Pitfalls in using case-control studies for the evaluation of the effectiveness of breast screening programmes

Because of socioeconomic and lifestyle characteristics, women not participating in mammography screening are at a greater risk of dying from breast cancer than participating women, and this risk difference is not related to screening participation. Many observational studies control for this self-selection bias using a method derived from pharmacological randomized trials. We assessed whether the correction for self-selection is valid when breast cancer mortality is decreasing because of reasons other than screening. We estimated the relative risk of breast cancer death in women likely and not likely to participate from a hypothetical cohort based on main Swedish randomized trials on mammography screening. We set the participation to 75%. We then applied the correction for self-selection on various scenarios of changes in breast cancer mortality because of screening and nonscreening factors. An average breast cancer mortality relative risk of 0.60 of participants versus nonparticipants was estimated before starting the screening. When no factor other than screening affected the risk of breast cancer death, the correction for self-selection led to the correct relative risk estimate. When mortality decreased because of reasons unrelated to screening, correction for self-selection failed to adjust for differences in risk between participants and nonparticipants, and the corrected relative risks suggested that decreases in breast cancer deaths were because of screening, even though screening had no effect on risks. The case-control design is not a valid method for the evaluation of the effectiveness of breast cancer screening when factors other than screening influence trends in breast cancer mortality.