Breast cancer risk following radiotherapy for Hodgkin lymphoma: modification by other risk factors

Young women (aged ≤ 30 years) treated with chest radiotherapy for Hodgkin lymphoma (HL) have among the highest breast cancer incidence rates of any population, outside familial cancer syndromes. Compared with the general population, breast cancer risks have been elevated 4- to 56-fold in treated women.^1,2  In a recent international case-control study,³  young women who received a radiation dose of 4 Gy or higher to the area of the breast in which cancer developed had a 2- to 8-fold increased risk compared with women who received lower doses. Similar findings were reported in a Dutch investigation of women aged 40 years or younger treated for HL,⁴  nearly 80% of whom were included in the international study. Yet, many young women who receive chest radiation doses of up to 60 Gy do not develop breast cancer, suggesting that endogenous or exogenous exposures mitigate risk. In other radiation-exposed populations, factors that have influenced breast cancer relative risk for a given dose include age at exposure and attained (current) age, with the highest relative risks for both evident at the youngest ages.^5,6  Relative risk rarely has been elevated among women exposed after 50 years of age,^5-8  a possible surrogate for menopause.

Women who develop de novo breast cancer before the age of 50 years or before menopause have a risk factor profile that differs somewhat from that of women diagnosed at older ages or after menopause.^9,10  A family history of breast cancer and a benign breast disease diagnosis are generally stronger risk factors for breast cancer in younger (< 50 years old) than in older women.^11,12  In addition, reproductive factors such as age at first full-term pregnancy and number of live births commonly have a relatively minor effect on breast cancer risk at young ages, while they are established risk factors for older-onset cancer.¹³  Compared with nulliparous women, parous women are likely to have an increased risk at young ages, possibly concentrated within 3 to 10 years of pregnancy, and a reduced risk at older ages.^13-16  Similarly, women in the lowest quartile of body mass index generally have an increased risk of breast cancer before age 50, compared with heavier women, but a reduced risk thereafter.¹⁷  The consistency of such findings has led to the consideration of pre- and postmenopausal breast cancer as diseases with distinct etiologies.

In several investigations, these and other established risk factors for breast cancer have been examined as possible modifiers of risk among women exposed to radiation.^18-21  Although none emerged as strong or consistent risk modifiers, identification may have been hampered by small numbers of cases, or by inclusion primarily of women who were either postmenopausal or long-term breast cancer survivors.

In this investigation, we examined whether breast cancer risk factors might modify risk among a group of young women (≤ 30 years old) who received radiation to the breast during HL treatment.

A detailed description of the study methods has been published.³  General population features are summarized in Table 1. A cohort of 3 817 female 1-year survivors of HL, diagnosed at 30 years of age or younger between January 1, 1965, and December 31, 1994, was followed through September 30, 1999, in 6 population-based cancer registries (Iowa, Denmark, Finland, Sweden, Canada, and 4 affiliated tumor registries in the Netherlands: The Netherlands Cancer Institute, the Dr Daniel den Hoed Cancer Center, Leiden University Medical Center, and Catharina Hospital Eindhoven). Record linkage methods were used to identify 105 cohort members diagnosed with primary breast cancer. At least 2 controls for each confirmed case were selected by stratified random sampling, matching on registry, age at HL diagnosis (± 3 years), calendar year of HL diagnosis (± 5 years), and survival after HL for at least as long as the breast cancer case, resulting in a total of 266 matched controls. The median age at HL diagnosis was 22 years (range, 13-30 years; Table 1), and the median age at breast cancer diagnosis was 41 years (range, 27-57 years). The study was exempted from review by the National Cancer Institute (NCI) Institutional Review Board because it used only existing and anonymized data.

