Cancer mortality in the 50-69 year age group before and after screening

In our paper, ¹ we asked the simple question: did breast cancer mortality decline more in the screening epoch in the age group 50-69 than in other age groups not invited to screening? In the 50-69 age group, mortality was 98 per 100,000 person-years in 1974-1988, and 71 per 100,000 in 1995-2004. The corresponding figures for the age group 70+ were 171 and 172 per 100,000. The relative risk of mortality for the 50-69 age group relative to other ages was therefore 0.72 (71/98 - 172/171). Relative to the age group <50, the risk was 0.81. The overall age-adjusted relative risk from the Poisson regression was 0.72. That is, there was a 28% greater reduction in mortality in the 50-69 age group. Professor Gøtzsche and colleagues ² question our inclusion of the 65-69-year age group, when screening only began in this age group in the early 2000s. They seem not to understand that screening does not prevent breast cancer deaths instantly, but some years after the screening takes place, and so also will prevent some deaths in the years above the age range for screening.

We did not claim that there has not been a reduction in mortality in women aged 40-49, nor did we claim that there were no reductions in mortality before the screening programme could have had an effect. These are evident and likely due to improvements in therapy, but they do not rule out a further effect of screening. The analysis of Gøtzsche and colleagues considers only the single year of highest mortality and the single year of lowest, which is unreliable. Our analysis included all years from 1974 to 2004, which is more correct.

Ductal carcinoma in situ and ‘overdiagnosis’

Our estimate from the Two-County trial showed that the ratio of number of breast cancer deaths prevented to overdiagnosed cases including ductal carcinoma in situ (DCIS) was more than 2:1, the same qualitative conclusion as our analysis of the UK programme. We did not have English data on DCIS in the prescreening epoch with which to estimate the trends for the analysis, but on the basis of the incidence of DCIS in the screening epoch, even if we assumed that a DCIS case was four times more likely to be overdiagnosed than an invasive case, the deaths prevented in the programme in England would still outnumber overdiagnosed cases. Gøtzsche and colleagues also incorrectly state that we assume that all incidence increase with screening represents earlier diagnosis. After taking account of trends in incidence not attributable to screening, we estimated that most of such an increase represents earlier diagnosis and that a small proportion is due to overdiagnosis. There was indeed a deficit in incidence in comparison to expected numbers in women above the age range for screening, contrary to the assertion of Gøtzsche and colleagues.

We have already pointed out the illogical argument of Jørgensen and Gøtzsche ³ on overdiagnosis. Their 57% estimate would imply that 36% of cancers in the relevant age range were overdiagnosed in the 1990s. We noted that in the earlier years of the UK screening programme, only around 37% of tumours were screen-detected, with the remaining 63% divided between interval cancers, cancers diagnosed in non-attenders and cancers in women yet to receive their first invitation. This is well documented,^{4, 5} but Jørgensen and Gøtzsche seem unwilling to accept it. According to their estimate of overdiagnosis, therefore, almost all breast cancer cases diagnosed by screening were overdiagnosed, which is clearly false. Gøtzsche and colleagues ² attempt to counter this argument by asserting that ‘Duffy et al. claim that only 37% of breast cancers are screen-detected’, which we did not, and by quoting two websites to the effect that in 2006, 68% of breast cancers in the relevant age group were screen-detected. As stated above, we quoted the well-documented fact that in the early years of the UK programme, this was the case. In the second place, the proportion in 2006 is irrelevant to Jørgensen and Gøtzsche's assertion ³ which relates to the 1990s. Thirdly, the 68% refers to cancers only among those who actually attend for screening. In total, 25-30% of women in the invited age range do not attend. In fact, even in more recent years, ⁴ the total proportion of screen-detected cancers is considerably lower than asserted by Gøtzsche and colleagues. ² In scientific terms, ‘it is well known that…’ is not a particularly persuasive argument. Thus, we see no reason to abandon our estimates of overdiagnosis in favour of the less than credible claims of Jørgensen and Gøtzsche. ³

Trial data are empirical

Gøtzsche et al. ² say that we ‘seem to imply that the trial data are not empirical.’ Of course they are; it is therefore regrettable that the Cochrane review discards the empirical mortality results from the trials in favour of a conjectured mortality reduction of 15% which has no basis in empirical data. ⁶ This is one of the many reasons why we, and others, find the Cochrane review unconvincing.

Potential bias

Professor Gøtzsche has had longstanding negative views about screening, ⁸ predating his interest in mammography. We appreciate that secondary research inevitably contains a major interpretative component, but there appears to be a tendency on the part of Gøtzsche and colleagues towards preoccupation with what they consider SHOULD happen after the introduction of a breast cancer screening programme, based on their views and assumptions. This may be the reason for the factual errors in Gøtzsche and colleagues’ arguments about overdiagnosis, as noted above, and may explain the use of the unobserved 15% estimate of the mortality reduction in the Cochrane review, smaller than the 20% actually observed. We suggest that more emphasis be given to what actually DID happen. When we introduced a screening programme in England, breast cancer mortality fell by 28% more than it would have in the absence of screening (similar to that expected from the trial evidence), and the number of deaths prevented outnumbered the overdiagnosed cases by more than two to one.