Prognostic factors in children and adolescents with acute ... - Nature

Leukemia (2000) 14, 1201–1207  2000 Macmillan Publishers Ltd All rights reserved 0887-6924/00 $15.00 www.nature.com/leu

Prognostic factors in children and adolescents with acute myeloid leukemia (excluding children with Down syndrome and acute promyelocytic leukemia): univariate and recursive partitioning analysis of patients treated on Pediatric Oncology Group (POG) Study 8821 M Chang1, SC Raimondi2, Y Ravindranath3, AJ Carroll4, B Camitta5, MV Gresik6, CP Steuber6 and H Weinstein7 1

Pediatric Oncology Group Statistical Office, University of Florida, Gainesville, Florida; 2St Jude Children’s Research Hospital, Memphis, Tennessee; 3Barbara Ann Karmanos Cancer Institute and Wayne State University, Detroit, Michigan; 4Department of Human Genetics, University of Alabama at Birmingham, Alabama; 5Midwest Children’s Hospital, Milwaukee, Wisconsin; 6Baylor College of Medicine, Houston, Texas; and 7Massachusetts General Hospital and Harvard Medical School, Boston, Massachusetts, USA

The purpose of the paper was to define clinical or biological features associated with the risk for treatment failure for children with acute myeloid leukemia. Data from 560 children and adolescents with newly diagnosed acute myeloid leukemia who entered the Pediatric Oncology Group Study 8821 from June 1988 to March 1993 were analyzed by univariate and recursive partitioning methods. Children with Down syndrome or acute promyelocytic leukemia were excluded from the study. Factors examined included age, number of leukocytes, sex, FAB morphologic subtype, cytogenetic findings, and extramedullary disease at the time of diagnosis. The overall event-free survival (EFS) rate at 4 years was 32.7% (s.e. = 2.2%). Age ⭓2 years, fewer than 50 × 109/l leukocytes, and t(8;21) or inv(16), and normal chromosomes were associated with higher rates of EFS (P value = 0.003, 0.049, 0.0003, 0.031, respectively), whereas the M5 subtype of AML (P value = 0.0003) and chromosome abnormalities other than t(8;21) and inv(16) were associated with lower rates of EFS (P value = 0.0001). Recursive partitioning analysis defined three groups of patients with widely varied prognoses: female patients with t(8;21), inv(16), or a normal karyotype (n = 89) had the best prognosis (4-year EFS = 55.1%, s.e. = 5.7%); male patients with t(8;21), inv(16) or normal chromosomes (n = 106) had an intermediate prognosis (4-year EFS = 38.1%, s.e. = 5.3%); patients with chromosome abnormalities other than t(8;21) and inv(16) (n = 233) had the worst prognosis (4-year EFS = 27.0%, s.e. = 3.2%). One hundred and thirty-two patients (24%) could not be grouped because of missing cytogenetic data, mainly due to inadequate marrow samples. The results suggest that pediatric patients with acute myeloid leukemia can be categorized into three potential risk groups for prognosis and that differences in sex and chromosomal abnormalities are associated with differences in estimates of EFS. These results are tentative and must be confirmed by a large prospective clinical trial. Leukemia (2000) 14, 1201–1207. Keywords: childhood acute myeloid leukemia; prognostic factors; recursive partitioning analysis

Introduction Tailoring therapy by risk group has improved the overall outcomes of children with acute lymphoblastic leukemia (ALL) and is now a standard approach for treating this disorder.1 However, treatment failure has remained high for most children with acute myeloid leukemia (AML) despite various attempts to intensify therapy.2–10 Notable exceptions include children with Down syndrome (DS) who received high-dose cytarabine (HdA) early in consolidation therapy;11 these children have long-term event-free survival (EFS) estimates of

⬎75%. The use of all-trans-retinoic acid with chemotherapy has significantly improved the outcome of children with acute promyelocytic leukemia (APL).12 New therapeutic strategies are needed for improving the outcomes of children who have AML but do not have DS or APL. To facilitate the design of new therapeutic strategies, it is necessary to redefine prognostic groups for children with AML. Defining widely accepted risk groups has proved to be difficult because of the morphologic diversity, cytogenetic heterogeneity, frequent extramedullary disease (EMD), prior myelodysplastic prodrome, the different causes of AML (ie primary disease and secondary disease), and the inability to identify a subgroup with a clearly superior outcome. Nevertheless, emerging data indicate that there may be distinct cytogenetic and morphologic subgroups with vastly different outcomes.13,14 Studies of adults with AML have shown that the presence of t(8;21) or an inv(16) is a predictor of good outcome, presence of t(15;17) an intermediate outcome, and all other chromosomal abnormalities are predictors of a poor outcome.13,14 The greatest benefit of treatment with HdA is seen in patients with a t(8;21) or an inv(16).13 Recent studies also suggest that adult patients with a t(8;21) or an inv(16), whose disease entered remission with the first course of induction had much better outcomes than those whose disease achieved remission after more than one course.14 However, the data from studies of adults may not be applicable to children with AML because the distribution of specific cytogenetic subgroups is different in children than in adults. For example, AML with abnormalities of chromosomes 5 and 7 is more common in adults, and AML with 11q23 abnormalities is more common in children. Further, the occurrence of both DS and AML is unique to children. For children with AML, the presence of constitutional trisomy 21, ie DS, has emerged as the single best predictor of a good outcome. The ability to identify additional subgroups of pediatric patients according to their risk of relapse should improve our understanding of the pharmacobiology of response to therapy and should facilitate the development of more effective therapies. In this study, we identified subgroups by using retrospective univariate and recursive partitioning analyses of data from a large cohort of pediatric patients with AML treated on Pediatric Oncology Group (POG) Study 8821. Materials and methods

