Time trends of breast cancer survival in Europe in relation to incidence and mortality
Article first published online: 8 SEP 2006
Copyright © 2006 Wiley-Liss, Inc.
International Journal of Cancer
Volume 119, Issue 10, pages 2417–2422, 15 November 2006
How to Cite
Sant, M., Francisci, S., Capocaccia, R., Verdecchia, A., Allemani, C. and Berrino, F. (2006), Time trends of breast cancer survival in Europe in relation to incidence and mortality. Int. J. Cancer, 119: 2417–2422. doi: 10.1002/ijc.22160
- Issue published online: 26 SEP 2006
- Article first published online: 8 SEP 2006
- Manuscript Accepted: 8 MAY 2006
- Manuscript Received: 20 DEC 2005
- European Breast Cancer Network (EBCN). Grant Number: SPC.2002482
- Compagnia di San Paolo of Torino
- breast cancer;
- population registries;
- time trends;
Increasing breast cancer survival, observed in most western countries, is not easily interpreted: it could be due to better treatment, more effective treatment due to earlier diagnosis or simply lead-time bias. Increased diagnostic activity (e.g., screening) can inflate both incidence and survival. To understand interrelations between incidence, mortality and survival trends and their consequences, we analyzed survival trends in relation to mortality and incidence. Starting with observed survival from EUROCARE, mortality from WHO and using the MIAMOD method, we estimated breast cancer incidence trends from 1970 to 2005 in 10 European countries. To smooth out peaks in incidence and survival due to early diagnosis activity, survival trends were assumed similar to those observed by EUROCARE in 1983–1994. The following patterns emerged: (1) increasing survival with increasing incidence and declining or stable mortality (Sweden, Finland); (2) slight survival increase, marked incidence increase and slight mortality decrease (Denmark, the Netherlands and France); (3) increasing survival, marked decrease in mortality and tendency to incidence stabilization (UK); (4) marked survival increase, steady or decreasing mortality and moderate increases in incidence (Spain, Italy); (5) stable survival, increasing incidence and mortality (Estonia). In most countries survival increased, indicating a real advantage for patients when accompanied by decreasing or stable mortality, and attributable to improved cancer care (Sweden, UK, France, Italy and Spain). In Finland (with high survival), the Netherlands and Denmark, increasing mortality and incidence indicate increasing breast cancer risk, probably related to life-style factors. In Estonia, low and stable survival in the context of increasing incidence and mortality suggests inadequate care. © 2006 Wiley-Liss, Inc.
The risk of cancer at the population level is usually assessed by means of incidence and mortality statistics. The former are provided by cancer registries, which cover variable proportions of national populations; the latter are available for almost all countries from national statistics agencies.
For breast cancer, incidence rates are highly sensitive to the dissemination of early diagnosis techniques; and mass screening for breast cancer, set up in many European countries in the late 1980s and early 1990s, resulted in increased incidence particularly of early stage, in situ and microinvasive cancers.1, 2, 3, 4 Screening also improves outcomes for symptomatic cancers, which are diagnosed earlier and apparently treated more effectively in regions covered by screening than in those not covered.1, 4 A certain proportion of cases destined to remain clinically silent throughout life is also uncovered, again inflating survival and incidence (overdiagnosis). In the absence of treatment improvement, screening leads to earlier diagnosis with no advantage to the patient (lead-time bias). Breast cancer survival rates have been increasing wherever they have been collected and analyzed.5, 6
Breast cancer mortality is insensitive to these effects of screening since it reflects the risk in patients diagnosed over the period from death to 15 or 20 years earlier;7 although mortality statistics do reflect the overall effects of primary risk, diagnostic activity and treatment over that period.
From these considerations it is evident that simple inspection of incidence and mortality trends does not enable us to understand the population dynamics of breast cancer (i.e., interrelations between incidence, mortality and survival trends). In the absence of screening, stable or decreasing incidence with increasing survival and stable or decreasing mortality suggest a real advantage to patients; increasing or stable incidence with increasing mortality and stable survival suggest suboptimal care; increasing incidence and increasing or stable mortality suggest a risk increase even when survival is increasing and the need for etiological research, and prevention measures. However, the introduction of screening considerably complicates the interpretation of these trends for the reasons aforementioned.
