Cancer mortality trends in the EU and acceding countries up to 2015

M. J. Quinn1,+, A. d’Onofrio2, B. Møller3, R. Black4, C. Martinez-Garcia5, H. Møller6, M. Rahu7, C. Robertson8, L. J. Schouten9, C. La Vecchia10 and P. Boyle11

1 National Cancer Intelligence Centre, Office for National Statistics, London, UK; 2 Division of Epidemiology & Biostatistics, European Institute of Oncology, Milan, Italy; 3 Cancer Registry of Norway, Oslo, Norway; 4 Information & Statistics Division, NHS in Scotland, Edinburgh, UK; 5 Granada Cancer Registry, Spain; 6 Thames Cancer Registry, King’s College London, UK; 7 Department of Epidemiology & Biostatistics, Institute of Experimental & Clinical Medicine, Tallinn; and Estonian Centre of Excellence in Behavioural & Health Sciences, Estonia; 8 University of Strathclyde, Glasgow; and Scottish Centre for Infection & Environmental Health, UK; 9 Department of Epidemiology, Maastricht University, The Netherlands; 10 Mario Negri Institute of Pharmacological Research, and Institute of Medical Statistics, University of Milan; 11 Division of Epidemiology & Biostatistics, European Institute of Oncology, Milan, Italy

Received 23 May 2003; accepted 23 May 2003


    Abstract
 Top
 Abstract
 Introduction
 Patients and methods
 Results
 Discussion
 References
 
Background:

Examination of trends in cancer mortality in Europe over the past 30 years has shown that, after long-term rises, age-standardised mortality from most common cancer sites has fallen in the EU since the late 1980s. This study aimed to examine trends in the age-specific and age-standardised cancer mortality rates and numbers of cancer deaths up to 2020 for all cancers and various specific sites for all 15 EU countries, the 10 acceding countries, Bulgaria and Romania (currently applicant countries, along with Turkey), and Iceland, Norway and Switzerland of the four EEA countries.

Patients and methods:

Mortality rates were modelled as a function of age, calendar period and birth cohort. Birth cohort was calculated as age subtracted from calendar period.

Results:

As a consequence of the generally decreasing trends in the age-standardised rates, the best estimate is that there will be ~1.25 million cancer deaths in 2015, which is almost 130 000 (11%) more deaths than in 2000, but 155 000 (11%) fewer deaths than the 1.4 million projected in 2015 on the basis of demographic changes alone. The increases in the forecast numbers of cancer deaths in 2015 are proportionally larger in males than in females (13% and 10%, respectively) and proportionally larger in the acceding countries than in the current EU member countries (14% and 11%, respectively).

Conclusions:

Our forecasts are conservative best estimates of future cancer mortality. There is clearly scope for large improvements in survival, and hence reductions in cancer mortality, in some countries, through eliminating these differences using existing knowledge and treatment regimes.

Key words: cancer, Europe, forecasts, mortality, trends


    Introduction
 Top
 Abstract
 Introduction
 Patients and methods
 Results
 Discussion
 References
 
The European Code Against Cancer was originally produced and endorsed by the Committee of Cancer Experts in 1987. A revised version was adopted by the Committee in late 1994 [1]. The Code aims to reduce the cancer burden in Europe by issuing advice on the primary prevention and early detection of cancer. The aim of primary prevention is to reduce the incidence of cancer and hence also its mortality, while the aim of early detection of cancer is to reduce cancer mortality by improving survival.

At the time the Code was revised, an ambitious target reduction in cancer mortality was set for 2000 of 15%, compared with 1985, in all of the (then) 12 European Union (EU) countries. The actual overall decrease was 10% in men and 8% in women [2]. The target was met only in Austria and Finland in both men and women; in Luxembourg and the UK there were 15% reductions in men but not in women. In Greece and Portugal there were increases in the numbers of cancer deaths in both men and women.

