Trends in testicular cancer incidence and mortality in 22 European countries: Continuing increases in incidence and declines in mortality

Authors


Abstract

This study profiles testicular cancer incidence and mortality across Europe, and the effects of age, period and generational influences, using age-period-cohort modeling. Despite a 5-fold variation in incidence rates, there were consistent mean increases in incidence in each of the 12 European countries studied, ranging from around 6% per annum (Spain and Slovenia) to 1–2% (Norway). In contrast, declines in testicular cancer mortality of 3–6% per annum were observed in the 1980s and 1990s for the majority of the 22 countries studied, particularly in Northern and Western Europe. The mortality trends in several European countries were rather stable (Romania and Bulgaria) or increasing (Portugal and Croatia). Short-term attenuations in increasing cohort-specific risk of incidence were indicated among men born between 1940 and 1945 in 7 European countries. In Switzerland, successive generations born from the mid 1960s may have experienced a steadily declining risk of disease occurrence. While the underlying risk factors responsible remain elusive, the temporal and geographical variability in incidence may point to an epidemic in different phases in different countries—the result of country-specific differences in the prevalence of one or several ubiquitous and highly prevalent environmental determinants of the disease. Advances in treatment have led to major declines in mortality in many European countries from the mid 1970s, which has translated to cohorts of men at successively lower risk of death from the disease. Slower progress in the delivery of optimal care is however evident from the mortality trends in several lower-resource countries in Southern and Eastern Europe. The first beneficiaries of therapy in these populations may be those men born—rather than diagnosed—in the era of major breakthrough in testicular cancer care. © 2006 Wiley-Liss, Inc.

Testicular cancer accounts for 1–3% of all cancers in males in Western countries, but is the most common malignancy among young men (aged 15–34 years) in most European populations.1 The highest incidence rates are recorded in a number of countries in Northern (Denmark, Norway), Central (Germany, Switzerland) and Eastern Europe (Czech Republic).2 Incidence trends in almost all European populations are characterized by rapid increases in the last few decades,3, 4, 5, 6, 7 particularly in adolescent men and young adults.8

The etiology of testicular cancer is not well understood, and the underlying reasons for the steadily increasing incidence trends throughout Europe are largely unknown. Improving ascertainment and better diagnostic procedures cannot account for the estimated 3–5% rises in incidence per annum, as the course of the disease is rapidly fatal if left untreated. In addition, the consistent evidence of uniformly rising secular trends comes from a number of well-established European cancer registries with standardized procedures.4

It has been hypothesized that the risk of testicular cancer is, to a large extent, determined very early in life, perhaps in utero.9 Several perinatal factors, including low birth weight,10, 11, 12, 13 older maternal age,12, 14 prematurity,10, 11, 14, 15 low birth order,10, 12, 13, 14, 16, 17 have been associated with an increased risk of testicular cancer, although the evidence is not entirely consistent across studies. Testicular cancer is consistently associated with cryptorchidism, the most common congenital malformation of the male genital organs.18 Results for perinatal risk factors have been often interpreted in the light of the so-called estrogen hypothesis, which postulates a carcinogenic effect of an excess of sex hormones at the time of testicular differentiation.19 Maternal life-styles during pregnancy could also affect testicular cancer risk. In particular, ecologic studies have identified maternal smoking as a possible risk factor,15 although this hypothesis has not found support from analytical studies.20, 21

In contrast to incidence, testicular cancer mortality has been markedly declining in a number of European countries since the mid 1970s, because of the introduction of platinum-based chemotherapy schemes22 and best-practice tumor management.23 Echoing these improvements, the pooled 5-year relative survival estimate among European patients diagnosed in the early 1990s was over 90%, although striking differences across Europe were observed, with 5-year survival as low as 71% in Estonia.24 The reductions in mortality have thus not been uniform between countries, with slower and later declines seen in lower resource settings,25 in accordance with the high cost of appropriate treatments, or inadequate patient referral systems.26

To describe the impact of testicular cancer across Europe, our study systematically assesses the effects of age, period and generational influences on time trends of testicular germ-cell cancer incidence in 12 countries and testicular cancer mortality in 22 countries. We contrast country-specific temporal patterns of incidence in light of the putative and known risk factors for the disease, and mortality trends, with particular reference to the introduction of effective treatment practices.

Data sources and methods

Incidence

Incident cases of testicular germ-cell cancer (ICD-O-2 9060-9102) and corresponding population datasets were extracted from the EUROCIM software package and database27 by registry, year of diagnosis and 5-year age group. A minimum requirement for a registry's inclusion in the analysis was their consecutive compilation in the last 3 volumes (6–8) of Cancer Incidence in Five Continents (CI5).1, 28, 29 This criterion was chosen as a general marker of each registry's data quality over time, given the editorial process involves a detailed assessment of the comparability, completeness and validity of the submitted incidence datasets. In addition, each dataset was required to span a minimum of 15 years to enable the fit of age-period-cohort (APC) models to 5-year time periods and 5-year age groups. Because of the computation difficulties in dealing with small numbers, Estonia and Iceland were not included in the analyses.