In the light of these consideration, we performed the present study with the aims of (a) analyzing breast cancer survival trends across Europe, in relation to mortality and incidence trends, and (b) to estimate underlying risk trends for the disease, taking account of the transient effects of screening.
To accomplish these aims, we estimated time trends for breast cancer incidence, survival and mortality in selected European countries, using a model to remove the variable and temporary effects of screening and other changes in diagnostic activity, and thereby permit investigation of changing risks of breast cancer with time.
Data sources were EUROCARE survival data for breast cancer patients diagnosed from 1982 to 1994 in 10 European countries,8, 9, 10 and mortality data from the World Health Organization (WHO).11 We modeled secular incidence and mortality trends in these countries from 1970 right up to 2005 using an established statistical method.12
Material and methods
The modelling procedure consisted of 3 stages, each applied to the data of cancer registry areas. First, breast cancer relative survival data (from EUROCARE-3, see below) were used to estimate the parameters of a multiplicative survival model13 that included age and period of diagnosis as main effectors. The effector “diagnosis period” was modeled as linear to remove sudden increases in incidence due to increased diagnostic activity. Relative survival can be interpreted as the proportion of patients surviving at a given time after cancer diagnosis, after adjusting for the general mortality of the population of same age and sex. To render the data from the different countries comparable, the relative survival figures were age-standardized to the age distribution of the EUROCARE-3 study population.
Second, incidence trends were estimated, using a back-calculation method, from the observed mortality (WHO11) and the survival as modeled above.
Third, short-term projections of incidence and mortality up to the year 2005 were estimated assuming relative survival to increase at the same rate, in each country, as that observed during the EUROCARE-3 study period (1978–1994). To render the incidence and mortality trends from the different countries comparable, they were age-standardized (age range 0–84 years) to the age distribution of the European population.
The incidence and mortality trends and projections were estimated using the Mortality-Incidence Analysis Model (MIAMOD) software12 specifically developed to model chronic irreversible diseases, and relying on assumed mathematical relationships between mortality, incidence and survival, for each given birth cohort.
To test the goodness of fit and reliability of the MIAMOD estimates, the incidence and mortality estimates for each registry or country were compared with the observed figures from the WHO mortality database11 and the EUROCIM database,14 respectively. EUROCIM is a database containing cancer incidence and mortality data from European population-based cancer registries, developed and made available to cancer registries by the International Agency for Research on Cancer (IARC).
Figures 1 and 2 show, for Scotland and Sweden, the results of the analyses carried out to test the goodness of fit of the models. Specifically the figures compare the observed incidence (as reported by EUROCIM) and that estimated by MIAMOD. The figures also show comparisons of observed and modeled mortality. The results of similar analyses for the other countries are available from the authors on request. Satisfactory goodness of fit was obtained for most cancer registry areas, over the entire period of observed incidence. However, for England, the estimated incidence was slightly higher than that of the observed incidence over the period 1975–1990, after which estimated and observed incidences were concordant.
The breast cancer survival data obtained from EUROCARE8, 9, 10 consisted of cases diagnosed in 1978–1994 and followed up to the end of 1999. The data were available by cancer registry, age class (15–44, 45–54, 55–64, 65–74, 75+), follow-up time and period of diagnosis (1978–1982, 1983–1985, 1986–1988, 1989–1991, 1992–1994).
Each country contributed with a different period of diagnosis, depending on when registration started. For the survival trend analyses, we used the period 1983–1994, since this was the period for which most EUROCARE registries had complete survival data.
As indicated previously, population and mortality data by country, age and calendar year, for the period 1970–1999, were taken from the WHO mortality database.11
The following countries were included: Finland, Estonia, Sweden, Denmark, England, Scotland, France, Italy, Spain and the Netherlands. The criteria for inclusion were a sufficiently long observation period for survival (at least from 1983–1985), combined with a satisfactory goodness of fit of incidence and mortality estimates and projections. For countries with local registries (France, Italy, Spain, England, and the Netherlands), data on survival, incidence and mortality were pooled and considered as national estimates for the purposes of this study.
