Risk of Leukemia after Chemotherapy and Radiation Treatment for Breast Cancer

Study Subjects

A case–control study was conducted in a cohort of 82,700 women with invasive breast cancer diagnosed from 1973 to 1985 who were reported to one of five population-based cancer registries (Connecticut, Iowa, and the metropolitan areas of Detroit, Atlanta [1975 to 1985], and Los Angeles). Because secondary leukemia does not develop until at least 1 to 2 years after therapy, only women who survived at least 18 months after the diagnosis of their breast cancer were eligible for study. Newly diagnosed cases of leukemia were identified from registry files. Cases of chronic lymphocytic leukemia, which has not been related to radiation exposure or the use of cytotoxic drugs, were excluded. Death records were searched for patients with breast cancer whose underlying cause of death was severe anemia or another blood disorder. All suspected leukemic conditions were evaluated and reclassified with the use of reports and slides of peripheral-blood samples, bone marrow aspirates, and biopsy specimens. The 90 eligible case patients included 84 with leukemias and 6 with myelodysplastic syndrome. The leukemic conditions were classified as follows: 74 were acute nonlymphocytic leukemia, 3 were acute lymphocytic leukemia, 7 were chronic myelogenous leukemia, 5 were refractory anemia with excess of blasts, and 1 was acute myelofibrosis. The 20 women from Connecticut with leukemic conditions have been described previously.³ For each case patient, three control patients were selected by means of random sampling from the cohort of women with breast cancer, and they were matched to the case patient on the basis of age and year of diagnosis of the initial tumor (exact year, when possible), race or ethnic group, and length of follow-up, which had to be at least as long as the interval between the diagnosis of breast cancer and the onset of the leukemic condition. Three controls each were found for 84 case patients, and 2 controls each were found for the other 6 case patients.

Abstraction of Medical Records

Information about surgery, radiotherapy, chemotherapy, hormonal therapy, and demographic characteristics was abstracted from registry records, hospital charts, and oncology-clinic records. Because chemotherapy for breast cancer is frequently administered directly from physicians' offices, an attempt was made to contact all private physicians associated with the patients' treatment. Information regarding primary treatment was abstracted from the records kept by at least one of the patients' physicians (or clinics) for 90 percent of the case patients and 92 percent of the controls, and for subsequent courses of therapy for 85 percent of the case patients and 89 percent of the controls. For 3 percent each of the case patients and controls, yearly follow-up reports from the individual hospital tumor registries describing details of radiotherapy and chemotherapy were used to confirm the treatment information obtained from the medical records. Treatments given within one year before the diagnosis of leukemia (or an equivalent date for the controls) were considered unlikely to contribute to the development of a second cancer and were excluded from analyses.

Information on the cumulative dose of alkylating agents was available from the patients' medical charts and physicians' records for 85 percent of the women (81 percent of the case patients and 88 percent of the controls) included in the dose–response analyses. For the remaining patients, the doses were estimated on the basis of the duration of therapy (10 percent) or the best estimate of its duration (4 percent) or were imputed from the median dose (1 percent). The effect on the results of the dose-estimation procedure was evaluated by recomputing the risk of leukemia after patients with incomplete dose information were excluded and after lower-quality dose estimates were increased and then decreased by 30 percent. The results were not materially changed with the use of either method.

Radiation-treatment records were photocopied and used to calculate the individual doses of radiation to 16 sections of active bone marrow as previously described.⁶ Radiotherapy was primarily administered by megavoltage linear accelerators (48 percent) or cobalt-60 units (48 percent). Typical treatments delivered an average dose to the tumor of 38 Gy from the supraclavicular lymph-node field, as well as 45 Gy from the tangential fields and 40 Gy from the mediastinal field (or chest field) alone or in combination, with an optional boost of 13 Gy to the axilla.

Statistical Analysis

Estimates of the relative risk of leukemia associated with specific treatments were calculated by comparing the case patients' history of exposure with that of their individually matched controls with conditional logistic-regression methods.^{7 , 8} Comparisons between treatment groups were based on likelihood-ratio tests. Two-sided P values and 95 percent confidence intervals were computed. Estimates of the risk associated with specific alkylating agents were restricted to patients who received the therapy for more than one month and were adjusted for radiotherapy with the use of a categorical variable (yes or no). The relation between the risk of leukemia and the dose of an individual alkylating agent was evaluated by grouping the cumulative dose into evenly spaced categories and computing relative risks between each category and the reference group of patients who had not been exposed to the treatment. These analyses were adjusted for the dose of radiation with a continuous variable. Continuous variables were used to adjust the radiation dose–response analyses for the cumulative dose of melphalan and cyclophosphamide and for the months of treatment with other alkylating agents. Tests of trend were made by assigning the midpoint of the dose group as the representative score. When the number of subjects was so small that the conditions required for convergence of the conditional logistic model were not met, the relative risks were computed with an unmatched logistic model.

