Trends in Mortality Rates from Malignant Melanoma in Sweden

To monitor mortality rates from malignant melanoma we analysed all patients in Sweden (6,324) who died of malignant melanoma in 1953 through 1987. Age-standardised rates per 105 increased from 1.1 to 4.0 in men and from 1.0 to 2.6 in women. The average annual increase levelled off in men from 4.6% during 1953-1967 to 2.0% in 1978-1987; and in women from 3.7% to 0%. Multivariate analyses showed that the change in rates for men was mainly due to a birth-cohort effect, whereas in women the rates changed similarly in all age-groups in accordance with a time-period effect. The risk of dying of malignant melanoma increased in men for birth cohorts up to 1932, whereas in women the rise continued for cohorts born as late as 1947. The best-fitted multivariate models were extrapolated to the year 2007, among men a slight increase in mortality rates seemed likely, whereas among women the rates will probably remain unchanged. The incidence of malignant melanoma has been increasing in white populations throughout the world during the last few decades (Jensen & Bolander, 1980; Hakulinen et al., 1986). No apparent change in diagnostic criteria can explain this increase (van der Esch et al., 1991). In many countries, malignant melanoma may become one of the commonest malignant tumours and it therefore constitutes a growing problem in cancer control. However, studies of trends in survival show that the prognosis of the disease has improved Sweden the relative risk of dying of malignant melanoma within 5 years after diagnosis decreased by 68% in men and 71% in women from 1960-1964 to 1980-1982 (Thorn et al., 1989). This improvement was larger than for any other solid tumour in Sweden (Adami et al., 1989). Probably the increase in public awareness of malignant melanoma has entailed earlier diagnosis and thus facilitated curative treatment. This explanation of the temporal trends in survival is also supported by an increasing proportion of smaller and thinner tumours in recent surveys of histopathological data (Shafir et al., 1982; English et al., 1986; Drzewiecki et al., 1990). Notwithstanding these improvements there is still steady rise in the death rates from malignant melanoma in many The present study was based on mortality rates from malignant melanoma in Sweden from 1953 to 1987. In addition to calcuations of age-standardised and age-specific rates, multivariate analysis was done to separate the effects of time period from those of birth cohort to further …

Trends in mortality rates from malignant melanoma in Sweden 1953-1987 andforecasts up  Summary To monitor mortality rates from malignant melanoma we analysed all patients in Sweden (6,324) who died of malignant melanoma in 1953 through 1987. Age-standardised rates per 105 increased from 1.1 to 4.0 in men and from 1.0 to 2.6 in women. The average annual increase levelled off in men from 4.6% during 1953-1967 to 2.0% in 1978-1987; and in women from 3.7% to 0%. Multivariate analyses showed that the change in rates for men was mainly due to a birth-cohort effect, whereas in women the rates changed similarly in all age-groups in accordance with a time-period effect. The risk of dying of malignant melanoma increased in men for birth cohorts up to 1932, whereas in women the rise continued for cohorts born as late as 1947. The best-fitted multivariate models were extrapolated to the year 2007, among men a slight increase in mortality rates seemed likely, whereas among women the rates will probably remain unchanged.
The incidence of malignant melanoma has been increasing in white populations throughout the world during the last few decades (Jensen & Bolander, 1980;Hakulinen et al., 1986). No apparent change in diagnostic criteria can explain this increase (van der Esch et al., 1991). In many countries, malignant melanoma may become one of the commonest malignant tumours and it therefore constitutes a growing problem in cancer control. However, studies of trends in survival show that the prognosis of the disease has improved (Shaw et al., 1977;Balch et al., 1983;Thorn et al., 1989). In Sweden the relative risk of dying of malignant melanoma within 5 years after diagnosis decreased by 68% in men and 71% in women from 1960-1964(Thorn et al., 1989. This improvement was larger than for any other solid tumour in Sweden . Probably the increase in public awareness of malignant melanoma has entailed earlier diagnosis and thus facilitated curative treatment. This explanation of the temporal trends in survival is also supported by an increasing proportion of smaller and thinner tumours in recent surveys of histopathological data (Shafir et al., 1982;English et al., 1986;Drzewiecki et al., 1990).
Notwithstanding these improvements there is still steady rise in the death rates from malignant melanoma in many countries (Jensen & Bolander, 1980;Lee et al., 1979;Lee, 1985). The only exception is Queensland, Australia, where a stabilisation of mortality rates was apparent in males after 1960-1964and in females after 1965-1969(Holman et al., 1980. The present study was based on mortality rates from malignant melanoma in Sweden from 1953 to 1987. In addition to calcuations of age-standardised and age-specific rates, multivariate analysis was done to separate the effects of time period from those of birth cohort to further explain the trends. The resulting multivariate models were also extrapolated to estimate the mortality rates until the year 2007.

