**4. Discussion**

The rationale underlying effective breast cancer screening is that the screening intervention provides lead time; cancers will be found earlier in the screened group. In the CNBSS trials, this would cause an initial excess of cancers in the MP arms and that excess would persist (a) as long as women in one arm were screened, while in the other they were not, or (b) where the lead time in one trial arm was greater than in the other. Both of these conditions would be expected to occur in CNBSS1 because of the use of only one initial physical examination in the UC arm. The second condition would be in effect in CNBSS2 where less lead time would be expected in the PE arm at each screen as compared to that which received mammography plus clinical examination. At some point, however, it would be expected that the cancer incidence in the control arms would begin to catch up and the excess would diminish. As seen in Figure 3, this is exactly what was observed for invasive cancers in the older women in CNBSS2, but this was not seen in CNBSS1. If the excess cancers in the MP group were due to overdetection during the screening period, it would be expected that the absolute excess should remain constant after the intervention. From the continuing increase in the excess in CNBSS1 with time, we can deduce that factors other than overdetection are at play.

The most obvious explanation is that more women with breast cancer or who were more likely to develop breast cancer were recruited into the MP trial arms. The first factor would explain part of the initial excess in cancers in the MP arms in both trials; the second could contribute to the failure of the UC group in CNBSS1 to "catch up" in the number of invasive cancers over successive years.

Another explanation is there was some causal factors associated with screening in the MP arms that would contribute to more cancers. For example, it has been suggested that the additional radiation exposure from X-rays or the compression of the breast in the mammography might be responsible [27]; however, no credible evidence has emerged to support these hypotheses [28–30].

Finally, the thoroughness of cancer ascertainment during the post-screening follow up could be different between the MP and control arms. This possibility was suggested by Baines et al. as a possible contributor to the overestimation of overdetection [21]. This is plausible, especially in CNBSS1, because, by having only a single initial screen, women in the UC group were likely to have much less interaction with the Study than those in the MP arm.

#### *4.1. Excess of In Situ Cancers*

The CNBSS publications presented data on the number of invasive plus in situ cancers found by screening and otherwise, but did not separately indicate the number of in situ cancers found during the period of screening [23,24]. On the other hand, in their overdiagnosis publication, Baines et al. did present both the total number of invasive cancers and invasive plus in situ cancers in each study arm for CNBSS1 and CNBSS2 at the end of the screening period and for times out to 20 years beyond that [21]. The curves in Figure 3 demonstrate the excess in cancer detection in the MP group over the control group in each trial. The differences between the two curves are also plotted and at Year 0 indicate the excess number of in situ cancers detected in the MP arms during the screening period. This was 33 cancers in CNBSS1 (10% of cancers) and 42 in CNBSS2 (11.1% of cancers).

Following the period of screening intervention, if further screening does not occur, the detection of in situ cancers in both arms should fall dramatically. Under these conditions, it is unlikely that cancers in the control arm corresponding to those in the mammography arm that were detected as in situ cancers (and created that excess) will be detected when they are in situ. If the cancers progress, it is more likely that they will subsequently be detected as invasive cancers during the post-screening follow-up of the control group. If they do not progress to become invasive, it is likely that they will never be detected if the control group remains unscreened after t = 0. In this latter case, they contribute to overdetection.

This may explain why, as seen in Figure 3, the curves for total cancers and for invasive cancers remain essentially parallel; the initial excess in in situ cancers remains constant up to 20 years post-screening. All other factors being equal, this would sugges<sup>t</sup> that many of the excess in situ cancers in the MP arms of CNBSS did not progress to become invasive and that much of the overdetection reported in that study is associated with in situ cancer. In situ cancer is not homogeneous [31] and it is possible that in CNBSS, particularly with concerns due to poor image quality, many of those detected were of the less aggressive phenotypes.

It is interesting that, from Baines et al., by subtracting the number of invasive cancers from the total number of cancers reported, it appears that over the 20 years after the screening intervention, only four and two additional in situ cancers were added in the MP and UC arms, respectively, to those detected during the screening period in CNBSS1. For CNBSS2, only a single in situ cancer was added in each arm over that period. This almost certainly implies that there was no effective post-intervention follow-up of in situ events, even though some degree of screening must have occurred in women in both arms during that time. Possibly cancer registries used for ascertaining cancers arising after t = 0 y simply did not record in situ cancers or possibly the overall ascertainment process had gaps.