The excess risk (the excess number of cases of leukemia per 10,000 patients) within the first 10 years after the diagnosis of breast cancer was estimated by multiplying the relative risk minus 1 by the expected number of leukemias per year, as calculated in our previous study³ of patients with breast cancer treated with chemotherapy (0.56 case of acute nonlymphocytic leukemia expected per 10,000 women-years at risk), and then by 8.5, which is the number of years at risk (assuming a latent period of 1.5 years before the onset of leukemia). For example, a twofold relative risk of leukemia associated with cyclophosphamide therapy would correspond to an excess of 4.76 leukemias per 10,000 patients over a 10-year period: (2 - 1) × 0.56 × 8.5.

Table 1. Table 1. Characteristics of Case Patients with Breast Cancer Who Had Leukemia or Myelodysplastic Syndrome and Their Matched Controls.*

Most of the women in our study were over 50 years of age when breast cancer was diagnosed and were treated during the mid-to-late 1970s (Table 1). The mean length of time between the initial diagnosis of breast cancer and the development of leukemia was 5 years (range, 1.7 to 12.5). Case patients were more likely than controls to have regional node involvement (67 percent vs. 35 percent). More than twice as many case patients as controls were treated with chemotherapeutic regimens that included an alkylating agent (53 percent vs. 19 percent). High-dose regional radiotherapy was given to 42 percent of the case patients and 27 percent of the controls. All but four patients had surgery after the initial diagnosis.

Table 2. Table 2. Risk of Leukemia and Myelodysplastic Syndrome Associated with Alkylating-Agent Therapy and Radiotherapy for Breast Cancer.*

There were significant relative risks of leukemia and myelodysplastic syndrome associated with the use of alkylating agents (relative risk, 6.5) and radiotherapy (relative risk, 1.8) (Table 2). The numbers of excess leukemias after drug therapy were concentrated in the subgroups of patients with acute nonlymphocytic leukemia and myelodysplastic syndrome (relative risk, 8.7), and the risk of erythroleukemia, a subtype of acute nonlymphocytic leukemia, was substantially increased (14 cases; relative risk, 21.4). The use of alkylating agents was not linked to increases in chronic myelogenous leukemia or acute lymphocytic leukemia; in fact none of the 10 case patients with these types of leukemia had been treated with chemotherapy. All subsequent analyses included only cases of acute nonlymphocytic leukemia and myelodysplastic syndrome, referred to as "leukemia" for simplicity of presentation.

Table 3. Table 3. Relation of Radiation and Alkylating Agents to the Risk of Acute Nonlymphocytic Leukemia and Myelodysplastic Syndrome.*

Treatment with radiotherapy alone, alkylating agents alone, or alkylating agents in combination with radiotherapy was linked to a significantly increased risk of leukemia (Table 3). Patients receiving both systemic drug therapy and radiation were at greatest risk (relative risk, 17.4). Treatment with alkylating agents alone or in combination with radiotherapy resulted in significantly higher risks of leukemia than radiotherapy alone (P = 0.001 and P<0.001, respectively). Patients treated with chemotherapy and radiation did not have a significantly different risk from those given alkylating agents alone in an analysis classifying patients as exposed or unexposed to these agents (P = 0.24). However, in a multivariate model that took into account the type and amount of alkylating agents received and the dose of radiation, the risk associated with combined alkylating-agent therapy and radiotherapy was significantly higher than that for alkylating agents alone (P = 0.02).

Table 4. Table 4. Risk of Acute Nonlymphocytic Leukemia and Myelodysplastic Syndrome, According to the Type of Alkylating Agent Administered.*

Patients treated with chemotherapy most often received regimens that included cyclophosphamide alone (30 percent of case patients and 74 percent of controls) or melphalan alone (50 percent of case patients and 19 percent of controls) (Table 4). Cyclophosphamide was usually given in combination with methotrexate and fluorouracil. Melphalan was most often administered as a single agent or less commonly in combination with fluorouracil. The risk of leukemia after treatment with cyclophosphamide alone (no other alkylating agent) was significantly increased (relative risk, 3.1), but was substantially lower than that with melphalan alone (relative risk, 31.4). The difference in the leukemogenic effect between the two drugs was significant (P<0.001). Too few patients received other alkylating agents alone (thiotepa or chlorambucil) to estimate their risks separately. Patients treated with more than one alkylating agent generally received long-term therapy and had a risk of leukemia comparable to that of women given melphalan alone (relative risk, 30.5); six of the seven case patients in this group received melphalan during at least one course of treatment.

