On the age-dependent association between cancer of the breast and of the endometrium. A nationwide cohort study.

The association between breast and endometrial cancer was investigated in a cohort consisting of 60,065 subjects (99% of all women in whom a first breast cancer was diagnosed in Sweden in 1960-63 and 1968-81). Complete follow-up until 1981 revealed a total of 260 endometrial cancers, as against an expected number of 151.1 (relative risk (RR) = 1.72; 95% confidence limits (CL) 1.46; 1.87). RR increased steadily from close to unity in women younger than 50 at breast cancer diagnosis to 2.40 (CL 1.97; 2.93) in those 70 years of age and older. The excess number of endometrial cancers occurred primarily during the first five years of follow-up (RR = 2.07; CL 1.79; 2.38). A common causal agency for breast and endometrial cancer is more likely to lie in environmental than in genetic factors and other observations in the same population do not support that such factors are related to characteristics of the women's reproductive histories.

Breast and endometrial cancers are generally believed to have aetiologic factors in common, primarily of a dietary and endocrine origin (Dunn, 1975;Cole & Cramer, 1977;Howe et al., 1984;Henderson et al., 1982;Willett & MacMahon, 1984). Descriptive epidemiologic studies have revealed high correlations between breast and endometrial cancer with regard to incidence and mortality rates. Such observations have been reported from international investigations in both low-and high-risk countries and in migrant populations (Dunn, 1975) and also from different parts of high-risk countries such as Canada (Howe et al., 1984) and the United States (Hoover et al., 1975;Winkelstein et al., 1977).
The possibility that these demographic correlations reflect common genetic or environmental aetiologic factors has gained some support from cohort studies in which an increased risk of endometrial cancer has been found in breast cancer patients (Schoenberg et al., 1969) and vice versa (Schottenfeld & Berg, 1971;Bailar, 1963;MacMahon & Austin, 1969). The knowledge that can be derived from these analytic studies is generally uncertain, however, because of small numbers of cases, contradictory findings in blacks and whites (Newell et al., 1974), and limited information as to the possible dependence of an association between breast and endometrial cancer on age at diagnosis of the first primary (Bailar, 1963;MacMahon & Austin, 1969). The aim of this investigation was to confirm and extend in a larger cohort our recent finding that the association between breast and endometrial cancer is seemingly restricted to older women (Adami et al., 1984). The availability of a national cancer registry, reliable incidence figures for the studied population and opportunities for complete long-term follow-up facilitated our analysis.

Material and methods
The cohort Since 1958, when the National Swedish Cancer Registry was started, all newly diagnosed malignant tumours have had to be reported to the Registry. This obligation rests upon both the physician and the pathologist or cytologist who confirms the diagnosis on surgically removed tissues, biopsies, cytologic specimens or at autopsy. As a result, -95% of the cases entered in the Registry are notified from two sources. The overall frequency of underreporting has been shown to be -5% (Mattsson, 1977) and the completeness of registration of breast and endometrial cancer has been assessed from death certificates to be 98 and 95% respectively (Mattsson & Wallgren, 1984).
The cohort was based on all women reported as having a first breast cancer diagnosed in 1960 through 1963 and 1968 through 1981. The four-year period of our previous study (Adami et al., 1984(Adami et al., ), 1964(Adami et al., through 1967 excluded, since one major aim was to rule out the possibility that the excess risk previously found was due to chance (type I error) i.e. to a random high relative risk among the large number of tumour sites and subgroups that were analysed. Patients were included irrespective of whether they had had any other malignant disease reported prior to the breast cancer.
A total of 61,341 women with breast cancer were notified to the Cancer Registry during the period of study. We excluded 583 women with an incomplete national registration number (a number which permits exclusive identification), precluding computerized linkages and followup in the Cancer Registry and other registers. The breast cancer of a total of 693 patients had been diagnosed at autopsy and these patients were therefore not included in the study. As a result, 60,065 women with the age distribution shown in Table I were available for complete follow-up and comprised the study cohort.
Person-years at risk The cancer file is linked annually to the Register of Causes of Death, which covers the entire Swedish population. The date of deathwhich was the only information used in this analysisand the causes of death are then transferred. At the time of this study the Register was complete up to December 31 1981, giving a follow-up period of 0-20 completed years. The cohort was also linked through the national registration numbers with a national register covering all persons who emigrate. The person-years at risk were calculated from the date of the breast cancer diagnosis. The date of the first diagnosis was used in those 3023 women whomost probably because of a metachronous bilateral diseasewere registered more than once for breast cancer. The end of the observation time was defined as the date of endometrial cancer diagnosis, emigration or death, or the closing date of follow-up. By this latter time 55 (0.1%) patients had emigrated and 28,362 (47.2%) had died. The proportions of patients in the cohort observed for at least 5 and 10 completed years were 38.5 and 14.7% respectively. The number of person-years at risk is shown in Table I. H.-O. ADAMI et al.
E.vpected incidence of endometrial cancer Official statistics from the Swedish Cancer Registry (Swedish Board of Health and Welfare. Stockholm 1963-1985) provided age-specific incidence rates for each year of observation. The expected number of persons with endometrial cancer in the cohort was obtained by multiplication of person-years for different 5-year age groups for each year of observation by the corresponding agespecific incidence rates.
Observed incidence of endometrial cancer The register of the cohort was linked through the national registration numbers to the entire Cancer Registry covering the whole Swedish population for the period 1960 through 1981. All endometrial cancers diagnosed during the same month as the breast cancer or during any subsequent month were regarded as second primary cancers. A detection bias might have exaggerated the incidence of endometrial cancer that was diagnosed at the same time as the breast cancer or at autopsy. The cases in which this happened were therefore analysed separately.