Figure 3 summarizes the time trends of 5-year overall relative survival from 1983 to 1994 by four 3-year intervals in the 10 European countries included in the study. Pooled survival of all 10 countries is also plotted. The survival data in Figure 3 and subsequent figures and tables are basically observed data from EUROCARE, although they are age-standardized and also adjusted to take account of general mortality in each age and sex-matched population (i.e., relative survival). From 1983–1985 to 1992–1994, relative survival increased in all countries, except Estonia where a small decrease was observed (−2%). Countries with low survival at the beginning of the study period generally had more marked survival increases with time than countries with high survival at the outset. By 1992–1994, 5-year relative survival had reached or exceeded 80% in Sweden, Finland, France, Italy and Spain; was between 75 and 78% in Denmark, Scotland, England and the Netherlands; and was 62% in Estonia.
Figure 4 shows time trends of incidence, mortality and survival for each country; the graphs on the left show estimated age-standardized incidence and mortality time trends per 100,000 from 1970 to 2005; the graphs on the right show overall age-standardized 5-year relative survival from 1983 to 1994 by four 3-year intervals. The top graphs (Fig. 4a) show trends in countries with highest incidence (Finland, Sweden, Denmark and the Netherlands). These countries were characterized by markedly increasing incidence, moderately increasing survival and stable (Denmark and the Netherlands), slightly increasing (Finland) or decreasing (Sweden) mortality.
The second row of graphs (Fig. 4b) shows trends for Estonia, the only country where survival did not increase. For Estonia, estimated incidence and mortality were initially lower than in other northern European countries, but increased markedly over the study period.
The third row (Fig. 4c) shows trends for the UK (England and Scotland). During the 1970s, the estimated UK incidence was lower and mortality higher than that in northern European countries. From the end of 1980s to beginning of 1990s, survival increased dramatically, modeled mortality decreased markedly and estimated incidence leveled off and started to decrease. Over the EUROCARE study period (1983–1994), relative survival increased considerably (+16%).
Figure 4d shows breast cancer trends in the southern–western European countries of France, Italy and Spain, where the estimated incidence was relatively low in the 1970s, but increased markedly in subsequent decades. The estimated mortality also increased, but in more recent decades leveled off and has now started to decline. Trends in France are similar to those in Finland, with relatively high estimated incidence and a moderate increase in relative survival. By contrast, Italy and Spain started from low estimated incidence rates which increased markedly up to the mid 1990s, and were accompanied by steep increases in relative survival.
Table I shows estimated time trends by country for the most recent period (2001–2005). The estimates assume that survival increased at the same rate as observed over the EUROCARE study period (1983–1994). Specifically the table shows estimated age-standardized incidence and mortality in 2005 (per 100,000), 5-year age-standardized relative survival in 1992–1994 (latest available in EUROCARE-3), mean annual change in incidence and mortality between 2001 and 2005, and mean annual change in the relative excess risk (RER) of death. The last column shows the year in which organized breast cancer screening programme initiated, and indicates whether the programme was national or regional.
|Country||Estimated rate per 100,000 in 2005; Age 0–84 years||Five-year relative survival in 1992–1994 (EUROCARE)||Mean annual change 2001–2005 (%)1||Estimated annual change (%) in relative excess risk of death2||Start of organized screening3|
Incidence was estimated as high in Finland, Sweden, Denmark and the Netherlands (range 110/100,000 in Denmark to 138/100,000 in the Netherlands); mortality ranged from 21/100,000 (Sweden) to 33/100,000 (the Netherlands). Finland and Sweden had the highest survival, 82 and 83%, respectively. Trends were as follows: incidence was estimated to increase in the range 0.3%/year (Denmark) to 1.7%/year (Finland and the Netherlands); mortality was estimated to increase in Finland (0.2%/year) and decrease in the other 3 countries (range −0.3%/year to −0.5%/year). RERs of death decreased in the range −4% (Finland) to −2% (the Netherlands).
In Estonia, estimated incidence was 82/100,000 with estimated increase of 2.5%/year. Mortality was fairly high at 32/100,000 and estimated to increase at 2.5%/year. Survival was the lowest of the included countries (61%) and the RER of death was estimated to increase very slightly (0.4%/year).
In Scotland and England, estimated incidence was lower and mortality similar to that in the neighboring northern countries (Finland, Sweden, Denmark and Netherlands), while survival was much lower (74% in both Scotland and England). Estimated trends from 2001 to 2005 were: slight declines in incidence (by −0.9%/year and −0.6%/year, respectively) and clear decreases in mortality (−2%/year), with the RER of death decreasing markedly (−4%) in both.
