Core‐binding factor acute myeloid leukemia with t(8;21): Risk factors and a novel scoring system (I‐CBFit)

1. INTRODUCTION

Acute myeloid leukemia (AML) with rearrangements involving genes encoding subunits of core‐binding factor (CBF), a group of DNA‐binding transcription factor complexes composed of α and β subunits, shares similar pathogenesis and clinical features and is considered as a distinct subset in AML.1, 2, 3, 4 Translocation(8;21)(q22;q22) and inv(16)(p13q22), the most frequent cytogenetic abnormalities occurring in CBF‐AML, lead to the creation of the fusion genes RUNX1/RUNXT1 and CBFB/MYH11 that disrupt, respectively, the α and β subunits of CBF, dysregulate hematopoiesis, and thus contribute to leukemogenesis.5

Although the prognosis of CBF‐AML is better than other subtypes of AML, approximately 30%‐40% of the patients still relapse and may require allogeneic hematopoietic cell transplantation (HCT).6, 7, 8 A scoring system to predict who has a higher risk of relapse at the time of diagnosis may be clinically valuable to guide decision‐making. There have been only a few studies attempting to develop a scoring system for poor outcomes of CBF‐AML (eg, relapse and disease‐free survival [DFS]).6, 8 The relative rarity of CBF‐AML (approximately 15%‐20% of AML cases) in adults9 and its relatively good prognosis may have limited these efforts. A useful prognostic system requires a large sample size and long follow‐up time including all treatment data. This is challenging, even for large registries or cooperative groups. For example, the Center for International Blood and Marrow Transplant Research (CIBMTR) only has data of patients with CBF‐AML receiving a HCT, while US cooperative groups may have too few patients with a long follow‐up to examine outcomes after HCT. Moreover, recent studies clearly indicate that AMLs with t(8;21) (q22;q22) and AMLs with inv(16) (p13q22) are two different diseases regarding patient and disease characteristics.2, 6, 8, 10, 11, 12, 13, 14 Each cytogenetic subgroup therefore should be evaluated separately to develop a specific prognostic scoring system.

In this multicenter study, we created an extensive database including US and European centers for CBF‐AML patients with t(8;21) (q22;q22), and developed and validated a significant risk scoring system with high predictive probabilities.

2. METHODS

Eleven centers in the US and Europe collaborated to collect data on 550 CBF‐AML patients. Two‐hundred and forty‐seven of these patients had t(8;21)(q22;q22) and are the subject of this report. Inclusion criteria were as follows: (a) AML patients with t(8;21)(q22;q22) or RUNX1‐RUNX1T1 confirmed by the reporting institutions; (b) cases diagnosed between July 1996 and January 2017. Data were uniformly collected by completing a predesigned data spreadsheet. The data form included the following: patient characteristics (age, sex, race); disease characteristics (date of diagnosis, white blood cell count [WBC] at diagnosis [×10⁹/L], cytogenetics, KIT D816V mutational status, primary or secondary AML); therapy characteristics (induction regimens and their number, consolidation regimens, and number of cycles); HCT (autologous or allogeneic, donor type, remission status at HCT); and events (relapse, death, or alive at last contact). Patients’ data were anonymously transferred to University of Minnesota where the main database was created and managed. This study was approved by the Institutional Review Board Human Subjects Committee at the University of Minnesota.

3. RESULTS

The characteristics of the test and validation groups combined are provided in Table 1. Patients were mostly male, were Caucasian, and had a median age of 47 years, and 17.4% had secondary AML. Additional cytogenetic abnormalities were frequently observed (58.7%), and 44.5% of patients had a hypodiploid or hyperdiploid clone. KIT D816V mutation was present in 11.3% of patients (17.8% of the patients tested), and any KIT mutations were detected in 16.6% of patients (25.7% of the patients tested). There was no association between KIT mutation (positive, negative, nontested) and WBC (Figure S1).

Complete remission (CR) was achieved in the vast majority of patients (92.7) (Table 1). Relapse occurred in 67 patients (27.1%) at a median of 10.6 months (range 1‐65.5 months). AlloHCT was performed in 61 patients (24.7%): 31 with CR1 (12.5%), 19 with ≥CR2 (7.6%), and 10 (4.0%) with active leukemia (all relapsed after CR). AlloHCT in CR1 was performed at a median of 6 months (range 2 to 13.1 months) from diagnosis and 4 months (1.1‐12 months) from the date of CR1. The median follow‐up was 64 months (0.5 to 1378 months).

The risk factors and risk ratios from a logistic regression model are presented in Table 2. Older age, higher WBC at diagnosis, KIT D816V mutation, and a pseudodiploid karyotype were associated with higher risks of death or disease relapse. Race, sex, and primary vs. secondary AML had no impact.

The risk of death or relapse within 2 years associated with the covariates retained in the predictive risk score is shown in Table 2. The concordance statistic (a measure of the model fit, also called the area under curve (AUC), or area under the receiver operating characteristics (ROC) curve for the predictions) is 0.756 (Figure S2). The optimal risk score is found by computing the following linear score:

The full set of results of the validation study along with the sensitivity analysis results (the highest of the conditional probabilities was negative predictive value [NPV], 80%) are presented in Table S1. When I‐CBFit > 0, a patient is classed as being at high risk of death or relapse within 2 years. DFS rate at 2 years was 76% for patients with a low‐risk I‐CBFit score compared with 36% for those with a high‐risk I‐CBFit score (P < 0.0001). Low‐ vs high‐risk OS at 2 years was 89% vs 51%, P < 0.0001 (Figures 1 and 2).

