Background: Efficacy of breast screening may differ in practice from the results of randomized trials. We report one of the largest case-control evaluations of a screening service.
Methods: Subjects included 491 breast-cancer deaths affecting 45-80-year-old South Australian females during 2002-2005 (diagnosed after BreastScreen commencement) and 1,473 live controls (three per death) randomly selected from the State Electoral Roll after birth-date matching. Cancer Registry and BreastScreen records provided cancer and screening details. Risk estimates were calculated by BreastScreen participation, using conditional logistic regression. Interpretation was assisted by a population survey of risk factor prevalence by BreastScreen participation in 1,684 females aged > or =40 years.
Results: The relative odds (OR) (95% confidence limits) of breast-cancer death in BreastScreen participants compared with non-participants were 0.59 (0.47, 0.74). Compared with non-participants, the OR was 0.70 (0.47, 1.05) for women last screened through BreastScreen more than 3 years before diagnosis of the index case, and 0.57 (0.44, 0.72) for women screened more recently. The OR of 0.47 (0.34, 0.65) for women screened more frequently in the pre-diagnosis phase was lower than the 0.64 (0.50, 0.82) for other screened women. The overall OR of 0.59 approximated 0.70 when corrected for the screening self-selection bias observed in five randomized trials. However, multivariable analysis of survey data did not indicate a lower prevalence of breast-cancer risk factors among BreastScreen participants, suggesting that this correction may be inappropriate.
Conclusions: Participation in screening was associated with a breast-cancer mortality reduction of between 30 and 41%, depending on assumptions about screening self-selection bias. A downward mortality risk by recency of last screen prior to cancer diagnosis, and frequency of recent screening, is consistent with a screening effect.