Statistical methods
The relative risk was defined as the ratio of observed numbers of cases of endometrial cancer to expected numbers. The 95% confidence limits (CL) of the relative risk were then calculated on the assumption that the observed number of cases follows the Poisson distribution (Bailar & Ederer, 1964).  (Table II). The relative risk for all women 70 years and older was 2.40 (CL 1.97-2.93) which is the same figure as in our previous report (Adami et al., 1984). The cumulative numbers of expected and observed cases of endometrial cancer for each year of observation is presented in Figure I and the relative risk in relation to the duration of follow-up is shown in Table III. The increased risk was greatest in the first years after the breast cancer diagnosis. Analyses by each year revealed relative risks of 2.51, 1.60, 2.11, 1.79 and 2.05 consecutively during the first 5 years of observation. The highest risk was incurred during the first year after diagnosis by patients 70 years of age or older, with 37 observed versus 9.89 expected cases of endometrial cancer (RR=3.74; CL 2.64-5.16).
Years after breast cancer diagnosis Two possibilities of bias were specially analyzed. Firstly, the diagnosis of a breast cancer might have increased the likelihood of having the date of diagnosis advanced for a limited period of time in some women whose endometrial cancer was already symptomatic. A total of 28 (38%) of 66 endometrial cancers which occurred during the first year of observation were in fact diagnosed during the same month as the breast cancer. Exclusion of these 25 cases would not, however, have altered the general pattern of the results or the regular trend towards an increase in relative risk with increasing age.
Secondly, the question of ascertainment bias was addressed. The incidence of endometrial cancer among breast cancer patients might have been overor underestimated as a result of a higher or lower rate of autopsy in such patients than in the general population from which the incidence figures were derived. However, as shown in Table IV, the proportion of endometrial cancers diagnosed at autopsy was of approximately the same low magnitude in both groups of women.

