Relation between breast cancer mortality and screening effectiveness: systematic review of the mammography trials

The mammography screening trials have shown varying results. This could be because screening was better in some trials than in others at advancing the time of diagnosis. If so, more cancers would be identified in such trials relative to the control group, and fewer of the cancers would have reached an advanced stage. I performed a systematic review of the mammography screening trials using metaregression. Finding many cancers was not related to the size of the reduction in breast cancer mortality (p = 0.19 after seven and p = 0.73 after 13 years of follow-up). In contrast, finding few cancers in stage II and above predicted a larger reduction in breast cancer mortality (p = 0.04 and p = 0.006). This expected association was also found for node-positive cancers (p = 0.008 and p = 0.04). However, a screening effectiveness of zero (same proportion of node-positive cancers in the screened group as in the control group) predicted a significant 16% reduction in breast cancer mortality after 13 years (95% confidence interval, 9% to 23% reduction). This can only occur if there is bias. Further analyses uncovered bias in both assessment of the cause of death and of the number of cancers in advanced stages. Consequently, the differences in the reported reductions in breast cancer mortality cannot be explained by differences in screening effectiveness. Given that the size of the bias was similar to the estimated screening effect, screening appeared ineffective.