Table 5. Table 5. Risk of Acute Nonlymphocytic Leukemia and Myelodysplastic Syndrome, According to the Estimated Dose of Alkylating Agents and the Duration of Treatment.*

Table 5 presents the risk of leukemia according to the estimated cumulative dose and the duration of cyclophosphamide and melphalan therapy when each was used as the sole alkylating agent. The doses of cyclophosphamide were generally higher for case patients (median dose, 22,300 mg; range, 5700 to 54,800) than for controls (median dose, 16,700 mg; range, 1800 to 55,100), and the risk was found to increase significantly over categories of cumulative dose. The risk among patients treated with doses of more than 20,000 mg was 5.7 times that of women not treated with alkylating agents (95 percent confidence interval, 1.6 to 20.6). The duration of therapy was also strongly associated with the risk of leukemia. Patients who received cyclophosphamide for less than 12 months had little detectable increase in the incidence of leukemia. The risk rose substantially with treatments that lasted longer than 12 months. Women treated for 18 months or more had a sevenfold increased risk of leukemia.

Case patients also received higher cumulative doses of melphalan (median dose, 530 mg; range, 230 to 1040) than controls (median dose, 300 mg; range, 80 to 420). There was evidence of a steeply graded dose–response curve, with risks climbing to 133-fold for doses of more than 350 mg (data not shown). The duration of treatment with melphalan was directly related to the risk of leukemia, especially in patients treated for 15 months or more. Three controls and no case patients received melphalan for less than 12 months. Analyses of cyclophosphamide and melphalan treatment with other dose groupings and after adjustment for the stage of breast cancer yielded similar results.

The cumulative dose, duration of treatment, and intensity (average dose per month) of alkylating-agent therapy were evaluated separately and jointly. The cumulative dose appeared to be the strongest determinant of risk; however, the dose and the duration of chemotherapy were strongly correlated in these data. The intensity of the dose did not appear to influence the risk of leukemia in a model accounting for cumulative dose.

Table 6. Table 6. Risk of Acute Nonlymphocytic Leukemia and Myelodysplastic Syndrome, According to the Dose of Radiation to the Total Active Bone Marrow.*

Women treated with regional radiotherapy received a mean dose of 7.5 Gy averaged over the total active bone marrow (median dose, 7.2 Gy). There was a significant trend in the risk of leukemia with increasing doses of radiation, after adjustment for the effects of alkylating agents (Table 6). Patients receiving doses of 9.0 Gy or more were subject to a sevenfold elevated risk. A similar dose–response pattern was found when analyses were restricted to radiotherapy given within the first 12 months after the initial diagnosis. The risk of leukemia was especially increased among the six case patients and four controls given radiation as both initial and subsequent therapy (relative risk, 15.5; 95 percent confidence interval, 2.6 to 92). Excluding patients treated with alkylating agents from the analysis lowered the overall risk due to radiation somewhat (relative risk, 1.9; 95 percent confidence interval, 0.7 to 5.0); however, a significantly increased risk persisted among women who were treated with radiotherapy alone and exposed to more than 9 Gy (relative risk, 10.4).

The risk of leukemia that was associated with alkylating-agent therapy was lower among women in whom breast cancer was diagnosed before the age of 50 years (relative risk, 2.4) than among women who were at least 50 at the time of diagnosis (relative risks, 13.0, 15.0, and 10.5, for age groups 50 to 59, 60 to 69, and ≥70 years, respectively), but this difference was not significant (P = 0.49 by the test of homogeneity). The patterns of risk according to age were similar after melphalan and cyclophosphamide therapy. Because of the concentration of patients treated with melphalan during the mid-1970s, women given a diagnosis of breast cancer from 1975 to 1976 had an exceptionally high risk of leukemia. A threefold increase in the risk was evident during the later calendar years (1979 and after) when practically all patients treated with alkylating agents received cyclophosphamide-based regimens. An excess number of leukemias linked to treatment with alkylating agents was observed for patients from each cancer registry.

