Endometrial cancer is the most common cancer of the female genital tract in Western countries. Monitoring the incidence is important for health care planning and the identification of risk factors. We present an age-period-cohort analysis of incidence trends of endometrial cancer in Norway from 1953 to 2007 and compare the incidence trends with those in 3 other Nordic countries. Based on the observed trends, we have predicted endometrial cancer rates in Norway in 2015 and 2025. In women at postmenopausal age (≥55 years), the annual incidence increase was 2.1% (95% CI: 0.9%, 3.4%) from 1988 to 1997 and 1.7% (95% CI: 0.6%, 2.8%) from 1998 to 2007. In younger women, there was an annual reduction of 0.6% (95% CI: −2.3%, 2.2%) from 1988 to 1997, followed by an annual increase of 1.7% (95% CI: −0.4%, 3.9%) from 1998 to 2007. The secular changes are likely to reflect both cohort and period effects. Our prediction estimates by 2025 suggest an incidence increase in the range of 50 to 100%, relative to the observed incidence in 2005. There has been a strong and consistent incidence increase in endometrial cancer in the Nordic countries over the last 50 years. The increase has been most pronounced in postmenopausal women, but in the last decade, rates have increased also in women younger than 55 years. The prediction for the next 20 years suggests that endometrial cancer rates will dramatically increase unless effective preventive strategies are implemented.
Endometrial cancer is the most common cancer of the female genital tract in many Western countries, and some of the highest incidence rates are observed in European populations.1 Temporal incidence trends appear to differ by menopausal status, and in Norway, as in other Nordic countries, the increasing incidence of postmenopausal endometrial cancer contrasts with the declining rates observed in premenopausal women.2
Monitoring the incidence is important for health care planning purposes and for the identification of risk factors. The planning of health services is an integral component of cancer control programmes,3 and prediction of future rates is therefore of great interest to society.
The etiology of endometrial cancer is unknown, but the main hypothesis has emphasized exposure to high levels of circulating estrogens in conjunction with inadequate levels of progesterone.4 The use of unopposed estrogen replacement therapy in postmenopausal women is associated with increased risk,5 whereas use of combined oral contraceptives (COCs) confers a degree of protection.6 Other risk increasing factors include nulliparity,7–10 obesity and diabetes,11–16 while cigarette smoking17 and delayed childbearing18 may be inversely associated with endometrial cancer risk.
In this study we present an age-period-cohort (APC) analysis of the observed incidence trends of endometrial cancer in Norway from 1953 to 2007, comparing these trends with those in 3 other Nordic countries (Sweden, Denmark and Finland). Based on the Norwegian trends, we provide 2 model-based scenarios of the future burden of endometrial cancer in Norway, predicting the number of new cases and incidence rates in 2015 and 2025.
Material and Methods
Incident cases of corpus uterine cancer (ICD-10 C54) were extracted from the Cancer Registry of Norway for the years 1953–2007, with corresponding person-years at risk within age groups obtained from the population data available at Statistics Norway (http://www.ssb.no). The Norwegian cancer registration system is in accordance with international standards and has an estimated completeness of around 98% overall, and above 99% for cancers of the corpus uteri for the years 2001–2005.19 A high level of ascertainment is considered the result of mandatory reporting from all hospitals, laboratories and general practitioners in Norway, in combination with the routine trace-back systems in place.19
The vast majority of cancers of the uterine corpus are endometrial adenocarcinomas, and hereafter we refer to the disease as endometrial cancer.
To assess the time trends, the age distribution of the world standard population20, 21 was used to calculate age-standardized incidence rates per 100,000 by 5-year time periods (1953–1957, 1958–1962, …, 2003–2007). Truncated age-standardized rates and cumulative risk for the age groups 0–54 and 55–79 years were also calculated, with lifetime risk estimated for all ages up until 80, on assuming an absence of competing causes of death. A comparison of 5-year moving averages of the age-standardized rates in Denmark, Finland and Sweden were extracted from the NORDCAN website22 to serve as a comparison.