The effect of the latest (third) version of the European Code Against Cancer [3] can also be measured by examining cancer mortality. Cancer incidence can measure the effect of cancer prevention—but not of early detection, because the latter may cause an increase in cancer incidence as has happened in several countries with breast screening using mammography. The principal aim of early detection is to improve survival. Cancer mortality reflects the combined effects of changes in incidence and survival.

Examination of trends in cancer mortality in Europe over the past 30 years [48] has shown that, after long-term increases, age-standardised mortality from most common cancer sites has fallen in the EU since the late 1980s. Overall, for males the decreases have been >10% for cancers of the lung, intestines and bladder, and ~5% for cancers of the oral cavity, and pharynx and oesophagus. For women, the decreases were >20% for cancers of the intestines and uterus (mainly cervix), and 7% for breast cancer. For both sexes, persisting declines have been observed of around 30% for stomach cancer and 10% for leukaemias. Mortality from other common neoplasms, including pancreas for both sexes, prostate and ovary, has been fairly stable. The major unfavourable trend has been in female lung cancer. In central and eastern Europe, however, cancer mortality tended to rise for most cancer sites, mainly in those related to tobacco, until the mid-1990s, and then levelled off or declined thereafter [8].

Even if age-specific cancer mortality rates were to remain unchanged in the future, the total number of cancer deaths would vary with changes in both the size and the age distribution of the population. Age-standardised mortality rates have to be used to adjust for these fluctuations. To measure the effect of the third version of the European Code Against Cancer on cancer mortality, reliable estimates of the number of cancer deaths are needed for the near future, taking the recent trends in mortality rates and projected populations into account.

Our study examined trends in the age-specific and age-standardised cancer mortality rates and numbers of cancer deaths up to 2020 for all cancers and various specific sites for all 15 EU countries, the 10 acceding countries, Bulgaria and Romania (currently applicant countries, along with Turkey), and Iceland, Norway and Switzerland of the four EEA countries. We used the most recent cancer mortality data available and proven statistical models. This paper covers results for only the EU and acceding countries up to 2015; a more extensive and detailed paper is in preparation.


    Patients and methods
 Top
 Abstract
 Introduction
 Patients and methods
 Results
 Discussion
 References
 
Mortality data were obtained from the WHO database [9]: data for long periods were available for all 15 EU countries and for five of the 10 acceding countries, including the Czech and Slovak Republics, for which data were combined to give the necessary time span of data for input to the statistical models (see below). Only recent data were available, however, for Estonia, Latvia, Lithuania and Slovenia (more historical data for the three Baltic countries and Slovenia exist and will become available shortly). No data were available for Cyprus. Population estimates in 5-year age groups from the 1950s up to 2000 for each country were also obtained from the WHO database [9]. Corresponding population projections up to 2020 were obtained from the UN database [10]. Mortality rates were modelled as a function of age, calendar period and birth cohort. Birth cohort was calculated as age subtracted from calendar period. Since the data were aggregated into 5-year age groups and 5-year calendar periods, the birth cohorts are synthetic and partly overlapping.

Historically, the age–period–cohort (APC) model [11] has been widely used for making cancer mortality predictions. The model can be written as: Rap = exp (Aa + D · p + Pp + Cc), where Rap is the mortality rate in age group a in calendar period p, Aa is the age component for age group a, D is the common drift parameter [12], Pp is the non-linear period component of period p, and Cc is the non-linear cohort component of cohort c. This Poisson regression model often gives a good fit when modelling cancer mortality rates. However, the multiplicative relationship between the rate and the covariates produces predictions in which the rates may grow exponentially with time. Another problem is that current trends extrapolate more reliably for the near future than for more distant periods. A third problem is that if there is a recent sharp change in the trends, the model projects only the average increase from the whole observation base. Following Møller et al. [13], these problems were handled in the following way. (i) To level off the exponential growth in the multiplicative model, a power link was used instead of the log link. (ii) Only 75% and 50% of the linear trend was projected for the second and third 5-year prediction periods, based on the belief that trends would eventually tend to flatten. (iii) To make the predictions more dynamic, if the rates displayed significant curvature in the prediction base, the trend in the last 10 years was used as the drift component to be projected.