Table I provides details of the cancer registries included in the analysis of incidence trends. In France, Spain, Italy and Switzerland, a number of regional registries were aggregated to obtain an estimate of the national incidence. As the span of data available from regional registries varied, the aggregation maximized the registration period, while ensuring as many of the regional registries were involved in the national estimation.

Table I. Testicular Germ Cell Incidence: Populations Included in The Trends Analysis, Regular Trend, and Goodness-of-Fit Statistics for Best-Fitting APC Model by European Area
European areaCountryPeriod available (no. of 5-yearperiods)Incident cases1Male Person-years2ASR3Rank4Overall trend 1993–97 (95% confidence interval)5APC model6Residual deviance7d.f.7p-value7
  • 1

    Mean annual number of incidence cases 1993–97 in age group 15–54, except Czech Republic (1995–99).

  • 2

    Mean annual male population 1993–97 in age group 15–54, except Czech Republic (1995–99), expressed in million person-years at risk.

  • 3

    Truncated (ages 15–54) age standardised rate (TASR) 1993–97 (using European standard), except Czech Republic (1995–99).

  • 4

    Ranked in descending order of TASR.

  • 5

    Mean estimated annual percentage change based on the drift 1983–97 in age group 15–54, except Czech Republic (1985–99).

  • 6

    The most parsimonious final model providing a good fit over the whole period available. A, Age; AD, Age + Drift; AC, Age + Drift + Cohort; AP, Age + Drift + Period; APC, Age + Drift + Period + Cohort.

  • 7

    To determine the goodness-of-fit, the deviance was compared with the chi-squared distribution on the degrees of freedom (d.f.) determined by the model. A p-value of <0.05 denotes the full APC model does not yield an adequate fit.

  • 8

    Aggregation of England, Scotland.

  • 9

    Aggregation of Florence, Varese Province, Parma Province, Ragusa Province, Turin.

  • 10

    Aggregation of Catalonia, Tarragona; Granada, Murcia, Navarra, Zaragoza.

  • 11

    Aggregation of Bas-Rhin, Calvados, Doubs, Isere, Somme, Tarn.

  • 12

    Aggregation of Basel, Geneva, Neuchatel, St.Gall-Appenzell, Vaud, Zurich.

NorthernDenmark1979–1998 (4)2631.516.711.9 (1.0–2.7)AC10.1140.76
 Finland1955–1999 (9)611.54.2112.9 (1.1–4.9)APC30.7420.90
 Norway1953–1997 (9)1551.311.931.1 (0.1–2.3)APC41.7420.48
 Sweden1964–1998 (7)2012.48.292.6 (1.6–3.7)AC33.7350.53
 United Kingdom81978–1997 (4)135915.18.682.6 (2.2–3.0)AC15.0140.38
EasternCzech Republic1985–1999 (3)3273.110.844.0 (3.1–4.9)AC7.370.40
 Slovakia1968–1997 (6)1381.68.663.7 (2.4–5.1)APC28.0240.26
SouthernItaly91983–1997 (3)771.35.9101.2 (−0.5–3.0)A16.1160.45
 Slovenia1985–1999 (3)620.610.355.9 (3.6–8.4)AD15.4150.42
 Spain101983–1997 (3)310.93.2125.9 (2.6–9.7)AC11.170.13
WesternFrance111978–1997 (4)1031.28.671.9 (0.5–3.4)AC6.4140.96
 Switzerland121983–1997 (3)1400.815.921.2 (0.1–2.4)AC3.170.88

Mortality

Testicular cancer mortality data (ICD-9 186) were extracted from the WHO mortality databank by European country, year of death and 5-year age group (restricted again to men aged 15–54 years), alongside national population data from the same source. Two restrictions for inclusion were applied. First, as with incidence, datasets spanned at least 15 years, and second, the analysis was restricted to trends in mortality from 1968, in order to focus on how the effects of improving treatment, starting from 5 to 10 years later, subsequently impacted on the observed trends (see Assumptions on period slopes for mortality). Table II provides information on the national data from 22 countries that met the criteria: the time-span varied from 3–7 five-year periods. Because of their small numbers, Iceland, Luxembourg, Malta and Slovenia were not included in the subsequent analyses.