Correspondence: Y Ravindranath, Children’s Hospital of Michigan, 3901 Beaubien Boulevard, Detroit, Michigan 48201, USA; Fax: 313745-5237 Received 11 October 1999; accepted 16 March 2000

Treatment protocol The treatment regimen and the overall results of the POG 8821 study have been published.8 Remission induction ther-

Prognostic factors in childhood AML M Chang et al

1202

apy consisted of one course of DAT (daunorubicin, cytarabine and 6-thioguanine) followed by a high dose of cytarabine (3 g/m2 every 12 h for six doses) (HdA). All patients then received one course of etoposide plus azacytidine (VP/AZ). Randomization was performed after treatment with HdA. Allogeneic bone marrow transplant (allo-BMT) was offered to patients who had a matched sibling donor. Those not eligible for alloBMT were randomized to receive either intensive chemotherapy (sequential courses of D/HdA, DAT, VP/AZ, HdA, DAT, VP/AZ) or autologous bone marrow transplantation (ABMT) with 4-hydroperoxycyclophosphamide-purged marrow after a preparative regimen of busulfan (4 mg/kg/day for 4 days) and cyclophosphamide (50 mg/kg/day for 4 days).

Patients Between June 1988 and March 1993, 666 patients less than 21 years of age with previously untreated AML entered the protocol after informed consent. Ten patients were ineligible, and seven others were not evaluable for response. Of the remaining 649 patients evaluable for treatment response, 478 (74%) had complete cytogenetic data. For the purposes of this analysis, 34 children with DS (one child with DS had a t(15;17) and 55 other cases with a t(15;17) were excluded for reasons stated above. Data from the remaining 560 eligible patients were used for the analysis; of these, the central FAB review was available in 479 (in 74 cases central laboratory did not receive slides and in seven others the slides sent were not satisfactory for review); extra medullary disease (EMD) status was unspecified in 68/560 cases; 428/560 (76%) had adequate cytogenetic analysis (diagnostic marrow/blood samples were not submitted for cytogenetics in 95 cases and in 37 others samples were unsatisfactory for a chromosome analysis).

Cytogenetic analysis and FAB classification Whereas comparison of outcomes achieved by ABMT and intensive chemotherapy was the primary aim of POG 8821, a secondary aim was to identify prognostic factors associated with outcome. All cytogenetic studies were done in the central reference laboratories at the University of Alabama at Birmingham or St Jude Children’s Research Hospital. Detailed descriptions of the cytogenetic findings are reported elsewhere.15 Morphology, which was categorized according to the French–American–British (FAB) classification system, was centrally reviewed at the Baylor College of Medicine.

Definitions A patient’s disease was considered to be in remission if either M1 marrow (fewer than 5% blast cells) or M2a marrow (fewer than 15% blast cells) was observed after the two induction courses. Event-free survival (EFS) was calculated from the date of registration until the date of first relapse or the date of death.

Statistical analysis The covariates in the study were age at the time of study entry (subdivided at 2.0, 5.0, 8.0, 10.0, and 12.0 years), number of leukocytes (subdivided at 10, 15, 25, and ⭓50 × 109/l), sex, Leukemia