Table 7. Table 7. Risk of Leukemia and Myelodysplastic Syndrome, According to the Time since the First Treatment with Alkylating Agents.

The pattern of the risk of leukemia according to the length of time since the first treatment with alkylating agents is shown in Table 7. The influence of continued exposure with subsequent courses of chemotherapy was minimized by the exclusion from these analyses of women who received multiple courses of chemotherapy or alkylating agents for more than 30 months. The risk of leukemia was elevated throughout most intervals, but was most prominent within two to seven years after the initiation of therapy. The risk appeared to approach normal levels after seven years (relative risk, 1.8). Women who had received the last drug treatment within two years of the diagnosis of leukemia (or an equivalent date for controls) had an immediate high risk (relative risk, 18.1; 95 percent confidence interval, 6.4 to 51) (data not shown). Patients in this group tended to receive chemotherapy for longer periods and were primarily treated with melphalan. The risk remained elevated for at least five years after the last drug treatment, and declined thereafter.

Alkylating agents were administered as a subsequent course of treatment for 14 case patients and 5 controls. There was little difference in the risk between patients receiving subsequent therapy but no initial therapy (relative risk, 9.6) and those given initial therapy only (relative risk, 7.4). The risk was highest among women who had received alkylating agents for both primary and subsequent courses of therapy (relative risk, 25.7).

No increase in risk was associated with treatment with fluorouracil (67 patients), methotrexate (58 patients), prednisone (20 patients), doxorubicin (13 patients), or vincristine (13 patients) after an adjustment for the effects of alkylating agents and radiation; however, few patients received these drugs in the absence of alkylating agents.

We studied patients with breast cancer treated in the general community during the 1970s and early 1980s, a period when adjuvant chemotherapy was introduced and later came into wide use. Women treated with alkylating agents had an eightfold increase in the risk of acute nonlymphocytic leukemia and myelodysplastic syndrome. The risk was significantly increased within three years after the initial treatment with chemotherapy. Melphalan therapy accounted for most of this excess risk, with the relative risk increasing 30-fold. Cyclophosphamide was less leukemogenic, with a more moderate threefold increase in the risk. The risks were especially heightened among women treated with multiple alkylating agents during several courses of treatment and with alkylating agents administered in combination with high-dose regional radiotherapy.

In interpreting these results, consideration should be given to changes in breast-cancer therapy over the past decade. Melphalan is rarely used today for adjuvant therapy, cyclophosphamide is given at lower doses and for shorter periods than in the past, and the use of adjuvant radiotherapy after mastectomy has largely been replaced by conservative surgery with localized radiotherapy. However, because women in the study received therapy over a wide range of doses, we were able to provide estimates of risk at the lower dose levels currently used to treat patients with breast cancer. The relation between dose and effect found in this study may also be applicable to other therapeutic settings.

There are few quantitative data on the ability of cyclophosphamide to induce leukemia at the lower doses currently used for adjuvant therapy. In our study we observed a significant trend of increasing risk with increasing doses of cyclophosphamide. The risk of leukemia was especially high among women who received cumulative doses exceeding 20,000 mg. These results are consistent with those of a previous study of breast cancer⁵ and studies of ovarian cancer^{9 , 10} that found elevated risks at higher doses. Data from clinical trials of breast cancer have not revealed an excess incidence of leukemia after adjuvant chemotherapy including cyclophosphamide, probably because the small numbers of patients involved were not sufficient to detect small risks of leukemia at the low doses administered.^{11 12 13} In our study, with a source, population of more than 82,000 women with breast cancer, we were able to confirm that low risks of leukemia (e.g., approximately twofold) are associated with treatment with cyclophosphamide at levels currently used in adjuvant therapy. Iatrogenic leukemia, however, is a rare occurrence after breast cancer, and only about 5 of 10,000 patients treated for 6 months with a cyclophosphamide-based adjuvant regimen might be expected to have leukemia within 10 years of the diagnosis of breast cancer.

Melphalan given as primary therapy for ovarian cancer and multiple myeloma has been reported to have a much higher leukemogenic potential than cyclophosphamide.^{9 , 10 , 14} The present study confirmed these findings among patients with breast cancer who were receiving adjuvant therapy and estimated that there was a 10-fold difference in risk between melphalan and cyclophosphamide. When melphalan was used as the sole alkylating agent, there was a steep gradient in risk with increasing cumulative dose. The risk for doses exceeding 350 mg was more than 100-fold. The magnitude of the overall risk associated with melphalan was similar to that found among patients with breast cancer who were enrolled in clinical trials.⁴ Fisher et al. found no relation to the dose,⁴ however, probably because there was strict uniformity of treatment in the randomized trials.

