Survival Following Relapse in Children with Acute Myeloid Leukemia: A Report from AML-BFM and COG

Simple Summary Acute myeloid leukemia in children remains a difficult disease to cure despite intensive therapies that push the limits of tolerability. Though the intent of initial therapy should be the prevention of relapse, about 30% of all patients experience a relapse. Hence, relapse therapy remains critically important for survival. This retrospective analysis of two large international study groups (COG and BFM) was undertaken to describe the current survival, response rates and clinical features that predict outcomes. We demonstrate that children with relapsed AML may be cured with cytotoxic therapy followed by HSCT. High-risk features at initial diagnosis and early relapse remain prognostic for post-relapse survival. Current response criteria are not aligned with the standards of care for children, nor are the count recovery thresholds meaningful for prognosis in children with relapsed AML. Our data provide a new baseline for future treatment planning and will allow an updated stratification in upcoming studies. Abstract Post-relapse therapy remains critical for survival in children with acute myeloid leukemia (AML). We evaluated survival, response and prognostic variables following relapse in independent cooperative group studies conducted by COG and the population-based AML-BFM study group. BFM included 197 patients who relapsed after closure of the last I-BFM relapse trial until 2017, while COG included 852 patients who relapsed on the last Phase 3 trials (AAML0531, AAML1031). Overall survival at 5 years (OS) was 42 ± 4% (BFM) and 35 ± 2% (COG). Initial high-risk features (BFM 32 ± 6%, COG 26 ± 4%) and short time to relapse (BFM 29 ± 4%, COG 25 ± 2%) predicted diminished survival. In the BFM dataset, there was no difference in OS for patients who had a complete remission with full hematopoietic recovery (CR) following post-relapse re-induction compared to those with partial neutrophil and platelet recovery (CRp and CRi) only (52 ± 7% vs. 63 ± 10%, p = 0.39). Among 90 patients alive at last follow-up, 87 had received a post-relapse hematopoietic stem cell transplant (HSCT). OS for patients with post-relapse HSCT was 54 ± 4%. In conclusion, initial high-risk features and early relapse remain prognostic. Response assessment with full hematopoietic recovery following initial relapse therapy does not predict survival. These data indicate the need for post-relapse risk stratification in future studies of relapse therapies.


Introduction
The prognosis of children with acute myeloid leukemia (AML) has improved in recent decades, with current overall survival rates of approximately 70% [1][2][3][4][5][6]. Most international study groups currently utilize treatment regimens that include four to five courses of intensive myelosuppressive chemotherapy or intensive chemotherapy followed by hematopoietic stem cell transplantation (HSCT) for high-risk (HR) patients [7,8]. Despite intensive regimens that push the limits of tolerance, relapse rates as high as 30% have improved little over the past twenty years [7]. Post-relapse therapy remains critical for survival in childhood AML [5,[9][10][11].
In this retrospective analysis of children with AML in first relapse, we report postrelapse response, survival estimates and prognostic variables from the BFM registry and recent COG Phase III trials. These data represent the largest available relapse AML datasets analyzed for post-relapse survival.

Patients
Datasets included patients treated in the United States, Canada, Australia, New Zealand, Germany, Austria, Czech Republic and Switzerland.
Included are children with documented first relapse between April 2009 and December 2017 (n = 197). The primary analysis is limited to those patients who experienced a relapse after closure of the last I-BFM Relapsed trial Acute Myeloid Leukemia 2001/01. However, the previously described 513 patients enrolled before 04/2009 on Relapsed Acute Myeloid Leukemia 2001/01 [9] are included for comparison. COG relapse cohort. All patients enrolled on COG AAML0531 (NCT00372593 [22]) and AAML1031 (NCT01371981 [6]) who relapsed are included (n = 852: AAML0531 n = 358, and AAML1031 n = 494). Eligibility, therapy and results for these trials have been previously reported [6,22]. Only patients with DS-AML enrolled on AAML0531 are excluded from this post-relapse analysis. Residual disease (RD) was evaluated by central flow cytometry as previously described [23,24] at the end of one cycle of initial induction therapy in 765 patients (90%). Disease characteristics are well-characterized at diagnosis, but post-relapse data collection in the COG cohort is limited to survival.
National ethics committees and institutional review boards approved all studies and patients or guardians provided written informed consent. The retrospective analysis and all included studies were performed in accordance with the Declaration of Helsinki.