Table II. Testicular Cancer Mortality: Populations Included in The Trends Analysis, Regular Trend and Goodness-of-Fit Statistics for Best-Fitting APC Model by European Area
European areaCountryPeriod available (no. of 5-year periods)Deaths1Male Person-years2ASR3Rank4Overall trend 1980-5 (95% confidence interval)APC model6Residual deviance7d.f.7p-value7
  • 1

    Mean annual number of deaths in most recent 5-year period in age group 15–54.

  • 2

    Mean annual male population in most recent 5-year period in age group 15–54, expressed in million person-years at risk.

  • 3

    Truncated (ages 15–54) age standardised rate (TASR) in most recent 5-year period in age group 15–54 (using European standard).

  • 4

    Ranked in descending order of TASR.

  • 5

    Mean estimated annual percentage change based on the net drift in age group 15–54, in most recent two decades: Austria (1981–2000); Belgium (1981–1995); Bulgaria (1980–1999); Croatia (1986–2000); Czech Republic (1986–2000); Denmark (1984–1998); Finland (1980–1999); France (1984–1998); Germany (1985–1999); Greece (1984–1998); Hungary (1981–2000); Ireland (1984–1998); Italy (1984–1998); Norway (1984–1998); Poland (1982–1996); Portugal (1980–1999); Romania (1981–2000); Slovenia (1985–1999); Spain (1984–1998); Sweden (1984–1998); Switzerland (1980–1994); The Netherlands (1980–1999); Uniited Kingdom (1980–1999); mean estimated annual percentage change based on the net drift in most recent 20-year period.

  • 6

    Refers to the most parsimonious final model providing a good fit to the trends over the whole period available: A, Age; AD, Age + Drift; AC, Age + Drift + Cohort; AP, Age + Drift + Period; APC, Age + Drift + Period + Cohort.

  • 7

    To determine the goodness-of-fit, the deviance was compared with the chi-squared distribution on the degrees of freedom (d.f.) determined by the model. A p-value of <0.05 denotes the full APC model does not yield an adequate fit.

NorthernDenmark1969–1998 (6)131.50.86−3.4 (−5.8 to −0.6)APC31.4300.40
Finland1970–1999 (6)51.50.86−4.0 (−6.5 to −1.1)AP39.2420.59
Ireland1969–1998 (6)41.00.86−5.7 (−9.2 to −1.1)AP31.1420.89
Norway1969–1998 (6)61.30.67−4.6 (−8.0 to −0.1)AP47.8420.25
Sweden1969–1998 (6)72.40.58−5.1 (−8.1 to −1.3)AP40.1420.55
United Kingdom1970–1999 (6)6816.50.58−5.0 (−5.6 to −4.4)APC29.7240.19
EasternBulgaria1970–1999 (6)342.31.41−0.1 (−1.4 to 1.2)AP35.5350.45
Czech republic1986–2000 (3)393.11.32−3.0 (−4.5 to −1.4)AC8.770.27
Hungary1971–2000 (6)342.91.23−2.3 (−3.4 to −1.3)AC18.2280.92
Poland1980–1994 (3)11410.71.04−1.1 (−2.2 to 0.0)AD24.4150.06
Romania1981–2000 (4)486.50.95−0.1 (−1.2 to 1.0)A29.7240.19
SouthernCroatia1986–2000 (3)111.30.584.4 (−0.3 to 10.0)A13.0160.68
Greece1969–1998 (6)92.90.49−2.9 (−5.9 to 0.8)AC33.2280.23
Italy1969–1998 (6)4616.20.49−4.0 (−5.3 to −2.6)AP55.2420.08
Portugal1980–1999 (4)112.80.492.0 (−0.7 to 5.1)AC22.7140.06
Spain1974–1998 (5)2911.40.49−0.9 (−3.0 to 1.4)AD42.0310.09
WesternAustria1971–2000 (6)122.30.49−4.5 (−5.9 to −2.9)APC29.8300.48
Belgium1971–1995 (5)72.80.310−4.8 (−7.7 to −1.3)AD43.5390.28
France1969–1998 (6)7816.40.310−3.5 (−4.5 to −2.3)AP45.1420.34
Germany1985–1999 (3)14823.20.310−6.2 (−6.8 to −5.6)APC108.36<0.01
The Netherlands1970–1999 (6)224.70.310−3.2 (−4.5 to −1.8)APC32.1300.36
Switzerland1970–1994 (5)172.00.211−5.6 (−7.4 to −3.5)AP43.0350.17

