Cytarabine dose reduction in patients with low-risk acute myeloid leukemia: A report from the Children’s Oncology Group

Abstract

Background:

The optimal number of chemotherapy courses for low-risk (LR) pediatric acute myeloid leukemia (AML) is not known.

Objective:

To compare outcomes for four (21.6 g/m² cytarabine) versus five (45.6 g/m² cytarabine) chemotherapy courses for LR-AML using data from Children’s Oncology Group (COG) AAML0531 and AAML1031.

Methods:

We compared relapse risk (RR), disease-free survival (DFS), and overall survival (OS), and the differential impact in LR subgroups for patients receiving four versus five chemotherapy courses. Cox (OS and DFS) and risk (RR) regressions were used to estimate hazard ratios (HR) to compare outcomes.

Results:

A total of 923 LR-AML patients were included; 21% received five courses. Overall, LR-AML patients who received four courses had higher RR (40.9% vs. 31.4%; HR = 1.40, 95% confidence interval [CI]: 1.06–1.85), and worse DFS (56.0% vs. 67.0%; HR = 1.45, 95% CI: 1.10–1.91). There was a similar decrement in OS though it was not statistically significant (77.0% vs. 83.5%; HR = 1.45, 95% CI: 0.97–2.17). Stratified analyses revealed the detrimental effects of cytarabine dose de-escalation to be most pronounced in the LR-AML subgroup with uninformative cytogenetic/molecular features who were minimal residual disease (MRD) negative after the first induction course (EOI1). The absolute decrease in DFS with four courses for patients with favorable cytogenetic/molecular features and positive MRD was similar to that observed for patients with uninformative cytogenetic/molecular features and negative MRD at EOI1, though not statistically significant.

Conclusions:

Our results support de-escalation of cytarabine exposure through the elimination of a fifth chemotherapy course only for LR-AML patients who have both favorable cytogenetic/molecular features and negative MRD after the first induction cycle.

1 |. INTRODUCTION

Intensification of cytarabine- and anthracycline-based chemotherapy regimens, advancing supportive care, and improved risk stratification have increased overall survival (OS) from less than 20% to 65% in children with acute myeloid leukemia (AML) over the last several decades.^1–5 The total number of chemotherapy courses varies between cooperative groups, but cumulative doses range from approximately 33 to 47 g/m² of cytarabine and 300 to 450 mg/m² of anthracyclines for patients treated with chemotherapy alone.^6–11

The Medical Research Council (MRC) AML12 trial previously compared five versus four courses where, depending on risk stratification and the availability of a matched related donor, the final course was either high-dose cytarabine or an allogeneic stem cell transplantation (SCT).^12,13 In children, the cumulative cytarabine exposure on AML12 was either 10.6 g/m² (four courses) or 34.6 g/m² (five courses) and no improvement in event-free survival or OS was observed with a fifth course.

Based on the MRC AML12 results, the Children’s Oncology Group (COG) modified their standard treatment regime on the AAML1031 trial by eliminating the fifth chemotherapy course (high-dose cytarabine/L-asparaginase) from the standard backbone utilized in the prior trial, AAML0531.^2,14 This modification reduced the cumulative cytarabine exposure from 45.6 g/m² (five courses) to 21.6 g/m² (four courses) for patients with low-risk (LR) disease, defined as those with either favorable molecular/cytogenetic markers, or uninformative molecular/cytogenetic markers but negative minimal residual disease (MRD) at the end of induction I (EOI1).

While such a dose reduction would ideally be tested in a randomized clinical trial (RCT), an adequately powered RCT comparison is not feasible in children particularly given the frequency of LR subgroups. Therefore, COG prospectively planned to evaluate cytarabine de-escalation in a combined analysis of data from AAML1031 (four courses) and AAML0531, which used the historic five-course backbone. These analyses were intended to identify potential differential effects in LR subpopulations as a step toward more personalized therapy for children with AML.

