Screening status, tumour subtype, and breast cancer survival: a national population-based analysis.

PURPOSE

We examined whether demographic and tumour characteristics (including subtype) were different for women with breast cancer diagnosed via mammography screening compared with women with interval breast cancers, lapsed attenders of the screening programme and non-participants of the screening programme. In addition, we explored whether there were survival differences between the groups, taking into account lead time bias.

METHODS

We used linked data from National Cancer Registry Ireland and the national breast screening programme BreastCheck. Multinomial logistic regression was used to test the association of covariates with screening status. For survival analysis, we corrected the survival time for screen-detected cases for lead time bias, examined Kaplan-Meier curves and then used Cox regression to investigate differences in survival by screening status.

RESULTS

Subtype (HER2 over-expressing, triple negative), stage (III/IV), grade (poor), having co-morbidities, area of deprivation, smoking status and age were associated with having interval cancer or being a non-participant of the screening programme in the multivariable model. After correcting for lead time bias, and adjusting for variables associated with screening status, there was no evidence that risk of breast-cancer death for women with screen-detected cancer was different from women with interval cancer (HR = 0.76, 95% CI 0.56-1.03), non-participants (HR = 1.07, 95% CI 0.84-1.37) and lapsed attenders (HR = 0.97, 95% CI 0.65-1.45).

CONCLUSIONS

Screening status was strongly associated with subtype and this association persisted after adjustment for covariates including tumour stage and grade. After correcting for lead-time bias and adjusting for stage, subtype, grade and socio-demographic variables, no significant survival difference was demonstrated for women with screen-detected cancer in the 5-year period post-diagnosis. Since we are adjusting for stage, subtype and other variables, the lack of difference between these groups would be expected but has not been demonstrated in studies which do not correct for lead time bias.