Our results indicate a significant decrease in BC mortality over the past five years, a nonsignificant slight increase in incidence and a stage shift towards early stages over the past fifteen years with some proportions in the range of the accepted levels given by EU guidelines.
Strengths and Limitations
This observational study was conducted in the population of Tyrol. Mortality and incidence data were collected on a population level. Mortality data were provided by Statistics Austria. The quality of death certificates was very important for the conclusions drawn. In general, mortality statistics in Austria have been of high quality for decades [
8]. Coding of cause of death is done according to international guidelines by specialists who attend international benchmarking exercises. As already stated above, death certificates are written by specially trained doctors.
Data on BC incidence on a population level are provided by the Cancer Registry of Tyrol, which is a member of IACR and whose data are published in Cancer Incidence in Five Continents, thus giving some evidence for good quality of incidence data [
9,
10,
15]. Figures on completeness of incidence data show that for BC, in the past decade the proportion of death certificate-notified cases was 3.2% and the proportion of death certificate only cases 1.4% [
7]. Both proportions allow the conclusion that completeness is good as compared to international data.
In addition, the proportion of cases with unknown or unspecified stage is less than 5% in age groups 40 to 69. When we analyse incidence data on a population level, we always encounter some cases that lack detailed information for various reasons. Since year of diagnosis 2004, the cancer registry includes a variable for mode of detection. However, the information is very incomplete, because in many cases we cannot obtain sufficient information from the hospital discharge records.
The model we fitted for analysis of the time trend shows very good model fit. This means we can trust the time trend parameters and can therefore draw reliable conclusions from the model. Moreover, the staging information used to describe stage shift should be reliable.
The main limitation is the lack of a screening database. Consequently, we do not have detailed information on screening performance parameters. Another weakness is that we have only some limited information on coverage from a micro-census performed in 2006/2007 and from a publication by Frede [
5]. Both sources indicate a coverage of 70%. In Catalonia, Spain, Baré et al [
16] reported also a coverage of 70% prior to introducing a screening programme and the authors investigated reasons for non-participation which can be very helpful in improving coverage. On the other side, it is known that self-reporting overestimates true coverage,[
4] and a more realistic estimation could be a coverage of about 50%. This would fit to first preliminary data from the organised programme in Tyrol (data not shown). The lack of information on coverage and, of course, also on many other screening details was one of the reasons for changing the screening system, because we are convinced that a detailed knowledge of screening parameters is essential to draw valid conclusions in future. For staging distribution, the only source of information is the Cancer Registry dataset, whose focus was not to obtain information on screening indices but to concentrate on cancer cases.
Time trend for mortality and incidence data, model fit
We applied an APC model that takes age, period and cohort effects into account and models time trends that differ from a linear trend. Such models are widely used in epidemiology, see for example [
17,
18]. Each of the models we applied for both mortality and incidence fits well on its own, and all parameters allowing judgment of model fit are reasonably good. Also, the graphs showing observed and predicted rates give additional evidence that the model describes the data very well and hence that we can rely on estimated parameters (graphs not shown). In summary, the time trends given by the models should adequately describe the situation we observe.
Concerning the decrease in BC incidence in recent years, Ravdin et al. [
19] hypothesized for the USA that the reduction in hormone replacement therapy (HRT) is the main cause of the rapid decrease in BC incidence seen in the USA from 2003 to 2004. Also in Tyrol, we observe a decrease in BC incidence only in age groups 50+, although the decrease is not as sharp as in the USA. According to local experts, it is likely that also in Tyrol, part of the decrease in breast cancer incidence in the age group 50+ between 2004 and 2006 is due to a reduction in HRT.
The main question remains whether the significant 26% reduction in BC mortality over the past five years as compared to 1992-1996 is associated with opportunistic mammography screening. Both randomised trials and data from population-based organised mammography screening programmes provide clear evidence that organised mammography screening can reduce BC mortality. This was also communicated at an IARC international expert conference [
20]. The extent of mortality reduction differs in detail, but in general is estimated to be between 20% and 25% [
21‐
32]. However, for population-based organised programmes it is necessary to distinguish between various factors influencing BC mortality [
33]. Some authors[
33] estimate that a great part of BC mortality reduction (approximately 2/3 of reduction in England and Wales) is related to improvements in therapy, mainly the introduction of tamoxifen. For the USA, Berry et al. [
34] found a range of 28% to 65% (median 46%) for the proportion of BC mortality reduction attributed to screening by modelling this proportion by seven independent investigators. For Tyrol, this would imply a mortality reduction of 7% to 17% attributable to screening. In addition, when comparing BC mortality trends between countries, stage distribution and differences in therapy also have to be discussed as factors influencing BC mortality at the population level.
