Complex statistical techniques cannot overcome weak methodology in the evaluation of breast cancer mortality trends

While underlying cause of death is the result of a standardized coding process, the mention of breast cancer as a cause of death, as used by Mukhtar et al.¹ is not. Further, due to lead time, it increases the numbers of women ‘dying’ from breast cancer, especially in older women, who are more likely to die of other causes.

In their join point regression analysis, mortality rates peaked in 1985 and declined prior to 1988, the year NHSBSP was initiated. Yet, in Supplementary Table 3 showing the overall reduction in average mortality, the pre- versus the post-screening decline is greater for the age group 50–64 than for all other age groups combined. This also is evident in age-specific mortality rates in England for years 1985–1994 and for the last year available (2010).²

In a trend analysis, mortality rates should always be considered together with incidence rates, which were increasing in England before and after the start of service screening, whatever the reason – as for example HRT use. Cohort effects, menopausal hormone therapy, greater surveillance for genetic breast cancer have influenced incidence rates and mortality rates. In 40–49-year-old women the effect of early diagnosis exists, especially in 45–49 (substantial numbers of women have their first NHSBSP screen before age 50 and many others have private screening). Remarkably, here visual inspection of the data is much more informative and does not support 1985 as the only or most important turning point, irrespective of the results of the join point analysis. The analysis of trends has well-known challenges and limitations.^2,3 Conclusions about the effectiveness of screening programmes should be supported by more valid approaches using individual data.

Competing interests

None declared