FAB subtype of AML, results of cytogenetic analyses (t(8;21) or inv(16), normal chromosomes, and other chromosomal abnormalities), presence of blast cells in the cerebrospinal fluid (CSF), and evidence of extramedullary disease (EMD) other than in the CSF. These covariates were analyzed by univariate methods and by recursive partitioning and amalgamation methods to assess their degree of association with EFS. The partitioning and amalgamation analysis16–18 is based on the Cox regression model19 that includes three independent variables, an indicator for one of the binary covariate listed above and two time-dependent treatment indicators. The first treatment indicator is set to zero and becomes one when the patient undergoes ABMT; the other indicator is set to zero and becomes one when the patient undergoes allo-BMT. In this analysis, an estimate of the hazard ratio (defined as ratio of the instantaneous risks of treatment failure for two groups of patients: those with the factor and those without it) associated with each binary (yes/no) covariate is computed, and the P value for the test that this ratio differs from unity is determined. This process is repeated for each binary covariate described above. The model includes the two treatment indicators so that the estimated hazard ratios and P values are adjusted for treatment effect. If at least one of the hazard ratios differs from unity at the 0.01 level, the patient population is partitioned into two groups by the covariate with the smallest estimated hazard ratio among those covariates meeting the criterion of P ⬍ 0.01. In the next attempt, each of the two groups is analyzed individually in exactly the same way in an attempt to subdivide each group. After each partition step, the subgroups are ranked from the highest risk to the lowest risk of treatment failure and adjacent subgroups in this ranking are combined if the Cox proportional hazards are not significantly different from unity, that is P ⬎ 0.05. This procedure continues until no further partitioning can be achieved. The advantages and disadvantages of this procedure as compared with those of the stepwise Cox regression model have been discussed in Harris et al.17 The EFS curves were determined for the final risk groups (good, intermediate, and poor) by the Kaplan–Meier method.20 Results In a previous report describing POG 8821, it was shown that there was no statistically significant difference in overall survival or EFS of patients in either of the two randomized treatment groups.8 The two treatment groups were similar with regard to FAB subtypes, age, sex, initial WBC, presence or absence of EMD, cytogenetic subgroups and the proportion of patients achieving remission after the first or second course of induction therapy. The remission rates of the 560 patients who were evaluated for response on POG 8821 were compared by using the chisquare test (Table 1). Four hundred and eighty or 85.7% achieved remission as defined (438 (91%) M1 and 42 (9%) M2a). The disease-free survival from date of response in M1 and M2a categories were not significantly different (two-sided P = 0.78) and therefore the two groups are analyzed together for all other response evaluations. The data suggest higher remission rates were seen for patients ⭓2 years old (87.5%, P = 0.026), females (88.9%, P = 0.053), and patients with a t(8;21) or an inv(16) (96.5%, P = 0.002). Conversely, lower rates of remission were seen for patients with the M5 subtype of AML (78.4%, P = 0.028) and for patients with chromosomal abnormalities other than t(8;21) and inv(16) (83.3%, P =


Table 1

1203

Characteristics, remission rates and 4-year EFS rates of 560 children with AML

Characteristic

Rate of remission induction (%)

Chi-square test P valuea

4-year EFS rate % (s.e. %)

Log rank test P value

All patients

480/560 (85.7)

Age ⬍2 Years ⬎2 Years

94/119 (79.0) 386/441 (87.5)

0.026

22.9 (4.2) 35.3 (2.5)

0.0004

Leukocytes (×109/l) ⬍10 10–15 15–25 25–50 ⬎50

166/188 (88.3) 54/59 (91.5) 52/61 (85.2) 72/85 (84.7) 136/167 (81.4)

0.25 0.24 1.0 0.87 0.065

37.2 27.9 26.2 35.5 30.4

0.12 0.86 0.22 0.45 0.12

Cytogenetics t(8;21) or inv(16) Normal Other abnormalities

83/86 (96.5) 93/109 (85.3) 194/233 (83.3)

0.002 0.75 0.047

49.3 (6.2) 43.1 (5.0) 27.0 (3.2)

0.0003 0.030 ⬍0.0001

Sex Male Female

249/300 (83.0) 231/260 (88.9)

0.053

30.1 (2.9) 35.7 (3.2)

0.092

EMD No EMD CSF only Non-CFS EMD

386/441 (87.5) 22/23 (95.7) 22/28 (78.6)

0.82 0.34 0.24

34.4 (2.5) 34.8 (9.9) 21.6 (8.6)

0.18 0.91 0.043

FAB classification M1/M2 M4 M5 M7 Other

187/213 (87.8) 85/94 (90.4) 69/88 (78.4) 26/32 (81.2) 45/52 (86.5)

0.35 0.19 0.028 0.43 1.0

34.4 36.1 21.2 28.1 43.0

0.11 0.26 0.0002 0.14 0.24

32.7 (2.2)

(3.8) (6.6) (5.8) (5.8) (3.9)

(3.7) (5.3) (4.7) (8.4) (7.5)

EFS, event-free survival; AML, acute myeloid leukemia; s.e. standard error; EMD, extramedullary disease; CSF, disease present in the cerebrospinal fluid; Non-CSF EMD, extramedullary disease present except in cerebrospinal fluid; FAB, French–American–British. a The P values were computed by the two-sided test comparing the rates of remission induction and EFS between two groups with and without the specific entry. For example, P = 0.12 when the EFS of patients with fewer than 10 × 109/l leukocytes were compared with the EFS of patients with more than 10 × 109/l leukocytes. The P values for remission rates were computed by exact conditional chi-square test and the P values for EFS were computed by log rank test.

