SEARCH

SEARCH BY CITATION

Keywords:

  • age–period–cohort models;
  • skin melanoma;
  • mortality;
  • trends;
  • epidemiology

Abstract

  1. Top of page
  2. Abstract
  3. Patients and methods
  4. Results
  5. Discussion
  6. References

Recent statistics cohort analysis of mortality reveals that skin melanoma rates are dropping in the younger cohorts. Therefore, the aim of this study is to provide up-to-date information and to analyze recent changes in skin melanoma mortality trends in Spain during the period of 1975–2004 using joinpoint regression and age–period–cohort models. Between 1975 and 2004, the age-standardized (overall) mortality rates for skin melanoma increased from 0.3 to 1.3 per 100,000 person-years for males and from 0.2 to 0.8 per 100,000 person-years for females, with an estimated annual percentage of change of 4.8 and 4.3%, respectively. In men and women, the best fit was found for the full model, which simultaneously considered the effects of age, period and cohort. The risks among both males and females increased in each successive birth cohort born between 1895 and 1950. Thereafter, the risks declined through the most recent birth cohort born in 1970. Examination of the mortality trends by age group and birth cohort revealed that the recent less rapidly rising (men) or stabilizing (women) age-adjusted skin melanoma mortality rates in Spain were a result of declining mortality in the younger age groups and more recent birth cohorts. The particularly favorable trends in young people suggest that a further decline in mortality from skin melanoma in Spain is likely to occur within the next few years. © 2007 Wiley-Liss, Inc.

In Europe, ∼26,100 males and 33,300 females are diagnosed each year with melanoma, and around 8,300 males and 7,600 females die because of it. It is the 8th most commonly diagnosed cancer in females and 17th in males.1

In Northern Europe, where incidence rates are high, mortality rates seem to be leveling off since the mid-1990's, especially in younger age groups. In contrast, in Southern and Eastern Europe rates are increasing steeply in all age categories.2, 3

Spain has one of Europe's lowest melanoma incidence and mortality rates.4 In a previous report,5 we analyzed skin melanoma mortality trends in Spain during the period 1975–2001, showing that mortality rates increased in the last few decades, although this rising trends are now leveling off in middle-aged adults (35–64 years), following a similar tendency to that observed in other countries.3, 6

Skin melanoma mortality trends can be attributed to changes due to methods of classifying cause of death, short-term effects of early detection and the effects of new medical procedures, which can reduce melanoma mortality near the time of death (period effects), and changes that are associated with the generation to which people belong (cohort effects).7 In many countries a so-called age–cohort pattern was observed in both sexes. This means that starting with some specific birth cohort the mortality is increasing or decreasing for successive cohorts rather than for successive time periods.2, 8

Therefore, the aim of this study is to provide up-to-date information and to analyze recent changes in skin melanoma mortality trends in Spain during the period of 1975–2004 using age–period–cohort models.

Patients and methods

  1. Top of page
  2. Abstract
  3. Patients and methods
  4. Results
  5. Discussion
  6. References

Age- and sex-specific skin melanoma deaths (International Classification of Diseases [ICD], 8th and 9th revisions, code 172; 10th revision, code C43 for post-1999 deaths) were taken from the official vital statistics published by the National Statistics Institute for the years 1975–2004.

For each sex, truncated rates (35–64 years) and overall age-standardized (world population) mortality rates were computed by the direct method, and are expressed as rates/100,000 men or women. A joinpoint regression was fitted to provide an estimated annual percentage change (EAPC) and to detect points in time where significant changes in the trends occur. The joinpoint regression model describes continuous changes in rates and uses the grid-search method to fit the regression function with unknown joinpoints. In this model, the annual rates over a given period of time are examined and the points in time when the direction of the trends changes significantly are detected.9 The computation of mortality rates and their standard errors were performed using a spreadsheet (Microsoft® Excel). Joinpoint analyses were performed using the “Joinpoint” software from the Surveillance Research Program of the US National Cancer Institute.10

We used a graphical approach to visually detect age, period and cohort effects of the skin melanoma mortality. We tabulated the data into 11 five-year groups (30–35 to 80–84), 6 five-year periods (1975–1979 to 2000–2004) and 16 overlapping 10-year birth cohorts, identified by central year of birth from 1895 to 1970. We plotted age-specific rates by year of death (period), and year of birth- (cohort) specific rates by age group, for men and women separately. Age effects were identified whenever age-specific rates were consistently different for an age group over a range of periods or birth cohorts. Period effects were observed if rates for all age groups changed by period. Cohort effects were observed if age-specific rates were not parallel across periods, or were elevated for all ages of the same birth cohort.

