Determinants of the number of mammography units in 31 countries with significant mammography screening

In the 2000s, most of the female population of industrialised countries had access to mammography breast cancer screening, but with variable modalities among the countries. We assessed the number of mammography units (MUs) in 31 European, North American and Asian countries where significant mammography activity has existed for over 10 years, collecting data on the number of such units and of radiologists by contacting institutions in each country likely to provide the relevant information. Around 2004, there were 32 324 MU in 31 countries, the number per million women ranging from less than 25 in Turkey, Denmark, the Netherlands, the United Kingdom, Norway, Poland and Hungary to more than 80 in Cyprus, Italy, France, the United States and Austria. In a multivariate analysis, the number of MUs was positively associated with the number of radiologists (P=0.0081), the number of women (P=0.0023) and somewhat with the country surface area (P=0.077). There is considerable variation in the density of MU across countries and the number of MUs in service are often well above what would be necessary according to local screening recommendations. High number of MUs in some countries may have undesirable consequences, such as unnecessarily high screening frequency and decreased age at which screening is started.

Biennial mammography screening is considered to reduce breast cancer mortality by 25% in women aged 50-69 years (IARC, 2002). In women aged 40-49 years, annual screening seems to reduce breast cancer mortality by 15-17% (Moss et al, 2006). Since the beginning of this millennium, most women living in industrialised nations have had access to mammography screening. Therefore, for instance, in 2005, X70% of women aged 50-69 years participated in mammography screening in the Netherlands, France, Norway, the United Kingdom and the United States (OECD, 2007). However, there is considerable variation among countries (and sometimes also between counties or provinces) in mammography screening, including the age groups that are recommended for screening and those for which it is reimbursed by health insurance, and in the frequency of mammography (IARC, 2002;Lynge et al, 2003;Smith-Bindman et al, 2003;Yankaskas et al, 2004;USPSTF, 2008). Attendance can be by invitation from a screening programme, selfreference, a doctor's referral or through a combination of these three. Variation in all these factors may influence the number of mammography units (MUs) in countries. The objective of this study was to estimate the number of MUs in European, North American and Asian countries where significant mammography screening activity has existed for over 10 years.

MATERIALS AND METHODS
For 34 countries, using address lists obtained from the International Agency for Research on Cancer (IARC) and through internet searches, we gathered a list of potential sources of information. For some countries, the data were readily available in published reports or on websites; it was nonetheless verified through direct contact with the sources. We wrote to all potential sources of information we identified, asking for information on (i) the total number of MUs (analogic and digital) and (ii) the total number of radiologists, with numbers specialising in mammography.
The letter clearly stated that data sent to the IARC would be used to make a comparison among countries. If a contact could not provide relevant data, he or she was asked to provide us the details of an appropriate institution or to forward our letter directly to this institution.
We considered an MU to be any X-ray machine used for breast examination, through either analogical or digital modalities. As the same equipment could serve for both diagnosis and screening purposes, we made no distinction between MU declared as serving these purposes or reported as being part of a national screening programme or a medical facility (e.g., hospital, breast clinic, private radiology practice).
Between March and December 2006, we had contact with 229 potential sources of data, many of which forwarded our request to more appropriate data source (details can be obtained from the authors). We received data from 123 institutions or companies. When we obtained data from several sources for one country, we gave priority to radioprotection institutes, as registration of X-ray-emitting devises is compulsory in all countries. Sometimes, however, governmental radioprotection offices are established at a sub-national rather than at a national level precluding the identification of any single body having the relevant information for the entire country. When radioprotection institutes did not answer, or were not available at a national level, we turned to alternative sources of information. When several sources responded, we used the one most likely to be aware of MU in the country. Information from social security offices was usually not considered, as for these institutions, a clinic or a radiological facility is usually considered as a single 'mammography centre' although it may comprise more than one mammography unit. When dissimilar data from at least two a priori reliable sources were received for a country, we verified the information by re-sending the letter to these sources and, when possible, to other contacts. If for a country, no source of MU data was found, we used data from the European Coordination Committee of the Radiological and Electronical Industry (COCIR, 2003) or from the Organisation for Economic Cooperation and Development (OECD, 2007). If the number of radiologists in a country could not be obtained, we used data from the European Association of Radiology (EAR, 2005). Data selected for each country are listed in Table 1.
As only five countries (Finland, Ireland, the Netherlands, Sweden and the United Kingdom) provided separate counts of MU used in national screening programmes and in other medical The complete list of institutions contacted in each country and the 122 institutions or companies that sent data can be obtained from the authors. b Data obtained from OECD (2007).
Number of mammography units P Autier and DA Ouakrim facilities, we did not use these in our analysis. Some countries gave data on digital MU; given the rapid changes in digital mammography equipment during the 2000s, it was considered premature to provide these statistics. We collected information of country breast screening practice through literature search (e.g., Lynge et al, 2003;Yankaskas et al, 2004) and information gathered at the IARC. This information was not requested to institutions contacted for the number of MUs, as it was often known to be unavailable.
For each country, we computed the number of MUs divided by the number of women in 2005. The population data source was the Population Division of the Department of Economic and Social Affairs of the United Nations (ESA, 2007). For defining the number of MUs that would be necessary in a country, we took as a basis the Netherlands and the United Kingdom, two countries with national mammography screening programmes, where screening outside the national programme is rare and where a participation of at least 70% of the population was reached in 1995 in the United Kingdom (women aged 50 -64 years, triennial screening) (ACBCS, 2006) and in 1997 in the Netherlands (women aged 50 -69 years, biennial screening) (Otto et al, 2003). Computations in Table 2 are based on data from the Netherlands because triennial screening schedule exists only in the United Kingdom. We assumed three sets of recommendations: (i) biennial screening of women 50 -69 years old, (ii) annual screening for women aged 40 -49 years and of biennial screening at 50 -69 years and (iii) annual screening at 40 -69 years. The last scenario corresponds to recommendations made in the United States by the American Medical Association, the American College of Radiology and the American Cancer Society (USPSTF, 2008). In the first, second and third scenarios, about 20, 46 and 66 MU per million women would be necessary, respectively.
Using least square linear regression, we fitted a multivariate model for the prediction of the number of MUs according the to number of women of all ages, of radiologists and of country surface. We fitted another model for European Union Member States to examine the relationship between the number of MUs and the percentage of women who had a mammography in the last 12 months. The latter data were taken from a survey done in the European Union in 2006 that reported the percentages of women 50 years old and over who had a mammography examination in the last 12 months, regardless of whether it was for screening or for diagnostic purposes (Eurobarometer, 2007). The survey distinguished between examination done after receiving an invitation to attend the screening programme and that through woman's own initiative and that through a doctor's initiative.
This study has been approved by the Institutional Review Board of the IARC.