0.047). Cox regression analysis (adjusted for treatment) of each univariately significant (P ⬍ 0.05) prognostic indicator for EFS showed that higher rates of EFS were univariately associated with age ⭓2 years (P = 0.0032), fewer than 50 × 109/l leukocytes (P = 0.049), a t(8;21) or an inv(16) (P = 0.0003), normal chromosomes (P = 0.031) (Table 2). Lower rates of EFS were associated with the M5 subtype of AML (P = 0.0003) and chromosomal abnormalities other than t(8;21) and inv(16) (P ⬍ 0.0001). These results were mostly consistent with those from the log rank tests without adjustment for treatment (Table 1). The outcome in the group without the cytogenetic data Table 2

was examined separately. The 4-year EFS for the 132 cases without cytogenetics was 22.3% compared to 35.3% for those with cytogenetics (P = 0.011). There were no significant differences between the two groups with regard to the frequency of various FAB subgroups, age ⬍ or ⭓2 years, WBC at diagnosis, induction/remission deaths or post-remission treatment assignment. Thus the reasons for the observed difference in EFS are not clear. A tree diagram summarizing the results of the recursive partitioning and amalgamation analysis is shown in Figure 1. Data from patients from whom inadequate samples were

Relationship between pretreatment characteristics and EFS in 560 children with AMLa

Factor

Age Leukocyte (× 109/l) M5 subtype t(8;21) or inv(16) Normal chromosomes Other abnormalityb

Factor indicating better prognosis

P value

Hazard ratio better/worse

95% Confidence interval (HR)

⬎2 years ⬍50 − + Normal −

0.0032 0.0492 0.0003 0.0003 0.0309 0.0001

0.699 0.802 0.612 0.544 0.732 0.530

0.551–0.887 0.644–0.999 0.470–0.798 0.391–0.756 0.551–0.677 0.415–0.677

EFS, event-free survival; AML, acute myeloid leukemia; HR, hazard ratio. a Results determined from Cox regression analysis, adjusted for treatment, for univariately significant prognostic indicators of EFS. b Chromosomal abnormalities other than t(8;21) and inv(16). Leukemia


1204

Figure 1 Recursive partitioning and amalgamation analysis of EFS in 560 children with AML. At the top of each branching point, the branching variable is given with the P value and hazard ratio (HR). The groups, from left to right, are ranked from the highest risk to the lowest risk of treatment failure.

obtained for cytogenetic studies (n = 132) were excluded from further recursive and amalgamation analysis. Among the factors that were univariately associated with EFS and had P values ⬍0.01, chromosomal abnormalities other than t(8;21) and inv(16) had the smallest estimated hazard ratio. Therefore, the patient sample was partitioned into two groups at the first step: group 1 (n = 233), in which patients had chromosomal abnormalities other than t(8;21) and inv(16); and group 2 (n = 195), in which patients had a t(8;21), an inv(16), or normal chromosomes. Among patients (n = 233) with chromosomal abnormalities other than t(8;21) and inv(16), there was no additional significant prognostic variable. Among patients (n = 195) with a t(8;21), and inv(16), or normal chromosomes, the covariates significantly associated with higher rates of EFS were age greater than 2 years (P = 0.041), leukocytes less than 10 × 109/l (P = 0.037), female sex (P = 0.0056), and negative EMD (P = 0.037) (Table 3). Among the four variables, only sex met the significance criterion of P ⬍ 0.01. Therefore, the group of patients with t(8;21), inv(16), or a normal karyotype was subdivided into two groups: males and females. Partitioning at the second step yielded three groups of patients which were ranked from highest risk to lowest risk of treatment failure: patients with chromosomal abnormalities other than t(8;21) and inv(16); male patients with t(8;21), and inv(16), or a normal karyotype; and female patients with t(8;21), inv(16), or a normal karyotype. The first group and the second group could not be amalgamated, because P = 0.0061. No additional factors with prognostic significance were found within any of the three groups. Table 3 = 195)

Thus, the recursive partitioning and amalgamation analysis of the data could place patients into one of three different prognostic groups on the basis of risk factors for treatment failure. The group with a poor prognosis, which included patients with chromosomal abnormalities other than t(8;21) and inv(16), had a 4-year EFS rate of 27.0% (s.e. = 3.2%) (Figure 2). The group with an intermediate prognosis, which included male patients with t(8;21), inv(16), or a normal karyotype, had a 4-year EFS rate of 38.1% (s.e. = 5.3%). The group with a good prognosis, which included female patients with t(8;21), inv(16), or a normal karyotype, had a 4-year EFS rate of 55.1% (s.e. = 5.7%). The distribution of patient characteristics (ie age, M5, the number of leukocytes) among the three groups was examined. The number of leukocytes was similarly distributed among the three groups. The group with the poorest prognosis tended to have more patients with the M5 subtype and more younger patients. There were 58 patients with M5 in the group with the poorest prognosis, five in the group with the intermediate prognosis, and four in the group with the better prognosis. When the data from the patients with the M5 subtype were excluded, the results of analysis with the age-adjusted Cox regression model19 indicated that the ranks for the risk of treatment failure of the three groups were unchanged, and that the risks of treatment failure among the three groups were still significantly different. This finding suggests that the partitioning of patients into the three groups may have independent prognostic value with respect to EFS.