An empirical evaluation [14] showed that all these modifications resulted in better predictions for cancer incidence in the Nordic countries in the period 1993–1997. The power model used in the Nordic incidence predictions has therefore also been used for the present mortality predictions: Rap = (Aa + D · p + Pp + Cc)5, where Rap, Aa, Pp and Cc are defined as in the multiplicative model.

Mortality rates were directly age-standardised using the World standard population [15]. The predicted numbers of cancer deaths in the five periods following the last observed year were calculated by first predicting the age-specific mortality rates for the relevant 5-year periods, multiplying them by the population projections for those periods, and then interpolating the numbers of deaths for individual years (e.g. 2015).

Predictions of age-specific cancer mortality were made up to 2020. But taking into account both the need for a planning horizon over which mortality trends might be influenced by actions taken as a result of the promulgation of the third version of the European Code Against Cancer [3], and the greater reliability of the predictions over a slightly shorter time interval, this paper concentrates on the results up to 2015.

The above methods were applied to data for all 15 EU member countries, and for five of the acceding countries. Projections of cancer mortality based on rates for 2000 were made for Estonia, Latvia, Lithuania and Slovenia, but forecasts using the above methods were not made, owing to the current lack of a sufficiently long-term series of data. The projected numbers of cancer deaths using the rates in 2000 were therefore also considered to be the forecast trends. There was a total of ~20 000 cancer deaths in these four countries in 2000, just over 10% of the total in the 10 acceding countries, and only 2% of the total in all 25 countries. The mortality rates for 2000 and the forecast mortality trends for Greece were applied to Cyprus; the inclusion of the resulting cancer mortality figures has negligible effect on the overall totals or trends.

Similar projections based on mortality rates in 2000 and forecasts using APC models were made for cancers of the lung, breast (in females), prostate, colorectal and stomach, for non-Hodgkin’s lymphoma, and for ‘all other’ cancers. These results are not presented in detail here.


    Results
 Top
 Abstract
 Introduction
 Patients and methods
 Results
 Discussion
 References
 
If the current age-specific cancer mortality rates remained unchanged in the future, the numbers of cancer deaths would increase substantially if the total number of people in the population increased, or if the demographic ageing of the population resulted in proportionally larger numbers of people in the older age groups where cancer incidence and mortality rates are highest. In fact, the total population in the 25 countries—currently just over 450 million—is projected to fall very slightly, by <2 million (<0.5%), by 2015. There will, however, be an enormous demographic shift towards the elderly, with >20% more people aged 65 years and over, and 50% more people aged 80 years and over, by 2015 (Table 1).


View this table:
[in this window]
[in a new window]
 
Table 1. Projected changes in populations in the EU and acceding countries from 2000 to 2015
 
The results from the forecasting of age-specific cancer mortality rates indicate that in most EU and acceding countries, the age-standardised rates would decrease quite markedly; both the timing and the extent of the decreases vary considerably among the countries. Figure 1 shows the trends for males; the overall pattern for females was closely similar. EU countries where the trends in all cancer deaths were forecast to be fairly stable included Denmark, Greece, Portugal, Spain and Sweden (although with one of the lowest rates) in males, and The Netherlands in females. Among the acceding countries, trends were predicted to remain flat in Hungary in males—at the highest level of all the 25 countries—and in Poland in females (Figure 1).



View larger version (24K):
[in this window]
[in a new window]
 
Figure 1. Cancer mortality trends in the EU and acceding countries, males. (A) Forecast trends decreasing. (B) Forecast trends stable. *Directly age-standardised using the World standard population.

