Therapy-related myelodysplasia and acute myeloid leukemia after Ewing sarcoma and primitive neuroectodermal tumor of bone: a report from the Children's Oncology Group

Ewing sarcoma and primitive neuroectodermal tumor of bone are closely related, highly malignant tumors occurring in children, adolescents, and young adults. Since treatment with surgery or local radiation alone is usually associated with a very high fatality rate,¹  adjuvant chemotherapy has been extensively used in this population, with chemotherapy for patients with Ewing sarcoma initially being based on 4 drugs: doxorubicin, cyclophosphamide, vincristine, and dactinomycin.^2-5  More recently, treatment with ifosfamide, with or without etoposide, has demonstrated a therapeutic advantage in patients who had relapsed after standard therapies for Ewing sarcoma.^3,6-9 

These promising results led Grier et al¹⁰  to assess whether addition of ifosfamide and etoposide to the standard regimen would improve survival in patients with newly diagnosed disease. Between 1988 and 1992, patients with or without metastatic disease were treated according to a cooperative group (Children's Cancer Group/Pediatric Oncology Group) therapeutic protocol: INT-0091. Patients were assigned randomly at study entry to receive standard chemotherapy (regimen A) with doxorubicin, vincristine, cyclophosphamide, and dactinomycin (VAdCA), or experimental therapy (regimen B) consisting of these 4 drugs alternating with courses of ifosfamide and etoposide (VAdCA/IE). Between 1992 and 1994, patients with metastatic disease were nonrandomly assigned to a high-intensity chemotherapy regimen (regimen C). Reported results for patients treated on regimen A or B revealed that the addition of ifosfamide and etoposide to a standard regimen significantly improves the outcome for patients with nonmetastatic Ewing sarcoma, with an overall survival of 72%.¹⁰  The outcome for patients with metastatic disease continues to be dismal despite intensification of therapy.

Therapy-related myelodysplasia and acute myeloid leukemia (t-MDS/AML) are well-documented complications in patients with Hodgkin lymphoma^11,12 ; acute lymphoblastic leukemia^13,14 ; and breast,^15,16  ovarian,¹⁷  and testicular¹⁸  cancer, with clearly defined associations with exposure to alkylating agents and topoisomerase II inhibitors. Little information exists regarding the risk of t-MDS/AML in patients treated for Ewing sarcoma.

Patients treated according to therapeutic protocol INT-0091 received high doses of alkylating agents and topoisomerase II inhibitors, potentially placing them at an increased risk for t-MDS/AML. We therefore aimed to determine the incidence of t-MDS/AML and associated risk factors in individuals treated for newly diagnosed Ewing sarcoma on a CCG/POG therapeutic protocol (INT-0091).

INT-0091 was opened to all member institutions of CCG and POG between 1988 and 1992. Patients accrued to this study were 30 years of age or younger at diagnosis with primary bone tumors diagnosed as Ewing sarcoma, peripheral neuroectodermal tumors, or primitive sarcomas of bone. Prior anticancer therapy other than surgery for diagnosis was not allowed. To participate, patients or their guardians gave written informed consent according to institutional and US National Cancer Institute guidelines and in accordance with the Declaration of Helsinki. The protocol was approved by the institutional review boards at all participating centers.

Patients were assigned randomly at study entry to receive standard chemotherapy (regimen A) with doxorubicin, vincristine, cyclophosphamide, and dactinomycin (VAdCA) or experimental therapy (regimen B) comprising those 4 drugs alternating with courses of ifosfamide and etoposide (VAdCA/IE). Randomization was stratified according to presence of metastases. This study was designed initially to include patients with and without metastatic disease at presentation.

In both regimens A and B, the planned VAdCA courses comprised vincristine 2 mg/m² (2-mg maximum dose), doxorubicin at 75 mg/m² per dose, and cyclophosphamide at 1200 mg/m². Dactinomycin at 1.25 mg/m² was substituted for doxorubicin when a total doxorubicin dose of 375 mg/m² was reached. In regimen B, ifosfamide and etoposide courses consisted of ifosfamide at 1800 mg/m² per day for 5 days and etoposide at 100 mg/m² per day over 5 days. Courses of chemotherapy were administered every 3 weeks for a total of 17 courses. Hematopoietic cytokines were not available at the start of the protocol; however, filgrastim and sargramostim, approved during the course of the study, were used at the discretion of treating physicians.