Figure 2 Kaplan–Meier estimates of EFS of patient subgroups. Group 1: females with normal chromosomes, t(8;21), or inv(16) (n = 89); Group 2: males with normal chromosomes, t(8;21), or inv(16) (n = 106); Group 3: patients with chromosomal abnormalities other than t(8;21) and inv(16) (n = 223).

Relationship between pretreatment characteristics and EFS in patients with AML with normal chromosomes, t(8;21), or inv(16) (n

Factor

Factor indicating better prognosis

P value

Hazard ratio better/worse

95% Confidence interval (HR)

Age Leukocytes (× 109/l) Sex EMD

⭓2 years ⬍10 Female Absence of EMD

0.041 0.037 0.0056 0.037

0.0553 0.602 0.570 0.522

0.313–0.977 0.375–0.969 0.383–0.849 0.283–0.962

EFS, event-free survival; AML, acute myeloid leukemia; HR, hazard ratio; EMD, extramedullary disease. Leukemia


Discussion Several groups, including POG, have previously published the results of univariate analyses of prognostic factors in childhood AML.4,5,13,14,21–23 These studies included children with DS and patients with APL. The Berlin–Frankfurt–Munster (BFM) group used morphology, cytogenetic evaluation, and immunophenotyping to define two risk groups.4 The standard risk group included those with FAB M1 plus Auer rods, patients with the M2 subtype and with fewer than 20 × 109/l leukocytes, and patients with the M3 subtype or the M6 subtype of AML. The group with unfavorable risk included all other patients, including those for whom late remission was achieved. In the current analysis, for reasons cited in the Introduction, we excluded patients with DS and APL. The series of pediatric patients with AML included in this analysis constitute the largest group to have undergone central FAB review and centrally performed cytogenetic studies. All children were treated with the same induction therapy. Although the postremission therapy (intensive chemotherapy, ABMT or alloBMT) differed, the model used for the analyses of prognostic factors compensated for this difference. This study confirms the value of cytogenetic analysis for predicting EFS. Chromosomal abnormalities other than t(8;21) and inv(16) indicated a poor prognosis, and the t(8;21) and the inv(16) indicated a better prognosis. Patients with a t(8;21) or an inv(16) had higher EFS rates than those with normal chromosomes did, but the difference was not significant. In addition, age ⭓2 years and fewer than 50 × 109/l leukocytes were favorable factors, and M5 was an adverse factor. Previous POG studies found that age less than 2 years was a predictor of good outcome,3,8 but in the present study, this factor was a predictor of poor outcome. This contrast in findings is in part due to the exclusion of patients with DS in this analysis. Within the subgroup of 88 cases with M5 20 (23%) had t(9;11), seven (8%) had t(11;19), 10 (10%) had other 11q23 abnormalities, 21 others (24%) had abnormal cytogenetics other than t(8;21) or inv(16), nine had normal cytogenetics and in 21 cytogenetics were not done. The outcome of various 11q23 subgroups did not differ significantly.15 The relatively small numbers within each group and the competing risk factors of each of the post-remission treatment arms preclude meaningful speculation with regard to the lack of a better outcome in t(9;11) subgroup compared to other 11q23 variants. In this study, the 4-year EFS for M7 cases was 28% and is higher than the recent data from CCG who reported a 21% EFS in non-DS M7.24 Among our M7 cases four had t(1,22), 11 had other abnormal cytogenetics (no t(8;21) or inv(16)); five had normal cytogenetics and 12 had no data. As in the case of subgroups of M5, the reasons for better outcome are not apparent. The recursive partitioning and amalgamation method used for the multivariate analysis revealed three risk groups of patients. The group with a good prognosis (4-year EFS = 55.1%) included females with t(8;21), inv(16) or a normal karyotype. The group with the intermediate prognosis (4-year EFS = 38.1%) included males with t(8;21), inv(16) or a normal karyotype. The poor prognosis (4-year EFS = 27.0%) group included patients with chromosomal abnormalities other than t(8;21) and inv(16). The Nordic group reported that among children with AML, females had significantly higher EFS rates than did males.21 In our study, the 4-year rate of EFS was 35.7% for females and 30.1% for males, but the difference was not statistically significant (P = 0.092). Further analysis, however, revealed that