2 |. METHODS

2.1 |. Study population

The study population included patients with LR AML treated on either AAML1031 or AAML0531 who survived in first remission without relapse to the start of the fourth chemotherapy course. For these analyses, LR was defined per COG AAML1031.¹⁴ Specifically, patients were considered to have LR disease if they met either of the following conditions: (a) favorable cytogenetic/molecular features (i.e., inv(16)/t(16;16), t(8;21), NPM1, CEBPα) regardless of high-risk disease features (i.e., monosomy 7, monosomy 5, del5q) or MRD status at EOI1; or (b) uninformative cytogenetic/molecular features (i.e., absence of all of the aforementioned favorable or high-risk cytogenetic/molecular features), but negative MRD (i.e., MRD <0.1%) at EOI1. High allelic ratio FLT3/ITD+ patients were excluded from analyses. Information on the cytogenetic/molecular features required for the LR classification was captured consistently on both trials. Patients randomized to receive gemtuzumab ozogamicin (GO) on AAML0531 were excluded to mitigate confounding ² In contrast, LR patients treated on AAML1031 were included in the study population irrespective of randomization (i.e., standard chemotherapy ± bortezomib) because outcomes were comparable between treatment arms.¹⁴

2.2 |. Exposure

On AAML1031, LR patients received a four-course chemotherapy backbone including two induction courses of cytarabine/daunorubicin/etoposide and two consolidation courses: cytarabine/etoposide, followed by cytarabine/mitoxantrone. On AAML0531, LR patients were scheduled to receive a five-course chemotherapy backbone that included the same four courses as AAML1031 plus an additional fifth course comprised of high-dose cytarabine/L-asparaginase (Table S1). Four courses equated to approximately one-half of the cumulative cytarabine exposure of five courses (i.e., 21.6 vs. 45.6 g/m²), with no difference in the anthracycline exposure.

In the primary analyses, patients were analyzed according to the number of chemotherapy courses received and the start of the fifth course was assessed as a time-varying exposure to avoid immortal and immune person-time bias.¹⁵ Although five courses were planned per-protocol on AAML0531, there were a subset of AAML0531-enrolled patients who completed frontline treatment after four courses of chemotherapy. A secondary intent-to-treat analysis in which patients were analyzed according to the number of planned chemotherapy courses based on their given treatment protocol (AAML1031: four courses, cumulative cytarabine dose = 21.6 g/m²; AAML0531: five courses, cumulative cytarabine dose = 45.6 g/m²) was also performed. This analysis aimed to assess the sensitivity of the primary results to possibility that the subset of AAML0531 patients who received only four courses of chemotherapy may not have proceeded to a fifth chemotherapy course because of a less favorable prognosis.

2.3 |. Outcomes

Outcomes of interest included relapse risk (RR), disease-free survival (DFS), OS, and treatment-related mortality (TRM). RR was defined as time to relapse, where deaths without a prior relapse were considered competing events. DFS was defined as time to relapse, secondary malignancy or death, whereas OS was defined as time to death. TRM was defined as time to death as first event during on-protocol therapy or within 30 days of completing protocol therapy, where relapses prior to death were considered as competing events. The initiation of the fourth chemotherapy course marked the start of follow-up for all outcomes. Due to the temporal differences between the AAML0531 (2006–2010) and AAML1031 (2011–2016), the total amount of follow-up time available for the compared groups differed considerably. In an effort to avoid potential bias due to differential follow-up given that the risk for the outcomes of interest likely changes over time, the end of the observation period for these analyses was set to a specific duration of time over which there was more comparable data available between groups. Specifically, the end of follow-up was considered as 1100 days from the beginning of course 4 (i.e., the median follow-up time among AAML1031 patients who were alive at last contact) and patients who did not experience the given outcome by this time in follow-up were censored.

2.4 |. Covariates

Patient information including date of birth, sex, race, Hispanic or Latino ethnicity, EOI1 MRD status, presence of specific favorable and unfavorable cytogenetic/molecular features, duration of the fourth course, as well as times to absolute neutrophil and platelet count recovery in course 4 were obtained from the AAML0531 and AAML1031 study databases.