For all patients, detailed data regarding HL treatment were used to estimate radiation dose to the area of the breast in which cancer developed and the comparable area in controls. In our previously published findings,³  breast cancer risk increased with radiation dose, with risk elevated 8-fold among women who received a breast dose of greater than 40 Gy. Women treated with alkylating agent chemotherapy (35% of breast cancer cases, 50% of controls) experienced a 40% reduction in breast cancer risk, and risk was reduced 60% among women who received a radiation dose of 5 Gy or greater to the ovaries (6% of cases, 15% of controls). Radiation doses less than 5 Gy to the ovaries were not associated with breast cancer risk.

Breast cancer risk factors were collected primarily using a structured data collection instrument to abstract medical records, and at a few sites, self-administered interviews⁴  or linkage to national cancer registries (to obtain family cancer history) were also used (Denmark, Finland). Each control was assigned a cutoff date, analogous to the breast cancer diagnosis date in the matched case, and only information on experiences prior to the cutoff date was included. To determine whether the results were sensitive to data collection methods, we conducted analyses in which data from each participating registry were consecutively excluded.

Conditional regression was used to calculate odds ratios (ORs) and 95% confidence intervals (CIs) for the relationship between breast cancer risk factors, radiation dose to the breast, and breast cancer risk among the matched case-control sets. Cutpoints for categoric variables were selected in part to allow calculation of statistical interactions on a multiplicative scale, using medians or quartiles of control distributions when possible. Only 1 case and 10 controls were not treated with radiotherapy. Thus, to obtain a sufficient number of unexposed women in subgroups for interaction analyses, the reference group was defined as women who had received a lower than 5 Gy breast dose (23 cases, 95 controls) (previous cutpoint of lower than 4 Gy³  did not allow interaction analyses). In some analyses, finite estimates of relative risk on a continuous dose scale could only be obtained using a modified dose variable that set doses lower than 5 Gy to 0. To evaluate interaction on a multiplicative scale, we fitted the model:

where exp (Σ_j α_j x_j) is the background risk, which may be adjusted for variables x_j, z₁ is an indicator variable for the breast cancer risk factor of interest; and z₂ denotes radiation dose. Departure from the multiplicative model was evaluated by testing whether γ (the “Interaction OR”) = 0 (exp γ= 1.0). Interaction was also evaluated on an additive relative risk scale by fitting the model:

The parameter γ is known as the RERI (relative excess risk attributable to interaction), or the ICR (interaction contrast ratio),^22,23  and equals the relative excess risk in those with both risk factors (RR11-1) that remains after subtracting the relative excess risk for each individual factor in the absence of the other (RR01-1) and (RR10-1). An ICR equaling 0 indicates an exact additive relation between the 2 risk factors, while an ICR greater than 0 suggests a greater than additive effect, and an ICR less than 0 implies a less than additive joint effect. We also calculated γ* = (β₂ +γ)/β₂, which is the ratio of radiotherapy (RT)–related excess relative risks (ERRs) when the risk factor is present compared with when it is absent. Although this case-control study cannot directly evaluate absolute risks, the ratio of the ERRs for those with and without the factor of interest expressed relative to a common baseline should be pertinent to this comparison. Multiplicative models were examined using SAS software version 8 (SAS Institute, Cary, NC), and additive relative risk models were examined using the Pecan module of the software package EPICURE (HiroSoft International, Seattle, WA).

Analyses regarding ever having a live birth, age at first and number of births, timing of births in relation to HL or breast cancer diagnoses, or oral contraceptive use were limited to women who did not receive HL treatment with alkylating agents (AAs) and who received a less than 5 Gy radiation dose to the ovaries, as those treatments can alter ovarian function,²⁴  induce menopause, influence childbearing, and also reduce breast cancer risk.^3,4,25 

Factors considered as potential confounders in the analysis included age at menarche, body mass index at HL diagnosis, age at menopause, menopausal status, ever having a live birth, age at first and number of live births, timing of live births in relation to diagnosis of HL or breast cancer, first- or second-degree family history of breast or ovarian cancer, and oral contraceptive use. Analyses were adjusted for HL treatment, including breast radiation dose, number of cycles of AA, and radiation dose of 5 Gy or higher to the ovaries, except as noted.