Definitions and Cohorts
Five-year estimate of the probability of post-relapse overall survival (pOS) was defined as time from date of first relapse to date of last follow-up or death from any cause. The 5-year estimate of event-free survival (pEFS) was defined as time from diagnosis at first relapse to the next event (second relapse, death of any cause, failure to achieve second remission or secondary malignancy) or date of last follow-up. Failure to achieve second remission was considered an event on day 0. Additional definitions are listed in Table S1.
A detailed response evaluation is included for a subset of uniformly treated BFM patients. The remission status data after first re-induction were derived from a bone marrow aspiration performed prior to a second re-induction. Second re-induction commenced at the discretion of the treating investigator and did not require hematopoietic recovery [25,26].

Statistical Analyses
Statistical analyses were performed with SAS (SAS Institute version 9.4, Cary, NC, USA). Median follow-up after diagnosis of first relapse was 4.2 years (0.3-10.3 years) in the AML-BFM cohort and 4.6 years (0-11.1 years) for the COG cohort. The Kaplan-Meier method was applied to estimate probabilities of survival. EFS and OS were compared with the log-rank test. Cumulative incidence functions of early death or relapse were constructed according to Kalbfleisch and Prentice. The Cox proportional hazards model was used for multivariate analysis of outcomes. We selected risk factors that have been significant in univariate analysis in one or both study groups for the multivariate analysis: risk group at initial diagnosis (inv(16)(p13.1q22), t(8;21)(q22;q22.1), high-risk), nonresponse at initial disease, time to relapse, date of relapse diagnosis and age at relapse. Proportions were compared between groups using the Chi-square or Fisher's exact test. p values < 0.05 were considered significant. Living patients were censored at date of last follow-up. Data were frozen at 03/27/2020 (BFM) and 03/31/2020 (COG).

Patient Characteristics and Previous Treatment
BFM cohort-Using a combined cytomolecular and response-guided risk stratification at initial disease (Table S1), 68 patients (38%) were classified as high-risk, 81 (45%) were intermediate and 31 (17%) standard risk (unclassified, n = 17). Forty-six percent of patients experienced a first relapse within one year of diagnosis. Fourteen percent of the patients (n = 28) had an HSCT prior to relapse. More details are shown in Table S2.
COG cohort-All relapsed patients were reclassified using the AAML1031 risk stratification (Table S1). Across the two studies, 608 patients (72%) were classified as low-risk at initial diagnosis and 237 (28%) high-risk (no data, n = 7, 1%). Among the high-risk patients, 194 (82%) had residual disease detected by flow cytometry at the end of one cycle of initial induction chemotherapy. Early relapse (within 1 year of diagnosis) occurred in 500 patients (59%), while 352 patients relapsed more than 1 year from diagnosis (41%). One hundred seventeen patients (14%) received an HSCT prior to relapse, 670 (77%) had no prior HSCT and 65 (8%) have insufficient data. Additional patient characteristics are shown in Table S3.

Post-Relapse Therapy
BFM cohort-After closure of the last relapse trial in 2009 [9], BFM guidelines recommended fludarabine, cytarabine, liposomal daunorubicin (DNX-FLA) followed by FLA and HSCT. Subsequently, 81% (n = 156) of all patients received the treatment with DNX-FLA(G) with or without second FLA. Fourteen percent of the patients had an alternate second cycle, primarily due to nonresponse in relapse and often including Gemtuzumab Ozogamicin (GO) or Clofarabine. Four percent of the patients received palliative care only (Table 1). Most patients who received an intensive re-induction treatment proceeded to transplant (81% of all patients and 87% of patients receiving DNX-FLA). Ten percent (n = 15) of the transplanted patients had a prior HSCT (Table 1). Among the 90 patients alive at last follow-up, 87 had a post-relapse HSCT, two are unknown and one patient alive did not receive an HSCT. Of note, the rate of HSCT following relapse increased over time. Sixty-nine percent of the patients included in the first time period of the previous relapse trial AML 2001/01 were able to proceed to HSCT, while 82% percent of the patients had an HSCT in the recent time interval (p(chi) = 0.0286; Table S4). Table 1. Treatment and Response (BFM cohort).