Adjuvant therapy with tamoxifen was routinely introduced in Tyrol around 1985. We do not collect detailed information on BC therapy in the Cancer Registry, but we have an overall variable for adjuvant hormonal therapy. When we analysed the effect of adjuvant hormonal therapy in a COX model adjusted for age and stage, an overall effect of 13% was seen. With regard to time trend in survival rates, over the past fifteen years we observed an increase in relative five-year survival rates split by staging groups according to UICC (5% increase in stage I, 13% in stage II, and 5% in stage IV). Both observations are consistent with an estimated therapy effect on survival of between 10% and 15%, which is in line with the UK estimate [
21]. Furthermore, as compared with EU guidelines, we miss some of the accepted levels (coverage, proportion of small cancers, proportion of II+ cancers, proportion of node-negative cancers). In conclusion, we estimate that less than half of the mortality reduction should be due to screening. This would mean that the screening effect is less than 13% and that, consequently, the opportunistic screening programme does not realise the potential of organised programmes, namely a mortality reduction of 20% - 25%.
However, when we compare BC data for Tyrol with quality indicators for mammography screening, we must remember that the BC data we analysed included all BC cases diagnosed in the population of Tyrol, not only those detected by opportunistic mammography screening. For example, Paci et al. [
35] show a proportion of 53% for II+ breast cancer in the total population as opposed to 29% in the screen-detected subgroup.
Vutuc et al. recently analysed BC mortality in Austria [
6]. The authors argue that BC screening is a plausible explanation for BC mortality reduction and doubt that a change in screening policy (meaning changing from opportunistic screening to an organised programme) would significantly improve the situation in Austria. We agree that BC screening is indeed one possible explanation for BC mortality reduction. However, if we take into consideration the fact that we have no detailed information on diagnostic performance or coverage for opportunistic BC screening in Austria, we feel it is absolutely necessary that detailed information on mammography screening be collected, at least for several years. We need to know all the well-established quality indices for BC screening[
2] before we can draw a final conclusion on how to proceed with mammography screening in Austria.
Interestingly, the greatest reduction in BC mortality was observed in the age group 40 to 49. This differs somewhat from international data, where doubts still prevail on the efficacy of mammography screening in the age groups below 50, see for example [
36,
37]. Surrogate performance indicators like stage shift, cancer size less than 1 cm and proportion of node-negative cancer also showed a clear tendency towards better performance in the age group 40 to 49 as compared to the age groups 50 to 59 and 60 to 69. In addition, during the past decade, these indicators improved more quickly in the age group 40 to 49 (details not shown). One possible explanation is the wide-spread use of sonography as an adjunct to mammography in Tyrol. It has been shown by various authors that the additional use of sonography can improve cancer detection rates, especially in younger women and women with dense breasts. The relative percentage of carcinomas found in supplemental breast ultrasound examinations as a fraction of the total number of detected cancers was reported by four studies, with a mean percentage of 22.5% (15%-34%) [
38].
In opportunistic screening in Tyrol, sonography was offered to women with dense breasts (ACR density grades 3 and 4) and with inconclusive findings on mammography [
39]. In addition, women in the younger age groups are likely to go more frequently to their general practitioner or gynaecologist, which results in higher coverage by opportunistic screening [
3].
The discussion in the USA after publishing the revised recommendation by U.S. Preventive Services Task Force [
40,
41] shows that it is very challenging and hard to understand by women to remove a service that was recommended for several years. Without well founded data, we feel it is not justified to stop screening in age class 40-49. We are collecting detailed data and will evaluate the balance between goods and harms during the next years.
Some of the EU recommendations like double reading and making an appointment for mammography when inviting women will not be part of the organised programme in Tyrol. Thus, further investigation will be needed to prove whether mammography screening has an effect on BC mortality, even in the absence of these EU recommendations.