Women with a first- and/or second-degree family history of breast or ovarian cancer had a 2.5-fold increased breast cancer risk (95% CI, 1.2-5.3), compared with those without such history (Table 2). Differences between cases and controls with regard to other breast cancer risk factors generally resembled those identified in previous studies of premenopausal women (Table 2). Breast cancer risk was reduced among women who received AA chemotherapy or radiation of 5 Gy or higher to the ovaries, which may induce early menopause,^3,26  and the lower risk among exposed women who were postmenopausal at study end should be interpreted in that context. Since cancer treatment may affect fertility in patients with HL, only women who did not receive AA or radiation of 5 Gy or higher to the ovaries (n = 68 patients, 125 control subjects) were included in analyses regarding reproductive factors or oral contraceptive use. These factors, however, had little influence on breast cancer excesses.

Overall, women who received a breast radiation dose of 5 Gy or higher had a 2.7-fold increased breast cancer risk, relative to women who received lower doses (Table 2). For most risk factors in Table 2, there was no statistically significant interaction between the risk factor and radiation in relation to breast cancer risk, in either multiplicative or additive RR models (data not shown). However, the effect of a 5 Gy or higher breast radiation dose on risk differed between women with a first- or second-degree family history of breast or ovarian cancer (OR = 0.8, 95% CI 0.2-3.4), and those without a family history (corresponding OR 5.8, 95% CI 2.1-16.3; Interaction OR = 0.1; 95% CI 0.03-0.8; interaction P = .03) (Table 3). To assess whether the cutpoint at 5 Gy influenced this finding, risk modification by radiation dose also was analyzed on a continuous scale (Gy). Women who reported a first- or second-degree relative with breast or ovarian cancer also had a lower odds ratio estimate per increasing unit of breast dose (Gy) than women without such history (Interaction OR = 0.93 per Gy; 95% CI 0.89-0.99; interaction P = .005). The lower odds ratio persisted when breast or ovarian cancer family history was confined to first-degree relatives (interaction P = .005), and when examined only among study sites that used cancer registry data or interviews to ascertain family history (interaction P = .04). The interaction did not change substantially when data from any of the 6 participating cancer registries were omitted.

Since the nature of the relationship between radiation, other risk factors, and breast cancer risk has not been characterized in patients with HL, additive relative risk models incorporating family history were also evaluated. The joint effects of family history (first or second degree) and radiation dose were somewhat less than would be expected from adding the 2 risk factors, for either a dose cutpoint of 5 Gy or higher or radiation dose on a continuous scale (Gy), although neither interaction was significant. However, when family history was defined to include only first-degree relatives, the ERR added with each Gy (above 5 Gy) was –.16 (95% CI, –.40-+.23) among women with a family history, and +.17 (95% CI, .05-.51) among women without such history (ICR –.33; 95% CI, < –.33-+.03) (additive interaction P = .07), suggesting a joint effect that might be less than adding the 2 risk factors, and a possible difference in dose-response between the 2 groups. The ratio of these ERRs (γ*) was –0.9 (95% CI, < –0.9-+1.2), that is, the absolute excess risk in women with a family history was estimated to be smaller and unlikely to exceed by more than a factor of 1.2 the absolute excess risk in women without such history. Thus, the additive and multiplicative RR models were generally consistent in suggesting that the relative risk associated with radiation exposure was not higher among women with a family history than among other women, and might be lower.