sex was a significant prognostic factor for patients with t(8;21), inv(16), or normal chromosomes but was not a prognostic factor for patients with chromosomal abnormalities other than t(8;21) and inv(16). The relationship between sex and chromosomal alterations needs to be confirmed and elucidated in future investigations. The molecular/biochemical bases for the consistently superior outcome in AML with t(8;21) and inv(16) in several studies and conversely the poor outcome in 11q23 cases has not been elucidated yet. Both t(8;21) and inv(16) cause disruption of the normal AML1 gene product and core binding factor (CBF) mediated transcription of genes involved in the myeloid differentiation.25,26 ALL with the TEL/AML1 fusion product due to t(12;21) also has a highly favorable outcome.27 Thus, cytogenetic events, involving the CBF pathway, while predisposing to leukemogenesis, apparently also confer a heightened sensitivity to chemotherapeutic agents analogous to the unique responsiveness of APL to ATRA,12 and the increased sensitivity of the AML blast cells from DS children to cytarabine and daunorubicin.28 In vitro data suggest that the AML1/ETO fusion product of t(8;21) may trap histone deacetylase (HDAC) resulting in transcriptional repression and HDAC inhibitors have been shown to cause differentiation in t(8;21) cell lines.29,30 Hence it is becoming increasingly clear that within AML, each of the cytogenetic subgroups may have unique pharmacogenetic alterations which are susceptible to specific targeted drug therapy. A risk group definition based on cytogenetic findings, is therefore likely to identify if a given therapeutic intervention benefits or adversely affects each of the subgroups. Our partitioning analysis results show that the primary prognostic factors are cytogenetic events and that clinical variables have a modulating effect. Large samples are usually needed to obtain stable results by the partitioning and amalgamation method or by any other multivariate analysis. Because of the moderate sample size (n = 560) of the current study and the lack of cytogenetic evaluation in nearly a quarter of patients, these results should be considered tentative until confirmed in future clinical trials. Stratification of patients by such a risk group definition may allow better identification of subgroups that benefit from a particular intervention (eg cranial irradiation, timed intensification of induction therapy, ABMT, or biologic response modifiers).

1205

Acknowledgements We thank JC Jones for editorial assistance and, Julie Nucci, Martha Sopfe and Pearl Barber for secretarial assistance. The participating Pediatric Oncology Group (POG) institutions and their Principal Investigators are listed in the appendix. This work was supported in part by Grants from the National Institute of Health and the National Cancer Institute, Bethesda, MD (CA-30969, CA-29139, CA-31566, CA-29691, CA-25408, CA-32053, CA-03161, CA-29293, CA-41573, CA-21765) and the American Lebanese Syrian Associated Charities (ALSAC).

References 1 Smith M, Arthur D, Camitta B, Carroll AJ, Crist W, Gaynon P, Gelber R, Heerema N, Korn ML, Link M, Murphy S, Pui CH, Pullen J, Reamon G, Sallan SE, Sather H, Shuster J, Simon R, Trigg M, Tubergen D, Uckun F, Ungerleider R. Uniform approach to classiLeukemia


1206 2 3

4

5

6

7

8

9

10

11

12

13

14

15

Leukemia

fication and treatment assignment for children with acute lymphoblastic leukemia. J Clin Oncol 1996; 14: 18–24. Weinstein HJ, Mayer RH, Rosenthal DS, Coral FS, Camitta EM, Gelber RD. Chemotherapy for acute myelogenous leukemia in children and adults: VAPA update. Blood 1983; 62: 315–319. Steuber CP, Civin C, Krisher J, Culbert S, Ragab A, Ruymann FB, Ravindranath Y, Leventhal B, Wilkinson R, Vietti TJ. A comparison of induction and maintenance therapy for acute nonlymphocytic leukemia in childhood: results of a Pediatric Oncology Group study. J Clin Oncol 1991; 9: 247–258. Creutzig U, Ritter J, Schellong G. Identification of two risk groups in childhood acute myelogenous leukemia after therapy intensification in the study AML-BFM-83 as compared with study AMLBFM-78. AML-BFM Study Group. Blood 1990; 75: 1932–1940. Creutzig U, Harbott J, Sperling C, Ritter J, Zimmermann M, Loffler H, Riehm H, Schellong G, Ludgwig WD. Clinical significance of surface antigen expression in children with acute myeloid leukemia: results of study AML-BFM-87. Blood 1995; 86: 3097–3108. Woods WG, Kobrinsky N, Buckley J, Neudorf S, Sander J, Miller L, Barnard D, Benjamin D, DeSwarte J, Kalousek D. Intensively timed induction therapy followed by autologous or allogeneic bone marrow transplantation for children with acute myeloid leukemia or myelodysplastic syndrome: a Children’s Cancer Group pilot study. J Clin Oncol 1993; 11: 1448–1457. Ravindranath Y, Steuber CP, Krischer J, Civin CI, Ducore J, Vega R, Pitel P, Inoue S, Bleher R, Sexauer C. High-dose cytarabine for intensification of early therapy of childhood acute myeloid leukemia: a Pediatric Oncology Group study. J Clin Oncol 1991; 9: 572–580. Ravindranath Y, Yeager AM, Chang MN, Steuber CP, Krischer J, Graham-Pole J, Carroll A, Inoue S, Camitta B, Weinstein HJ. Autologous bone marrow transplantation versus intensive consolidation chemotherapy for acute myeloid leukemia in childhood. New Engl J Med 1996, 334: 1428–1434. Woods WG, Kobrinsky N, Buckley JD, Lee JW, Sanders J, Neudorf S, Gold S, Barnard DR, DeSwarte J, Dusenbery K, Kalousek D, Arthur DC, Lange BJ. Timed-sequential induction therapy improves postremission outcome in acute myeloid leukemia: a report from Children’s Cancer Group. Blood 1996; 87: 4979– 4989. Hann I, Stevens RF, Goldstone AH, Rees JK, Wheatley K, Gray RG, Burnett AK. Randomized comparison of DAT versus ADE as induction chemotherapy in children and adults with acute myeloid leukemia. Results of the Medical Research Council’s AML trial (MRC AML-10). Adult and Childhood Working Parties of the Medical Research Council. Blood 1997; 89: 2311–2318. Ravindranath Y, Abella E, Krischer JP, Wiley J, Inoue S, Harris M, Chauvenet A, Alvarado CS, Dubowy R, Pitchery AK. Acute myeloid leukemia (AML) in Down’s syndrome is highly responsive to chemotherapy: experience on Pediatric Oncology Group AML Study 8498. Blood 1992; 80: 2210–2214. Degos L, Dombret H, Chomianne C, Daniel MT, Miclea JM, Chastang C, Castaigne S, Fenaux P. All-trans-retinoic acid as a differentiating agent in the treatment of acute promyelocytic leukemia. Blood 1995; 85: 2643–2653. Bloomfield CD, Lawrence D, Byrd JC, Carrol A, Pettenati MJ, Tantravahi R, Patil SR, Davey PR, Berg DT, Schiffer CA, Arthur DC, Mayer RJ. Frequency of prolonged remission duration after highdose cytarabine intensification in acute myeloid leukemia varies by cytogenetic subtype. Cancer Res 1998; 58: 4173–4179. Grimwade D, Walker H, Oliver F, Wheatley K, Harrison C, Harrison G, Rees J, Hann I, Stevens R, Burnett A, Goldstone A. The importance of diagnostic cytogenetics on outcome in AML: analysis of 1612 patients entered into the MRC AML, 10 trial. The Medical Research Council Adult and Children’s Leukemia working Parties. Blood 1998; 7: 2322–2333. Raimondi SC, Chang MN, Ravindranath Y, Behm FG, Gresik MV, Steuber CP, Weinstein HJ, Carroll AJ for the Pediatric Oncology Group. Chromosomal abnormalities in 478 children with acute myeloid leukemia: clinical characteristics and treatment outcome