Discussion
In this investigation essentially the same methods were applied as in our previous study in which they were critically reviewed (Adami et al., 1984). Several characteristics of the design indicate that the internal validity is acceptable. Virtually all cases in a defined population could be included and subjected to complete follow-up. The endometrial cancers observed and the number expected were both derived from the National Cancer Registry, which has a low frequency of undernotification (Mattson, 1977;Mattson & Wallgren, 1984). The risk of false positive resultswhich inclusion of a subgroup with a random high outcome might entailwas minimized by excluding the period 1964 through 1967 during which our previous cohort was recruited (Adami et al., 1984). Breast cancers diagnosed during that period were thus used only to genermte the hypothesis which was further tested in a differcnit imiaterial in the present study. Still, the size of the cohort and the duration of follow-up provided us with a sufficiently large number of cases to make the risk of false negative findings reasonably low. The possibility of a higher detection rate of endometrial cancer in breast cancer patients than in the general population due to closer medical surveillance cannot be definitely excluded. The assumption that such a bias should operate particularly in older women is, however, contradicted by the absence of a similar trend towards increased risk at higher ages for cancer of the ovaries, colon and rectum (Adami et al., 1984). In addition, vaginal bleedingwhich is the first evidence of endometrial canceris a dramatic event in postmenopausal women. The impact of patient and doctors delay should therefore be small.
In Sweden radiation-induced menopause has been used as an adjuvant treatment only occasionally in pre-and perimiienopausal women with breast cancer and would thus not affect the risk of endometrial cancer in older women. Hormonal treatment has been used for palliation only in advanced cases with too short survival time for an oestrogen induced endometrial cancer to develop. A confounding effect of radiation to the pelvic area or of oestrogen treatment is therefore unlikely (Ewertz et al., 1984).
The present data confirm our earlier observation that the risk of developing endometrial cancer is increased in breast cancer patients and that this characteristic pertains primarily to older women. This finding, which is in accordance with an earlier report based on a small number of observed cases (MacMahon & Austin, 1969) was extended, furthermore, by the observation of a regular increase in risk with age, the relative risk in women older than seventy years being two to three times higher than that in women under fifty. Moreover, there is seemingly a strong temporal correlation in the occurrence of malignant disease at these two sites; the excess risk was largely confined to the first five years after the breast cancer diagnosis and a substantial proportion of the endometrial cancers become manifest clinically within one year of observation.
The interpretation of these findings is not straightforward. However, the strength and age-dependence of the increased risk and the temporal correlation in the occurrence of the disease suggest that breast and endometrial cancers have at least one aetiologic factor in common, that this factor is not operative at premenopausal ages but becomes increasingly more important after the age of fifty, and that the initiation or promotion of neoplastic growth is synchronized so that the cancers develop to clinical size at about the same period of time.
In principle, causative factors can be genetic or environmental. Earlier studies have shown that familial occurrence of breast cancer is not a common feature in this population. Among women who had a sister with breast cancer the risk of cancer at this site was doubled, whereas the relative risk among those whose mother had had the disease was non-significantly increased to 1.4 (Adami et al., 1981). We concluded from these findings that environmental rather than genetic factors might account for the aggregation of breast cancer within certain families and exert their effect more homogeneously on relatives from the same generation than on those from consecutive generations. With this evidence that genetic determinants only play a minor roleif anyin the occurrence of breast cancer in general, it seems unlikely that they would be of importance for the cases in which endometrial cancer also occursespecially given that these cases seem to occur mainly in relatively old women (Table II) which appears not to be true of familial breast cancer (Anderson, 1974;Adami et al., 1981). Our discussion therefore has to be focused on the possible nature of common environmental factors for breast and endometrial cancer in older women.
The aetiology of cancer in endocrine target organs has traditionally been soughtwith limited progress so farwvithin the paradigm of hormonal expressions of characteristics of reproductive life or, more recently, in dietary habits (Dunn, 1975;Cole & Cramer, 1977;Howe et al., 1984;Henderson et al., 1982;Willett & MacMahon, 1984). It seems unsolved whether the elevated risk of breast cancer in women of low parity and in those of high age at first birth represents a causal relationship (Kelsey & Hildreth, 1983;Adami et al., 1980). Moreover, there is no indication from our data or from others that this association becomes more pronounced with increasing age.
In endometrial cancer, on the other hand, a causal relationship with certain characteristics of the reproductive history (Ewertz et al., 1984;Kelsey & Hildreth, 1983;Pettersson et al., 1986) excess endogenous oestrogens (Henderson et al., 1982) and with oestrogen therapy (Ewertz et al., 1984;Persson et al., 1986;Zeil, 1982) seems to be more firmly established. We have recently found that the relation between a long menstruation spandefined as the period between menarche and menopause, adjusted for anovulatory periods during parity and lactationand an increased risk of endometrial cancer is negatively correlated with age at diagnosis (Pettersson et al., 1986). The term mcnstruation span has been introduced to reflect the number of menstrual cycles in a woman's life and thus summarizes several factors which proposedly may influence the risk of both breast and endometrial cancer. Nevertheless, the enhanced risk of endometrial cancer secondary to a long menstruation span virtually disappears after the age of seventy (Pettersson et al., 1986), when the association between breast and endometrial cancer reaches its maximum.
In conclusion, there is no obvious common risk factor related to reproducive life which could give a reasonable explanation for the age-dependent association between cancer of the breast and cancer of the endometrium. There is some support, rather, for the view that in this population such factors are of little importance for the occurrence of at least endometrial cancer in older women.
Supported by grants from the Swedish Cancer Society.