After it was apparent that accrual was sufficient to analyze patients without metastases at presentation as a separate group, the study was amended to allow piloting of high-dose chemotherapy in patients with metastatic disease only (regimen C). The chemotherapeutic protocol consisted of 2 blocks: one with vincristine (2 mg/m²), doxorubicin (90 mg/m²), and cyclophosphamide (2200 mg per square meter); and the second with ifosfamide (2800 mg/m² per day for 5 days) and etoposide (100 mg/m² per day for 5 days). The 2 blocks were alternated for the duration of therapy. All patients enrolled on regimen C, however, received G-CSF as part of protocol therapy.

Local control for all patients consisted of radiation therapy, surgery, or both at week 12 at the treating physician's discretion. For patients who received radiotherapy alone, the initial tumor volume was treated to 4500 cGy, followed by reduction in treatment volume to the postchemotherapy, preradiotherapy tumor extent for a total dose of 5580 cGy. Patients with residual tumor after surgery also were irradiated using these dose-volume guidelines for gross residual disease; for microscopic residual disease, irradiation was limited to 4500 cGy to the original volume with a 1-centimeter margin. At completion of local control therapy, maintenance therapy was started.

The treatment schema is summarized in Figure 1.

Follow-up information, current through August 2000, is used for this analysis. Data on the timing and type of local control measures were obtained for patients without metastasis at diagnosis.

For patients in whom t-MDS/AML developed, date of diagnosis, histologic characteristics, and cytogenetics were recorded. Pathology reports were obtained from treating institutions and reviewed to verify diagnoses. Each patient diagnosed with t-MDS/AML was further classified according to whether the t-MDS/AML demonstrated 11q23 abnormalities, had a deletion of all or part of chromosomes 5 or 7, had neither genetic abnormality, or could not be classified because of lack of cytogenetic analysis.

The time under observation for each patient was calculated as the time from study enrollment until diagnosis of t-MDS/AML, date of death, or the date patient was last seen, whichever came first. A patient who experienced t-MDS/AML was considered to have experienced an event; in all other cases, the patient was considered censored at the date of last contact.¹⁹  Analyses including treatment for local control used the time at risk accumulated from the date of the start of maintenance. A patient who experienced t-MDS/AML or died prior to the start of maintenance was excluded from the analysis. Only patients without metastatic disease were included in the analyses related to local control.

The cumulative incidence of t-MDS/AML was calculated and compared across characteristics of interest by the method proposed by Gray.²⁰ 

The expected number of cases of t-MDS/AML was calculated using the public use dataset from the Surveillance, Epidemiology and End Results (SEER) program of the US Centers for Disease Control and Prevention (Atlanta, GA). For every person, the number of person-years at risk contributed to each SEER-defined 5-year age group was calculated from the time of enrollment until diagnosis of t-MDS/AML, death, or the date the patient was last seen, whichever came first. The expected number of cases was estimated as the sum across all age groups of the number of person-years contributed to that age group times the SEER AML rate in each age group. The standardized incidence rate (SIR) was calculated as the ratio of the observed number of cases to the expected number of cases. The 95% confidence interval for the SIR was calculated assuming the number of cases was distributed as a Poisson distribution.

The association between various patient characteristics and risk for t-MDS/AML was assessed using the proportional-hazards regression model. Patient characteristics included in the regression model included age at diagnosis (0 to 9 years [referent group], 10 to 17 years, and 18 years of age or older), sex, and therapeutic regimens (regimens A [referent group], B, and C). For patients with nonmetastatic disease, measures of local control (surgery only or no treatment [referent group], radiation only, and surgery with radiation) were also included. The coefficients of the proportional-hazards regression provide asymptotically unbiased estimates for the effects associated with the patient characteristics on risk for development of t-MDS/AML at any time after enrollment.

The relationship between filgrastim and the risk for t-MDS/AML among patients enrolled on regimen C was examined, using a time-dependent covariate, proportional-hazards regression analysis. Time to t-MDS/AML was calculated as the time from enrollment until the diagnosis of t-MDS/AML, death, or last contact, whichever was earliest. If the patient was diagnosed with t-MDS/AML, the individual was considered to have experienced an event; otherwise, the patient was considered censored at last contact.

The risk for t-MDS/AML was modeled according to time at risk and total filgrastim dose at that time as follows: λ(t) = λ₀(t)exp{βZ(t)}, where Z(t) is the cumulative dose of filgrastim up to time t. β represents the increment or decrement in log-relative risk for each unit increase in filgrastim.

Excess absolute risk was calculated as an additional indicator of impact of cancer diagnosis and therapy on the cohort when compared with the general population. Excess absolute risk was determined by subtracting the expected number of malignancies in the cohort from the observed number, dividing the difference by person-years of follow-up, and multiplying that value by 1000.