In addition to family history, breast cancer relative risk following radiation exposure also appeared to differ somewhat by parity status. Among women who had experienced a live birth, those who received a breast radiation dose of 5 Gy or higher had a 3.5-fold elevated breast cancer risk (95% CI, 1.4-8.9) compared with women who received lower doses, while among women who had not had a live birth, the corresponding OR was 1.1 (95% CI, 0.3-4.7) (Interaction OR 3.1; 95% CI, 0.6-17.2; interaction P = .20) (Table 4). In an analysis examining breast radiation dose on a continuous scale, women who had experienced a live birth had a greater increase in breast cancer relative risk with increasing radiation dose than women who had not (Interaction OR 1.06; 95% CI, 1.01-1.12; interaction P = .04). This finding is unlikely to be explained by altered ovarian function due to treatment because women who received AA or ovarian doses of 5 Gy were excluded from these analyses. We examined whether ovarian doses lower than 5 Gy could have altered ovarian function among included women, possibly accounting for the findings. Breast cancer risk was not reduced among women who received ovarian doses of 0.2 to 0.92 Gy or 0.93 to 4.99 Gy (tertiles) compared with those who received ovarian doses lower than 0.2 Gy. In addition, the interaction between a live birth and breast radiation dose (Gy) persisted even among women who received ovarian radiation doses lower than 1 Gy (63% of patients, 65% of control subjects; data not shown). The interaction was not altered when data were excluded from any contributing cancer registry except the Netherlands (largest contributor); among remaining registries the statistical power was lower, and the relationship was similar in direction and magnitude, but no longer significant. In additive RR models, the effect on risk of a breast radiation dose of 5 Gy or higher or per Gy (continuous dose variable) did not differ among women who had a live birth versus those who had not (additive interaction P = .31 and .20, respectively).

Among women who had a live birth before HL diagnosis, breast cancer risk was not influenced by the relative timing of most recent birth in relation to HL diagnosis (Table 5). Women whose most recent live birth after HL diagnosis was within 60 months (control median) had a 2.6-fold elevated breast cancer risk, relative to those whose most recent birth occurred at a longer interval (who thus were also fertile after HL treatment). The timing of live births in relation to breast cancer diagnosis/study cutoff date had no appreciable influence on breast cancer risk.

No previous study to date has examined the influence of breast cancer risk factors and radiation dose in relation to breast cancer risk among young women treated for HL. Extensive efforts were made to estimate radiation dose to the area where the breast tumor occurred, and that received by the ovaries, and to collect details regarding cytotoxic drug treatment. Information regarding factors that might modify breast cancer risk in young women is important, as the cumulative incidence may approach 20% by 45 years of age among women treated before 17 years of age.² 

Our data suggest that among women with a first- or second-degree family history of breast or ovarian cancer, the additional increase in relative risk of breast cancer with increasing radiation dose probably does not exceed that of women without such history, and it may be lower. The combined effect of family history and breast radiation dose (on a categoric [≥ 5 Gy] or continuous [Gy] scale) was less than that expected from multiplying the 2 risk factors, thus we were able to reject a multiplicative model in favor of a submultiplicative model. In additive RR models, the combined effects were in the direction of less than additive. While information regarding breast or ovarian cancer family history may have been more completely ascertained among breast cancer cases than controls, misclassification that differs by case-control status, when assessing an interaction, usually biases the interaction risk estimate toward the null (OR = 1.0),²⁷  and is unlikely to account for the observations. Our findings also are supported by the persistence of altered breast cancer risk among women from sites that collected information from cancer registries or questionnaires, and among women with a first-degree family history only.