First Relapse Treatment of Pediatric AML n = 197
Chemotherapy
The survival of the 157 patients who received an HSCT following relapse was 52 ± 4% ( Figure S3A). Only one long-term surviving patient is alive without a post-relapse HSCT. This patient received Sorafenib and donor lymphocyte infusions (DLIs).
Among 157 patients who underwent HSCT following first relapse, 15 patients had a prior HSCT in first remission (93% for HR disease at initial diagnosis). The pOS was 28 ± 13% (n = 15) vs. 55 ± 4% in patients with first HSCT (n = 142; p(KM test) = 0.06; Figure S3B). When limiting the same analysis to just the HR patients, the pOS was 31 ± 14% for patients receiving a second transplant at relapse (n = 14) compared to a pOS 47 ± 9% for patients receiving their first transplant following relapse (n = 34, p(KM test) = 0.75; Figure S3C).
The 5-year pOS for the entire COG cohort is 35 ± 2%. For patients relapsing after treatment on AAML0531, the 5-year pOS is 33 ± 3% and for AAML1031 the 5-year pOS is 37% ± 2% ( Figure 1C). For patients relapsing between 2013 and 2017, 5-year pOS was 40% ± 3% ( Figure 1D Time to relapse. Survival of BFM patients experiencing early relapse within one year from initial diagnosis was significantly reduced (pOS 29 ± 5%, n = 91 vs. 55 ± 5%, n = 106; p < 0.0001; Figure 2A). Patients with a time to relapse that was less than 6 months showed a comparable outcome to patients relapsing within 6-12 months (n = 19, pOS 37 ± 11% vs. n = 72, pOS 27 ± 5%; p = 0.55; data not shown). Although the treatment year was not significant in multivariable analysis, there is a trend towards improvement in survival that is limited to patients with a late relapse ( Figure S4A,B). In the COG cohort, the 5-year pOS was 25 ± 2% for patients relapsing within one year from initial diagnosis (n = 352) and 51 ± 3% for patients relapsing later (n = 500, p < 0.001; Figure 2B).
Response to initial induction. Response to induction therapy was evaluated differently for COG and BFM data sets; however, poor response in both groups predicts poor survival. Nonresponse to therapy for initial disease (≥10% blasts after first or ≥5% after second induction) in the BFM cohort translated into a dismal prognosis after relapse compared to those who responded well at initial diagnosis (pOS 0 ± 0%, n = 12 vs. pOS 45 ± 4%, n = 185; p = 0.031; Figure 2E and Table S5). In the COG cohort, 765 of the 852 (90%) were evaluated for residual disease (RD) by central flow cytometry at the end of one cycle of initial induction therapy. In total, 222 (29%) had RD. The 5-year pOS following relapse for patients who were RD-positive at the end of initial induction was 24 ± 3% (n = 222) and 41 ± 2% for those who were RD-negative (n = 543, p < 0.001) ( Figure 2F).
Response and event-free survival after first relapse (BFM cohort). Within the BFM dataset, age, gender, white blood cell count or initial de novo treatment protocols were similar between the DNX-FLA group (n = 156). The remaining group had a higher proportion of patients with high-risk disease, nonresponse to initial therapy and early relapse (Table S2), several of whom received only palliative therapy at relapse.