APC model

The incidence and mortality data were tabulated as birth cohorts in 10-year intervals by subtracting the midpoints of 5-year age groups (15–19, 20–24, …, 50–54) from the corresponding 5-year periods, with each resulting cohort overlapping by exactly 5 years. We assumed that the rates were constant within 5-year age classes a = 1, 2, …, A and 5-year periods of diagnosis p = 1, 2, …, P, leading to a likelihood for the observations that is proportional to Poisson likelihood for the counts, with the log of the person-years of risk specified as an offset. The magnitude of the rates was described by a full APC model:

equation image

which can be fitted under the application of generalized linear model theory,30 with the birth cohort derived from period and age such that c = pa for c = 1, 2, …, C with C = A + P − 1. The parameters αa, βp and γc refer to the fixed effects of age group a, period p and birth cohort c. The models were fitted using Stata 8.31 Tests for the overall slope and separate effects of period and cohort curvature were obtained using the standard analysis of deviance of nested models, as suggested by Clayton and Schifflers.32, 33

To allow a systematic evaluation of the trends across countries, the results are presented using the full APC model, and the nonidentifiability problem was highlighted by partitioning the age, period and cohort effects in terms of their linear and curvature component parts, according to the method of Holford.34, 35 Holford showed that, while the overall slopes are unrestricted, they do not vary independently, given that the 3 linear slopes from an arbitrary APC model (indexed L) can be represented by αL = αL + ρL, βL = βL − ρ and γL = γL + ρ, where αL, βL and γL are the true values for the slopes according to age, period and cohort, and ρ is an unknown constant that may result in increasing or decreasing trends of each slope.34, 35 The drift, the sum of the period and cohort slopes, βL + γL, is therefore estimable35 and used in this study to describe the overall direction and magnitude of the time trend in each country. For incidence, the recent drift was estimated on the basis of the most recent 15 years of data available; for mortality, the drift was obtained from the most recent 20-year period, in order to capture the full impact of the introduction of successful therapy (see Assumptions on slopes for mortality). To identify plausible period and cohort effects in the incidence and mortality trends, we postulated different specifications of the range of the period and cohort slopes, as outlined here.

Assumptions on incidence slopes

For incidence, we took into consideration both the possibility of period-specific increases, as would be expected in the event of improving diagnostic procedures or increasing ascertainment with time, and the importance of birth cohort influences, possibly due to the changing prevalence and distribution of known and putative risk factors that impact on cancer rates in successive generations. The substantial contribution of cohort influences in explaining testicular cancer incidence trends in Europe has been consistently demonstrated in previous reports,3, 6, 36, 37 and thus, we a priori assumed that the overall linear slopes of period and birth cohort were positive, and specified scenarios for which the cohort component accounted were (i) all of the regular trend, and (ii) half of the regular trend. The possible values of the cohort slopes (γL) were thus bounded so that equation image, leaving the period slopes (βL) to range between zero and half of the net drift defining the corresponding linear slopes as equation image. Age parameters were similarly bound between 2 estimable functions.

Assumptions on slopes for mortality

For mortality, we postulated 2 specifications for the period slope that mirror those for incidence. The first scenario attempted to capture the period-related declines in testicular cancer mortality due to the introduction of effective therapy and care starting in the early to mid 1970s, initially in high-resource European countries. The second specification took account that the regular trend is related to the underlying incidence (and its generational influences), as well as case-fatality. On the basis of these requirements, we present 2 sets of parameter estimates that constrain the period component βL to take either (i) all of the regular trend or (ii) half of the regular trend. The boundary values of the period slopes were equation image, and accordingly, the range of linear slopes for cohort γL were equation image.

Each postulation of the period slope provided an identifiable range of the age and cohort slopes. The effects for the individual categories of each effect were generated by adding together the linear and curvature components. For example, the ath age effect can be expressed as αa = (a − (A + 1)/2) × αL + φa, with φa representing the departures from the linear trend, and βL and γL, the slopes for period and cohort, defined in the same way.35

Results

Background risk, age-adjusted trends and recent drift: incidence

During the most recent 5-year period available, there was a five-fold variation in incidence in the 12 European countries (in Table I), with rates ranging from around 3 per 100,000 in Spain through to more than 15 per 100,000 in Denmark and Switzerland. Increases in incidence during the period 1983–1997 were observed in all countries studied. The extent of the increase varied considerably, although no clear relation between the level of incidence and the magnitude of the recent trend was apparent (Fig. 1). The average increases per annum varied at least 6-fold (Table I), with the most rapid inclines in Spain and Slovenia, estimated to be almost 6% per year on average, compared with overall increases of 1–2% per annum in Norway, Switzerland, Italy, France and Denmark. There was a suggestion of a recent peak in several countries, most evidently in Switzerland and Norway, during the 1990s.

Figure 1.

Trends in truncated age-standardized (15–54 years) incidence and mortality rates (European standard) for selected countries. Rates are based on 5-year aggregates and corresponding to the period available, as described in Table I.

