Leukemia Following Chemotherapy for Ovarian Cancer

Table 1. Table 1. Distribution of Cases of Leukemia Following Ovarian Cancer, According to Study Center.

The study was carried out by a collaborative group comprising 11 population-based cancer registries in Europe and Canada and two large oncology hospitals in Europe, both of which have maintained cancer registries for decades (Table 1).

Study Subjects

The collaborating registries and hospitals all routinely carry out long-term follow-up of patients with ovarian cancer. The occurrence of second cancers was identified through a comparison of records (in the registries) or through active follow-up (in the hospitals) and was confirmed histologically whenever possible. The case patients, defined as patients who were given a diagnosis of leukemia at least one year after receiving a diagnosis of ovarian cancer, were identified from the records maintained by the participating institutions of patients given diagnoses of ovarian cancer on or after January 1, 1960. Women who received diagnoses of myelodysplastic syndrome⁸ at least one year after their diagnoses of ovarian cancer were also included as case patients.

For each case patient, three matched control subjects were sought (four in the case of the German Democratic Republic Cancer Registry, which began the study somewhat earlier with a slightly different protocol). First, all patients were identified who had received diagnoses of ovarian cancer in the same registry or hospital as the case patient and had survived free of a second cancer for at least as long as the interval between the diagnoses of ovarian cancer and of leukemia in the case patient. Among the patients in this group, those who were most similar to the case patient with respect to year of birth and then to year of diagnosis of ovarian cancer were chosen as controls. If the controls were equally well matched, a random choice was made. The maximal acceptable differences between case patients and their matched controls were three years for year of birth and five years for year of diagnosis of ovarian cancer, unless no patients satisfied the requirements, in which event the most similar available patient was chosen as a control.

The purpose of matching was to eliminate the potentially confounding effect of factors that are of little interest in themselves but that may be related to both the risk of leukemia and the type of treatment for ovarian cancer. The stage of ovarian cancer was not a matching factor, because its relation to the risk of leukemia has not been established.

Data Abstraction

For each case patient and the corresponding matched controls, the full medical record of the diagnosis and treatment of ovarian cancer was sought. For the case patients, the record of the leukemia diagnosis was also obtained. The records were abstracted on specially prepared forms that requested information on the patient's date of birth; the morphology, stage, and date of diagnosis of the ovarian cancer; and all cytotoxic chemotherapy and radiotherapy given for the ovarian cancer. For the case patients, the record of the diagnosis of leukemia was also obtained, including relevant pathology reports, and the morphologic features and date of diagnosis were recorded. For each course or cycle of chemotherapy, the details abstracted included the name and total dose of each drug used and the dates of the period during which it was administered. For each patient treated by radiotherapy, the most detailed record of radiotherapy available was obtained.

For each course of external-beam radiotherapy, the details abstracted included the size and location of the field, the energy level of the radiation, and the dose to the tumor. For brachytherapy, data on the anatomical site of the implant, the activity and type of the isotope, and the duration of implantation were abstracted.

Radiation Dose

To provide an average dose of radiation to the total active bone marrow, the bone marrow was partitioned and weighted according to the method of Cristy.⁹ In a region treated by external beams, the dose to the bone marrow was assumed to be the same as the tumor dose. Outside the treated region, the dose was estimated on the basis of measurements in a water phantom.¹⁰ In the case of brachytherapy, the dose to the bone marrow was calculated with a computer program in routine clinical use.¹¹

Statistical Analysis

Standard conditional-likelihood methods for the analysis of matched case–control studies were used.¹² These methods can provide estimates only of relative risk, not of absolute risk, and thus they require the choice of a reference category of treatment against which the risk of other types of therapy is to be compared. In this study, the comparisons were made either in relation to patients treated with surgery only or to those treated without chemotherapy — that is, with surgery or radiotherapy only.

The treatment of the case patients was considered only in the period between the diagnosis of ovarian cancer and the diagnosis of leukemia. For each control subject, therapy was considered only in an equivalent interval, starting with the date on which ovarian cancer was diagnosed.

Although chemotherapy with combinations of drugs has been widely used in the treatment of ovarian cancer over the past decade, the case patients and controls in this study were generally treated with single agents. Furthermore, the combinations of drugs used rarely involved more than one alkylating agent. Nevertheless, some patients were treated with multiple potentially leukemogenic agents in the same course of therapy, and we therefore defined two categories of combination therapy. The first was the combination of doxorubicin hydrochloride and cisplatin, and the second included all other combinations of two or more alkylating agents or drugs identified as carcinogenic to humans or laboratory animals.⁷ Combinations that included only one such drug were treated as single agents.

