Risk of Primary Non–Breast Cancer After Female Breast Cancer by Age at Diagnosis

The high incidence of breast cancer in developed countries (1) combined with the favorable survival for the disease (2) makes breast cancer survivorship issues an important topic. Female survivors from breast cancer have been shown to be at increased risk of being diagnosed with additional primary cancers with risks consistently reported as higher among women diagnosed at younger ages relative to older ages (3–15). Younger breast cancer patients more often have unfavorable tumor characteristics and are more frequently treated with chemotherapy (16). Thus, age-related differences in the risk of second cancers may be partly explained by an excess of treatment-induced cancer among the youngest patients. Since it is possible to follow young patients over extensive periods of time, young patients tend to be more frequent among those with long-term follow-up and among those diagnosed in earlier calendar periods. Long-term follow-up (3, 8, 9) and early calendar years (8, 9) have been found to be associated with a higher risk of second non–breast cancer, but previous large-scale cancer registry–based studies have generally reported crude risks for second cancer by age, i.e., risks that were not adjusted for length of follow-up and calendar year.

We aimed at testing the hypothesis that the relative risk (RR) of subsequent non–breast cancer according to age at breast cancer diagnosis is dependent on length of follow-up and/or calendar period of breast cancer diagnosis by using multivariate models applied to a cohort of approximately 300,000 women with breast cancer from Denmark, Finland, and Norway. We also investigated how much of the age effect may be explained by age-related differences in treatment by adjusting the models for treatment. In addition, we assessed how the relative and the absolute risks of a subsequent cancer throughout life depend on the age at breast cancer diagnosis.

Women with breast cancer at all stages at ages above 20 years were identified in the population-based Cancer Registries in Denmark (1943–2006; refs. 17, 18), Finland (1953–2006; ref. 19), and Norway (1953–2003; ref. 20). These Nordic Registries are national and are based on reports of cancer from multiple sources which increase both the completeness and the quality of the data. In addition to malignant neoplasms, the registries also include papillomas of the urinary tract and histologically benign tumors of the central nervous system and meninges, and these tumors were included in the analyses. In situ cancers such as carcinoma in situ of the cervix are also reported to the registries but were not included in the analyses. There are different registration procedures for non-melanoma skin cancer and contralateral breast cancer in the 3 registries, and therefore, breast cancer patients were not followed up for these types of cancer. From each Cancer Registry, the following variables were collected: date of birth, vital status, date of breast cancer diagnosis, and initial treatment of breast cancer in 3 broad categories (radiotherapy, chemotherapy, and hormonal therapy). Initial treatment was defined as treatment initiated within 4 months of the date of diagnosis. Information on histology was not obtained from the registries, because it has not been registered for the entire study period in Denmark and Norway. Extent of tumor at time of diagnosis was also not obtained.

Information on all primary cancers (except contralateral breast cancer and non-melanoma skin cancer) occurring among women in the breast cancer cohort was obtained by a search through each of cancer registry files. Those diagnosed with cancer (other than non-melanoma skin cancer) prior to a breast cancer diagnosis were excluded from the study. For all malignancies diagnosed after a breast cancer diagnosis, the following variables were obtained from the registries: date of diagnosis and type of cancer [International Classification of Diseases, 7th revision (ICD-7) or 10th revision (ICD-10)]. The subsequent cancers were grouped according to possibly causative factors into:

• radiotherapy-related: salivary glands, esophagus, lung, pleura, bone, connective tissue, and thyroid gland (9);

• chemotherapy-related: leukemia (21);

• tamoxifen-related: endometrial cancer (22);

• BRCA-gene–related: ovarian cancer (23)

• alcohol-related: mouth, pharynx, liver, and larynx (24); and

• overweight/obesity-related: colorectum, gallbladder and bile ducts, pancreas, and kidney (25, 26).

Several cancer types are likely to be associated with more than 1 of the aforementioned risk factors, but we did not allow overlap between the groups. Cancer sites belonging to treatment-related sites and other groups (e.g., esophageal and endometrial cancer) were included among the treatment-related sites. Colorectal cancer which is associated with both overweight/obesity and alcohol was placed in the overweight/obesity group, because the evidence for this association is stronger than that for alcohol (25).

