Skip navigation

Breast cancer screening


A number of readers have asked for Bandolier's views on the recent breast cancer screening furore, in which a meta-analysis [1] asks some pertinent questions about whether screening is justified. It is difficult to jump into a complicated area and provide all the answers, but the paper itself is worth reviewing because it makes one think.


The paper found eight randomised trials of breast screening with mammography. Authors of the original reports supplied more information when requested. In particular key questions were asked about whether the assignment methods were concealed so that no one could foresee whether the next woman would be screened or not. Again, randomisation should produce groups with identical characteristics, like age, so that the underlying risk of breast cancer was the same in screened and unscreened groups.


Out of the eight randomised trials examining the effects of screening on over 450,000 women, in six randomisation failed to produce similar groups. These six had other deficiencies, with four having women unaccountably missing from the analysis.

The two trials which were adequately randomised failed to show any difference for deaths from breast cancer between screened and unscreened women (Table). Screening 10,500 women for breast cancer would produce one extra breast cancer death, with confidence intervals from preventing one death in 2100 women to causing one death in 1500 women.

Table: Breast cancer deaths in screened and unscreened women
  Breast cancer deaths    
Trials Unscreened (%) Screened (%) Relative risk (95%CI) Number needed to screen (95%CI)
Adequately randomised 177/66150 (0.27) 183/66013 (0.28) 1.04 (0.85 to 1.25) -10570 (2138 to -1522)
Inadequately randomised 725/142052 (0.51) 654/182179 (0.36) 0.75 (0.67 to 0.83) 661 (506 to 950)
Pooling all trials 902/208157 (0.43) 837/248192 (0.34) 0.80 (0.73 to 0.88) 1040 (755 to 1672)

The other six trials were much more favourable to screening (Figure 1), and consistent in the extent of benefit. Combined they had a number needed to screen to prevent one breast cancer death of 661 (506 to 950). Pooling all eight trials produced a number needed to screen to prevent one breast cancer death of 1040 (755 to 1672).

Figure 1: Breast cancer deaths in screened and unscreened populations. Filled circles are properly randomised trials


The review makes an extremely good case that six of the breast cancer screening trials were flawed. In some cases, they seem so flawed as to be unreliable. This is particularly important when the absolute benefit of screening is so small. The percentage of women dying of breast cancer was always below 1% in the studies, and the difference in breast cancer death rates between studies (0.23 to 0.78%, average 0.43%) was many times greater than the difference effected by screening (-0.02 to 0.23%, average 0.1%). Figure 2 shows this graphically. It also suggests - it can be put no stronger than this - that reduced breast cancer deaths because of screening are only apparent when the baseline death rate is perhaps 0.4% or higher.

Figure 2: Reduction in breast cancer deaths and baseline risk of breast cancer death without screening. Filled circles are properly randomised trials

Giving the correct answer on breast screening is difficult. The evidence on overall reduction in death rate from randomised trials and from epidemiological studies falls short of showing much evidence of effect. The trials are suspect. The enthusiasts insist that the goalposts have moved, and that we may be playing a different game. The morbidity caused by numerous false positives remains relatively unstudied.

Perhaps the best we can do for now is to tell women the truth and let them decide. Tell them that one woman in every 1000 who undergoes breast screening may be prevented from dying from breast cancer, but there may be no benefit at all. Women should demand the best unbiased evidence on the benefits and harms of screening.


  1. PC Gøtzsche, O Olsen. Is screening for breast cancer with mammography justifiable? Lancet 2000 355: 129-34.
previous or next story in this issue