Background risk, age-adjusted trends and recent drift: mortality

The ratio of testicular cancer incidence to mortality ranged from 8:1 in the Czech Republic to over 30:1 in Switzerland, with a clustering of rates within region more apparent than that was noted for incidence (Table II). Death rates in the most recent 5-year period were generally highest in Eastern Europe, with Bulgaria, the Czech Republic, Hungary, Poland and Romania holding the top 5 positions, with high-risk Denmark in sixth place (Table II). In further contrast to incidence, decreases in testicular cancer mortality were observed in 19 of the 22 populations in the most recent 2 decades (Fig. 1), with declines of 3–6% seen throughout Northern and Western Europe, as well as in Italy and the Czech Republic. Elsewhere in Eastern Europe (e.g. in Romania and Bulgaria), the magnitude of the declines were negligible (Table II), while in the South, nonsignificant increases in the overall mortality trend of 2% and over 4% were observed in Portugal and Croatia, respectively (Table II).

The declines in mortality rates started first in Denmark, Norway and the U.K in the 1970s, followed, a few years later, in Sweden, Finland and France. The lower level of recent mortality declines in Eastern Europe partially reflects a tendency for their respective downturns to have occurred mainly in the last decade of observation, from the late 1980s.

APC modeling

Incidence trends related to period and birth cohort.

The full APC model or a submodel explained a sufficient amount of variation in each population (Table I). Cohort effects dominated in the majority, with cohort curvature significantly improving the fit in 10 of the 12 countries studied, with Italy and Slovenia being the exceptions (Table III). The age-cohort model adequately explained the variation in 7 countries (Table I). Only in 3 countries (Finland, Norway and Slovakia) did non-linear period effects significantly improve the fit (Table III).

Table III. Period and Cohort Curvature Over and Above Net Drift (Incidence)
European AreaCountryPeriod curvatureCohort curvature
Δ Deviance1Δ d.f.2p-value3Δ Deviance1Δ d.f.2p-value3
  • 1

    The difference in the deviance of the Age+Drift model and the model with the non-linear effects of Period or Cohort added.

  • 2

    The difference in the degrees of freedom of the Age+Drift model and the model with the non-linear effects of Period or Cohort added.

  • 3

    To determine the statistical significance of the non-linear effect, the change in deviance was compared with the chi-squared distribution on the change in degrees of freedom between the models.

NorthernDenmark1.720.4330.09<0.01
Finland16.870.0233.114<0.01
Norway18.470.0157.814<0.01
Sweden2.850.7252.112<0.01
United Kingdom2.420.3022.290.01
EasternCzech Republic1.310.2539.48<0.01
Slovakia27.54<0.0135.711<0.01
SouthernItaly0.210.6912.280.14
Slovenia0.0110.949.980.27
Spain1.710.2016.180.04
WesternFrance0.420.8319.990.02
Switzerland1.810.1829.58<0.01

Figure 2 shows the corresponding period and cohort risk parameters. It is evident that, even when half of the regular trend is attributed to the period of diagnosis, successive generational increases in risk can be seen in almost all European countries. The rises were fairly uniform and rapid with successive generations in Finland and the U.K, and in the Czech Republic and Slovakia.

Figure 2.

Age period cohort parameters based on assumptions on period and cohort slopes: incidence trends by country within European area (Panels 1–5: E Europe; 6–11: N Europe; 12–16: S Europe; 17–22: W Europe). Solid line: assumption of zero period slope (drift taken up entirely by cohort); dashed line: assumption of equal and same-direction slopes for period and cohort (drift attributed equally to period and cohort).

There was some evidence of a short-term dip in cohort-specific risk (regardless of the attribution of drift) in most other European countries, followed by rapid accelerations in risk thereafter. This was seen most evidently in men born around 1940–1945 in Denmark, Norway, and possibly, Sweden, and also in France, Italy, Slovenia and Spain, although the data are based on fewer years of observation for these countries. There was also a suggestion that successive generations of Swiss men, born after the early to mid 1960s, may have experienced some declines in risk of testicular germ-cell cancer.

Mortality trends related to period and birth cohort.

A deviance analysis of the mortality trends (Table II) indicates that the submodels or the full APC model provided an adequate fit in every country, excepting Germany. Period and cohort curvature significantly improved the fit in 14 and 13 populations, respectively (Table IV). Downward trends in mortality rates were seen in most Northern and Western European countries (and Italy) from the mid 1970s onward, translating (in spite of the rapid increases in incidence observed) to generation-specific decreases in risk of death for men born after 1940 (Fig. 3).

Figure 3.

Age period cohort parameters based on assumptions on period slope: mortality trends by country within European area (Panels 1–2: E Europe; 2–7: N Europe; 8–10: S Europe; 11–12: W Europe). Solid line: assumption of zero cohort slope (drift taken up entirely by period); dashed line: assumption of equal and same-direction slopes for period and cohort (drift attributed equally to period and cohort).