DFS at 2 years was 80% for patients with I‐CBFit low risk not undergoing alloHCT in CR1, was 82% for patients with I‐CBFit low risk undergoing alloHCT in CR1, was 33% for patients with I‐CBFit high risk not undergoing alloHCT in CR1, and was 67% for patients with I‐CBFit high risk undergoing alloHCT in CR1, P = <0.0001 (Figure 3). OS at 2 years was 91% for patients with I‐CBFit low risk regardless of alloHCT in CR1, was 52% for patients with I‐CBFit high risk not undergoing alloHCT in CR1, and was 73% for patients with I‐CBFit high risk undergoing alloHCT in CR1, P < 0.0001 (Figure 4).

4. DISCUSSION

In this large study with a long follow‐up, we were able to create and validate the risk scoring system we are calling the “International CBF group index for t(8;21)” (I‐CBFit) in t(8;21) AML. We show that older age, higher WBC at diagnosis, and KIT D816V mutation were risk factors associated with treatment failure (relapse or death). In addition, we found that pseudodiploidy was also a risk factor in t(8;21), a novel finding. I‐CBFit had a high NPV (80%) and a modest specificity and accuracy for DFS, and the NPV was even higher for the prediction of OS.

Current treatment guidelines for CBF‐AML with t(8;21) do not recognize heterogeneity in these patients, and thus, all t(8;21) AML patients generally receive the same induction and consolidation treatments. This might be appropriate for patients with a low‐risk score who are predicted to have nearly an 80% chance of extended DFS. On the other hand, high‐risk score patients may benefit from more intensive approaches in CR1. Current guidelines do not identify patients needing alloHCT in CR1. This new model may clarify this uncertainty, especially identifying patients who do not require intensive consolidations (eg, alloHCT) in CR1 given its high NPV. Although patients receiving alloHCT in CR1 was limited, when we analyzed the impact of alloHCT it seemed that patients with an I‐CBFit low‐risk score had similar DFS and OS regardless of alloHCT.

KIT mutations have been reported in 15%‐46% of adults patients with t(8;21) CBF‐AML.13, 15, 16, 17, 18 KIT D816V mutations were reported in 4%‐28% and strongly associated with poorer DFS (6%‐48%).13, 16, 19, 20 In pediatric populations, KIT mutations clustered in exon 17 and exon 8 were identified in 20‐30% of the CBF‐AML patients,21, 22, 23 yet its effect on prognosis is not agreed upon.22, 24 A meta‐analysis indicated KIT mutation increased relapse risk (RR at 2 years 1.76 [95% CI: 1.45‐2.12]) and decreased OS 1.35 (95% CI: 1.09‐1.66).25

Chromosomal abnormalities secondary to t(8;21), mostly involving loss of a sex chromosome, ‐Y in men and ‐X in women, trisomy 8, and deletion of the long arm of chromosome 9 [del(9p)] are frequently reported.6, 8, 14, 18 In our patients, additional cytogenetic abnormalities were common, as in other reports.14, 18, 26 Sex chromosome loss was reported as favorable for two‐year event‐free survival (66.9% vs 43.0%, P = 0.031).18 In contrast, DFS was shorter for male patients with loss of the Y chromosome. In another study,8 loss of a sex chromosome was associated with increased CR rates in CBF‐AML.14 We found no particular chromosomal abnormality to be associated with poor outcome. However, consistent with findings of Krauth et al18 loss of a sex chromosome had a modestly favorable, but not significant effect on DFS. We also found that the chromosome number was important, with patients with pseudodiploid karyotypes having worse outcome compared with those with hypodiploidy or hyperdiploidy.

Higher WBCs were found to be associated with poorer outcomes.8 Schlenk et al8 described a scoring system using two factors, high WBC, and low platelet counts, to be prognostic. Low platelet count was also a poor prognostic factor in a CALGB/Alliance study.6 In our study, we did not find a correlation between KIT mutation and WBC.

An earlier CALGB/Alliance study showed that age was associated with shorter overall survival (OS).6 In a more recent CALGB/Alliance study,27 3‐year OS rate was 61% for adults younger than 60 years vs only 47% for those at least 60 years old. Appelbaum et al14 showed that age is associated with a shorter OS.

We were able to collect data over a two‐decade period and believe this long time period does not adversely impact the validity of the study as (a) the type and number of induction or consolidation therapies did not have an impact on outcomes and (b) the most widely used treatments (7 + 3 in induction phase and high‐dose cytarabine in consolidation phase) have not changed over this time. Although this is a retrospective study, we find the data robust and substantial given the lengthy time period of patient follow‐up. In fact, long‐term follow‐up allowed complete evaluation in this relatively good prognostic disease.

Another limitation is that molecular abnormalities, including mutations in the KIT and FLT3 genes, were not uniformly tested. As a result, information on KIT mutational status is missing in approximately one‐third of the patients. However, KIT mutation was associated with significantly decreased survival compared with KIT wild type, whereas outcomes of patients with the KIT mutational status untested fell between outcomes of patients with KIT mutations and those with wild‐type KIT; this might be expected given that some but not all untested patients would have mutated KIT. This strongly supports the adverse effect of a KIT mutation.

This new scoring system, I‐CBFit, uses known and novel risk factors to provide a binary prediction of the risk of death or relapse within 2 years. Importantly, all factors and thus the scoring system can easily be determined at diagnosis. Although its validation by other studies is needed, I‐CBFit can contribute to current treatment of patients with t(8;21) and tailor consolidation treatments for individual patients in the spirit of precision medicine to identify those who do not need intensified management including alloHCT during CR1.

CONFLICT OF INTEREST

The authors have no conflict of interest relevant to the study to disclose.

Supporting information

Ustun C, Morgan E, Moodie EEM, et al. Core‐binding factor acute myeloid leukemia with t(8;21): Risk factors and a novel scoring system (I‐CBFit). Cancer Med. 2018;7:4447–4455. 10.1002/cam4.1733