Discussion
Survival data from pediatric patients treated within the AML-BFM protocols between 2004 and 2017 and COG Phase 3 trials between 2006 and 2018 were analyzed retrospectively. The 5-year pOS for 197 BFM relapse patients was 42 ± 4% and the 852 COG patients 35 ± 2%. Among the 156 BFM patients who received DNX-FLA following relapse, the 5-year pOS was 44%. When compared to an analysis of patients treated on the AML 2001/01 trial (5-year-pOS of 34%) [9], there is a trend towards improved survival over time, as well as increasing rates of post-relapse HSCT. This increase of patients proceeding to HSCT may account for improvements in survival. Since re-induction therapy has not improved, it is most likely that more children are receiving HSCT at relapse because of improved supportive care and donor availability.
Limitations within this retrospective review include the non-compulsory treatment schedule in the BFM dataset and the missing post-relapse treatment and response data in the COG dataset. Nonetheless, the results of our large retrospective analysis have important implications for future treatment planning. Previously published prognostic factors following a first relapse include time to relapse, treatment response at relapse, initial cytogenetics and HSCT in CR1 [9][10][11][12][13][14][17][18][19][20]. In the current study, multivariable analysis identified relapse within a year of diagnosis in both datasets and initial high-risk disease classification in the COG datasets are predictive of poor overall survival. Importantly, high-risk assignment in the COG cohort includes patients with detectable RD by flow cytometry at the end of initial induction therapy.
In the BFM registry, 156 patients (80%) received DNX-FLA at the discretion of the treating investigator. This homogeneously treated population permitted a retrospective analysis of response. Nonresponse rates after relapse re-induction therapy (21%) were higher at relapse than nonresponse rates reported in de novo AML studies (range of 3-18%) [1][2][3][4][5][6]27,28]. The outcome of children with no response after relapse (≥5% leukemic blasts after second cycle of re-induction) was poor but significant (5-year pOS of 27%), suggesting that HSCT has a role post-relapse even in the absence of a complete remission. Sixty-one percent of patients achieved a CR, CRp or CRi after up to two cycles or reinduction chemotherapy with DNX-FLA +/− FLA. Traditionally, the International Working Group (IWG) criteria for response require an absolute neutrophil count of 1 × 10 9 /L and platelets of 100 × 10 9 /L for a CR. These criteria imply that the absence of count recovery is prognostic. While retrospective studies in adult patients with de novo AML have indicated that outcomes are superior for patients with response and full hematologic recovery compared to those with Cri/CRp [29,30], such a claim has never been validated in children with AML. Within the BFM cohort, absence of full hematopoietic recovery following standard relapse re-induction therapy does not predict survival in children at relapse ( Figure 3B). This is comparable to what is observed in a recent study with adult patients following a first relapse or refractory disease. Overall survival was similar in patients who achieved CR with full hematologic recovery vs. those with incomplete hematologic recovery [31].
The Cheson criteria, introduced in 1990 to assess response in adult de novo AML, should be reconsidered as the standard for response evaluation for children with AML [25,26,30], and perhaps adults with AML as well (as reviewed in Bloomfield et al. [32]). In the AML-BFM and COG studies, it is common to continue intensification of treatment without waiting for full hematopoietic recovery [6,8,22]. Our data again confirm that clinicians prioritize maintaining therapy intensification rather than waiting for a hematopoietic regeneration. Continuation of treatment without count recovery has produced favorable outcomes in North American and European trials for children with newly diagnosed AML [1,33]. The IWG/Cheson criteria are likely to underestimate response in children as they are not aligned with the standards of care for children, nor are the count recovery thresholds to achieve a CR meaningful in children with relapsed AML. It will be important in future studies of novel relapse therapies for children that the defined response criteria reflect standard treatment strategies specifically used for children.

Conclusions
The findings of our large retrospective analysis have important implications for future trials. Within this international cohort we confirmed that initial risk stratification and time to relapse are prognostic for post-relapse survival. We also declare, for the first time in children, that lack of a full hematopoietic regeneration to the thresholds required by the IWG response criteria is neither necessary for, nor predictive of, survival at first relapse in children. AML in children remains a difficult disease to cure despite intensive therapies that push the limits of tolerability. Though the intent of any initial therapy should be the prevention of relapse, nearly half of all relapse patients will still survive. Refinements in post-relapse care continue to show benefit in overall survival across cooperative groups. As we continue to evaluate innovative therapies to improve survival without adding cumulative short-and long-term toxicities, it is important to reevaluate poor risk features that predict survival at relapse and appropriate definitions for response in children.