 
Results for the individual cancer sites indicated that the overall trends were largely dependent on the decreasing rates in lung cancer mortality in males, and in breast cancer mortality in females (data not shown).

The overall projections of the numbers of cancer deaths, based on the rates in 2000, and on the forecast trends from the APC models for males and females from 2000 to 2015 are illustrated in Figure 2. The effect of the demographic shift towards the elderly clearly outweighs that of the forecast decreasing mortality rates (Figure 2).



View larger version (19K):
[in this window]
[in a new window]
 
Figure 2. Cancer deaths in the EU and acceding countries, 2000–2015.

 
The observed numbers of all cancer deaths in 2000, and the projected and forecast numbers in 2015, are shown, separately for the 15 current EU member countries combined and the 10 acceding countries combined, in Table 2. The total number of cancer deaths in the 25 countries in 2000 was ~1.1 million, with just over 0.6 million in males and just under 0.5 million in females. The large majority of these deaths (almost 85%) occurred in the 15 EU countries. On the basis of mortality rates in 2000, the expected number of cancer deaths in 2015 would be ~1.4 million, an increase of 25% or just over 280 000 deaths.


View this table:
[in this window]
[in a new window]
 
Table 2. Cancer deaths in the EU and acceding countries, 2000 and 2015 (thousands)
 
As a consequence of the generally decreasing trends in the age-standardised rates, the best estimate is that there will be ~1.25 million cancer deaths in 2015, which is almost 130 000 (11%) more deaths than in 2000, but 155 000 (11%) fewer deaths than the 1.4 million projected in 2015 on the basis of demographic changes alone (Table 2). The increases in the forecast numbers of cancer deaths in 2015 are proportionally larger in males than in females (13% and 10%, respectively) and proportionally larger in the acceding countries than in the current EU member countries (14% and 11%, respectively).

The largest contributions to the increased numbers of cancer deaths in the EU countries in both males and females in 2015 are forecast to occur in France, Germany, The Netherlands, and Spain—together accounting for >75 000 (around 75%) of the total increase of 100 000 deaths; there will be little or no increase in cancer deaths in either of the other two countries with large populations, Italy and the UK. In the acceding countries, the increase of around 15 000 cancer deaths in Poland accounts for almost 60% of the forecast total increase of around 26 000 deaths.


    Discussion
 Top
 Abstract
 Introduction
 Patients and methods
 Results
 Discussion
 References
 
The reliability of the projected numbers of deaths, using the mortality rates in 2000, and the forecast numbers from the APC models, depends heavily on the quality, consistency and comparability of the mortality data on which they are based. Quite apart from any problems of random variation, which are greater in smaller than in larger countries, there are complex issues about the process of death certification and the validity of the information in various countries [1620]. In general, for most common neoplasms, including lung, stomach, colorectal and breast, death certification is sufficiently reliable to permit meaningful inferences about the trends for most of the EU countries, as well as for the major central European countries, particularly for deaths below the age of 75 years. Greater caution is required for a few other countries—including new national entities, for which no comparison with previous calendar periods is available—and cancer sites for which diagnosis and certification may be substantially influenced by the availability of newer diagnostic techniques and/or by the accuracy of death certification. These factors may have had considerable influence on overall recorded cancer mortality in various parts of Europe.

The reliability of any forecasts of future trends based on historical mortality data also depends heavily on the appropriateness of the statistical model(s) used and the assumptions that are either made explicitly or are built into the statistical processes. We used the classical APC model but with a power link to avoid the problems of wildly (exponentially) growing rates, which may occur with the standard multiplicative model [11, 12]. In addition, as it is not realistic to assume that current trends will continue to the same extent throughout a future 20-year period, we applied a gradual reduction in the drift component after the first 5 years. The amount of the damping is to a certain degree arbitrary. This has, however, been shown—by application to older data on cancer incidence, for which the forecasts were compared with actual outcome [13, 14]—to produce more reliable forecasts. It is reasonable to assume that the equivalent adjustments we made to the statistical models would give similar results for cancer mortality as have been demonstrated for incidence.