Subjects and methods
Deaths from malignant melanoma The Swedish National Cause of Death Register systematically collects information on dates and causes of deaths based on all Swedish citizens who have died during a given calendar year, including those who have died outside the country (Statistics Sweden, 1954. The cause of death is registered in accordance with the International Classification of Diseases (World Health Organization, 1967). The underlying cause of death and the contributory cause of deathif a malignant diseasewere registered in the Cause of Death Register up to 1980. After that year only a single underlying cause of death was registered. A total of 6,324 deaths from malignant melanoma was registered in the Cause of Death Register from 1953 through 1987. These cases were distributed by sex, age and 5-year periods as shown in Table   I.

Statistical methods
Age-standardised mortality rates were calculated for men and women annually. The direct method of standardisation (Fleiss, 1981) was used with the Swedish population of 1970 as reference. A log-linear regression model, which implies a constant annual percentage change, was used to estimate the temporal trends of the rates. Age-specific mortality rates were estimated as the average rate per year during each 5-year time period, starting with 1953-1957 and ending with 1983-1987, using the age-groups <30, 30-39, 40-49, 50-59, 60-69 and > 70 years of age. In the multivariate analysis the number of deaths was assumed to be Poisson-distributed, with a mean which depends on multiplication of the explanatory variables of age, period and cohort. The full model may be formulated as Lijk= Nijk exp (@ij7k), where N is person-years and mi, Pj and ?k are the effects of age, period and cohort. The model was estimated by the maximum likelihood method, using the GLIM software package (Baker & Nelder, 1978). Submodels, such as a combination of age and period and a combination of age and cohort, were fitted in addition to the full model. The special case when the effects of period or cohort on the logarithmic rates in age-period and age-cohort models was assumed to be linear was also considered. In that case, it was impossible to separate the period effects from the cohort effects, and the combined linear effect is denoted 'drift' . The model fit was evaluated in terms of the deviance, which has an asymptotic chi-square distribution. By determining the difference in deviance, various models can be compared. When the deviance is close to the degrees of freedom of the model, the fit may be considered adequate.
Correspondence: M   For the analysis, 13 five-year age-groups (ranging from ages 20-24 to 80-84 years) and 7 five-year calendar periods from 1953 to 1987 were defined. In all 19 ten-year overlapping birth cohorts were used, starting at 1868-1877 and ending at 1958-1967. The risk associated with a certain age-group (or period or cohort) relative to a chosen reference group was obtained by exponentiation of the parameter (for example, exp (i).
Forecasts based on the multivariate models were produced for four 5-year periods in the future. The estimates of the unknown period effects and cohort effects were obtained by linear regressions based on arbitrarily chosen numbers of the most recent period and cohort values. These analyses were done on the logarithms of the sets of values, the form in which the underlying model is linear (Osmond, 1985).

Age-specific rates
In men younger than 50 years of age, there was no increase in the mortality rates after 1968. In contrast, men aged 50 years or more showed increasing rates during the entire study period (Figure 2).
In women aged 30 to 50 years, the rates increased slightly, except during the last five-year time period, 1983-1987. Women older than 50 years had steadily increasing mortality rates, except those older than 70 years who had decreasing rates during the last 5-year time period (Figure 3).