RESULTS
Of the 34 countries studied, we could not find data on the number of MUs in three and on the number of radiologists in seven countries. Data on the number of MUs were thus available for 31 countries, and data on the number of radiologist were available for 27 countries. Germany was the only country for which we could not obtain data more recent than 2001.
Around 2004, there were 32 324 MU in 31 countries where significant mammography screening was established. The number of MUs per million women ranged from 13 in Turkey to 100 in Austria (Table 3). There were less than 25 MU per million women in Turkey, Denmark, the Netherlands, the United Kingdom, Norway, Poland and Hungary, whereas there were more than 80 in Cyprus, Italy, France, the United States and Austria. Sixteen countries had more than 46 MU per million women, and seven had more than 66 MU per million women.
Acquisition of digital mammography equipments was most noticeable in Austria, Finland, France, Norway, Switzerland, Japan and the United States, but data are not shown as the change from analogical to digital mammography is now taking place rapidly in a number of countries.
Eleven countries reported the number of radiologists specialised in mammography examination (Table 3), ranging from 7% in South Korea to 62% in Canada. In spite of the great variability in the proportion of radiologists reported as being specialised in mammography examination, a positive correlation existed between the total number of radiologists and the number of radiologists specialised in mammography examination (Pearson r coefficient ¼ 0.80, P ¼ 0.0024). We then examined how female population size, the number of radiologists and country surface influenced the number of MUs by fitting a linear regression (Table 4). Both female population size and the number of radiologists predicted the number of MUs, whereas country surface was a less good predictor. More complex models, including for instance variables related to age groups being actually screened (when available) or population density, were not better predictors of the number of MUs.
In Member States of the European Union, the number of MUs was a good predictor of attendance to mammography screening when attendance was due to self-reference or due to doctor's prescription, but not after invitation by a breast cancer screening programme (Figure 1).

