Benefits and Harms of Mammography Screening

Magnus Løberg; Mette Lise Lousdal; Michael Bretthauer; Mette Kalager


Breast Cancer Res. 2015;17(63) 

In This Article


Mammography screening inevitably entails increased breast cancer incidence[36] due to earlier detection of cancers that would otherwise have been diagnosed later in life and due to diagnosis of cancers that would not have been identified clinically in someone's remaining lifetime. The latter category is commonly referred to as overdiagnosis. Theoretically, overdiagnosis can occur because the tumor lacks potential to progress to a clinical stage, or even regresses,[37] or because the woman dies from other causes before the breast cancer surfaces clinically. In reality, these three alternatives cannot be reliably disentangled. In any of the three scenarios the individual woman would be diagnosed and treated with no possible survival benefit. Hence, overdiagnosis represents a substantial ethical dilemma and burdens the patient and the health care system. Treatment for breast cancer includes surgery, radiotherapy, chemotherapy, and antiestrogen treatment. Risk of death from cardiovascular disease is increased in women treated with radiotherapy,[38] and adjuvant treatment may be cardiotoxic (for example, taxanes, anthracyclines, or trastuzumab).[39] It is possible that overtreatment causes increased mortality by other causes besides breast cancer. This may explain why there is no reduction in measurable overall mortality with screening mammography[9] (Figure 2).

Figure 2.

Scenarios for different outcomes of screening mammography. (A) Screening is ineffective. (B) Screening is effective. (C) Screening leads to overdiagnosis. (D) Screening leads to overdiagnosis that causes death from side effects of treatment.

Overdiagnosis does apply to both carcinoma in situ and invasive cancer; the lifetime risk of progression of carcinoma in situ to invasive breast cancer is unknown, but probably less than 50%;[40] and the lead-time is longer for in situ than invasive cancers. Thus, it is logical and intuitive that carcinoma in situ can be overdiagnosed. However, pathological verified invasive cancers can also be overdiagnosed. This contradicts what most clinicians were taught in medical school, and can be hard to understand for both clinicians and the public. One way of looking at this challenge is by using the 'iceberg model':[40] the development of cancer is a lengthy and complex process, where unrepaired genetic instability and changes in tumor microenvironment could lead to distinct, heterogeneous subpopulations of abnormal cells. Cancer can be envisioned as an iceberg of disease, where the visible tip above the waterline comprises the most aggressive lesions - those that produce symptoms and clinical disease. The majority of our body of knowledge concerning the natural history of malignancies comes from observations from these 'top-of-the-iceberg', symptomatic lesions above the waterline.[40] Underneath the water's surface, however, there might be multiple, indolent cancer subpopulations of cells. These subpopulations will look like cancer to the pathologist if detected through screening.[40] Early detection (such as mammography screening) dives under the surface and picks up silent lesions. The natural history of these asymptomatic lesions has not been studied and is therefore essentially unknown, but many of these may be indolent over time and never generate symptoms or disease without screening.

Estimates of Overdiagnosis

Precise estimation of overdiagnosis is a complicated and difficult task. There is no perfect analysis that would be universally applicable to this problem. Consequently, recent studies show a large variation in the estimated overdiagnosis of breast cancer, from none to 54%.[41] In studies based on statistical modeling to adjust for lead-time, estimates of overdiagnosis are consistently below 5%.[42,43] In contrast, observational studies have published higher estimates, between 22 and 54%,[37,41,42] depending on the use of the denominator.[44] In Table 2, we present the amount of overdiagnosis and reduction in mortality estimated with different denominators (incidence/death from breast cancer in different age groups). It clearly shows that different denominators (rows 2 to 4 in Table 2) result in different amounts of overdiagnosis and mortality reduction. Thus, it is important that benefits and harms of mammography screening are presented using similar denominators (in Table 2).