Table IV. Period and Cohort curvature Over and Above Net Drift (Mortality)
European AreaCountryPeriod curvatureCohort curvature
Δ Deviance1Δ d.f.2p-value3Δ Deviance1Δ d.f.2p-value3
  • 1

    The difference in the deviance of the Age+Drift model and the model with the non-linear effects of Period or Cohort added.

  • 2

    The difference in the degrees of freedom of the Age+Drift model and the model with the non-linear effects of Period or Cohort added.

  • 3

    To determine the statistical significance of the non-linear effect, the change in deviance was compared with the chi-squared distribution on the change in degrees of freedom between the models.

NorthernDenmark21.95<0.0134.612<0.01
Finland18.55<0.0113.9120.31
Ireland23.05<0.0111.6120.48
Norway26.15<0.0121.5120.04
Sweden35.25<0.0123.5120.02
United Kingdom35.34<0.0181.311<0.01
EasternBulgaria15.54<0.0115.3110.17
Czech Republic0.410.5519.180.01
Hungary18.04<0.0188.711<0.01
Poland2.910.0911.480.18
Romania8.820.0112.990.17
SouthernCroatia0.410.544.780.79
Greece1.340.8621.5110.03
Italy48.55<0.0117.9120.12
Portugal1.220.5618.990.03
Spain7.530.0610.1100.43
WesternAustria44.45<0.0134.312<0.01
Belgium5.640.235.7110.89
France85.55<0.0132.212<0.01
Germany1.310.2548.68<0.01
The Netherlands20.55<0.0128.412<0.01
Switzerland40.24<0.0129.311<0.01

The period-specific trends elsewhere in Southern Europe are difficult to interpret. The reduction in risk by calendar period in Greece and Spain since the 1970s has led to a discontinuation in the increases in risk of death, amongst affected birth cohorts. In contrast, the risk of death appears to have increased in Portugal and Croatia in consecutive cohorts born after 1950. In Eastern Europe, period-specific declines are most evident in Hungary (where a sufficient span of data is available) and the Czech Republic. The declines in Bulgaria and Romania are seen at least a decade later than in Northern Europe, with cohort-specific declines suggested only in the latter country, and only among generations born recently.

Discussion

This study describes the temporal patterns of testicular germ-cell cancer incidence and testicular cancer mortality in European countries, with particular reference to the importance of cohort influence (on incidence) and period effects (on mortality). Similar multicountry analyses of incidence in Europe have been compiled previously,5, 6 although this report extends the analysis to 12 countries, including several in Southern and Eastern Europe. The variability in the geographical and temporal patterns within Europe is extensive: a 5-fold variation in incidence rates was observed, and there was a steady rise in incidence across populations that varied from 1 to 6% per annum. Rates rose most rapidly in low-risk Spain and intermediate-risk Slovenia, with Switzerland being the exception to the increasing profile, for which rates have been extremely high but stable over several decades,38 with recent declines additionally suggested in this study. It can be speculated that a leveling off of incidence in high-risk countries may represent a mature phase in the epidemic, relative to lower-risk countries for which this phenomenon might be considered to be at an early phase, with further increases anticipated in forthcoming years.

Attention was first drawn to the increases in testicular cancer occurrence in England and Wales39 and Denmark40 half a century ago, yet the underlying causes are still largely unknown. The large variation in testicular cancer incidence both across the European countries and within each population over time could point to one or several ubiquitous and highly prevalent environmental agents being responsible, and moreover, must vary in prevalence according to population.

Regardless of temporal and geographical disparities, the age-incidence curves of testicular cancer are well-known to be largely invariate, implying that the age window of susceptibility to strong determinants of testicular cancer is likely equivalent in different populations.9 One of the strongest lines of evidence relates to the importance of factors acting prenatally or early in life that may initiate the process of testicular carcinogenesis. The strong causes involved in the development of carcinoma in situ, the precursor lesion of all germ-cell tumors,41 appear identical to the strong causes of testicular cancer.9 Carcinoma in situ most probably occurs during the first trimester of pregnancy,9 and the associations between testicular cancer and genital malformations and prenatal factors suggest that the strong causes of carcinoma in situ and, subsequently, of testicular cancer act prenatally. Increased estrogen exposure in utero has been related to increasing abnormities in the development and functioning of the testis,19 and a number of prenatal and perinatal exposure-related factors have been implicated for testicular cancer in analytical studies.