Restricting the age range to 35 to 79 years, 10-year synthetic cohorts were defined for women born in 1874–1883, 1879–1898, …, 1964–1973 on subtracting the midpoints of 5-year age groups from the corresponding midpoints of 5-year aggregates of calendar time. The observed trends were presented as rates versus birth cohort by age, and rates versus calendar period by age, with quasi-parallelism of the age-specific curves on either time scale, an indication of their respective influence on the temporal pattern.
Cohort effects are established if changes in rates are seen in successive birth cohorts. They commonly arise through a changing prevalence and distribution of one or more key risk factors for the disease under study. Period effects are characterized by an immediate or fixed-delayed change in the incidence for all age groups, and may therefore act as surrogate measures of exposures that quickly change rates across all age groups. They may include changes in classification criteria, the availability of new diagnostic tests, or specific interventions that affect rates in all studied age groups.
To formally assess the importance of the nonlinear effects of period and cohort and the goodness of fit of submodels, a standard analysis of deviance of nested APC models was performed, treating age, period and cohort as factors.23, 24
The APC model cannot estimate the individual linear components of the age, period and cohort effects due to their linear dependence.25 Thus, we present 2 unique sets of estimates from the full APC model, one with an allocation of the net drift (the identifiable sum of the period and cohort slopes) to birth cohort, and another assigned to calendar period. By setting the linear component of the period (cohort) effects to zero, we present the cohort (period) effects, hypothesizing that they predominate the underlying trends. The method still enables presentation of the nonlinear period (cohort) effects in each set of estimates. The model analysis was performed using the APC functions available in the library Epi (version 1.0.8) in R,26 and specifically the apc.fit command. Both period and cohort effects are presented as rate ratios, with the age-specific rates relative to the corresponding median cohort or period. Stata 10 was used for data management and the plotting of the observed trends.27
Future predictions for 2015 and 2025
To predict incidence trends for four 5-year periods (2008–2027), APC models were fitted that allowed 2 different scenarios of the disease burden by extrapolating the 4 most recently observed 5-year periods (1988–2007): Scenario A involved use of (i) a power function to level off the growth, thus avoiding the overestimation of cases associated with the method based on the multiplicative model, and (ii) projection of the recent linear trend based on the last 10 years (1998–2007) of observed data, attenuated by 25, 50 and 75% in the second (2013–2017), third (2018–2022) and fourth (2023–2027) prediction period, respectively. In a previous comparison of 15 different prediction methods, it was shown that the underlying model provided among the best estimates of future cancer burden.28 The predicted rates and numbers based on scenario A represent a conservative future pattern. An alternative scenario B used the same specifications as in (i) above, but assumed no attenuation of the linear trend (a constant drift) in future periods. For both scenarios we present the numbers of new cases and incidence rates for the midpoints of the second and fourth period, in 2015 and 2025, respectively.
The numbers of new cases were predicted by multiplying the projected incidence rates by official national population forecasts for future years obtained from Statistics Norway for 2015 and 2025 by the 5-year age group, based on the “medium variant” and assumptions on future fertility, life expectancy, internal mobility and net immigration. Differences predicted numbers of new cases, relative to those observed around 2005 are presented on partitioning the changes into those due to demographics (population aging and growth) and those due to changing risk (rates).
Trends by calendar period in Norway and the Nordic countries
Endometrial cancer rates in women younger than 55 years tend to be higher in Norway than in other Nordic countries, but at older ages, rates in Norway, Sweden and Finland have converged during the last 2 decades (Fig. 1). The cumulative risk (prior to 80 years) has increased over time and is estimated to ∼2.5% for women diagnosed between 2003 and 2007. This estimate suggests that 1 in 40 women will be diagnosed with endometrial cancer during their lifetime.
Between 1988 and 2007, the annual incidence increase in endometrial cancer was 1.7% among Norwegian women under 80 years of age (Table 1), however, the temporal increase differed by age group. In women 55–79 years, the annual increase was roughly the same during the 2 decades; 2.1% (95% CI: 0.9%, 3.4%) from 1988 to 1997 and 1.7% (95% CI: 0.6%, 2.8%) from 1998 to 2007. In women younger than 55 years, there was an annual reduction of 0.6% (95% CI: −2.3%, 2.2%) from 1988 to 1997, followed by an annual increase of 1.7% (95% CI: −0.4%, 3.9%) from 1998 to 2007. Estimated as cumulative incidence, about 1 in 250 women was diagnosed with endometrial cancer in Norway before the age of 55 years during 2003–2007.