Five hundred seventy-eight patients with Ewing sarcoma or primitive neuroectodermal tumor of bone were included in this analysis. Of these, 262 were randomly assigned to regimen A, and 256 to regimen B that received ifosfamide and etoposide. An additional 60 patients with metastatic disease were nonrandomly assigned to regimen C. The cumulative doses of the relevant therapeutic exposures prescribed for each therapeutic arm are detailed in Table 1.

The cohort was observed for a median of 8.03 years (range, 0-11.8 years). Eleven patients developed t-MDS/AML, including 4 patients with deletion of chromosome 5 or 7, 2 patients with 11q23 abnormality, and 1 patient who had both types of cytogenetic abnormalities (11q23 and deletion of chromosome 5/7). Cytogenetic reports were not available for 4 patients with t-MDS/AML. One patient developed secondary acute lymphoblastic leukemia (ALL), and 4 patients developed solid nonhematopoietic malignancies. The secondary ALL presented T-cell characteristics phenotypically. The solid nonhematopoietic malignancies included osteosarcoma (n = 1), malignant fibrous histiocytoma (n = 2), and ovarian cancer (n = 1). These 5 patients were censored at the time of development of second malignancies and are not included in the current report beyond that point. Characteristics of the patient population and treatment received are summarized in Table 2.

As shown in Figure 2, the cumulative incidence of t-MDS/AML developing in this cohort approached 2% at 5 years (95% confidence interval [CI], 0.6% to 3%), with a plateau at 7 years. Figure 3 shows the cumulative incidence of t-MDS/AML by therapeutic regimen. Patients randomized to regimen A or B demonstrated the 5-year cumulative incidence of developing t-MDS/AML to be 0.4% (95% CI, 0% to 1%) and 0.9% (95% CI, 0% to 2%), respectively, with no statistically significant difference in the incidence of t-MDS/AML between the 2 arms. However, patients assigned to regimen C demonstrated a cumulative incidence of 11% (95% CI, 2% to 19%) at 5 years for developing t-MDS/AML. Of the 6 patients with t-MDS/AML on regimen C, 4 demonstrated deletions of chromosome 5 or 7, 1 demonstrated both deletion of chromosome 5 or 7 and 11q23 abnormality, while cytogenetics were unavailable for the sixth patient.

At the time of the analysis, the cohort had accrued approximately 2926 person-years of follow-up. The observed and expected cases of t-MDS/AML, calculated on the basis of age- and sex-specific rates for AML, are shown in Table 3. Overall, a total of 0.09 cases would have been expected, and 11 were observed (SIR, 127.68; 95% CI, 71.96-228.49). Although, significantly elevated risks were observed in patients placed on regimen A (SIR, 77.63; 95% CI, 28.08-226.94) and regimen B (SIR, 47.41; 95% CI, 14.58-171.28) when compared with the general population, the difference between the 2 arms was not statistically significant. However, among patients treated on regimen C, the risk of developing t-MDS/AML was 1126-fold that of the general population (SIR, 1126.34; 95% CI, 527.50-2451.66), and this risk was statistically significantly higher than that for the cohorts of patients treated on regimens A and B. The absolute risk for t-MDS/AML was 3.76 cases per 1000 patients per year and ranged from 1.40 cases per 1000 patients per year for regimen B to 33.64 cases per 1000 patients per year for patients assigned to regimen C.

Only 1 of the 11 cases of t-MDS/AML developed among patients who had suffered a relapse of their original disease necessitating additional chemotherapy before diagnosis of t-MDS/AML. That patient received 3 additional regimens after first recurrence. Two regimens consisted of retrieval therapy including ifosfamide and etoposide, and the third regimen included hematopoietic-cell transplantation including a conditioning regimen with busulfan and cyclophosphamide. Including relapse as a competing risk did not appreciably change the estimated risks or conclusions of the analysis.

Results of the multivariable analysis (shown in Table 4) demonstrated that treatment according to regimen C was the only independent risk factor associated with the development of t-MDS/AML (relative risk, 15.9; 95% CI, 3.84 to 65.82) when compared with regimen A. In particular, radiation did not emerge as a significant risk factor for the development of t-MDS/AML in this population. The time-dependent analysis of delivered dose of filgrastim for patients enrolled on regimen C (where such data were recorded on data forms) did not indicate an increased risk for t-MDS/AML with increasing dose of filgrastim (P = .23).