The follow-up period started at the first day of the month following the month when the breast cancer was diagnosed and continued until date of death, emigration, or end of study [December 31, 2006 (Denmark and Finland) or December 31, 2007 (Norway)].

Person-years and observed cancers after the breast cancer diagnosis were assigned to strata defined by 5-year intervals of attained age and calendar year. For each country, cancer incidence rates among the general population in corresponding age and calendar year intervals were multiplied with the strata-specific accumulated person-years to estimate the number of expected cancer cases. Crude RR estimates for subsequent cancer were calculated as standardized incidence ratios (SIR) with 95% CIs, assuming Poisson distributed number of observed cancers. Multivariate log linear Poisson regression models were used to estimate RRs of subsequent cancer according to age at breast cancer diagnosis (<40, 40–49,….,70+ years) with adjustment for calendar period of breast cancer diagnosis (1943–1952,…., 2003–2006), length of follow-up (<1, 1–4, 5–9, and 10+ years), radiation therapy (yes, no, and unknown), chemotherapy (yes, no, and unknown), and hormonal therapy (yes, no, and unknown). Competing risks have been considered by comparison of cancer rates with censoring of the patients when they die or emigrate. To investigate whether risk by age at breast cancer diagnosis was similar throughout the study period, we tested the interaction between age at breast cancer diagnosis entered as a continuous variable in 1-year age groups and the calendar periods defined previous. The GENMOD procedure in SAS version 9.1 was used for the Poisson regression analyses (SAS Institute). Finally, we computed the excess absolute risk (EAR) as the observed minus the expected number of subsequent cancer cases divided by the accumulated person-years.

We identified a total of 304,703 women with breast cancer as the first primary cancer. The number of patients in each country is shown in Table 1 along with the distribution of patients by age at breast cancer diagnosis, calendar period of diagnosis, and type of breast cancer treatment. The women were followed for an average of 8.8 years (>0–62 years). Among 21,502 women with a new malignancy during follow-up, 19,833 (92%) had 1 malignancy, 1,544 (7%) had 2, 117 (0.5%) had 3, and 8 (0.04%) had 4 malignancies. Thus, a total of 23,304 new primary cancers (excluding contralateral breast and non-melanoma skin cancer) were diagnosed during follow-up, whereas 20,330 cancers were expected resulting in an overall SIR of 1.15 (95% CI = 1.13–1.16) and with SIRs of 1.30 (95% CI = 1.24–1.36) less than 1 year after breast cancer diagnosis, 1.05 (95% CI = 1.02–1.07) 1 to 4 years after diagnosis, 1.17 (95% CI = 1.14–1.20) 5 to 9 years after diagnosis, and 1.17 (95% CI = 1.15–1.20) 10 or more years after diagnosis.

In the crude SIR analysis, the risk of developing a new primary cancer clearly decreased with increasing age at breast cancer diagnosis, and this pattern was largely unchanged after adjustment for calendar period of breast cancer diagnosis, length of follow-up, and country of residence (Table 2). Comparable risk patterns by age were obtained when data from each of the countries were analyzed separately and when starting follow-up 1 year after the breast cancer diagnosis (data not shown). The decreasing linear effect of age at breast cancer diagnosis was similar for all calendar periods (P = 0.32; data not shown). The association between breast cancer and subsequent cancers by age at breast cancer diagnosis was largely unchanged after further adjustment for radiotherapy, chemotherapy, and hormonal therapy (Table 2). Given the evidence of a strong association between young age at breast cancer diagnosis and risk of ovarian cancer, we omitted this cancer and still found a clear decrease in risk of a subsequent primary cancer with increasing age at breast cancer diagnosis.

We found an inverse relationship between EAR and age at diagnosis, with EAR decreasing from 252 per 100,000 person-years among patients below age 40 years at breast cancer diagnosis to 183 per 100,000 person-years among patients aged 40 to 49, 130 per 100,000 person-years among patients aged 50 to 59, 90 per 100,000 person-years among patients aged 60 to 69, and −67 per 100,000 person-years among patients above age 70 years at breast cancer diagnosis.