An important finding of the current study is that radiotherapy that delivers high doses to bone marrow in the chest wall and other areas may add appreciably to the risk of leukemia after chemotherapy in a manner that is consistent with a multiplicative relation. Patients with breast cancer who were treated with radiotherapy and alkylating agents were found to have a significantly higher risk than women receiving alkylating agents alone. The literature is conflicting regarding the risk after combined therapy. Two recent investigations of Hodgkin's disease^{15 , 16} concluded that the increased risk after chemotherapy is not affected by concomitant radiotherapy; however, other studies have reported the opposite result,¹⁷ and considerable uncertainty surrounds this issue.¹⁸ Previous evaluations of patients with ovarian cancer have consistently reported that combined treatment with chemotherapy and radiation is not associated with a greater risk of leukemia than treatment with chemotherapy alone.^{9 , 10} Our study is unique in having substantial numbers of patients with leukemia in each treatment category, coupled with a large comparison group of women who received neither radiation nor chemotherapy; thus, we were able to estimate the separate and combined effects of regional radiotherapy and chemotherapy.

After adjusting for the effects of alkylating agents, we found a twofold increase in the risk of acute leukemia after regional radiotherapy. We found strong evidence of increasing risk with an increasing dose of radiation. A sevenfold increase in risk was associated with average bone marrow doses of more than 9 Gy. The small effect at low doses of radiation was of interest, since current treatment practices involving localized radiation to the breast after conservative surgery deliver substantially lower doses to the bone marrow. Our earlier, but smaller, study of patients with breast cancer in Connecticut who were treated between 1935 and 1972 found no evidence of an excess number of leukemias occurring after radiotherapy¹⁹; however, increased risks have been reported among patients with breast cancer treated with radiation who were enrolled in clinical trials.^{4 , 20} Aside from the effects of chance, changes in the type and techniques of radiotherapy over time may account for some of these differences, since earlier treatments used lower-voltage machines and delivered doses to the active marrow that were about 25 percent lower. Radiotherapy has also been found to cause a twofold increase in the relative risk of leukemia in a large study of patients with cervical cancer,²¹ although there was no evidence of a marked increase in risk at the highest doses of radiation.

By matching controls with case patients, the study design incorporated an adjustment for age and calendar year at the initial diagnosis and for the time to the onset of leukemia. Although the risk associated with alkylating agents was lower among women who were younger than 50 years of age than among those who were at least 50, this difference was not significant. Previous studies of patients with breast cancer found that the risk of leukemic conditions after melphalan therapy did not vary with age,⁴ but higher risks have been reported among younger women treated with cyclophosphamide.⁵

The temporal pattern of drug-induced leukemia is of considerable interest, in particular the level of risk remaining several years after the initial treatment. Among patients treated with a single course of chemotherapy, the risk of leukemia rose rapidly within two years after the initiation of therapy and then appeared to decrease after seven years. Few patients were followed for more than 10 years. A high risk was seen among women who were current or recent users of alkylating agents (<2 years), with a decline seen thereafter. These results indicate that drug-induced leukemia is an early effect, with an elevated risk appearing soon after exposure, and that the excess risk decreases to low levels after 7 to 10 years.

No cases of chronic myelogenous leukemia or acute lymphocytic leukemia in this study had been treated with alkylating agents, and the risk for these subtypes appeared to be substantially lower than the risk for acute nonlymphocytic leukemia and myelodysplastic syndrome. These findings suggest that there may be differences in the mechanisms of carcinogenesis between exposure to cytotoxic therapy and radiation, since studies of atomic-bomb survivors and other groups exposed to radiation indicate that there is a substantial excess of all subtypes of leukemia except chronic lymphocytic leukemia.²² The high risk of erythroleukemia after therapy with alkylating agents appears to be a distinguishing feature of drug-induced leukemia and contrasts with results from practically all radiation studies, except those involving thorium dioxide (Thorotrast). This agent is an alpha-particle emitter that irradiates the marrow throughout life and causes a high number of erythroleukemias.²³

Adjuvant chemotherapy for breast cancer is now a well-established treatment of proved effectiveness. The risk of leukemia after cyclophosphamide-based therapy administered according to regimens in common use today is small and unlikely to be a factor in most treatment decisions. On the other hand, the possibility that radiation in combination with alkylating agents greatly enhances the leukemic potential of the treatment is an important source of concern that deserves further study.