2.5 |. Statistical analyses

Patient characteristics were summarized by the number of chemotherapy courses received (four or five) for the full study population and separately for the LR subpopulations. Categorical characteristics were compared using chi-square test or, in the event of sparse data, Fisher’s exact test. Medians for continuous characteristics were compared using the Mann-Whitney test.

Cox regression models estimated hazard ratios (HR) and corresponding 95% confidence intervals (CI) comparing OS and DFS between four relative to five chemotherapy courses. Competing risk regression models estimated HR (95% CI) for comparisons of RR. As Kaplan-Meier methods do not accommodate time-varying exposures, the Cox model-predicted survival probabilities are presented and were used to plot the OS and DFS curves for the primary as-treated analyses. For the secondary intent-to-treat analyses, Kaplan-Meier methods were utilized to obtain the estimated RR, DFS, and OS. Stratum-specific analyses were performed to evaluate potential heterogeneity of treatment effects by LR classification conditions. A descriptive assessment of four versus five courses among patients with favorable cytogenetics and negative EOI1 MRD with non-central nervous system extramedullary disease (EMD) was also performed. To confirm robustness of results to any unexpected confounding by bortezomib exposure, sensitivity analyses were performed excluding patients who were treated on the experimental arm of AAML1031. All reported p-values are two-sided.

3 |. RESULTS

3.1 |. Study population and patient characteristics

A total of 923 pediatric patients with LR AML who were treated on either AAML0531 (n = 225) or AAML1031 (n = 698) met all inclusion criteria for the current analyses. Twenty-one percent (n = 191, all treated on AAML0531) received five courses of frontline chemotherapy, while the remaining 79% (n = 732, 698 treated on AAML1031 + 34 treated on AAML0531) received four courses. Distributions of demographic and clinical characteristics were comparable between four and five courses, both overall (Table 1) and within strata of LR classification condition (Table S2), suggesting little evidence of confounding; therefore, unadjusted comparisons are presented.

TABLE 1.

Characteristics of the low-risk AML study population, overall and by number of treatment courses received

		Frontline chemotherapy courses received
Characteristic	Total	Four courses	Five courses	p-Value
Total, N	923	732	191
Sex
Male	489 (53.0)	384 (52.5)	105 (55.0)	.535
Female	434 (47.0)	348 (47.5)	86 (45.0)
Age at diagnosis, years
<2	179 (19.4)	139 (19.0)	40 (20.9)	.543
2–10	311 (33.7)	244 (33.3)	67 (35.1)	.650
11–15	253 (27.4)	196 (26.8)	57 (29.8)	.397
16–20	164 (17.8)	137 (18.7)	27 (14.1)	.140
≥21	16 (1.7)	16 (2.2)	0 (0)	.054
Median (range)	9.9 (0.0–29.5)	10.2 (0.0–29.5)	9.00 (0.09–19.1)	.075
Race
White	658 (80.1)	519 (79.2)	139 (83.2)	.249
Black or African American	102 (12.4)	85 (13.0)	17 (10.2)	.328
Asian	46 (5.6)	38 (5.8)	8 (4.8)	.612
American Indian or Alaskan Native	9 (1.1)	7 (1.1)	2 (1.2)	1.000
Native Hawaiian or other Pacific Islander	7 (0.9)	6 (0.9)	1 (0.6)	1.000
Ethnicity
Hispanic or Latino	170 (18.9)	132 (18.5)	38 (20.5)	.531
Not Hispanic or Latino	728 (81.1)	581 (81.5)	147 (79.5)
BMI at start of course 4
Median (range)	18.7 (11.7–50.5)	18.7 (11.7–50.5)	18.7 (12.7–40.4)	.963
Number analyzeda	744 (80.6)	585 (79.9)	159 (83.2)
MRD at end of induction I
Negative (MRD <0.1%)	808 (92.6)	665 (93.3)	143 (89.4)	.090
Positive (MRD ≥0.1%)	65 (7.5)	48 (6.7)	17 (10.6)
Favorable cytogenetic/molecular risk features
None	397 (44.0)	312 (43.8)	85 (44.5)	.866
Any	506 (56.0)	400 (56.2)	106 (55.5)
t(8;21)
Yes	199 (22.0)	156 (21.9)	43 (22.5)	.851
No	705 (78.0)	557 (78.1)	148 (77.5)
inv(16)/t(16;16)
Yes	146 (16.2)	110 (15.4)	36 (18.9)	.254
No	758 (83.9)	603 (84.6)	155 (81.2)
NPM
Yes	91 (9.9)	74 (10.1)	17 (8.9)	.614
No	831 (90.1)	657 (89.9)	174 (91.1)
CEBPA
Yes	71 (7.7)	60 (8.2)	11 (5.8)	.257
No	850 (92.3)	670 (91.8)	180 (94.2)
Unfavorable cytogenetic/molecular risk features
Monosomy 7
Yes	1 (0.1)	1 (0.1)	0 (0)	NE
No	904 (99.9)	713 (99.9)	191 (100)
Monosomy 5/Del5q
Yes	0 (0)	0 (0)	0 (0)	NE
No	905 (100)	714 (100)	191 (100)
Course 4 duration in days, median (range)	49 (4–260)	49 (4–260)	46 (28–121)	.183
Hematologic recovery in course 4
Days to ANC recovery, median (range)	37 (9–157)	38 (9–157)	37 (15–92)	.740
Days to platelet recovery, median (range)	37 (17–124)	37 (17–124)	38 (18–105)	.580
Recovered both ANC and platelets, n	548	417	131