Table 1. Endometrial cancer incidence in Norway: numbers of new cases, person-years and cumulative risk 1988–1992 and 2003–2007, and estimated annual percentage change 1988–1997 and 1998–2007 for ages 0–79, 0–54 and 55–79
A similar decline among women at premenopausal age has taken place in the other Nordic countries, beginning in the early 1980s (Fig. 1). In Sweden, the decline started before the 3 other countries, and has also been stronger. In Finland, there was first a decreasing incidence followed by increasing rates in women younger than 55 years. Denmark represents an exception to the uniform patterns of increased postmenopausal incidence since around 1985. The rates in women 55 years or older declined during the 1980s and early 1990s; however, there has been an increase in incidence in Denmark from the late 1990s.
Age-period-cohort analyses of the incidence 1953–2007 in Norway
Age-specific rates by birth cohort and period of diagnosis are complex and difficult to interpret (Fig. 2). However, it is instructive to partition the rates into premenopausal and postmenopausal age groups. There was a decline in incidence in Norwegian women younger than 55 years both by calendar period and by cohort until 1998, followed by an incidence increase. In women 55 years and older, the rates have consistently increased among successive birth cohorts from the late 19th century and thereafter, and in consecutive periods of diagnoses from the early 1960s.
The full APC model was required in order to yield an adequate fit to the data (deviance = 26.6 on 27 degrees of freedom) (Table 2). All effects in the tabulated models were significant with the exception of nonlinear period effects over and above the drift, and on grounds of parsimony, the APC model can be considered the best-fitting model. Figure 3 presents the modeled trends by age, period and birth cohort based on the APC model and 2 parameterizations (see Methods). Whether we fix the linear trend for either period or cohort to zero, increasing rate ratios are observed by birth cohort in generations born from around 1880 through to 1925. In generations born from 1925 to 1945 the rates are reasonably stable, while for postwar cohorts, either parameterization indicates a decline in endometrial cancer rates among successive generations born after 1945. Assuming that the underlying linear trend is due to calendar period, the analysis of the period effects suggests that rates are uniformly increasing throughout the study period, with the observation of a possible acceleration in rates since around 1998.
Table 2. Analysis of deviance for nested APC models, Norway 1953–2007
Predicted incidence in 2015 and 2025
We present observed (1988–2007) and predicted (2008–2027) trends in age-adjusted rates in Figure 4. Rates in Norway are estimated to peak around 2020, and subsequently decline, assuming a future attenuation in trends (scenario A). With the constant drift (scenario B), rates would continue to increase almost linearly up to the mid-2020s. Table 3 shows the mean number of cases predicted in 2015 and 2025 by age on the basis of the 2 scenarios compared with the mean number observed in 2005. The differences are partitioned according to changing risk (rates) and demographic changes (population ageing and growth). On the basis of the more conservative scenario A, we predict around 873 new cases in 2015, which corresponds to an additional 200 diagnoses on top of the 650 cases observed in 2005, representing a 35% increase in endometrial cancer incidence. Changes in underlying risk (18%) and population aging (17%) contribute almost equally. According to scenario A, there will be around 1,016 new cases annually circa 2025, or 57% more cases than in 2005, with about one-third of the increase due to the ageing of the population. By 2025, there will be a larger proportion (22%) of women diagnosed with the disease at the age of 80 or older.
Table 3. Predicted mean annual numbers of endometrial incidence cases 2015 and 2025 by age group compared with the mean annual numbers observed in 2005, and for all-ages incidence, relative changes due to changing risk and changing demographics (population aging and growth)
Applying scenario B, assuming the recent linear increases in the past will continue in the future, we predict a doubling of the number of new cases by 2025, corresponding to 1,257 new cases. The proportion of cases diagnosed at 80 years or older in 2025 is estimated to 21% (compared to 18% in 2005). The shift toward more cases at older ages is not only related to the ageing of the population, but also to a relatively higher mean annual increase in incidence (of about 4% per annum) is observed from 1998 to 2007 in women 80 years and older (data nor shown).
