaccurate death certificates and lethal toxicities of cancer treatment don't make sense because both are rare and more common in older than younger women.

Only two explanations seem at all plausible. It is possible that false reassurance after a "clean" mammogram is causing some patients to delay seeking medical attention for patient-detected tumors; this could affect younger women more than older women because there are more "false" negative mammograms in young women and young women may have faster tumor growth rates. Second, the screened groups in these trials received different breast cancer treatment than the control groups. In all trials except HIP, a higher percentage of screened women than control women had breast conserving treatment (lumpectomies), with the biggest difference being among younger women. Perhaps deaths in women with a local recurrence (almost never fatal) were counted as breast cancer deaths and there are more local recurrences after lumpectomy. Also, a lower proportion of breast cancers in the screened groups had axillary lymph nodes with cancer and so fewer were treated with chemotherapy (which seems to have a survival benefit, especially in younger women).

When the child in the fairy tale yelled out that emperor was naked, everyone immediately realized the truth of the statement. I have not had a similar experience. However, I'm sure the child wasn't very popular with the townspeople after this event, so maybe it is just as well that the general public hasn't heard my yell.

Last modified $Date: 1994/12/01 20:23:33 $ by Evelyn Ophir, ophir@hsph.harvard.edu