BACKGROUND: The presence of competing events poses challenges for the interpretation of effect estimates. In an analysis that censored individuals at death from causes other than breast cancer (BC), the hazard ratio (95% CI) of BC death was 0.78 (0.63 to 0.95) for continuing breast cancer screening vs. stopping among women aged 70-74 years, and was 1.00 (0.83-1.19) among women aged 75-84 years. However, censoring at death is only one possible way to deal with competing events.
OBJECTIVES: To estimate two causal estimands for the effect of mammography screening on BC deaths: (1) the total effect of BC screening on BC mortality, which quantifies the effect through all causal pathways, including those possibly mediated by any competing event and (2) the direct effect of BC screening on BC mortality, which quantifies the effect that is not mediated by competing events.
METHODS: We emulated a (hypothetical) target trial to compare the effect of two BC screening strategies (continue annual screening vs. stop screening) on the 8-year risk of BC mortality in 1,058,013 Medicare beneficiaries who were 70 to 84 years, had a life expectancy of at least 10 years, no previous diagnosis of BC, and had received screening at baseline between 1999-2008. We censored women when they deviated from their assigned strategy and adjusted for time-varying confounders via inverse probability weighting. We estimated the total effect of "continue screening" vs. "stop screening" as the difference in the 8-year risks (cumulative incidences) of BC mortality, that is, deaths from other causes were included in the denominator of the risk (cumulative incidence). We estimated the direct effect as the difference in the 8-year risks after censoring women when they died for causes other than BC and estimating additional inverse probability weights for this censoring.
RESULTS: During follow-up, there were 61,586 deaths (1533 due to BC, 2.5%) under the "continue screening" strategy and 76023 deaths (1304 due to BC, 1.7%) under the "stop screening" strategy. Among women aged 70 to 74 years, the total and direct effects were -1.01 (-2.3 to 0.09) and -1.17 (-2.5 to 0.01) BC deaths per 1000 women respectively. Among women aged 75-84 years, the total and direct effects were 0.11 (-0.93 to 1.3) and 0.022 (-0.99 to 1.1) BC deaths per 1000 women respectively.
CONCLUSIONS: The estimates of the total effect of BC screening on BC mortality were similar to the estimates of the direct effect in each age group. This finding is consistent with the hypothesis that differences in the effect of BC screening by age group are not explained by competing events.