The age-specific incidence trends of endometrial cancer in Norway from 1953 through 2007 reveal interesting patterns that may be useful in generating hypotheses that are relevant for primary prevention. In women at postmenopausal age (55 years or older), there was a general increase in incidence over the entire period, but in younger women, patterns were not consistent. In women younger than 55 years, there was a general incidence increase until around 1980 that was followed by a decline until 1998, after which the rates again have increased until 2007.
Despite the high validity of the information recorded by the Norwegian Cancer Registry and the well defined Norwegian population, it is possible that increased reporting and improved diagnostic efforts have contributed to the increase in endometrial cancer incidence over time. An easily performed endometrial sampling method introduced in the 1990s may have led to earlier detection of endometrial cancer and thereby to a shift toward increased incidence among younger women. It is possible that the increased incidence in premenopausal and perimenopausal women could be explained by such a drift. However, it may also be explained by the underlying random variation due to small numbers that limit the statistical power of our estimates in this age group.
The lack of adjustment for the rate of hysterectomies in the population is a weakness of this study. Since all women are included in the denominator of our estimations—as opposed to including only women with the uterus in place—the true risk of endometrial cancer incidence may be underestimated in this population. Although the rate of hysterectomies has been historically low in Norway, there was a rapid increase in the 1990s. Recent estimates suggest that about 12% of Norwegian women will have a hysterectomy performed during their lifetime.29 It has been estimated that the incidence of endometrial cancer, adjusted for the prevalence of hysterectomy, may yield 30% higher rates than the crude rates without adjustment.30 However, the predicted future rates could be an overestimate given that increasing hysterectomy rates will reduce the number of women at risk of developing endometrial cancer. Another limitation may be the calculation of synthetic birth cohorts based on period and age. This creates a linear dependency between the time components and a nonidentifiability of the linear slopes for age, period and cohort. We were, however, able to report the estimates of drift, the sum of the period and cohort slopes, and we provide solutions based on an allocation of drift to either period or cohort. This dual presentation was considered necessary given the complexity of the trends and the inability to estimate the extent to which changes in key risk factors, including reproductive patterns, the use of hormone replacement therapy (HRT), use of oral contraceptives, and obesity, could have resulted in cohort or period-related linear changes. A change in use of oral contraceptives and HRT may affect highly specific age groups and time periods, whereas obesity may affect rates in women across all age groups over a fixed recent period.
The consistent increase in incidence observed in the youngest cohorts is most likely due to changes in reproductive factors, including earlier age at menarche and lower parity, which are associated with increased risk for endometrial cancer.7–9, 18, 31 Thus, the lowest risk in Norway was observed in the 1890–1894 birth cohort. It has previously been estimated that about 27% of the increase in cumulative age-adjusted incidence from 1955 to 1984 may be attributed to changes in childbearing patterns.32 Birth rates have stabilized at around 1.8 during the last 30 years (http://www.ssb.no), and maternal age at first birth has gradually increased during the last 40 years. However, studies of the impact of age at first or last birth on endometrial cancer risk have shown conflicting results7, 9, 10, 18 and therefore, we do not know whether delayed childbearing has contributed to the increase in incidence.
Menopausal treatment using estrogens unopposed by progesterone increases the risk of endometrial cancer in a dose-risk manner,5 and combined regimens of estrogen and progesterone were introduced to prevent this side effect. Nevertheless, small increases in risk have also been reported for the combined treatment33, 34 and this may account for the increasing rates observed in postmenopausal women born after 1940. In the late 1980s less than 6% of postmenopausal Norwegian women used HRT,35 but the use increased rapidly in the 1990s, and reached a peak of about 35% toward the end of the decade.36 At that time, the sale was dominated by combined estrogen-progesterone preparations. Since then, sales of HRT have decreased by more than 50%.37 It seems plausible that the recent decline could have attenuated the incidence increase and it is possible that the modest incidence reduction in the age group 55–59 years could reflect such an attenuation. Increased prevalence of other risk factors, including obesity and diabetes type 2, may have contributed to the persisting incidence increase in postmenopausal women.