16

17

18

19 20 21

22

23

24

25 26

27

28

29

30

a cooperative Pediatric Oncology Group study – POG 8821. Blood 1999; 94: 3707–3716. Ciampi A, Hogg SA, Kates L. Stratification by stepwise regression, correspondence, analysis, and recursive partitioning: a comparison of three methods of survival analysis with covariates. Comput Stat Data Anal 1986; 4: 183–204. Shuster JJ, Falletta JM, Pullen DJ, Crist WM, Humphrey GB, Dowell BL, Wharam MD, Borounitz M. Prognostic factors in children with T-cell acute lymphoblastic leukemia: a Pediatric Oncology Group study. Blood 1990; 75: 166–173. Harris MB, Shuster JJ, Carroll A, Look AT, Borowitz MJ, Crist WM, Nitschke R, Pullen J, Steuber CP, Land VJ. Trisomy of leukemic cell chromosomes 4 and 10 identifies children with B-progenitor cell acute lymphoblastic leukemia with a very low risk of treatment failure: a Pediatric Oncology Group study. Blood 1992; 79: 3316–3324. Cox DR. Regression models and life-tables. JR Stat Sec B 1992; 34: 187–220. Kaplan EL, Meier P. Nonparametric estimation from incomplete observations. J Am Stat Assoc 1958; 53: 475–481. Lie SO, Jonmundsson G, Mellander L, Silimes MA, Yssing M, Gustafsson G. A population-based study of 272 children with acute myeloid leukemia treated on two consecutive protocols with different intensity: best outcome in girls, infants and children with Down’s syndrome. Nordic Society of Paediatric Haematology and Oncology (NOPHO). Br J Haematol 1996; 94: 82–88. Zittoun RA, Mandelli F, Willemze R, de Witte T, Labar B, Resegotti L, Leoni F, Damasio E, Visani G, Papa G. Autologous or allogeneic bone marrow transplantation compared with intensive chemotherapy in acute myelogenous leukemia. European Organization for Research and Treatment of Cancer (EORTC) and the Gruppo Italiano Malattie Ematologiche Maligne del’Adulto (GIMEMA) Leukemia Cooperative Groups. New Engl J Med 1995; 332: 217–223. Keating MJ, Smith TL, Kantarjian H, Cork A, Walters R, Trujillo JM, McCredie KB, Gehan EA, Freireich EJ. Cytogenetic pattern in acute myelogenous leukemia: a major reproducible determinant of outcome. Leukemia 1998; 2: 403–412. Lange BJ, Kobrinsky N, Barnard DR, Arthur DC, Buckley JD, Howells WB, Gold S, Sanders J, Neudorf S, Smith FO, Woods WG. Distinctive demography, biology, and outcome of acute myeloid leukemia and myelodysplastic syndrome in children with Down syndrome: Children’s Cancer Group studies 2861 and 2891. Blood 1998; 91: 608–615. Lowenberg B, Downing JR, Burnett A. Acute myeloid leukemia. New Engl J Med 1999; 341: 1051–1062. Wang J, Hoshino T, Redner R, Kajigaya S, Liu J. ETO, fusion partner in t(8;21) acute myeloid leukemia, represses transcription by interaction with the human N-Cor/mSin3/HDACI complex. Proc Natl Acad Sci USA 1998; 95: 10860–10865. Shurtleff SA, Buijs A, Behm FG, Rubinitz JE, Raimondi SC, Hancock ML, Chan GC, Pui CH, Grosveld G, Downing JR. Tel/AML1 fusion resulting from cryptic t(12;21) is the most common genetic lesion in pediatric ALL and defines a subgroup of patients with an excellent prognosis. Leukemia 1995; 9: 1985–1989. Taub J, Huang X, Matherly L, Stout M, Buck S, Massey G, Becton D, Chang M, Weinstein H, Ravindranath Y. Expression of chromosome 21-localized genes in acute myeloid leukemia: differences between down syndrome and non-down syndrome blast cells and relationship to in vitro sensitivity to cytosine arabinoside and daunorubicin. Blood 1999; 94: 1383–1400. Wang J, Saunthararajah Y, Redner R, Liu J. Inhibitors of histone deacetylase relieve ETO-mediated repression and induce differentiation of AMLI-ETO leukemia cells. Cancer Res 1999; 59: 2766–2769. Kosugi H, Towatari M, Hatano S, Kitamura K, Kiyo H, Kinoshita T, Tanimoto M, Murate T, Kawashima K, Saito H, Naoe T. Histone deacetylase inhibitors are the potent inducer/enhancer of differentiation in acute myeloid leukemia: a new approach to anti-leukemia therapy. Leukemia 1999; 13: 1316–1327.


