Over-diagnosis in breast cancer screening

The article by Zackrisson et al is an important follow-up study of diagnosis of breast cancer in women in the Malmö mammographic screening trial.¹

In 2002 the International Agency for Research on Cancer concluded that population mammographic screening for women aged 50-69 years reduces the mortality from breast cancer by about one third.² But screening can also lead to over-diagnosis and over-treatment.³

Over-diagnosis arises from two distinct phenomena: anticipation of diagnoses and excess diagnoses. Anticipation is the earlier diagnosis of cancers that would otherwise have become symptomatic and presented later: this phenomenon is both expected and desirable.

Excess diagnosis relates to cases detected through screening that would otherwise never have presented. A few may be false positive histological diagnoses, and some will arise when women are diagnosed by screening but then die shortly afterwards from other unrelated causes. Screening may also detect slow growing cancers that would not become symptomatic within the normal life expectancy.

When a new cohort of women first attends mammographic screening their incidence of breast cancer increases substantially compared with that of an unscreened cohort, owing to both anticipation and excess cases. During subsequent screens the incidence remains around 30% higher than before screening.⁴ Excess cases are more important here because anticipation will relate only to new tumours that have become detectable since the previous screen. Finally, when screening ceases at around age 64-69, the incidence returns to a lower than expected rate.

The most informative analysis in Zackrisson et al's study is the comparison of the cumulative incidence of breast cancer in the screened and the non-screened groups of women born between 1908 and 1922 and randomised between 1976 and 1978.¹ This is a mature cohort with follow-up to 2001 when about 60% of the women had died. The main finding is a 10% increase in the lifetime occurrence of breast cancer (including cancer in situ) in the screened group.

The study's low statistical power precludes an exact estimate of over-diagnosis (95% confidence limits around the 10% estimate are 1% and 18%). Because some women randomised to screening were not screened and some women in the control group were, the intention to screen analysis leads to somewhat conservative estimates of over-diagnosis and of the reduction in breast cancer mortality (around 17%).

To put these numbers into perspective, let us for simplicity assume that they are both correct. In a population where the lifetime risk of breast cancer is 8% and the lifetime risk of dying from breast cancer from age 50 onwards is 2.5%, screening 250 women may prevent about one death from breast cancer. Screening would, however, also lead to the over-diagnosis of two cases. The woman whose death from breast cancer is prevented receives all the important benefit, whereas the two over-diagnosed women pay part of the price by becoming breast cancer patients and undergoing treatment. We cannot predict, however, which three women these will be.

The trouble is that although we can easily calculate these or alternative numbers based on different sets of data and assumptions, we cannot determine who the three women are. Ideally we should try to identify prognostic factors to distinguish the over-diagnosed cases and reduce the aggressiveness of their treatment. The first step towards this is to appreciate the reality of over-diagnosis and its likely magnitude. Zackrisson et al's study should inspire similar estimations of over-diagnosis in other populations, not only for breast cancer but also for colorectal cancer, prostate cancer, and other cancers, where organised screening or other diagnostic tests are being introduced.