Mortality reduction by widespread screening: not quite as effective as we all hope?

The recent 29-year follow-up analysis of the prospective Swedish Study of Tabar and colleagues [1] fueled the hope that mammographic screening can reduce breast cancer mortality. The study deserves careful scrutiny given the enormity of the societal costs involved.

The study compared the number of first breast cancer cases diagnosed over a 7-year screening period and the subsequent number of deaths from breast cancer over a 29-year follow-up period between an actively screened population (ASP) and a passively screened population (PSP). The actively screened population consisted of 77,080 women age 40-74 years at enrollment, who were offered an initial mammogram followed by a re-screening mammography every 24 months up to age 49 and every 33 months after age 49 up to age 74 years. Of the 77,080 women enrolled and offered screening, 85% attended the screenings. The passively screened population (PSP) was comprised of 55,985 demographically matched women, who were not offered mammography screening. No screens were performed after age 75. After 29 years of follow-up, the number and ratio of women from each group that had been diagnosed with breast cancer during the 7-year screening phase and of those who died of breast cancer during the subsequent 29 years of follow-up were compared.

Breast cancer was found in 1.8% of the women studied in each group. In the ASP (the group that was offered screening), as expected, more breast cancers were diagnosed in the women who actually opted for screening. Of those 85%, breast cancer was found in 2.18%. Overall, 1426 women were diagnosed with breast cancer in the ASP during the initial 7-year period; 351 of these (24.6%) had died of breast cancer after 29 years of follow-up. This death rate of 24.6% appeared much lower than the rate in the unscreened group (PSP), where 367 of 1042 diagnosed cases (35%) had died of breast cancer. Comparing cumulative cases of death in the ASP and the PSP, the authors also calculated the 'number needed to test' (NNT) to 'save one life' and suggested this number was either 414 or 519, depending on which adjudicating source (local or overview) was used to calculate cause of death.

The insightful comments of both Amos Pines and Syd Shapiro in Menopause Live on NNT [2] help to focus on the importance of measuring longevity benefits of widespread mammographic screening with and without considering oophorectomized women's additional risks. They also addressed the importance of reporting absolute risk rather than relative risk. We agree with them wholeheartedly that risks need to be expressed in a way that patients and providers can actually understand.

We offer an additional observation regarding screening and a comment on routine oophorectomy at the time of hysterectomy that we think are relevant to the discussion on how we should advise our patients.

 

Comment

Recent data show much higher rates of breast cancer overdiagnosis than previously estimated [3]. Tabar's 2011 paper dismissed the issue of overdiagnosis with a citation to a conclusion in another paper that overdiagnosis was outweighed by benefits claimed. But that citation was not explained and may be incorrect. In a recent paper, Jorgensen and colleagues [3] compellingly demonstrate significant overdiagnosis with the inception of widespread mammography screening. The authors estimate that 30-50% of early-stage breast cancers would never develop to clinical significance or would actually regress.

If overdiagnosis is high, the group diagnosed with cancer is larger, while the number of women who die of cancer remains the same, resulting in an apparent decrease in the mortality rate. It looks like mortality is reduced, when what actually happens is that the sample size is enlarged with non-lethal, clinically irrelevant cases. This is an artifact that exaggerates the benefit of mammograms. Comparison with the rate in a group not spuriously enlarged will suggest an apparent 'reduction' of the risk ratio. An example may help. When calculating the mortality rate with five deaths in 100 women with cancer, increasing the pool to 130 women - with the same five deaths - lowers the mortality rate from 5% to 3.8%, yielding an artificial mortality 'reduction' of 24%.

The impact of additional artifact error is compounded when mortality in a group of women that includes overdiagnosis due to the effects of screening is compared to mortality in a group of women without overdiagnosis (not screened for early-stage cancer). Division of the mortality rate of the one group by the mortality rate of the other leads to an exaggerated conclusion about the benefits of mammograms.

In 2009, Jorgensen and colleagues estimated the extent of overdiagnosis in publicly organized screening programs. Overdiagnosis was defined as the detection of cancers that will not cause death or symptoms. The conclusion of their rational analysis was: 'One in three breast cancers detected in a population offered organized screening is overdiagnosed'. It is widely recognized that many cases of carcinoma in situ in the breast never develop into potentially lethal invasive disease, but it is less well appreciated that screening for breast cancer also leads to overdiagnosis of early invasive cancer. Non-progressing invasive cancers are actually quite common [4]:

If one were to accept Jorgensen's conclusions, to calculate a true mortality rate of Tabar's screened group [1], we should ask how many lethal cancer cases 'X' occurred without including the postulated overdiagnosed cases that would never have come to medical attention without screening. If 30% of the 1426 cases diagnosed in the ASP group were overdiagnosis cases, as Jorgensen [3] postulates, then 1426 equals 1.30 X. X is then calculated by dividing 1426 by 1.30 = 1097, which is the number of women who actually had lethal breast cancers that would progress rather than regress or remain asymptomatic. In this case, the 351 dead should be divided by the 1097 lethal cases of potentially progressive breast cancer to yield an adjusted cumulative mortality rate after 29 years of follow-up of 32%, which is much closer to the 35% death rate in Tabar's PSP group (no screening offered). We therefore conclude that comparing mortality rates between those assigned to mammograms and those not assigned may distort the apparent reduction in mortality to the degree that overdiagnosis spuriously increases the number of early-stage, non-lethal cancers in the screened group.

Kalager and colleagues in 2010 analyzed the mortality issue differently by comparing current mortality rates to historical rates in both screened and unscreened women [5]. They concluded that the availability of screening mammography was associated with some reduction in the rate of death from breast cancer, but only about one-third of the total reduction was due to the screening itself.

Regarding routine oophorectomy at the time of hysterectomy, the assumption that oophorectomy in postmenopausal women undergoing hysterectomy is harmless is seriously questioned both by earlier [6, 7] and recent data [8]. Castration of women at hysterectomy, just as castration of men, has serious consequences, apparently at every age, as confirmed by a recent analysis of the Nurses' Health Study population by Parker and colleagues. It is no longer routinely recommended by the American College of Obstetricians and Gynecologists [9]. Of the 122,700 married nurses who were 30-55 years old in 1976, by 2009 fully 41% had undergone a hysterectomy, in 56% with an ovariectomy. After 24 years of follow-up of 29,380 post-hysterectomized nurses, the death rates from all cancers was 4.06% in those who retained their ovaries and 5.04% in the 16,345 who underwent bilateral oophorectomy. For all causes, the death rate was 9.53% for the women who retained their ovaries and 11.96% for those with oophorectomy. The percentage comparisons similarly favored ovarian conservation for death from breast, lung and colorectal cancers as well as for death from coronary heart disease and stroke. Only ovarian cancer death (at one-quarter of one percent) was lower in those without ovaries. But ovarian cancer was the lowest cause of death compared to every single other disease. Lacking familial risk, it appears not justifiable, given such data, to routinely oophorectomize all women at hysterectomy.