Note: Data presented as n (%) unless otherwise noted. Unknown values are not tabulated and were excluded from the denominator of computed percentages.

NE = comparison not tested.

Abbreviations: AML, acute myeloid leukemia; MRD, minimal residual disease;

^aPatients younger than 24 months or with discrepant height or weight data were excluded.

3.2 |. Comparison of treatment outcomes

Among the full study population, DFS was significantly lower (56.0% vs. 67.0%; HR = 1.45, 95% CI: 1.10–1.91; p = .009) with a parallel increase in RR (40.9% vs. 31.4%; HR = 1.40, 95% CI: 1.06–1.85; p = .019) following four relative to five chemotherapy courses (Table 2 and Figure 1). A similar relative difference of four versus five courses was observed for OS (77.0% vs. 83.5%, HR = 1.45, 95% CI: 0.97–2.17; p = .068), but the difference was not statistically significant. TRM was low during both course 4 (2.4%) and course 5 (1.9%), but qualitatively higher in the fourth course (HR = 1.32, 95% CI: 0.31–5.62; p = .709).

TABLE 2.

Comparison of outcomes by number of frontline chemotherapy courses received, overall and stratified by low-risk subpopulations

		Disease-free survival			Overall survival
Patient group	n	%a	HR (95% CI)	p-Value	%a	HR (95% CI)	p-Value
All low-risk patients
4 Courses	732	56.0	1.45 (1.10–1.91)	.009	77.0	1.45 (0.97–2.17)	.068
5 Courses	191	67.0	1 (reference)		83.5	1 (reference)
Low-risk patients with uninformative cytogenetic/molecular features but negative MRD at EOI1
4 Courses	326	42	1.94 (1.26–2.98)	.002	64.7	1.60 (0.94–2.72)	.082
5 Courses	70	63.9	1 (reference)		76.2	1 (reference)
Low-risk patients with favorable cytogenetics/molecular features - overall
4 Courses	400	68.0	1.11 (0.74–1.65)	.62	87.5	1.04 (0.53–2.02)	.912
5 Courses	106	70.5	1 (reference)		87.8	1 (reference)
Low-risk patients with favorable cytogenetics/molecular features and positive MRD at EOI1
4 Courses	48	50.7	1.59 (0.64–3.92)	.318	83.3	2.56 (0.31–21.0)	.382
5 Courses	17	65.1	1 (reference)		93.0	1 (reference)
Low-risk patients with favorable cytogenetics/molecular features and negative MRD at EOI1
4 Courses	339	70.6	1.02 (0.63–1.66)	.934	89.1	0.96 (0.42–2.20)	.928
5 Courses	73	71.0	1 (reference)		88.6	1 (reference)

Abbreviations: EOI1, end of induction I; MRD, minimal residual disease.

