Modelling the effect of breast cancer screening on related mortality using French data

This study aimed at modelling the effect of organized breast cancer screening on mortality in France. It combined results from a Markov model for breast cancer progression, to predict number of cases by node status, and from relative survival analyses, to predict deaths. The method estimated the relative risk of mortality at 8 years, in women aged 50-69, between a population screened every two years and a reference population.

Analyses concerned cases diagnosed between 1990 and 1996, with a follow-up up to 2004 for the vital status. Markov models analysed data from 3 screening programs (300,000 mammographies) and took into account opportunistic screening among participants to avoid bias in parameter's estimates. We used survival data from cancers in the general population (n=918, 7 cancer registries) and from screened cancers (n=565, 3 cancer registries), after excluding a subgroup of screened cases with a particularly high survival. Sensitivity analyses were performed.

Markov model main analysis lacked of fit in two out of three districts. Fit was improved in stratified analyses by age or district, though some lack of fit persisted in two districts. Assuming 10% or 20% overdiagnosed screened cancers, mortality reduction was estimated as 23% (95% CI: 4, 38%) and 19% (CI: -3, 35%) respectively. Results were highly sensitive to the exclusion in the screened cancers survival analysis. Conversely, RR estimates varied moderately according to the Markov model parameters used (stratified by age or district).

The study aimed at estimating the effect of screening in a screened population compared to an unscreened control group. Such a control group does not exist in France, and we used a general population contaminated by opportunistic screening to provide a conservative estimate. Conservative choices were systematically adopted to avoid favourable estimates. A selection bias might however affect the estimates, though it should be moderate because extreme social classes are under-represented among participants. This modelling provided broad estimates for the effect of organized biennial screening in France in the early nineteen-nineties. Results will be strengthened with longer follow-up.