Most statistical analyses were restricted to leukemia that was acute, nonlymphocytic, or both and to the myelodysplastic syndrome — conditions that have all been associated previously with either chemotherapy involving alkylating agents or exposure to radiation. For simplicity, these conditions are referred to subsequently as acute or nonlymphocytic leukemia. There were not enough cases of chronic lymphocytic leukemia for analysis.

In the statistical analyses, all relative risks were estimated with the computer program PECAN,¹³ and P values and 95 percent confidence intervals were obtained with the likelihood-ratio test,¹⁴ unless otherwise specified.

Table 1 shows the number of patients with ovarian cancer studied, the number of cases of leukemia identified in these patients, and the number of matched controls sampled, according to study center. Among the women in whom leukemia developed subsequently, the median age at diagnosis of ovarian cancer was 58 years. Most of the cases of leukemia (69 percent) occurred in women given the diagnosis of ovarian cancer in 1970 or later, and the last one entered in the study occurred in a woman whose leukemia was diagnosed in 1985. Of the controls, 74 percent were matched within two years of the case patient according to age at diagnosis of ovarian cancer, and 90 percent were matched within two years according to date of diagnosis. For 9 case patients, it was impossible to find 3 controls who satisfied the matching criteria, and a total of 12 controls were therefore missing.

Table 2. Table 2. Distribution of Cases of Leukemia Following Ovarian Cancer, According to Type of Leukemia and Years since the Diagnosis of Ovarian Cancer.

The cases of leukemia following ovarian cancer were classified according to the subtype of leukemia and the length of the interval between the diagnoses of ovarian cancer and leukemia (Table 2). When a specific subtype was assigned, the designation was based on a histologically confirmed diagnosis. Most cases (74 percent) occurred between two and nine years after ovarian cancer, and acute or nonlymphocytic leukemias represented 89 percent of the total. Nine of the 11 cases of myelodysplastic syndrome were reported from the Karolinska Hospital, Stockholm. Subsequent analyses involved only the restricted group of 101 case patients referred to as having acute or nonlymphocytic leukemia (Table 2) and their matched controls. Because of the small number of cases, no separate statistical analyses were attempted for other subtypes of leukemia or for the myelodysplastic syndrome.

Table 3. Table 3. Distribution of Case Patients and Controls with Acute or Nonlymphocytic Leukemia, with Relative Risks According to Overall Treatment Category.

Table 3 shows the number of case patients and controls who were treated with surgery, chemotherapy, radiotherapy, or both chemotherapy and radiotherapy. The relative risks, P values, and 95 percent confidence intervals for each category (with the surgery-only category as the base line) are also shown. The relative risk for chemotherapy alone (12) was slightly higher than the risk for radiotherapy and chemotherapy combined (10), although the difference was not significant (P = 0.41). Almost identical estimates of relative risk (12 and 9.3, respectively) were obtained when the case patients and controls from the participating hospitals were excluded. Because of the small and nonsignificant increase in risk associated with radiotherapy alone and the limited number of case patients treated with surgery alone, the groups of patients treated with surgery or radiotherapy only were combined to provide a larger reference group for the estimation of relative risk. The resulting estimates may thus be slightly conservative (biased toward 1.0) but will have narrower confidence intervals than those using surgery alone as a reference category.

Table 4. Table 4. Distribution of Case Patients and Controls with Acute or Nonlymphocytic Leukemia, According to Type of Chemotherapy.*

For each drug recorded as having been used by itself to treat ovarian cancer and for the two categories of combination therapy, Table 4 gives the number of patients who were ever treated, according to their status as case patients or controls. The alkylating agents chlorambucil, cyclophosphamide, melphalan, thiotepa, and treosulfan were the most frequently recorded chemotherapeutic agents. Combinations of drugs were used far less often during the observation period of the study.