Multivariate analysis
In both sexes, 'drift'-models (which includes linear effects of period and cohort) were significantly superior to simple age models in explaining the mortality rates. Further, in men, an age-cohort modelwhich in contrast to the age-drift model  -~~~~< 30 1953-1958-1963-1968-1973-1978- Time-period Figure 2 Age-specific mortality rates of malignant melanoma in men, Sweden, 1953Sweden, -1987 allows the cohort effects to be non-linearwas a significant improvement (P<0.01) on the age-drift model. An ageperiod model was also an improvement on the age-drift model (P <0.05). The full model (age + period + cohort) further lowered the deviance, however, the change was not significant (P = 0.07) compared to the age-cohort model. In women, the use of an age-cohort model did not significantly improve the age-drift model. An age-period model, however, was a significant improvement (P <0.01) on the age-drift model. The full model was not an improvement on the age-period model (P = 0.29) (Table II).
In the age-cohort model in men, the relative risk of dying of malignant melanoma increased continuously by birth cohort up to a seven-fold higher relative risk in those born  1953-1958-1963-1968-1973-1978-1983-57 62 67 72 77 82 87 Time-period Figure 3 Age-specific mortality rates of malignant melanoma in women, Sweden, 1953. ' Women 1868-1878-1888-18981908-1918-1928-1938-1948 1958-77  87  97  07  1 7  27  37  47  57  67 Birth-cohort Figure 4 Relative risk of dying of malignant melanoma in successive birth cohorts of men and women, Sweden, 1953Sweden, -1987 Reference birth cohort 1868-1877 (relative risk = 1). around 1932, compared to men born in 1878-1887; in laterborn cohorts the risk decreased gradually and markedly to a relative risk below four in the youngest cohort ( Figure 4). In women, however, the relative risk by cohort increased almost sevenfold up to the birth cohort 1943-1952 before declining (Figure 4). The age-effects in the age-cohort model had a nearly constant slope for both men and women.
The relative risk by period increased continuously throughout the study period for men up to 3.0 compared to the earliest time period 1953-1957. In women the risk increased stepwise up to 2.6 and did not change during the time period 1973-1977 as compared to 1968-1972 and during 1983-1987 as compared to 1978-1982 (Figure 5).
Extrapolations offuture mortality rates In men, the age-period model generated considerably higher estimates than the age-cohort model. In both models the predictions were rather insensitive to using two or five recent period values or cohort values as the basis for the extrapolation (Table III). Since the mortality trends in men have mostly followed a cohort pattern hitherto, this model was considered more reliable even for future predictions. Thus, an about 35% increase in age-standardised mortality rates was estimated for the 20-year period following 1987. However, the rate of increase will probably decline gradually from about 14% between the first two 5-year periods to less than 5% between the last two periods.
In women, predictions from the cohort model were identical when two and five recent cohort values were included. In contrast, the extrapolation of period effects which, according to the multivariate analysis may be more relevant in women, was very sensitive to the number of period values included. Estimates based on two period values (10 years of the observation) suggest that the maximum was reached in 1978 through 1987 and that future age-standardised rates will stabilise at annual rates close to 3 per 105. Predictions that  Sweden, 1953Sweden, -1987 Reference time period, 1953-1957 (relative risk = 1). proceed from five recent period values, on the other hand, are strongly influenced by the apparent stepwise increase in mortality rate from 1973-1977 to 1978-1982 (Figure 5). Consequently, this model predicts a continuing 80% increase from 1987 through the year 2007 (Table III).