In addition, we estimated the effects of age, period and cohort on the skin melanoma mortality rates by using a Poisson regression model. This model assumes that the number of deaths is a variable with a Poisson distribution that has a mean depending multiplicatively on the number of person-years and the explanatory variables age, time period and birth cohort.11, 12

Inherent in the age–period–cohort model is the nonidentifiable problem7: parameters for age, period and cohort are not simultaneously estimable because of the exact linear dependence of the regressor variables (birth cohort = period − age). Although several methods dealing with the nonidentifiable problem have been proposed, there is no consensus in the literature as to which method is optimal.13, 14 In this study, we chose the approach suggested by Decarli and La Vecchia14 using grouped data and a penalty function to solve the identifiable problem. All 3 “two-factor” models (age–period, age–cohort and period–cohort) are fitted. The penalty function measures the Euclidean distance between the two-factor model estimates and the family of three-factor model estimates (age–period–cohort), and chooses the full model that weighted the average of the 3 two-factor model solutions that minimizes the penalty function. The goodness-of-fit of models and comparisons between nested models were evaluated by means of the deviance. For models with the deviance close to its degrees of freedom, the fit was considered adequate. Changes in deviance between the 2 models were assumed to be χ2 distributed with degrees of freedom equal to the difference in the number of parameters in the 2 models. Models were fitted using the R software with appropriate macros.15

Results

  1. Top of page
  2. Abstract
  3. Patients and methods
  4. Results
  5. Discussion
  6. References

From 1975 (population size of Spain: 35.7 million) through 2004 (population size: 41.0 million), about 14,190 people died due to skin melanoma in Spain.

Between 1975 and 2004, the age-standardized (overall) mortality rates for skin melanoma increased from 0.3 to 1.3 per 100,000 person-years for males and from 0.2 to 0.8 per 100,000 person-years for females, with an EAPC of 4.8 and 4.3%, respectively (Table I).

Table I. Age-standardized melanoma mortality rates (ASR) and joinpoint analysis (Spain, 1975–2004) in males and females
 ASREAPC (95% CI)Trend 1Trend 2
197520041975–2004YearsEAPC (95% CI)YearsEAPC (95% CI)
  1. EAPC, estimated annual percent of change; CI, confidence interval.

Males
 Overall0.31.34.8 (4.0; 5.5)1975–19908.6 (7.3; 9.8)1990–20041.6 (0.6; 2.6)
 Truncated (35–64)0.52.14.8 (3.9; 5.7)1975–19918.7 (7.4; 10.1)1991–20040.5 (−1.0; 1.9)
Females
 Overall0.20.84.3 (3.6; 5.1)1975–19946.7 (5.8; 7.5)1994–2004−0.3 (−2.1; 1.4)
 Truncated (35–64)0.41.54.6 (3.7; 5.6)1975–19947.1 (5.7; 8.6)1994–2004−0.7 (−3.6; 2.4)

In males, as can be seen in Figure 1, there was a marked increase from 1975 to 1990 [EAPC 8.6%, 95% confidence interval (CI): 7.3; 9.8], after which mortality rates increased more slowly (EAPC 1.6%, 95% CI: 0.6; 2.6) (Table I). Age-adjusted (35–64 years) mortality rates increased steadily from 1975 to 1991 (EAPC 8.7%, 95% CI: 7.4; 10.1) and then leveled off (EAPC 0.4, 95% CI: −1.0; 1.9). Throughout the study period, age-standardized mortality rates for men exceeded those for women. In women, there was a marked increase from 1975 to 1994 in age adjusted (overall and truncated) mortality rates [EAPC 6.7%, 95% CI: (5.8; 7.5) and 7.1%, 95% CI: (5.7; 8.6), respectively], followed by a leveling-off [EAPC −0.3%, 95% CI: (−2.1; 1.4) and −0.6% 95% CI: (−3.6; 2.4), respectively].

thumbnail image

Figure 1. Mortality from skin melanoma in Spain, 1975–2004. ASR, age standardized rate.

Download figure to PowerPoint

Figure 2 depicts age-specific rates on a logarithmic scale with respect to the generation's mid-year of birth for males and females. Rates for all age groups have been joined by lines. This presentation enables graphical assessment of the effects of age, period and cohort on mortality rates.

thumbnail image

Figure 2. Age-specific mortality rates per 100,000 persons-year of skin melanoma combined by birth-cohort.

Download figure to PowerPoint

The goodness-of-fit of various Poisson regression models of the skin melanoma mortality is summarized in Table II. An age–drift model reduced the deviance of the basic age model. An age–period model was a further improvement on the age–drift model. The full age–cohort–period model gave an adequate representation of the data in both males and females. Comparison of the age–cohort model with the age–cohort–period model entailed an improvement indicating both cohort and period effects.

Table II. Goodness-of-fit test for different age-, period- and cohort specific models of cutaneous malignant melanoma in males and females in Spain, 1975–2004
ModelDegrees of freedomDeviance
MalesFemales
  • 1

    A: Test for trend (drift); B: test for irregular trends in mortality (above trend) attributable to period effects; C: test for irregular trends in mortality (above trend) attributable to cohort effects; D: test for cohort effects adjusted for period effects; E: test for period effects adjusted for cohort effects.