With regard to the countries of southern–western Europe: Italy and Spain had low incidence, the lowest mortality of the countries in this study (both 19/100,000), and high survival (close to 80%). Incidence, survival and mortality data for France were similar to those of the northern countries. With regard to trends from 2001 to 2005: mortality was estimated to decrease, in the range −0.5%/year (France) to −2.6%/year (Italy); incidence was estimated to increase in France (1.3%/year) and to remain stable in Italy and Spain; the RER of death was estimated to decline considerably (from −2.8%/year in France to −8%/year in Spain).
In most of the countries included in this study, breast cancer survival increased from 1983 to 1994. The single exception was Estonia, where a 2% decrease was observed. In Sweden, England, Scotland, Italy and the Netherlands, the increase in survival was accompanied by a decrease in mortality, indicating a real improvement of prognosis, not due simply to lead-time bias.
In addition to the increase in survival, and well-known decrease in mortality starting in the early 1990s,15, 16, 17, 18, 19 we estimated, in England and Scotland, a tendency to decreased incidence starting in the 1980s.
In Sweden, survival rates were among the highest in Europe. The moderate increase in survival (in the context of already high survival at the beginning of the study period) was accompanied by a steady increase in incidence, and a decrease in mortality beginning in about the mid 1970s. For Sweden our model did not estimate a decrease in incidence, which would be expected some years after screening has become established. The continued increase in incidence, accompanied by decline in mortality, suggests that screening (which started in the mid 1980s) was effective in reducing mortality, but also that diagnostic activity led to the identification of a fraction of less aggressive or even non invasive cancers, as also suggested by the results of a recent Swedish study.20
In Finland and the Netherlands, national breast cancer screening was initiated in the late 1980s or early 1990s.1, 4 However, decreased mortality was subsequently observed only in the Netherlands. The continuous increase in incidence may be due to the same factors as operating in Sweden; while the simultaneous increase in mortality in Finland suggests a real increase in risk in this population. Note, however, that in the early 1970s, both incidence and mortality were much lower in Finland than in Sweden (incidence 53/100,000 and 75/100,000, respectively; mortality 20/100,000 and 26/100,000, respectively). Survival rates, that also in Finland were among the highest in Europe, increased remarkably up to the early 1990s and stagnated afterwards. In Denmark the increase in both incidence and mortality, with only a slight increase in survival (which remains lower than the other northern European countries and many western European countries) suggests a real risk increase, not balanced by a significant improvement in care for breast cancer.
The trends in survival and mortality that emerged for the UK are most likely related to national breast cancer screening programmes, which started in the late 1980s and the introduction of recommendations from the 1984–1985 world overview of systemic therapy.21
Population-based breast cancer screening has been shown to reduce breast cancer mortality.2, 3, 19 However, the introduction of screening does not entirely explain these time trends. It has been argued that much of the reduction in breast cancer mortality in western countries is attributable to the substantial advances in treatment that have occurred over the 30 years.22 The trends in survival and mortality that emerged for the UK are most likely related to national breast cancer screening programmes, which started in the late 1980s, and the 1984–1985 world overview of systemic therapy,21 with consequent introduction of recommendations, which are likely to have had an immediate impact on survival (extended) and mortality (reduced).
The EUROCARE studies found that UK breast cancer survival was low in the 1980s and early1990s, with cancers diagnosed at more advanced stage than in other countries of western and northern Europe5; a UK survey had substantially the same findings.23 Steps additional to screening were implemented in response to this low UK survival, including measures to improve care and treatment and reduce the time between symptoms onset and treatment.24 These steps were probably responsible for the decrease in mortality observed soon after the introduction of screening.16
The slightly declining UK incidence estimated by our model is apparently in contrast with the increases in incidence now being reported by most UK cancer registries following the introduction of mass screening.7, 25 Note, however, that our model was designed to smooth out the sudden peaks in incidence that occur soon after the introduction of screening (see Figs. 1 and 2) and thereby estimate an average time trend. Thus, the results of our study suggest that, in the UK, the real breast cancer risk is stabilizing even though the observed incidence is still increasing as a result of screening.
