Self-Selection Bias in Randomized and Observational Studies on Screening Mammography: A Quantitative Assessment

Philippe Autier International Prevention Research Institute (IPRI), Lyon, 69002, FranceCorrespondence: Philippe Autier, International Prevention Research Institute (iPRI), Lyon, 69002, France, Email [email protected]: Observational studies aimed at evaluating the effectiveness of screening mammography are prone to self-selection due to differences in personal characteristics between women attending and those not attending screening. A method based on a quantity Dr has been promoted to correct for this bias, Dr being the risk of breast cancer death in a group of women not attending screening compared to the risk of breast cancer death in a population without screening.Objective: To estimate the amount of self-selection in observational studies aimed at evaluating screening mammography effectiveness and to estimate Dr quantities needed to correct for this bias.Methods: A first step quantified self-selection and Dr quantities specific to Swedish randomized trials using the most recent publications. A second step estimated self-selection specific to cohort studies on screening mammography effectiveness using the relative risk of 0.54 for all-cause death from these studies and the relative risk of all-cause death of 0.98 reported in Swedish trials. Using self-selection estimated from cohort studies, the Dr quantity needed to correct observational studies on screening mammography effectiveness was estimated. In a last step, corrections for self-selection in observational studies on screening mammography were retrieved.Results: The self-selection bias was 2.10 in Swedish trials. Self-selection in cohort studies was computed as (0.98/0.54) = 1.78. The Dr quantity required to correct results of observational studies was 1.53. In 19 case-control and cohort studies on screening mammography effectiveness, the median Dr quantity used for correction purposes was 1.16 (IQR: 1.11– 1.28).Conclusion: Compared to women attending screening, the risk of breast cancer death was approximately two times greater in women not attending screening. This increased risk was independent of screening effects. Most observational studies have overestimated the effectiveness of screening mammography because they used Dr quantities that were too small to correct for self-selection.Plain Language Summary: Women attending and not attending mammography screening differ in several ways. Non-attending women have a higher risk of dying from breast cancer because they tend to be less health aware, more deprived, have more comorbidities, develop more aggressive breast cancer, and to be less compliant with therapies. This phenomenon is called self-selection. Consequently, observational studies (ie case-control and cohort studies) have nearly always found that women attending screening are at a lower risk of breast cancer death than women not attending screening. In a previous publication, we showed that methods used to date to control self-selection removed only a fraction of this bias. The objective of this study was to quantify how much of the changes in the risk of breast cancer death reported by observational studies on mammography screening was due to self-selection bias. To this end, we used a method allowing us to estimate the amount of self-selection in populations where women are invited to screening. The method was based on the fact that screening mammography cannot influence causes of death other than breast cancer. Self-selection was first quantified using most recent results of Swedish randomized trials on screening mammography, and then in cohort studies that estimated the reduction in the risk of breast cancer death associated with attendance to screening mammography. Our study found that compared to women who attended screening, women who did not attend screening had an approximately 2-fold increased risk of breast cancer death. This increased risk was independent of screening effects on the risk of breast cancer death. Hence, observational studies conducted to date have overestimated the health benefits of mammography screening.Keywords: breast, cancer, screening, selection, bias