LAUSR

In their recent report, using a novel methodology, Tabar et al estimate the reduction in mortality from breast cancer among screened and unscreened women in Sweden. By looking at the incidence of fatal breast cancers over time and according to the screening status, they skirt the problems of sensitivity and overdiagnosis. They compare the incidence of fatal breast cancer in women who accepted an invitation for a screening mammogram and women who did not. On the basis of incidence rates of 62.3 per 100,000 in the unscreened population and 25.1 per 100,000 in the screened population, they conclude that screening “in the real world” reduces breast cancer mortality by 60%, which is even greater than the rate seen in randomized trials. We believe that the observed difference is not the result of screening but is due to an unrecognized selection bias. Women who attended mammography clinics were selected to have no cancer at the study onset, whereas women in the comparison group might well have been diagnosed with breast cancer in the past. Consider a women who has breast cancer and receives an invitation in the mail to attend a screening. It is expected that she will not go. However, her death would be considered an event in the comparison group (non-attendees) in Tabar et al’s study. Consider that there were 126,511 deaths from breast cancer in the Surveillance, Epidemiology, and End Results registry recorded between 2004 and 2014. Of these women, 88,322 (70%) were diagnosed before 2004. A similar analysis was reported by Coldman et al, who, using the same method, reported a 40% reduction of breast cancer mortality due to screening in Canada for women who attended screening, but it likewise was later shown to be erroneous.