Background: Ongoing breast cancer screening programs can only be evaluated using observational study designs. Most studies have observed a reduction in breast cancer mortality, but design differences appear to have resulted in different estimates. Direct comparison of case-control and trial analyses gives more insight into this variation. Here, we performed case-control analyses within the randomized UK Age Trial.
Methods: The Age Trial assessed the effect of screening on breast cancer mortality in women ages 40-49 years. In our approach, case subjects were defined as breast cancer deaths between trial entry (1991-1997) and 2004. Women were ages 39-41 years at entry. For every case subject, five control subjects were selected. All case subjects were included in analyses of screening invitation (356 case subjects, 1,780 controls), whereas analyses of attendance were restricted to women invited to screening (105 case subjects, 525 age-matched controls). Odds ratios (OR) were estimated with conditional logistic regression. We used and compared two methods to correct for self-selection bias.
Results: Screening invitation resulted in a breast cancer mortality reduction of 17% (95% confidence interval [CI]: -36%, +6%), similar to trial results. Different exposure definitions and self-selection adjustments influenced the observed breast cancer mortality reduction. Depending on the method, "ever screened" appeared to be associated with a small reduction (OR: 0.86, 95% CI: 0.40, 1.89) or no reduction (OR: 1.02, 95% CI: 0.48, 2.14) using the two methods of correction. Recent attendance resulted in an adjusted mortality reduction of 36% (95% CI: -69%, +31%) or 45% (95% CI: -71%, +5%).
Conclusions: Observational studies, and particularly case-control studies, are an important monitoring tool for breast cancer screening programs. The focus should be on diminishing bias in observational studies and gaining a better understanding of the influence of study design on estimates of mortality reduction.