Aside from congenital malformations, of which cryptorchidism is the strongest and the most consistent determinant,18, 42, 43 certain prenatal factors have been reported in epidemiological studies with some consistency. These include premature birth,10, 11, 14, 15 low birth-weight,10, 11, 12, 13, 14 high birth-weight,10, 13 neonatal jaundice,13 exogenous estrogen use,11, 15 older maternal age12, 14 and first born.10, 12, 13, 14, 16 Other factors that have been reported are smoking during pregnancy,20, 21 subfertility,43, 44 exposure to viral infections45 and sedentary lifestyle.43

As has been consistently demonstrated,3, 6, 7, 37 generational influences appear largely responsible for the increasing incidence trends in Europe. Cohort curvature significantly improved the fit in all countries, except Italy and Slovenia, where age and age-drift models already provided a reasonable fit, respectively, possibly because of a lack of power to reject these simpler models.46 Trends in Finland, Norway and Slovenia required the full APC model, indicative of some period curvature being in operation, in addition to the cohort effects. Birth cohort effects can be viewed as a consequence of the changing prevalence of the risk determinants of the disease in successive generations, and generational increases of testicular germ-cell tumors are in accordance with the known biology of the disease, with possibly a role for external environmental factors mediated through exposure of the developing male embryo.47 The sharp rise in incidence observed around the onset of puberty implies a role of male sex hormones in the progression of germ-cell tumors.

A reduced incidence amongst a specific cohort born during the Second World War was observed in a number of countries, particularly in Denmark and Norway, as has been reported previously.6, 7, 37, 41, 48 It has been hypothesized that an altered supply of provisions in Denmark9, 37 may have impacted on consumption of a variety of foodstuffs and other commodities during the German military occupation. Interestingly, the pattern was also seen around the same time in Sweden, as previously reported,6 and in Italy, Slovenia, France and Switzerland. In Spain, a cohort with minimum risk was also identified, but at least a decade later, in the late-1950s. In the remaining countries (Finland, the U.K., Slovakia and the Czech Republic), no such break in the generational increases was evident. That the observation arises in many European countries in men born around the period of the Second World War suggests that modifications in lifestyle, possibly brought about by a war-related supply restriction at this time, were strong determinants of the disease, and that they acted very early in life, given the transitory nature of the phenomenon.9, 37 If a dietary factor is involved, this would probably concern an alteration in maternal diet affecting the offspring prenatally or postnatally. A recent study hypothesized that the mycotoxin Ochratoxin A, a contaminator of stored foods such as cereals and coffee, may be a causal factor.49

Despite the increases in incidence, decreasing mortality trends of between 3 and 6% per annum were observed throughout Northern and Western Europe, and in Italy, the Czech Republic and Hungary. The starting point and the rate of decrease in each country appears closely related to the dramatic improvements in the survival of young and middle-aged patients, following the introduction of cisplatin as a therapeutic agent for advanced germ-cell tumors.22 Notable declines were first observed in Denmark, Norway and the U.K in the early 1970s, followed soon after by Sweden, Finland and France. Further developments of cisplatin-based regimens, improvements in tumor imaging and surgical interventions of residual disease, together with a multidisciplinary approach to cancer care, have all contributed to the continuation of the declining mortality trends in the 1970s and 1980s.50 Thus, in spite of the generation-specific increases in incidence, the risk of death has been on the decrease in generations of men born after 1940 in higher resource countries.

In several Eastern Europe countries, where death rates are currently highest, the rate of decrease was of a relatively low order of magnitude, in part due to a later decline around the mid to late 1980s, at least a decade after Northern and Western Europe. The notable success of chemotherapy in terms of reductions in mortality were thus not uniformly seen across Europe, and slower and later declines in some lower resource countries imply that the high cost of appropriate treatments together with inadequate patient management systems are responsible for the high mortality rates and less favorable trends. Of particular and immediate concern were the increases in testicular cancer mortality in Portugal and Croatia of 2% and over 4% per annum, respectively. The trends were, in the majority of countries, based on small numbers, and particular interpretation was difficult at a more detailed level. The cohort analysis, however, clearly shows that the risk of death from testicular cancer has increased among men born in these countries since 1950.

Regarding our age-period-cohort analysis, the choice of slopes should be ideally founded on biological or epidemiological evidence; otherwise, if erroneous, a bias in all of the effects may be incurred.51 Selecting a range of slopes leaves some margin for error, allowing the researcher to contrast the age, period and cohort effects, based on their particular preference(s) for the fixed slope, with other less plausible specifications.51 Our approach to presenting trends estimates using the age-period-cohort model was predisposed a priori toward a cohort-based approach for incidence, given that both epidemiological evidence and biological mechanisms point to the importance of generational influences on disease occurrence. We thus circumvented the nonidentifiability problem and presented unique estimates of the period and cohort effects by firstly assuming a period slope of zero, implying that birth cohort influences were entirely responsible for the time trend; secondly, acknowledging the possibility of some increases in rates across all age groups over a period of time, the regular trend was attributed equally to period and cohort slopes.