Age551,065.8670.0
Age + drift54198.1128.1
Age + period5081.842.0
Age + cohort40119.190.0
Age + period + cohort3627.326.7
Difference between models1p values (males)p values (females) 
A. Model 2 vs. 1<0.001<0.001 
B. Model 3 vs. 2<0.001<0.001 
C. Model 4 vs. 2<0.001<0.001 
D. Model 5 vs. 3<0.0010.36 
E. Model 5 vs. 4<0.001<0.001 

The estimators obtained from the final age–period–cohort model are shown in Figure 3. The age effect is represented on a logarithmic scale in view of the fact that specific rates rise exponentially relative to this variable and are interpretable as age–specific death rates per 100,000 person-years, adjusted for period and cohort. The right-hand side depicts the cohort and period effects, averaged to unity using a linear scale; hence, these values can be interpreted as rate ratios or relative risks, with their weighted mean being taken as reference. The risks among both males and females increased in each successive birth cohort born between 1895 and 1950. Thereafter, the risks declined through the most recent birth cohort born in 1970. The period effect shows a less rapidly rising (men) or stabilizing (women) skin melanoma mortality in the last years.

thumbnail image

Figure 3. Age effect, birth-cohort and period of death effects estimated on the basis of the Poisson model. Skin melanoma mortality, Spain.

Download figure to PowerPoint

Discussion

  1. Top of page
  2. Abstract
  3. Patients and methods
  4. Results
  5. Discussion
  6. References

The risk of skin melanoma is associated to intense and intermittent exposure to sunlight, mainly in fair-skinned populations, and to ultraviolet radiation (UV) and sunburns particularly during childhood.16

From age–period–cohort analyses of mortality data, it is known that in countries with high melanoma incidence mortality rates increased starting from birth cohorts around 1880 and stabilized in cohorts born around 1940–1950 (e.g., in Australia, the rates peaked around the 1925 birth cohort in women and around the 1935 birth cohort in men17; in Sweden, the rates increased up to the 1947 birth cohorts in women and the 1932 birth cohorts in men18). In other countries, a birth cohort effect with declining rates in people born after 1950–1960 has been observed.19

Examination of the mortality trends by age group and birth cohort (Fig. 2) revealed that the recent less rapidly rising (men) or stabilizing (women) age-adjusted skin melanoma mortality rates in Spain (Fig. 1) were a result of declining mortality in the younger age groups and more recent birth cohorts. Cohort effects usually result from environmental and societal changes.

In many countries where incidence is high and where awareness and (secondary) prevention campaigns have often been organized, moderations in the increases in melanoma mortality rates are observed within a few years, followed by moderations in melanoma incidence in younger age groups after more than 10 years, corresponding with the observed birth cohort effects in the mortality rates.2, 20 In Spain, since 2000, the Spanish Academy of Dermatology has sponsored free annual national skin cancer prevention campaigns detecting more new cases and decreasing the number of advanced cancers.21, 22

The increase in cohort trend from 1895 to 1950 (Figs. 2 and 3) suggests that these birth cohorts experienced increased exposure to the etiologic agents (most likely intense and intermittent sun exposure particularly early in life). The use of sun protection methods (changes in the pattern of sunlight exposure and better UV protection with sunscreens)23 and early-stage diagnosis could be the most likely reasons for the downturn of mortality rates for cohorts born after 1950. If these trends persist, and efforts to improve awareness and increase early detection of melanomas are maintained, and assuming that other relevant factors associated with increased risk remains constant, mortality should decline in the next years.

The period effect shows a less rapidly rising (men) or stabilizing (women) skin melanoma mortality in recent years. This suggests that the decline might be due to mass phenomena that affected the Spanish population and exerted their effects with a short latency time. Quality of mortality statistics and changes in the health-care system are the most prominent among the possible candidates for these phenomena.

Although 3 different ICDs were used from 1975 through 2004 (ICD 8th, 9th and 10th revisions) this cannot explain the period effects, because there were no relevant changes in definition of skin melanoma.24 We can consider mortality figures a reliable indicator of the frequency of skin melanoma because of the accuracy of the information on death certificates25 as well as the similarity of the pattern observed for the truncated rates (Fig. 1), which are less affected by errors in death certification.26

Melanoma survival rates are strongly correlated with the stage of disease at diagnosis. When the melanomas are still localized, they can be excised and survival rates are generally very favorable. If, however, the melanomas are in more advanced stages, treatment is very difficult and survival rates are rather poor. Early detection and more frequent excision of pigmented lesions due to better access to health care, with a consequent improved survival (for men increased from 70.4% in 1987–1989 to 73.9% in 1990–1994 and for women from 84.1 to 89.8% over the same period)27, 28 and a growing public awareness of the dangers of excessive sunbathing are the most plausible explanations for the deceleration in mortality trends in recent years in Spain, as there were no major improvements in systemic melanoma treatments.

It is hoped that increased public and professional awareness and education in all areas relating to the prevention, detection and treatment of malignant melanoma will contribute to decreasing trends in the incidence and mortality from this cancer in the future.29

In conclusion, the change in skin melanoma mortality rates was best explained by age-, period- and to a less extent by cohort effects. The favorable mortality decline by birth cohort in the most recent birth cohort is an important indicator of a likely decline in mortality over the coming years.

References

  1. Top of page
  2. Abstract
  3. Patients and methods
  4. Results
  5. Discussion
  6. References