Correction for Self-Selection in Breast Cancer Screening. Comment on Dibden et al. Worldwide Review and Meta-Analysis of Cohort Studies Measuring the Effect of Mammography Screening Programmes on Incidence-Based Breast Cancer Mortality. Cancers 2020, 12, 976

SIMPLE SUMMARY: Studies looking at the reduction of breast cancer deaths from screenings that are not randomized may suffer bias because of “self selection”. This occurs when even before screening takes place, people who choose to be screened are already at lower risk of dying than those who do not participate. A publication by Dibden et al. applied a correction to remove the effect of self-selection. There is concern, though, that because the correction was developed using data from a group of women in Sweden, where the healthcare system and people’s attitudes and behavior may not be the same, it may not be suitable when applied to studies on women in other countries such as Canada. In fact, the authors of the Canadian study checked for and found no evidence of self-selection bias. ABSTRACT: Observational studies of cancer screening are subject to bias associated with the self-selection of screening participants for whom the underlying probability of cancer death may be different from those who do not participate. Dibden et al. reviewed data on mortality reduction from 27 observational studies of mammography screening expressed in terms of relative risk for women who were screened versus not screened. Results were given, both unadjusted and after application of a correction for self-selection. The correction was based on a constant (1.17)—the ratio of risks of death in screening non-attenders versus those not invited, derived from a Swedish study. For some of the studies this correction had a large effect in diminishing the measured mortality benefit associated with screening. In particular, application to The Pan-Canadian Study of Mammography Screening, a study whose authors had previously tested for and found no evidence of self-selection bias, caused the estimated benefit to decrease from 40% to 10%. The appropriateness of applying a correction based on a constant to a population whose healthcare environment and screening participation rates are very different from those from which it was derived is questionable.