DISCUSSION
This study shows the considerable variability in density of MU across countries, and the number of MUs in service often exceeds what would be necessary to fulfil local screening recommendations. Country-specific volumes of breast cancer screening activities were not examined because reliable quantitative data were not generally available (Lynge et al, 2003;Yankaskas et al, 2004). Similarly, age at screening and screening frequency could not be included in regression models. A strong discrepancy often exists between recommendations and actual practice. For instance, in France, biennial screening is recommended for women aged 50-74 years, whereas as many as 60% of French women aged 40-49 years reported at least one recent screening (Spyckerelle et al, 2002). Furthermore, recommendations may differ within the same country; according to health organisation in the United States, seven bodies have issued different recommendations on age and frequency of screening (USPSTF, 2008). The few data we had on the number of radiologists specialised in mammography examinations suggested that the total radiologists registered in a country could represent a reasonable approximation to those specialising in mammography. But the variability in radiologists specialising between countries probably reflects differences in what this entails. In some countries, geographical distances may lead to installation of more MUs for easier access to  , 2006), in 1997 in the Netherlands (women 50 -69 years old, biennial screening) (Otto et al, 2003) and in 2004 in Norway (women 50 -69 years old, biennial screening) (Vatten, 2007;Hofvind et al, 2007). e Year of inventory not specified by data source and assumed as being data valid for 2005. screening. The multivariate model we fitted showed borderline statistical association between country surface and the number of MUs, once the number of radiologists and of women was taken into account. However, similar densities observed in countries much larger than the Netherlands, Norway and the United Kingdom indicate that geographical factors cannot account for all the difference in density of MU. Hence, all countries considered together, both total female population and the number of radiologists established in the country were the essential determinants of the number of MUs, irrespective of country size. Our data are more recent than the COCIR report (COCIR, 2003) and cover more countries than the OECD reports (OECD, 2007). Good agreement was found between our data and OECD data, except that Spain, for which the OECD admitted that their data could be underestimated (we received data from the Sociedad Espanola de Diagnostico por Imagen de la Mama, see Table 1), and for Korea, where the OECD got data from the Health Insurance Review Agency, whereas our data came from the Korean Association for Radiation Protection (Table 1), which was probably more reliable than the former.
Examination of MU density in relation to the most recent mortality data (Héry et al, 2008a, b) shows no evidence of a correlation. In fact, until the late 1990s, breast cancer mortality remained practically unchanged in some countries with a high MU density (e.g., Belgium, France), whereas it decreased substantially in several countries with a low density of MU (e.g., the United Kingdom, the Netherlands).
Coverage of the female population at ages 50-69 years was not achieved in Turkey and Denmark in 2003, though in Turkey, the number of MU may have been underestimated (Voyvoda et al, 2007). In Denmark, in 2003, mammography screening was offered to about 20% of women aged 50-69 years, and there was practically no provision outside the national programme (Jensen et al, 2004). A participation of the target population to the screening programme of at least 70% was reached in 1995 in the United Kingdom (women 50-64 years old, triennial screening) (ACBCS, 2006), in 1997 in the Netherlands (women 50-69 years old, biennial screening) (Otto et al, 2003), and in 2004 in Norway (women 50-69 years old, biennial screening) (Hofvind et al, 2007;Vatten, 2007). The main differences between these three countries and most other countries were the higher screening frequencies and broader age groups to whom screening was offered, by national programmes or by doctors.
Sixteen of the 31 countries included had more than 46 MU per million women and five have about twice this density. These data suggest that in many countries the number of MUs is well above what would be necessary according to local screening recommendations, and oversupply of MU may exist, peaking in France, Cyprus, the United States, Austria and Italy. An oversupply of MU may have undesirable consequences (Brown et al, 1990), which are listed below. (i) Insufficient experience of radiologists in the interpretation of mammograms for optimal sensitivity and specificity (Smith-Bindman et al, 2005;Théberge et al, 2005). (ii) The broadening of age ranges in which mammography is offered, mainly women less than 40 years old. For instance, in Germany, 18% of first mammographies were in women below 30 years and 31% were in women aged 30-39 years (Klug et al, 2005). In United States and in France, 47 and 45% respectively of first mammographies were in women below 40 years (Spyckerelle et al, 2002;Colbert et al, 2004). (iii) An increasing frequency of mammography. (iv) Increased costs of screening because of the necessity to amortise and to pay the running costs of mammography centres.
The enforcement of the Mammography Quality Standard Act in the United States in 1992 did not notably reduce the number of MUs, but probably led to the creation of mammography facilities that could better apply quality assurance requirements (Fischer et al, 1998;Destouet et al, 2005).
The European Guidelines for Quality Assurance in Breast Cancer Screening and Diagnosis exist since 1993 (Perry et al, 2006). There are no data on the likely impact of these guidelines on the installation of MU in European countries. An essential feature of the European guidelines not present in the United States is the recommendation to implement regular invitations to women for mammography screening to maximise participation and regularity. The positive correlation in Europe between the number of MUs per million women and self-referred or prescribed participation in mammography screening (and not after invitation) suggests that globally speaking, screening attendance in the European Union is not related to invitations by the programmes but rather to the offering of mammography screening, which is itself tightly related to the number of radiologists. In this respect, in high MU-density countries, the introduction of an invitation-only programme could not absorb and support the costs of the already functioning mammography services. In such cases, such an introduction would not, therefore, improve participation and reduce avoiding unnecessary screening, including outside the recommended age range.

ACKNOWLEDGEMENTS
Ph Autier conceived the project, wrote letters to potential sources of information, supervised the work on data collection, made the statistical analysis and wrote the article. D Ait Ouakrim organised the lists of addresses, collected the data and managed the database. % of women 50 years old and over who had a mammography in the last 12 months A B Figure 1 Relationship between the number of mammography units and the percentage of women 50 years old and more in 21 countries of the European Union reporting a mammography done in the last 12 months related to (A) an invitation to attend mammography screening (Pearson's r coefficient ¼ 0.06, P ¼ 0.82); (B) own desire to have a mammography screening or prescribed by a doctor (Pearson's r coefficient ¼ 0.58, P ¼ 0.0074). Data on mammography use from Eurobarometer (2007).