The RR of subsequent cancer among patients diagnosed with breast cancer before age 40 years decreased by attained age until around age 50 years and was rather constant by increasing attained ages from around 50 to 75 years (Fig. 1A). The same pattern was seen for those diagnosed at older ages with the curve for each subsequent age group at a lower level. Each of the age groups also had their own level of EAR by attained age being higher, the younger the patients were at breast cancer diagnosis (Fig. 1B). The EAR tended to increase by attained ages from around 50 to 75 years for all age groups with the highest absolute risks found when patients were in their 60s and early 70s.

We found significantly increased SIRs for the groups of possibly treatment-related cancer sites, BRCA gene-, and overweight/obesity-related cancer sites (Table 3). The adjusted RRs decreased by increasing age at first breast cancer, though the slope of the decrease varied (Fig. 2A, B, D, and F). For the tamoxifen-related site, endometrial cancer, the risk only decreased for patients less than 50 years at diagnosis; at older ages at diagnosis, the risk increased by increasing age at breast cancer (Fig. 2C). Among other specified sites, the risk of a subsequent cancer decreased significantly by increasing age at diagnosis for stomach cancer, melanoma, and urinary bladder cancer and nonsignificantly for several other specified cancer sites (Table 3). These age-specific results were adjusted for calendar year, length of follow-up, and country; after further adjustment for treatment, the results for all 3 groups of possibly treatment-related cancer sites changed only slightly (data not shown). Exclusion of the first year of follow-up had no major impact on the risk patterns by age seen in Figure 2A to F (data not shown).

Age at breast cancer diagnosis remains a key determinant for the RR of developing a new primary non–breast cancer when calendar year of breast cancer diagnosis, length of follow-up, and breast cancer treatment have been taken into consideration. The age-specific variation in RR seems to persist throughout life. The risk pattern by age was observed during the last 5 decades, despite major changes in treatment and in prevalence of risk factors within the populations. A decrease in risk by increasing age at breast cancer diagnosis was seen for the majority of cancer sites–1 exception was endometrial cancer for which the risk increased by increments of ages above 50 years at diagnosis.

Our finding of a clear inverse trend in the SIRs for subsequent non–breast cancer by age at first breast cancer diagnosis is consistent with recent cancer registry-based studies from the United States and the Netherlands (3, 4). We adjusted the risk estimates by age at breast cancer diagnosis for length of follow-up and calendar period to separate the collinear effects of age, length of follow-up, and calendar period. The age effects were quite similar with and without such adjustment, suggesting that the age effect is rather independent of length of follow-up and calendar year. Also, initial treatment for breast cancer had little impact on the age effects in the study population. To our knowledge, only 1 study presented adjusted risk estimates; in this study that included a subset of the Danish breast cancer patients in the present study, age effects were adjusted for length of follow-up and treatment, and the risk pattern by age was similar to ours for the 3 cancer sites shown (11). We found a larger absolute excess of cancer cases among women who were young at diagnosis than women who were older at diagnosis, and therefore the decreasing RR is unlikely to be merely a consequence of low cancer rates in the young background population. Thus, age at breast cancer diagnosis seems to be a good indicator for susceptibility to develop a new cancer, and this level of risk seems to remain different for each age category throughout most of a lifetime.

For most of the cancer sites or groups of cancer sites, we found a decrease in risk by increasing age at breast cancer though the magnitude of decrease varied. We found a steep decrease in risk for ovarian cancer by age at breast cancer in accordance with earlier studies (3–5); deleterious mutations in the BRCA1 and BRCA2 genes predisposing to breast cancer and ovarian cancer are likely to be the underlying explanation for this observation (27).

Breast cancer patients treated with chemotherapy have previously been found to have an excess risk of leukemia (4, 28, 29). A prior study classified cancer of the esophagus, lung, pleura, thyroid, bone, connective tissue, and salivary glands as potentially radiotherapy-related sites based on a review of the literature (9). We found a moderate decreasing trend in risk by age at breast cancer diagnosis for these cancer sites combined and for leukemia. A similar pattern was seen in the Surveillance Epidemiology and End Results (SEER) data from the United States (3). Our results changed only slightly following adjustment for treatment, indicating that the pattern could not be caused by a higher frequency of treatment among young patients. However, the moderate decrease by age at diagnosis of breast cancer may possibly be a consequence of a higher sensitivity toward radiation or chemotherapy at younger ages at exposure, as suggested by results on radiotherapy treated 5-year survivors from SEER registries (30), and in parallel to what has been observed for radiotherapy and contralateral breast cancer (31, 32).