Some evidence supports the possibility that women with a family history of breast or ovarian cancer may have an altered response to radiation. Family history frequently reflects the effect of rare, highly penetrant alterations in genes such as BRCA1, BRCA2, TP53, and PTEN.²⁸  After 1-10 Gy radiation, cell lines deficient in BRCA1 or BRCA2 have demonstrated widespread cell death and a reduced capacity to repair DNA damage,^29,30  suggesting that in mutation carriers, unrepaired damaged cells might undergo cell death rather than serve as cancer-initiating cells. In our study, the area of the breast where the tumor occurred received a median dose of 24.8 Gy (median, 20.2 Gy to comparable area in control subjects), a dose likely to induce cell death in a substantial proportion of radiosensitive cells. The prevalence of BRCA1 or BRCA2 mutation carriers,^31,32  however, is too low to account entirely for our findings. BRCA1, BRCA2, and the putative breast cancer susceptibility gene mutated in ataxia telangiectasia (ATM) interact in the cellular response to radiation-induced DNA damage.^33,34  Individuals diagnosed with AT³⁵  and heterozygous carriers of ATM missense variants^36,37  may have an increased HL risk, and may be overrepresented in our population. In one study, HL survivors who carried ATM missense alterations were less likely to develop breast cancer following RT than noncarriers,³⁶  suggesting a possible differential effect of treatment, while ATM-truncating mutations have not been identified among HL survivors who developed breast cancer.^38,39 

Our data also suggest that the dose-response relationship between breast radiation exposure and subsequent cancer risk may be stronger, on a multiplicative but not an additive RR scale, among young women who have ever had a live birth. Women who received therapies associated with ovarian damage were excluded from this analysis, and we did not find indirect evidence of diminished fertility among included women. In animal experiments, mammary tumor incidence has been considerably higher among rats irradiated with 2.6 Gy while pregnant (92%) or lactating (90%-96%) than among virgin animals (26%-33%).^40-42  Increased production of prolactin during pregnancy and lactation has been implicated as 1 factor promoting mammary carcinogenesis: in another investigation, the incidence of mammary tumors in rats was 2% after low-dose irradiation alone, 41.6% if a prolactin-secreting pituitary transplant was given shortly afterward, and 24% if the prolactin-secreting transplantation was given 12 months after irradiation.⁴³  In some^13-16  but not all^44,45  epidemiologic studies, premenopausal breast cancer risk has been increased among parous women or specifically among women who gave birth within the previous 3 to 10 years, consistent with the possible growth-promoting effects of elevated gestational hormones. In the Nurses' Health Study, the increased risk among parous women seems to persist for at least 20 to 30 years following a first pregnancy.¹³  In our study, parous women who delivered infants within 60 months after HL had a further increased risk relative to other parous women. Although a critical postexposure period has not been identified, the hormonal milieu in the years following radiation exposure appears to act as a primary breast cancer determinant: women exposed after 40 to 49 years of age do not have an increased risk,^5,7,8  and younger women who also receive therapies that decrease ovarian function or induce menopause have a reduced risk.^25,26 

Our results should be considered in light of the study strengths and limitations. Breast cancer risk factor information was collected primarily from medical records; thus, patient data were missing less often than control subject data, although any misclassification should bias the interaction OR estimate toward the null (1.0).²⁷  Even though these analyses were conducted within the largest study to date of breast cancer following HL, sample sizes in various subgroups were small, limiting statistical power to detect differences in stratum-specific odds ratios. Given the unknown nature of the interaction between therapeutic doses of radiation and breast cancer risk factors, we tested for joint effects on more than one statistical scale, which may increase the probability of false-positive findings.⁴⁶  In addition, doses much lower than 5 Gy have been associated with increased breast cancer risk,^6,47,48  and interaction estimates should be attenuated with inclusion of such exposure in the reference group, thus interaction also was examined with RT dose on a continuous scale.

Women included in our study received very high radiation doses to the breast; thus, our results, if verified in similar populations, may not be generalizable to those who receive lower doses. Smaller radiotherapy fields and doses are used in current HL treatment protocols, and some AAs (ie, mechlorethamine) that affect fertility are now administered infrequently, and the effects of these newer treatment strategies have not been evaluated. Our results suggest that the additional increase in breast cancer relative risk after radiotherapy for HL is unlikely to be larger among women with a family history of breast or ovarian cancer than among other women. Consideration of breast cancer risk factors may offer insights regarding the increased breast cancer risk following radiotherapy for HL, and perhaps holds promise for identification of subgroups with altered susceptibility, and subsequent application of risk-adapted therapy.