Table 5. Table 5. Distribution of Case Patients and Controls with Acute or Nonlymphocytic Leukemia, with Relative Risks According to Type and Dose of Chemotherapy among Patients Who Received Only One Type.*

In the case of several drugs and combinations, there were enough case patients and controls who had received no other chemotherapy to allow a separate estimation of the relative risks, without the need to adjust for other types of chemotherapy. The results of these analyses are shown in Table 5. The median dose in the controls was used as a cutoff point to create high-dose and low-dose groups for each drug. In addition to the relative risks (with treatment with surgery and radiotherapy only as the base line), the table shows the number of case patients and controls in each group and the median dose among the controls in the low-dose and high-dose groups. For five drugs used as single agents, there was a clear and dramatic increase in risk between the two dosages. The increase was comparable in magnitude for four of the drugs. Cyclophosphamide appeared substantially less leukemogenic, the difference being statistically significant when cyclophosphamide was compared with the other four agents (P = 0.02). There were also two case patients and two controls who were treated only with the combination of doxorubicin hydrochloride and cisplatin, resulting in a relative risk of 6.5 (P = 0.05).

Among the patients who had been treated with radiotherapy only, the relative risks were 1.9 (P = 0.34) and 1.2 (P = 0.80) for estimated doses to the active bone marrow of less than 10 Gy and 10 Gy or more, respectively. In general, the patients treated with both chemotherapy and radiotherapy received less chemotherapy overall than those whose only treatment was chemotherapy (results not shown).

Of the 101 case patients, 39 received diagnoses of Stage III or IV ovarian cancer, as compared with only 38 of 304 controls. However, after allowance by multiple logistic regression for the leukemogenic effect of the categories of chemotherapy listed in Table 5, the estimated relative risks were 0.80, 4.9, and 1.5 for Stages II, III, and IV, respectively, as compared with Stage I. The relative risk for Stage III differed significantly from 1.0 (P = 0.004), but the risks for the other two stages were not significant at the 0.05 level. No attempt was made to evaluate the risk according to subtype of ovarian cancer.

Table 6. Table 6. Relative Risk of Chemotherapy among Case Patients and Controls with Acute or Nonlymphocytic Leukemia, According to Time since First and Last Treatments.

Table 6 shows the relative risks of chemotherapy as broken down according to the time since the first and last episodes of treatment. After the initial chemotherapy, the relative risk as compared with that in patients who received no chemotherapy increased steadily for about six years, and although it decreased when chemotherapy ceased, it remained elevated for at least eight years.

Finally, the effect of age at the time of diagnosis of ovarian cancer on the risk associated with chemotherapy was considered by estimating the relative risk separately for patients above and below the median age at diagnosis (57 years). For patients taking cyclophosphamide, there was little difference in relative risk, but the patients in the older group taking melphalan or chlorambucil had a relative risk that was nearly 10 times higher than that in the younger group.

The results in Tables 3, 5, and 6 remained essentially unchanged when the case patients with myelodysplastic syndrome and their matched controls were excluded. For example, the relative risks in Table 3 became 1.0, 1.7, 11, and 8.8, respectively, for the categories of surgery, radiotherapy only, chemotherapy only, and radiotherapy and chemotherapy combined.

The case–control design used in the present study has only recently been applied to the investigation of second cancers.^{6 , 15 , 16} Although the method does not readily produce estimates of absolute risk, approximate estimates could be obtained by applying the estimated relative risks to leukemia rates in the general population, taking into account the survival curve for patients with ovarian cancer.

This investigation was carried out after a previous cohort study¹⁷ in which the risk of a second cancer following ovarian cancer was estimated from the records of 11 population-based cancer registries listing more than 87,000 survivors of ovarian cancer. In that study, there were about four times as many cases of acute or nonlymphocytic leukemia following ovarian cancer as would have been expected in a population of healthy women of the same age during the same time.

On the basis of the percentages shown for the controls in Table 3, about 26 percent of the survivors of ovarian cancer were treated with surgery alone; 36 percent received radiotherapy only, with a relative risk of leukemia of 1.6, as compared with those receiving surgery alone; 17 percent received chemotherapy only, with a relative risk of 12; and 21 percent received both chemotherapy and radiotherapy, with a relative risk of 9.8. Thus, if the patients with ovarian cancer who were treated with surgery alone had no increased risk of leukemia as compared with the general population and if the same relative risks can be applied throughout the follow-up period, the predicted relative risk of leukemia in the cohort study would have been 0.26 + (1.6)(0.36) + (12)(0.17) + (9.8)(0.21) = 4.9, which is rather close to the fourfold increase seen in that study.¹⁷ It thus seems reasonable to attribute the excess risk of leukemia among survivors of ovarian cancer overwhelmingly to their chemotherapy, either alone or in conjunction with radiotherapy, rather than to the disease itself or to some related factor.