For mortality, a priori evidence suggested that the presentation of the trends should incorporate the well-known benefits from treatment, which should show up as period-related effects. Mirroring the approach to incidence, we assumed firstly that the regular trend was a result of period influences, setting the cohort slope to zero. A second set of presented estimates accounted for the possibility of generational influences related to disease occurrence; the regular trend was then apportioned evenly to the slopes of period and cohort, as was for incidence.

The testicle is a visible and palpable organ, and so, the origin of the tumor is usually evident, notably among young and middle-aged men. Hence, misclassification and underascertainment of registration are less of an issue than for most other malignancies. Palpable availability of the testicle together with standardized therapeutic approaches and primary inguinal orchiectomy provides the basis for the high proportion of histologically verified tumors.

This study provides a description of the current trends in testicular cancer incidence and mortality in Europe. Uniform increases in testicular germ-cell cancer varied 6-fold in 12 countries for which the background risk ranged 5-fold, with the importance of cohort effects clearly discernible. The underlying risk factors responsible for the increases remain elusive, though the extent of variation perhaps lends some support to the idea of an epidemic in different phases in different countries. Fortunately, advances in therapy and the management of testicular cancer since the mid 1970s have led to large declines in mortality in some European countries, despite the unabated increases in incidence. More disturbing, in lower-resource countries, has been the apparently slower progress toward delivery of optimal care reflected in the time trends of mortality. In countries like Bulgaria and Romania, the first beneficiaries of therapy appear to be men born—rather than diagnosed—in the era of this major breakthrough in oncology.

Acknowledgements

The work of Lorenzo Richiardi was partially supported by the Regione piemonte—Ricerca Finalizzata and project “Oncology” Compagnia di San Paolo/FIRMS. The following European cancer registries (Director in parentheses) are participating investigators, having contributed their incidence data, as well as their expertise in commenting on the final manuscript: Czech Republic—Czech National Cancer Registry, Prague (Dr. Jana Ajmová); Denmark—Danish Cancer Society, Copenhagen (Dr. Hans H. Storm); Finland—Finnish Cancer Registry, Helsinki (Dr. Timo Hakulinen); France—Registre Bas Rhinois des Cancers, Strasbourg (Dr. Michel Velten), Registre Général des Tumeurs du Calvados, Caen (Dr. J. Macé-Lesech), Registre des Tumeurs du Doubs, Besançon (Dr. Arlette Danzon), Registre du Cancer de l'Isère, Meylan (Dr. François Ménégoz), Registre du Cancer de la Somme, Amiens (Mme Nicole Raverdy), Registre des Cancers du Tarn, Albi (Dr. Martine Sauvage); Ireland—National Cancer Registry, Cork (Dr. Harry Comber); Italy—Registro Tumori Toscano, Florence (Dr. Eugenio Paci), Registro Tumori Lombardia (Provincia di Varese), Milan (Dr. Paolo Crosignani), Registro Tumori della Provincia di Parma (Dr. Vincenzo De Lisi), Registro Tumori della Provincia di Ragusa, Ragusa (Dr. Rosario Tumino), Piedmont Cancer Registry, Turin (Dr. Roberto Zanetti); Norway—Cancer Registry of Norway, Oslo (Dr. Frøydis Langmark); Slovakia—National Cancer Registry of Slovak Republic, Bratislava (Dr. Ivan Plesko); Slovenia—Cancer Registry of Slovenia, Ljubljana (Dr. Maja Primic-Zakelj); Spain—Tarragona Cancer Registry, Reus (Dr. Jaume Galceran), Registro de Cáncer de Granada, Granada (Dr. Carmen Martínez Garcia), Registro de Cáncer de Murcia, Murcia (Dr. Carmen Navarro Sánchez), Registro de Cáncer de Navarra, Pamplona (Dr. E. Ardanaz Aicua), Zaragoza Cancer Registry, Zaragoza (Dr. Carmen Martos Jimenez); Sweden—Swedish Cancer Registry, Stockholm (Dr. Lotti Barlow); Switzerland—Krebsregister Basel-Stadt und Basel-Land, Basle (Dr. Gernot Jundt), Registre Genevois des Tumeurs, Geneva (Dr. Christine Bouchardy), Registre Neuchâtelois des Tumeurs, Neuchâtel (Dr. Fabio Levi), Krebsregister St Gallen Appenzell, St Gallen (Dr. Silvia Ess), Registre Vaudois des Tumeurs, Lausanne (Dr. Fabio Levi), Kantonalzürcherisches Krebsregister, Zürich (Dr. Nicole Probst); United Kingdom—National Cancer Intelligence Centre, London (Dr. Mike Quinn), Scottish Cancer Registry, Edinburgh (Dr. David Brewster).

Ancillary