^aEstimate at 1100 days from start of the fourth course of chemotherapy: negative MRD = MRD <0.1%; positive MRD = MRD ≥0.1%.

FIGURE 1.

Predicted disease-free (A) and overall survival (B) following four versus five courses of frontline chemotherapy in patients with low-risk acute myeloid leukemia (AML), full study population

Stratified analyses by LR subgroup (Table 2; Figures 2 and 3) revealed that the unfavorable effects of reducing the cumulative cytarabine exposure were most pronounced in the subpopulation of LR patients with uninformative cytogenetic/molecular features who were MRD negative at EOI1. In contrast, there were no statistically significant differences in outcomes for four versus five courses among patients with favorable cytogenetic/molecular features overall. However, further stratification of the latter group by EOI1 MRD status (Table 2 and Figure 4) demonstrated an absolute increase in DFS with five courses for patients with favorable risk features and positive MRD (14.4%) that was similar to that observed among patients with uninformative cytogenetic/molecular features and negative MRD at EOI1 (21.9%); the difference was not statistically significant. Exploratory analyses also suggested a potential benefit of a fifth course among patients with favorable cytogenetic/molecular features and negative MRD at EOI1 with EMD (Table S3); however, effect estimates were imprecise and not statistically significant.

FIGURE 2.

Predicted disease-free (A) and overall survival (B) following four versus five courses of frontline chemotherapy in patients with low-risk acute myeloid leukemia (AML), restricted to patients with favorable cytogenetic/molecular features

FIGURE 3.

Predicted disease-free (A) and overall survival (B) following four versus five courses of frontline chemotherapy in patients with low-risk acute myeloid leukemia (AML), restricted to patients with uninformative cytogenetics who were minimal residual disease (MRD) negative at end of induction I only

FIGURE 4.

Predicted disease-free (DFS) and overall survival (OS) following four versus five courses of frontline chemotherapy in acute myeloid leukemia (AML) patients with favorable cytogenetic or molecular features, stratified by minimal residual disease (MRD) status at end of induction I. DFS (A) and OS (B) curves for patients with MRD <0.1% at end of induction I; DFS (C) and OS (D) curves for patients with MRD ≥0.1% at end of induction I

Sensitivity analyses excluding patients treated with bortezomib on AAML1031 were consistent with the primary analyses (Tables S4 and S5). Intent-to-treat analyses also produced results that were comparable to those observed in primary as-treated analyses comparing four versus five courses (Table S6).

4 |. DISCUSSION

The optimal number of chemotherapy courses for pediatric AML patients treated with chemotherapy alone is not well understood. Our results demonstrate that the benefit of the fifth course of high-dose cytarabine in the COG backbone varies by disease characteristics and early treatment response, presenting a potential opportunity for more individualized therapy. Specifically, LR patients with both favorable cytogenetic/molecular features and negative MRD at EOI1 had comparable outcomes with four or five courses of chemotherapy. The fifth course of high-dose cytarabine has a 1.9% TRM rate, requires on average 35 additional days of hospitalization, and increases average inpatient costs by $93,000.¹⁶ Given these toxicity risks in the absence of discernable improvement in treatment outcomes, cytarabine deintensification through elimination of the fifth course is warranted.

In contrast, clinically meaningful and statistically significant improvements in DFS and OS with a fifth course of high-dose cytarabine among patients with negative EOI1 MRD but uninformative risk markers justifies the additional chemotherapy exposure in this subpopulation. The small number of patients with favorable cytogenetic/molecular features but positive EOI1 MRD (n = 65) resulted in limited statistical power. However, the observed DFS improvement with the fifth high-dose cytarabine course is comparable in magnitude to patients with uninformative cytogenetics/molecular features and negative MRD at EOI1. Assuming the observed estimate of DFS improvement is accurate, then treatment of approximately six such patients would prevent one relapse¹⁷; thus, cytarabine intensification may be similarly justified for this subpopulation. Exploratory analyses among the small subpopulation with favorable risk markers and negative MRD at EOI1 who presented with EMD (n = 48) also had limited statistical power, but suggest a possible benefit of a fifth course.