We found a weak U-shaped relationship between age at breast cancer and risk of endometrial cancer consistent with some previous studies (3, 4), whereas 1 study reported a clear increasing trend by increasing age (5). The increase in risk of endometrial cancer by age confined to older age groups is in contrast to the findings for most cancer sites, where risk decreased by age throughout the age spectrum or no trend by age was seen. The reason for this atypical excess at older ages is not clear, but different risk factors could be responsible for the excess seen in youth and later in life, respectively. In addition to treatment with tamoxifen that is associated with an increased risk of endometrial cancer (22), shared etiology related to hormonal exposure such as reproductive history and/or obesity (25) may play a role.

Even though alcohol intake is a risk factor for breast cancer (25), we found RRs below 1 for alcohol-related cancer sites for all age groups except age at diagnosis less than 40 years. The effect of alcohol may to some extent be underestimated due to possible underreporting of liver cancer as primary cancer, because the liver is a common metastatic site for breast cancer. A weak decreasing trend in risk of overweight/obesity-related sites by age at diagnosis of breast cancer was seen with no significantly increased risks for ages above 59 years at breast cancer diagnosis. Since increased body fatness is positively associated with postmenopausal breast cancer but negatively associated with premenopausal breast cancer (25), the observed pattern is not consistent with excess body fatness as the underlying explanation. The finding of a bidirectional association between breast cancer and malignant melanoma in combination with reports of a higher risk of melanomas among BRCA mutation carriers and a higher risk of breast cancers among carriers of mutations in the melanoma susceptibility gene, CDKN2A, has led to the suggestion of a genetic link between the 2 diseases (33). However, although we found a significant trend by age, the lack of a markedly pronounced excess of melanomas among the youngest breast cancer patients in our and other studies (3, 4) does not favor this explanation. Common risk factors for early-onset breast cancer and malignant melanoma may also explain the observation.

Our study included breast cancer patients from 3 national and population-based Nordic Cancer Registries with follow-up for new primary cancers up to 60 years after initial breast cancer diagnosis. By use of Poisson regression models, we assessed RR stratified by age at breast cancer diagnosis and adjusted for calendar period and length of follow-up. We restricted the study to Nordic Cancer Registries that include data on treatment to enable adjustment for treatment effects; however, it should be noted that some misclassification may occur because only initial treatment is registered and that type of treatment was missing for approximately 15% of patients, so we may not have fully adjusted for the treatment variables. Also, we did not have information on genetic, lifestyle, and reproductive risk factors for breast cancer. We restricted the study population to first primary breast cancer cases to avoid interference from treatment of another cancer preceding the breast cancer. Contralateral breast cancers were not included as an outcome, but follow-up continued independently of whether a contralateral breast cancer or a non–breast cancer occurred, because the background cancer incidence rates were multiple cancer rates. Any treatment of a second cancer may affect the risk of a third cancer; however, a relatively small part of breast cancer patients have more than 1 subsequent cancer diagnosis (3), so this is unlikely to have had a substantial impact on our results. Possible misclassification of metastases as new primary cancer and more intense screening for new cancers may lead to overestimation of risk for new primaries, although screening, e.g., for cervical cancer, could have the opposite effect as precancerous lesions are identified.

The findings of our large-scale study underline the importance of age at diagnosis of the first primary breast cancer as a predictor of developing a new primary non–breast cancer independently of the calendar period of diagnosis of the breast cancer and the time passed since diagnosis.

No potential conflicts of interest were disclosed.

We thank A. Bautz at the Institute of Cancer Epidemiology, Danish Cancer Society, and S. Larønningen at the Cancer Registry of Norway for computer assistance.

This work was supported by the Nordic Cancer Union (NCU; S-01/07).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.