It is well established that the use of COCs protects against endometrial cancer later in life. In the late 1980s, ∼21% of Norwegian women in the age group 20–44 years were regular users of COCs,38 and the proportion of users has subsequently increased.37 In Norway, the Wholesaler-based Drug Statistics report total COC consumption from the late 1960s, and the sales have increased from 84,790 DDD/day (defined daily dose) during 1970–1979 to 159,256 DDD/day in the period 1990–1999. The declining endometrial cancer rates at premenopausal age starting from the early 1980s may therefore, in part, be ascribed to COC exposure.
The increasing incidence of endometrial cancer in the last decade across all birth cohorts born after 1950 seems to be in conflict with the increased use of hormonal contraception. Therefore one may speculate whether other risk factors, such as increasing body mass may have contributed to the increase in rates. Body mass index is linearly associated with the risk of endometrial cancer.11 Overweight doubles the risk, and the risk may be up to 6-fold higher among very obese women (BMI ≥ 40). In Norway, population-based health surveys have documented increases in BMI during the 1980s and the 1990s.39, 40 Mean body weight has increased in all age groups younger than 70 years, and the largest increase (7.3 kg) has been observed in women 20- to 29-years-old. The consistent increase in body mass has been observed since 1985, and the proportion of obese (BMI ≥ 30) women in the age group 20–29 years had tripled between the 2 surveys (3.7% vs. 11.9%). The most recent data (2000–2003) from 5 Norwegian counties suggest that body mass continues to increase.41
The recent increase in endometrial cancer incidence among women at premenopausal age coincides the strong increase in BMI in young women. The increased use of COCs may have contributed to an attenuation of an underlying endometrial cancer epidemic caused by an increasing BMI in the population. A similar incidence increase has been observed in Finland, but not in Sweden. Also in Finland, mean BMI has increased in women younger than 30 years over the last 20 years,42 whereas in Sweden, an increase in mean BMI has not been observed.43, 44
The impact of BMI may be strongest for endometrioid adenocarcinoma,45 but few, if any studies have assessed the association of risk factors with different histological subtypes of endometrial cancer. We studied endometrial cancer as a single entity, and could not address the possibility that relevant risk factors may have different impact on different histological subtypes.
Previous predictions of endometrial cancer incidence in Norway were based on observations until 1997, and the results predicted a stabilization of rates thereafter.46 For the Nordic countries combined, a decline by 15% from 1993–1997 to 2018–2022 has been predicted. Our analyses suggest that previous predictions were too conservative, and based on attenuated future trends (scenario A) we predict a continuing increase of 57% until 2025 compared to the observed rate in 2005. This scenario implies a decreased exposure to underlying risk factors over time. There are to date no studies that have evaluated if the risk increase due to BMI depends on age or duration of exposure. If the BMI effect depends on duration, the full BMI effect is yet to be seen. The less optimistic scenario B in this study that predicts higher rates of endometrial cancer incidence implies unchanged trends in exposure to underlying risk factors in the years beyond the observation period. It is in accordance with our previous finding that even a modest increase in population body mass will lead to an increase in endometrial cancer incidence.11 We therefore believe this scenario may provide a more realistic prediction of future endometrial cancer incidence should recent trends hold into the future. There are to date no reliable predictions of the development of BMI in Norway, but an increasing and substantial proportion of adolescents are overweight or obese.47 The proportion of endometrial cancers due to obesity in Europe has been estimated to about 40%,48 and by preventing obesity, a considerable proportion of endometrial cancer cases could be avoided.
In conclusion, this study provides evidence of a consistent increase in the incidence of endometrial cancer during the last 50 years. The increase has been more pronounced in postmenopausal women, but in the last decade the same trend has been observed in premenopausal and perimenopausal women. Our long-term predictions imply that the burden of endometrial cancer will continue to increase in the forthcoming decades. Further research that provides a better appreciation of the underlying risk factors and their prevention will therefore remain a major objective in coming years.