The large increases in breast cancer incidence found in Italy and Spain should be seen in relation to the low levels of incidence at the beginning of the study period, and may not be entirely due to increasing risk. Italy was characterized by declining mortality, also reported by other studies26, 27 and better survival than the European pool9, 10 over the entire EUROCARE study period, suggesting acceptable levels of care.
In Spain, despite a remarkable increase in survival, mortality increased markedly from 1970 to 2005 with most of this increase occurring before the 1980s, whereas mortality was estimated to decrease slightly from 2001 to 2005. These trends suggest a real increase in risk not offset by improvements in treatment or early diagnosis during the 1970s and 1980s. Changes in lifestyle and diet over time in Spain may be responsible for this increased breast cancer risk, which however may have declined in the most recent years. France too was characterized by a marked increase in incidence from 1970; however, the corresponding increase in mortality was slight, consistent with the results of a recent French Study.28 As in Italy, the negligible increase in mortality in France together with increased incidence and high survival suggest good levels of care.
Our model estimated increasing incidence in France and stabilizing incidence in Italy, although survival increased and mortality decreased in these countries, as it did in the UK. Since the model relies heavily on underlying assumptions of mortality and survival trends, one would expect decreasing or stabilizing incidence in France as well. This did not happen because the increase in survival was less marked (than in the UK), probably because of the high baseline survival, and the decrease in mortality was also not as marked.
In Spain our model estimated slightly decreasing incidence (less than estimated for the UK), while the mortality decrease was approximately half of that seen in the UK, and the increase in survival was higher than in all the countries analyzed. Note also that national screening had not been implemented in France, Italy or Spain over the study period29 and hence there could be no acute increase in survival as seen in the UK.
The situation uncovered in Estonia—increasing incidence and mortality, and stable survival—is rather worrying, and somewhat in contrast with the results of a recent study which reported a slowing of the increase in mortality in new EU member states of the former Eastern Bloc.30 Survival in Estonia was well below the European average. A study by our group indicated that Estonian breast cancer patients who survive 5 years or more after diagnosis are mainly those diagnosed with very early stage disease.5 This suggests that only patients with a favorable prognosis at diagnosis may have access to adequate care. The fact that survival decreased by 2% in the most recent period, while mortality increased steeply, suggests the access to adequate care worsened. This deterioration in outcomes is plausibly related to social and political changes beginning from 1991, which included the introduction of a market economy and an insurance-based health system.
It has been suggested that variations in the quality of case registration across European countries (particularly completeness) are so great as to render Europe wide comparisons problematic. However the impact on survival comparisons of variations in data quality can be estimated. The EUROCARE study showed that variable data quality only makes a small contribution to survival differences across Europe.8
To conclude, our study has estimated that the well-known geographic differences in the incidence of breast cancer are still present in 2005, with the lowest incidence in Spain and Italy, and highest incidence in the Netherlands, Denmark, Finland, Sweden and France. The high levels of incidence in these latter countries, despite the implementation and consolidation of early diagnostic activity in most, suggest that the most effective public heath interventions in breast cancer would be those that aim to reduce risk by encouraging changes in lifestyle.
- 8Survival of cancer patients in Europe: the EUROCARE study. Lyon: IARC Scientific Publications No. 132, 1995., , , , , .
- 9Survival of cancer patients in Europe: the EUROCARE-2 study. Lyon: IARC Scientific Publications No. 151, 1999., , , , , , , .
- 11WHO. Mortality. World Health Organization, 2005. Available at http://www3.who.int/whosis/menu.cfm
- 14European Incidence Database V2.3. ICD-10 dictionary. European Network of Cancer Registry, EUROCIM version 4.0. Lyon: IARC, 2001.
- 24Department Of Health. The NHS Cancer Plan. DOH, London, 2000. Available at http://www.doh.gov.uk
- 25Cancer trends in England and Wales 1950–1999. Studies on medical and population subjects No. 66. London: The Stationery Office, 2001., , , , .
- 29Breast cancer screening programmes in 22 countries: current policies, administration and guidelines. International Breast Cancer Screening Network (IBSN) and the European Network of Pilot Projects for Breast Cancer Screening. Int J Epidemiol 1998; 27: 735–42., , , , , , , , , .