1207

Appendix

Institution Alberta Children’s Hospital All Children’s Hospital Baylor Boston Floating Hospital Cancer Center of Hawaii Carolinas Medical Center Children’s Hospital Greenville System Children’s Hospital Michigan Children’s Hospital New Orleans/ LSU CCOP Children’s Memorial Hospital (Chicago) UC/San Diego Christ Hospital City of Hope Cook-Ft Worth Children’s Medical Center Dana-Farber Cancer Institute Dartmouth Hitchcock Duke University East Carolina University Emory University Fairfax Hospital Hackensack Medical Center Hurley Medical Center Joe DiMaggio Children’s Johns Hopkins University Kaiser Permanente/San Diego Kaiser/Santa Clara Keesler AFB Hospital Medical University of South Carolina Maine Children’s POG Operations Office POG Statistical Office Massachusetts General Hospital McGill University Medical College Virginia Miami Children’s Hospital Midwest Children’s Cancer Center Mount Sinai Medical School (NY) Naval Medical Center, SD Nemours Children’s Clinic Nemours/Orlando Oklahoma University Presbyterian Hospital Puerto Rico POG Rhode Island Hospital Roswell Park Cancer Institute

Grant number

CA-03161 CA-69177 CA-69177 CA-29691 CA-07431 CA-28439 CA-07431 CA-33625 CA-41573 CA-29293 CA-15525 CA-69177 CA-20549 CA-28476 CA-29691 CA-28476 CA-28439 CA-33603 CA-15898 CA-69177 CA-41573 CA-30969 CA-29139 CA-29293 CA-33587 CA-32053 CA-69428

CA-11233 CA-69177 CA-29293 CA-28383

Institution Rush-Presbyterian SPOG Basel SPOG Bern SPOG Geneva SPOG Lausanne SPOG Zurich SUNY Syracuse Sacred Heart Hospital San Antonio MPC & BDC San Jorge Children’s Hospital St Christopher’s Hospital St Francis Regional St Johns Hospital St Mary’s Hospital St Vincent Hospital Stanford University Tampa Children’s Hospital Tripler Army Medical Center University of Alabama University of Arizona University of Arkansas University of Florida University of Kansas University of Massachusetts University of Miami University of Mississippi Medical Center University of Missouri University of Rochester University of South Alabama University of South Florida University of Vermont University of Virginia UC/Davis UC/San Diego Southwestern Medical School UT/Galveston UT/San Antonio Wake Forest University School of Medicine West Virginia University, Charleston West Virginia University, Morgantown Walt Disney Memorial Cancer Institute Walter Reed Army Medical Warren Clinics Washington University Yale University

Grant number CA-07431

CA-29691 CA-33603 CA-25408 CA-33603

CA-69428 CA-15989 CA-05587

CA-29293 CA-28439 CA-33625 CA-03161 CA-15525 CA-15525 CA-11233 CA-05587 CA-69428

Leukemia