Discussion
Our analysis of mortality rates from malignant melanoma in Sweden 1953-1987 showed increasing rates in men and stabilising rates in women. The change in rates was best explained by cohort effects in men and by period effects in women. The relative risk of dying of malignant melanoma increased in men by birth cohort continuously up to men born around 1932, whereas in women the rise continued for cohorts born as late as 1947. In future, the increase in mortality rates will probably slow down for men, whereas in women predictions are more uncertain due to a stepwise increase in the late 1970s. Thus, mortality rates in women may stabilise around 3 per I05 or increase up to 5.7 per I05 during the next 20 years.
Mortality rates from malignant melanoma are affected by both the causative factors determining the incident number of cases and by factors related to the prognosis of the disease. One possible source of error affecting trends in mortality is the varying accuracy of the certified underlying cause of death. In Sweden, a comparison between diagnoses of skin cancer and malignant melanoma in the Stockholm Cancer   -year periods, in Sweden, 1953-1987, and 20 1958-1962 2.50 1.78 1963-1967 2.76 2.01 1968-1972 3.78 2.52 1973-1977 4.16 2.52 1978-1982 4.71 3.18 1983-1987 5.47 3.17 Extrapolated age-standardised ratesa Register and the certified underlying cause of death in the Cause of Death Register, during 1978, showed concordant diagnoses in 88% of the cases (Mattsson et al., 1984). To our knowledge, no such investigation has been performed for the years after 1978 or specifically for malignant melanoma. However, the reporting practices have probably improved and we consider it unlikely that the results of our study are seriously biased because of changes in reporting practices. Mortality rates from malignant melanoma in Sweden, adjusted to the European standard population for comparison (IARC, 1976), were lower than rates in Australia and New Zealand; they were similar to those in the United States and higher than the rates recorded in Canada, England and Wales (Lee et al., 1979;Venzon & Moolgavkar, 1984). In men, the mean annual percentage increases in mortality rates, from the 1950s to the 1970s, were almost identical in Sweden, Australia and New Zealand, whereas they were lower among men in England and Wales, Canada and the white population of the United States. Among women the increases in mortality rates were similar in Sweden, New Zealand, Australia, Canada, England and Wales but lower among white women in the United States.
Age-period-cohort models are superior to simple descriptive methods. It is possible to test whether a significant improvement is obtained when further factors are included in the model. It can be stated whether the full model is an improvement on an age-period or an age-cohort model. However, the individual parameters of the full model cannot be identified, this fact makes the interpretation of the results rather difficult and limits the use of the method. In our analysis, two-factor models (age-cohort or age-period) were found to be adequate and it was not necessary to identify the full model.
In contrast to earlier studies, in which it was mainly cohort effects that explained the changes in mortality rates in both sexes (Lee et al., 1979;Holman et al., 1980;Venzon & Moolgavkar, 1984), we found that period effects were more important in women. Similar results were obtained in a multivariate analysis of incidence rates of malignant melanoma in Sweden (Thorn et al., 1990). Thus the temporal changes in exposure to causative factorsas revealed by trends in incidenceare fairly well reflected also by the mortality rates. However, the increasing incidence trend has been more pronounced than the mortality trend particularly in women. This discrepancy can be explained by the large temporal improvement in relative survival from malignant melanoma documented earlier (Thorn et al., 1989).
In the forecasts, the number of values chosen as the basis for the extrapolations is crucial for the results predicted with the age-period model. The age-cohort model is less sensitive to the number of values included because the mortality is small in the youngest age-groups which are influenced by the extrapolated cohort-values. Consequently, the two alternative extrapolations in our study gave largely similar mortality rates in the cohort model, whereas in the period model the results were different, especially for women (Table III).
The overall achievements in control of malignant melanoma are best evaluated by studying the changes in the mortality rates. The planning of interventional strategiese.g., educational programs to reduce sun exposure and screening programs for earlier diagnosisshould be guided by estimates of the mortality rates in the future. Malignant melanoma currently accounts for about 1.5% of all cancer related deaths in Sweden (Statistics Sweden, 1990). Our study suggests that the mortality from malignant melanoma is likely to increase only slightly during the next 20 years. Under such circumstances, malignant melanoma will retain a limited quantitative role in the overall burden of death from cancer. This study was supported by grants from the Swedish Cancer Society.