Both the projections and forecasts of the number of cancer deaths up to 2015 are also highly dependent on the validity of the population projections. Major assumptions about birth rates, migration, and ‘all cause’ mortality have to be made in their production. For our purposes, however, the effects of the first two assumptions will be small, as <1% of cancer cases and deaths occur in children; and a large proportion of migrants are young adult ‘economic’ migrants, in whom cancer mortality is low.

We have not provided 95% confidence or prediction intervals around our forecasts of the number of cancer deaths, for three main reasons [13]. First, there are large uncertainties associated with the choice and specification of the statistical models. Secondly, the statistical variability in the forecasts depends heavily on the number of deaths on which they are based. Prediction intervals only include uncertainty about the parameters of the statistical models, which becomes very small when the numbers of deaths are large (as here for countries such as France, Germany, Italy, Poland, Spain and the UK, which have large populations), and the Poisson uncertainty in the future number of cases. Such intervals would give a false impression of the true range of possible future deaths because they do not include uncertainty about the continuation of current trends. Thirdly, the forecast number of deaths based on the age-specific trends will be affected by uncertainty in the ‘all cause’ mortality rates, which are built into the population projections.

Taking all of the above factors into account, our forecasts are conservative best estimates of future cancer mortality. There are, however, great uncertainties connected with forecasts of cancer rates, and they must be interpreted with caution.

There were 1.1 million cancer deaths in 2000. Our projections indicate that owing to a marked demographic shift towards the elderly, there would be 1.4 million such deaths in 2015 if the current (2000) mortality rates remained unchanged. If the numbers of cancer deaths in 2015 were to be no higher than in 2000, a reduction of 20% would be required. Around half of this reduction is likely to occur given the forecast decreasing age-specific mortality trends in most of the EU and acceding countries with large populations. The remaining reduction will only be achieved if smoking prevention and cessation programmes are expanded and are effective; if screening programmes for breast and cervical cancer with quality control procedures in compliance with EU guidelines are quickly extended to countries and areas that currently do not have them; if screening for colorectal cancer is proven in various pilots and is effectively introduced across the EU; and if future advances are made in treatments that significantly improve survival in the major groups of cancers. In addition, the results from the EUROCARE studies of cancer survival in Europe [2123] indicate (despite the limitations of the studies and some doubts about the geographical representativeness of some of the results) that there are extremely wide differences in cancer survival across Europe. There is clearly scope for large improvements in survival, and hence reductions in cancer mortality, in some countries, through eliminating these differences using existing knowledge and treatment regimes.


    Footnotes
 
+ Correspondence to: Dr M. J. Quinn, National Cancer Intelligence Centre, Office for National Statistics, London SW1V 2QQ, UK. Tel: +44-20-7533-5257; Fax: +44-20-7533-5103; E-mail: mike.quinn{at}ons.gsi.gov.uk Back


    References
 Top
 Abstract
 Introduction
 Patients and methods
 Results
 Discussion
 References
 
1. Boyle P, Veronesi U, Tubiana M et al. European Code Against Cancer. Eur J Cancer 1995; 31A: 1395–1405.[CrossRef]

2. Boyle P, d’Onofrio A, Maisonneuve P et al. Measuring progress against cancer in Europe. Has the 15% decline targeted for 2000 come about? Ann Oncol 2003; 14: In press.