The current treatment of Ewing sarcoma and primitive neuroectodermal tumors of bone includes intensive multiagent chemotherapy with topoisomerase II inhibitors and alkylating agents,¹⁰  potentially placing the exposed individuals at a high risk for t-MDS/AML. However, there is a paucity of information regarding the magnitude of risk of t-MDS/AML in patients with Ewing sarcoma exposed to these potentially leukemogenic agents.^21,22  In this cohort of patients treated for Ewing sarcoma on Children's Oncology Group therapeutic trial INT-0091, we found the cumulative incidence of t-MDS/AML to be 2% at 5 years. This incidence was comparable with that reported following conventional therapy for other diagnoses such as Hodgkin disease¹¹  and osteosarcoma.²³  Furthermore, the entire study cohort was at a 127-fold increased risk of developing t-MDS/AML.

Patients treated on regimens A and B received 375 mg/m² doxorubicin, while those on regimen C received 450 mg/m². Cyclophosphamide exposure was highest in patients treated on regimen A (20.4 g/m²); intermediate for regimen C (17.6 g/m²); and lowest for regimen B (9.6 g/m²). Patients on regimen A did not receive any ifosfamide, while those on regimen B received 90 g/m², and those on regimen C received 140 g/m². Similarly, patients on regimen A did not receive any etoposide, while those on the regimens B and C received 5 g/m² on identical schedules.

Taking into account the differences in exposure to potentially leukemogenic chemotherapeutic agents by treatment regimen, we made the following deductions regarding the role of these agents, as well as their cumulative doses in the development of t-MDS/AML in this population. The risk of t-MDS/AML was comparable for regimens A and B. However, patients in regimen A received more cyclophosphamide (20.4 g/m²) than those in regimen B (9.6 g/m²), but patients in regimen A did not receive ifosfamide. Not knowing the relative leukemogenic effect of cyclophosphamide versus ifosfamide, it can be concluded that exposure to cyclophosphamide at 20.4 g/m² with doxorubicin at 375 mg/m² (in the absence of etoposide and ifosfamide) placed patients at a risk of developing t-MDS/AML that was comparable with an exposure to cyclophosphamide at 9.6 g/m², doxorubicin at 375 mg/m², ifosfamide at 90 g/m², and etoposide at 5 g/m².

However, patients with Ewing sarcoma presenting with metastatic disease and treated on regimen C demonstrated a cumulative incidence of t-MDS/AML of 11% at 5 years from diagnosis, representing a 16-fold increased risk of t-MDS/AML when compared with patients treated on regimen A. Furthermore, patients treated on regimen C were at more than 1000-fold increased risk of t-MDS/AML when compared with the general population. These observations suggest that increasing the cumulative exposure of ifosfamide from 90 g/m² (regimen B) to 140 g/m², in combination with cyclophosphamide exposure from 9.6 g/m² to 17.6 g/m² and doxorubicin exposure from 375 mg per square meter to 450 mg/m², increased the risk of t-MDS/AML significantly. Furthermore, the large majority of t-MDS/AML developing in patients treated on regimen C demonstrated deletion of chromosomes 5 or 7, suggesting that exposure to alkylating agents was mechanistically involved in these malignancies.

In a previous study of 778 patients with osteosarcoma randomized to receive 45 g/m² of ifosfamide or no ifosfamide, with identical doxorubicin (450 mg/m²) and cisplatin (480 mg/m²) doses, we had demonstrated that the addition of ifosfamide at 45 g/m² did not significantly increase the risk of t-MDS/AML.²³  The current study demonstrates that increasing the dose of ifosfamide to 90 g/m² does not increase the risk of t-MDS/AML. However, the combination of ifosfamide at 140 g/m² with cyclophosphamide at 17.6 g/m² and doxorubicin at 450 mg/m² results in a dramatic increase in the risk of t-MDS/AML.

A recent study has demonstrated that short-term use of hematopoietic cytokines, in particular, filgrastim in the setting of intensive antileukemic therapy, may increase the risk of t-MDS/AML.²⁴  Enrollment on the therapeutic protocol described in the current study began in 1988, and hematopoietic cytokines were not available at the start of the protocol; however, filgrastim and sargramostim, approved during the course of the study, were used at the discretion of treating physicians after their approval. Information regarding exposure to these agents was not collected systematically and hence was not available for analysis for patients enrolled on regimen A or B. All patients enrolled on regimen C, however, received G-CSF as part of protocol therapy. The time-dependent analysis of delivered dose of filgrastim for patients enrolled on regimen C (where such data were recorded on data forms) did not indicate an increased risk for t-MDS/AML with increasing dose of filgrastim (P = .23).

In conclusion, the risk of t-MDS/AML following Ewing sarcoma is elevated compared with the general population. Given the relatively poor prognosis of t-MDS/AML, there is an urgent need to identify susceptible subpopulations by exploring the role of genetic susceptibility and gene-environment interactions, if we wish to continue to treat patients with high intensity, but potentially leukemogenic therapeutic agents.

This work was supported in part by U10CA 098543.