Our data contrast the reported pediatric experience on MRC AML12.^12,13 Several differences in study design and analytic approach may explain this discrepancy. While the AAML0531/AAML1031 analyses are restricted to LR risk de novo pediatric AML patients, the pediatric MRC AML12 comparison included acute promyelocytic leukemia (10%), high-risk AML (10%), and patients with either ALL or myelodysplastic syndrome (6%). Second, the MRC AML12 comparisons collapsed SCT with high-dose cytarabine as the fifth course. In addition, one-third of patients eligible for the consolidation randomization elected not to participate in the randomization. Thus, the final sample size of 135 patients per randomized arm may be subject to selection biases and has more limited statistical power than the AAML0531/AAML1031 study population.

Notably, our results are generally consistent with the recently reported MRC AML17 results that demonstrated that in adults increasing the cumulative exposures of cytarabine from 21.6 to 39.6 g/m² significantly reduced RR but did not improve OS due to effective salvage therapy following relapse.^18,19 Although not statistically significant, the benefit of the intensified consolidation in MRC AML17 was qualitatively more pronounced in patients with CBF-AML. While the overall results from the MRC AML17 analyses are consistent with the AAML0531/AAML1031 data, the subgroup analyses are not consistent. A combined, patient-level analysis of pooled MRC AML17 and AAML0531/AAML1031 data may help elucidate reasons for this discrepancy.

Our analysis has limitations, primarily the nonrandomized nature of the reported comparisons. Concerns with historical control data are three-fold: potential selection bias, misclassification, and confounding.^20,21 In our analyses, selection bias was minimized, as patients treated on either trial were required to meet identical eligibility criteria. Misclassification bias was mitigated through the harmonization of patient-level data included in the pooled analysis, specifically the use of a common LR disease classification and equivalent outcome characterization. While our findings may be susceptible to residual confounding, the similarities in the demographics and important clinical characteristics between patients treated with four or five courses of chemotherapy provides reassurance that the likelihood of such potential bias is modest. Still, unmeasured factors such as supportive care practices may have differed between the two consecutive trials. However, given the temporal improvements in supportive care, and that four courses was the standard on the more recent protocol, such an imbalance would likely have diminished the observed difference between four versus five courses. Despite a focus on the difference in cumulative cytarabine exposure between four and five courses, the fifth course included both high-dose cytarabine and asparaginase. Therefore, among the subpopulations who derived a benefit from the fifth course, we cannot rule out a potential contribution of asparaginase in addition to cytarabine.^22–24 Last, while the overall study population was large, the sample sizes for some of the stratified analyses were small, which may have precluded our ability to detect a true association. Careful consideration of these limitations, the limitations of prior MRC studies, as well as the toxicities of a fifth cycle of high-dose cytarabine and salvage allogeneic donor SCT are necessary for the clinical application of these study results.

In conclusion, data from AAML0531 and AAML1031 support de-escalation of frontline cytarabine exposure from 45.6 to 21.6 g/m² through the elimination of a fifth chemotherapy course for LR AML patients with favorable cytogenetic/molecular features who are MRD negative at EOI1. In contrast, AML patients with uninformative cytogenetics who are MRD negative at EOI1, and those with favorable cytogenetics who are MRD positive at EOI1 may benefit from a fifth cytarabine-containing chemotherapy course. Based on these data, COG revised the standard chemotherapy backbone of LR patients in the recently initiated Phase III clinical trial AAML1831. Given the extensive use of the COG protocols in the United States²⁵ and abroad, these changes will likely become the standard of care for many pediatric patients with AML.

Abbreviations:

AML

acute myeloid leukemia

CI

confidence interval

COG

Children’s Oncology Group

DFS

disease-free survival

EMD

extramedullary disease

EOI1

end of induction I

HR

hazard ratio

LR

low risk

MRC

Medical Research Council

MRD

minimal residual disease

OS

overall survival

RCT

randomized clinical trial

RR

relapse risk

SCT

stem cell transplantation

TRM

treatment-related mortality

DATA AVAILABILITY STATEMENT

Research data are not shared.