Overdiagnosis might be underestimated in the statistical modeling studies because they tested only one assumption at a time, based either on assumptions for the risk of progression from carcinoma in situ to invasive cancer,[42] or on sojourn time with adjustment for lead-time.[42,43] In statistical models based on sojourn time and lead-time, overdiagnosis has been disregarded in the estimation of lead-time, since the assumption of growth has been based on a progressive disease. This, however, is not the case for overdiagnosis where the disease is non-progressive or perhaps even regressive.[37] Thus, when using these estimates, overdiagnosis is likely to be underestimated.[48]

Since we do not have any direct, biological evidence of non-progression or regression of breast cancer, assumptions cannot easily be tested, and represent only a 'guess'. Evidence from observational studies is more convincing. The difference in the estimates from observational studies (22 to 54%) might be due to different assumptions of expected changes in breast cancer incidence due to changes in breast cancer risk factors, different follow-up time after introduction of screening, and differences in accounting for lead-time. After 25 years of follow-up, the Canadian National Breast Screening Study,[10] comparing physical breast examination with combined physical breast examination and annual mammography in women aged 40 to 59 years, found an excess of invasive cancer in the screening arm, resulting in 22% overdiagnosis. When the number of breast cancers detected at screening is used as the denominator (as in the Canadian study), the amount of overdiagnosis observed in the previous randomized trials is strikingly similar (22 to 24%)[10,49] and in line with the 30% reported in the Cochrane review of screening for breast cancer with mammography.[9] The amount of overdiagnosis might even be higher because DCIS, which accounts for one out of four breast cancers detected at mammography screening, was not included in these estimates.[10] If DCIS is a precursor of invasive breast cancer, we would expect a drop in incidence of invasive breast cancer after detection and removal of DCIS. There is no evidence for this. On the contrary, incidence rates keep increasing in countries with mammography screening.[50]

Given the uncertainty of the estimates from modeling and observational studies, we used the best available estimate of overdiagnosis from observational data from a US study where DCIS and invasive cancer were included, follow-up was more than 25 years after screening was initiated and no extensive untestable assumptions were made.[19] However, in the US there is no mammography screening program, and the rate of false positives is higher than in Europe and Australia. Thus, it might be possible that the amount of overdiagnosis differs between the US and Europe and Australia. Since none of the estimates of overdiagnosis from Europe or Australia were based on follow-up as long as in the US study, we choose to use the US estimate of 31% overdiagnosis (in line with what is observed in the randomized trials).[19] We estimated the number of overdiagnosed women based on the observed incidence of invasive breast cancer in women aged 50 to 69 years in the UK in 2007.[19,34,49] For 1,000 women invited to biennial mammography screening for 20 years from age 50, 15 will be overdiagnosed (Figure 1). Based on different meta-analyses and reviews of benefits and harms of mammography screening[9,22,32] and our best estimate,[19,34,35] we present a figure showing the different estimates of overdiagnosis and prevented deaths from breast cancer (Figure 3).

Figure 3.

Different estimates of overdiagnosed women and saved lives from breast cancer in different meta-analyses and trials. Euroscreen: estimates derived from a review of observational studies, where estimates of mortality reduction from case–control studies are included [32]. UK Independent review: estimates on relative effect derived from randomized trials of mammography screening and applied to UK national rates for women aged 55 to 79 years [22]. UK Observational: estimates based on 31% overdiagnosis [19] and 13 to 17% reduction in mortality from breast cancer [35] and applied to the observed incidence of invasive breast cancer (women aged 50 to 69 years) and mortality (women aged 55 to 74 years) in the UK in 2007 [34]; this resulted in 2 to 3 prevented deaths from breast cancer. Cochrane review: estimates from the randomized trials of mammography screening [9]. The Cochrane review does not assume the effect of mammography screening to last for 20 years as is assumed in the other estimates, but relates to what was observed in the randomized trials [9].

To be able to differentiate between potential lethal and non-lethal cancers, experimental studies have to be performed, preferably as an interdisciplinary cooperation between the biomedical and clinical communities. First, however, one has to accept that overdiagnosis does occur, and perhaps also change the terminology of non-lethal cancer to 'IDLE tumor' (InDolent Lesions of Epithelial origin), as recently suggested.[6]