By restricting our analysis to patients who received only one type of chemotherapy, we have been able to identify five single drugs as clearly leukemogenic —chlorambucil, cyclophosphamide, melphalan, thiotepa, and treosulfan. All these agents except thiotepa have been classified elsewhere as human leukemogens,⁷ but only in a few earlier studies has it been possible to demonstrate the leukemogenicity of more than one alkylating agent. In these studies, melphalan emerged as a far more potent leukemogen than cyclophosphamide.^{18 , 19} For all five agents, there was clear evidence of a dose–response relation with respect to leukemogenicity in our study when the patients were classified according to those who had received less or more than the median total dose of each drug or combination. In fact, the risks in the study may have been somewhat underestimated, if the matching according to registry and time of diagnosis resulted in a greater similarity between the case patients and the controls with regard to treatment than would otherwise have been observed. The possibility of overmatching cannot be evaluated in this study, but it was considered to be preferable to unadjusted confounding.

The leukemogenic effect of radiation in this study was close to that observed for radiotherapy in patients with cervical cancer,¹⁵ and the downturn in risk at doses to the bone marrow above 10 Gy supports the hypothesis that at high levels of exposure cell killing overwhelms the leukemogenic effect of radiation. The slightly lower risk in patients treated with both chemotherapy and radiotherapy than in those receiving chemotherapy alone can probably be attributed to differences in the amount of chemotherapy administered in the two groups. There was no suggestion of a multiplicative or synergistic joint effect of the two methods of treatment.

The relative leukemogenic potency of alkylating agents is difficult to characterize on the basis of data in humans.²⁰ Comparisons can be made in terms of either absolute drug dose (in grams, for example) or units of equal therapeutic or clinical effect. In the treatment of ovarian cancer, there is still no clear consensus on the relative effectiveness of different drugs, and it is therefore difficult to define standard doses. Risk per unit of weight of drug is thus a more generally useful measure of potency, since it can be used to calculate the predicted risk at any given dose level.

In crude comparisons of the risks associated with the administration of total doses of each agent below and above the median dose, only cyclophosphamide can be clearly distinguished as being far less potent than all the other drugs. However, when the dose is taken into account by dividing the excess relative risks (the relative risk minus 1.0) shown in Table 5 by the corresponding median doses (data not shown), melphalan was the most leukemogenic drug per gram at high doses, followed by thiotepa and chlorambucil. At low doses, thiotepa was the most potent. Treosulfan was the least leukemogenic drug at either dose level.

Four of the five drugs were substantially more potent at low doses than at high doses, implying that the increase in relative risk was less than linear with respect to dose. Several explanations could be put forward for this finding. One possibility is that for patients treated orally over long periods, the recorded doses were an overestimation of the true doses, simply because the patients stopped taking their tablets. Alternatively, the phenomenon may have been due to the effect of cell killing or saturation of metabolic activation at higher dosages, differences in dose rate, or lower sensitivity to leukemogenesis among patients tolerant of high doses.

An important new finding in the present study concerns the leukemogenicity of the combination of doxorubicin hydrochloride and cisplatin. Both drugs are used widely and effectively to treat a number of malignant diseases, and both have been recognized as carcinogens in animals.⁷ It thus appears that at least one of them is also a human leukemogen. Several of the drugs studied here are no longer in use. Although estimates of their leukemogenic effect have limited clinical importance, they are of interest to scientists studying aspects of the structure and activity of carcinogens.

Analyses of the temporal pattern of these leukemia cases show that the risk was at its highest four to six years after treatment for ovarian cancer. It dropped substantially after the cessation of therapy with alkylating agents, from a 13-fold increase immediately after cessation to about 6-fold five to eight years later. Increasing age at diagnosis of ovarian cancer seemed to increase substantially the relative risk due to certain types of chemotherapy in this study. This finding should be viewed with caution, in view of the small number of cases on which it is based.

The trend in treatment for ovarian cancer is toward more use of chemotherapy and in particular an increasing reliance on combinations of drugs. Despite some promising recent findings,²¹ there is limited evidence so far that the effectiveness of such chemotherapy improves with the intensity or number of agents, and our study clearly shows dramatic increases in the long-term risk of leukemia at high dosages. In the case of Hodgkin's disease, a substantial risk of leukemia following combination chemotherapy is clearly offset by enormous gains in survival.²² However, after ovarian cancer, the extent to which the increased risk of leukemia is offset is still unclear.