3. Boyle P, Autier P, Bartelink H et al. European Code Against Cancer and scientific justification: third version (2003). Ann Oncol 2003; 14: 973–1005.[Free Full Text]

4. Jensen OM, Estève J, Møller H, Renard H. Cancer in the European Community and its member states. Eur J Cancer 1990; 26: 1167–1256.[ISI][Medline]

5. La Vecchia C, Negri E, Levi F et al. Cancer mortality in Europe: effects of age, cohort of birth and period of death. Eur J Cancer 1998; 34: 118–141.[CrossRef][ISI][Medline]

6. Levi F, Lucchini F, Negri E et al. Cancer mortality in Europe 1990–1994, and an overview of trends from 1955 to 1994. Eur J Cancer 1999; 35: 1477–1516.[CrossRef][ISI][Medline]

7. Levi F, Lucchini F, Negri E, La Vecchia C. The decline in cancer mortality in the European Union, 1988–1996. Eur J Cancer 2000; 36: 1965–1968.[CrossRef][ISI][Medline]

8. Levi F, Lucchini F, Negri E et al. Mortality from major cancer sites in the European Union, 1955–1998. Ann Oncol 2003; 14: 490–495.[Abstract/Free Full Text]

9. World Health Organization Statistical Information System. WHO database of mortality and population estimates [on-line]; http://www3.who.int/whosis/mort (May 2003, date last accessed).

10. UN database of population projections [on-line]; http://esa.un.org/unpp (May 2003, date last accessed).

11. Osmond C. Using age, period and cohort models to estimate future mortality rates. Int J Epidemiol 1985; 14: 124–129.[Abstract]

12. Clayton D, Schifflers E. Models for temporal variation in cancer rates. II: Age–period–cohort models. Stat Med 1987; 6: 469–481.[ISI][Medline]

13. Møller B, Fekjær H, Hakulinen T et al. Prediction of cancer incidence in the Nordic countries up to the year 2020. Eur J Cancer Prev 2002; 11 (Suppl): S1–S96.[CrossRef]

14. Møller B, Fekjær H, Hakulinen T et al. Prediction of cancer incidence in the Nordic countries: Empirical comparison of different approaches. Stat Med 2003; In press.

15. Waterhouse JAH, Muir C, Correa P, Powell J (eds). Cancer Incidence in Five Continents, Vol III. IARC Scientific Publications No. 15. Lyon, France: IARC, 1976.

16. Coleman MP, Aylin P (eds). Death certification and mortality statistics: an international perspective. Studies on Medical and Population Subjects No. 64. London, UK: TSO 2000.

17. Garne JP, Aspegren K, Balldin G. Breast cancer as cause of death—a study over the validity of the officially registered cause of death in 2631 breast cancer patients dying in Malmö, Sweden 1964–1992. Acta Oncol 1996; 35: 671–675.[ISI][Medline]

18. Grulich AE, Swerdlow AJ, dos Santos Silva I, Beral V. Is the apparent rise in cancer mortality in the elderly real? Analysis of changes in certification and coding of cause of death in England and Wales, 1970–1990. Int J Cancer 1995; 63: 164–168.[ISI][Medline]

19. Percy CL, Muir CS. The international comparability of cancer mortality data: results of an international death certificate study. Am J Epidemiol 1989; 129: 934–946.[Abstract]

20. Percy CL, Miller BA, Gloeckler Ries LA. Effect of changes in cancer classification and the accuracy of cancer death certificates on trends in cancer mortality. Ann N Y Acad Sci 1990; 609: 87–97.[ISI][Medline]

21. Berrino F, Sant M, Verdecchia A, Capocaccia R (eds). Survival of Cancer Patients in Europe—The EUROCARE Study. IARC Scientific Publications No. 132. Lyon, France: IARC 1995.

22. Berrino F, Capocaccia R, Estève J et al. (eds). Survival of Cancer Patients in Europe—The EUROCARE Study II. IARC Scientific Publications No. 151. Lyon, France: IARC 1999.

23. Coebergh JWW, Sant M, Berrino F, Verdecchia A (eds). Survival of Adult Cancer Patients in Europe Diagnosed from 1979–1989: The EUROCARE II Study. Eur J Cancer 1998; 34: 2137–2278.[CrossRef][ISI]