Missclassification by exposure : a classic error in epidemiology
Analysing data on advanced breast cancer in the Netherlands, Autier et al. concluded that the “Dutch national mammography screening programme would have had little influence on the decrease in breast cancer mortality observed over the past 24 years” . As mortality data were retrieved including 2013, the 24 years started in 1989.
It is difficult to assess the impact of a national screening programme because there is no clear non-screened control group. It is, therefore, reasonable to look for proxies. Nevertheless, one may question the methods used by Autier et al.
Implementation of breast cancer screening started for women aged 50-69 years in the first two Dutch municipalities in 1987 and the programme became nationwide in 1997. In addition, women aged 70–74 years were invited over the years 1998 to 2001 . Therefore, the first point is to define breast cancer data for women exposed and non-exposed to screening. The oldest women invited to screening came from the first two Dutch municipalities and were 69 years old on January 1, 1987. By January 1, 1998, all women aged 70 years and above had been invited to screening. This means that a large part of the national breast cancer mortality data for women aged 50 years and above from 1989 onwards came from birth cohorts never or only partly invited to screening. Consequently, it is highly imprecise to consider all Dutch breast cancer mortality data from 1989 to 2013 for women aged 50 years and above as deriving from women exposed to screening.
Although the conclusion of Autier et al. referred to a 24 years period, the possible impact of screening was studied only for the period 1995–2012 “because the analysis showed that the mortality started to drop in 1995”. Autier et al. assumed that a possible effect of screening on breast cancer mortality could be mediated via the prevention of the 2% increase in breast cancer mortality that would have occurred if trends before 1995 had persisted until 2010–2012 and a decrease in the incidence of stage 2–4 cancers. As the latter rate decreased only by 4% from 1989 to 2012 for all women aged 50 years and above, it was assumed that screening could have caused only a 5% (= (4% x 2/3) + 2%) reduction in breast cancer mortality. However, it is completely unrealistic to expect that the Dutch screening programme could have affected the incidence of stage 2–4 cancers for all women aged 50 years and above from 1989 to 2012, as a major part of incident cancers in this period came from women never invited for screening.
A key factor in assessing the effect of screening is that a clear distinction is made between women exposed and non-exposed. The Dutch data do allow for this, as screening started at well-defined points in time across the Dutch municipalities. Following each municipality from the starting point, Sankatsing et al. found that the breast cancer mortality was reduced by 30% in women aged 55-74 and by 34% in women aged 75–79 since the introduction of the Dutch screening programme .
Misclassification by exposure is a classic error in epidemiology, the study by Autier et al. is heavily affected by this error.
1 Autier P, Boniol M, Koechlin A, et al. Effectiveness of and overdiagnosis from mammography screening in the Netherlands: population based study. BMJ 2017;359:j5224.
2 Sankatsing VDV, van Ravesteyn NT, Heijnsdijk EAM, et al. The effect of population-based mammography screening in Dutch municipalities on breast cancer mortality: 20 years of follow-up. Int J Cancer 2017;141:671–7.
Competing interests: No competing interests