Oncofusion

start / graph / epi-ibfm-childhood-kmt2a-aml-partner-risk-groups

Epidemiologystrong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.Q19-partner-outcome

I-BFM childhood KMT2A-r AML (n=1,130): the 3-prime partner defines two risk groups that differ by 24 EFS points

also called I-BFM-SG KMT2A-r AML fusion partner risk groups · van Weelderen 2023 JCO

In the largest study of children with KMT2A-rearranged myeloid leukemia, which partner gene KMT2A fused to split patients into two groups whose survival differed by about a quarter. Five specific partners marked a high-risk group, and that grouping is now written into how children are treated.

The same thing, for a scientist

The International Berlin-Frankfurt-Muenster Study Group assembled 1,130 children with KMT2A-rearranged AML diagnosed 2005-2016 and pre-assigned them to a fusion-partner-based high-risk group (6q27/KMT2A::AFDN, 4q21/KMT2A::AFF1, 10p12/KMT2A::MLLT10, 10p11.2/KMT2A::ABI1, 19p13.3/KMT2A::MLLT1; n=402, 35.6%) or a non-high-risk group (all other identified partners; n=728, 64.4%), using a classification derived from earlier retrospective series. The high-risk group had 5-year EFS 30.3% (95% CI 25.9-35.4) versus 54.0%, CIR 59.7% versus 35.2%, OS 49.2% versus 70.5%, all P<.0001. Whole-cohort rates were EFS 45.3% (42.3-48.5), CIR 44.0% (40.8-47.2), OS 62.8% (59.8-65.8). In multivariable analysis including end-of-induction-2 flow-MRD, the high-risk partner group remained independently associated with inferior EFS (HR 2.0, 95% CI 1.5-2.5), CIR (HR 2.1, 1.5-2.8) and OS (HR 2.1, 1.5-2.9), all P<.0001. This is the strongest evidence in this graph that 3-prime partner identity is prognostic, and it has been adopted operationally: COG incorporated these translocations into the initial risk stratification of AAML1831 (NCT04293562).

Go deeper
Which partners are high risk

The high-risk assignment was made a priori from a previously published classification, not discovered in this cohort 1. It comprises 6q27 (KMT2A::AFDN), 4q21 (KMT2A::AFF1), 10p12 (KMT2A::MLLT10), 10p11.2 (KMT2A::ABI1) and 19p13.3 (KMT2A::MLLT1). Everything else identified - including 9p22 KMT2A::MLLT3 and 19p13.1 KMT2A::ELL - went to the non-high-risk group. A KMT2A-other group of 126 patients (10.0% of the 1,256 eligible) was excluded because their partners could not be risk-assigned, and those excluded patients were younger and had higher WBC than the rest, so the analysed cohort is not the whole disease 1.

Why this matters for the menin-interface bet

Menin binds the KMT2A N-terminus, which every one of these fusions retains. If the retained N-terminus were the whole story, outcome should not depend on what is joined to it. Here it does, by 24 EFS points, independently of early response 1. Note the direction: KMT2A::AFF1 sits in the HIGH-risk group in myeloid disease, which inverts the intuition carried over from infant B-ALL where AFF1 is simply the commonest partner and is often treated as the reference case.

What this entry does not claim

This is a partner-group comparison, not a partner-by-partner one. The paper reports two arms, not five or ten curves, so it cannot say whether AFDN and MLLT1 behave alike inside the high-risk arm. It is retrospective across many national protocols over 12 years 1, so protocol and era are not controlled - see concept-partner-protocol-and-era-confound-kmt2a-outcome. And the grouping originates in 2, so the two are not independent confirmations of each other.

Facts and figures
metric5-year event-free survival by KMT2A fusion-partner risk group
value30.3
unit% (high-risk partner group)
comparator value54
comparator unit% (non-high-risk partner group)
populationchildren with KMT2A-rearranged AML diagnosed 2005-2016
n total1130
n high risk402
n non high risk728
regioninternational (I-BFM Study Group)
year2023
prespecifiedpartner risk grouping pre-assigned from a prior published classification; the cohort analysis itself is retrospective
adopted in trialNCT04293562 (COG AAML1831)
sourcePMID 36996387
whole cohort 5y
cir_pct44.0cir_pct: 44efs_pct45.3efs_pct: 45.3os_pct62.8os_pct: 62.8
whole cohort 5y — the numbers behind the chart
{
  "efs_pct": 45.3,
  "cir_pct": 44,
  "os_pct": 62.8
}
high risk partners
[
  "KMT2A::AFDN (6q27)",
  "KMT2A::AFF1 (4q21)",
  "KMT2A::MLLT10 (10p12)",
  "KMT2A::ABI1 (10p11.2)",
  "KMT2A::MLLT1 (19p13.3)"
]
How this could bite us back
  • If partner group predicts outcome independently of MRD, a menin inhibitor given as class therapy to all KMT2A-r children will be judged against a background risk differing by 24 EFS points between arms; an unstratified trial can be diluted or flattered by partner mix alone.
  • Adopting partner-based stratification, as AAML1831 has, means a future menin-inhibitor trial randomised within strata will have small per-partner arms, so the class claim will be hardest to test exactly where it matters.
  • The excluded KMT2A-other group (10% of eligible, younger, higher WBC) is precisely the population a rare-partner menin trial would enrol, and it has no outcome baseline at all.
Still open
  1. Do the five high-risk partners behave alike inside the high-risk arm, or is the group average hiding one very bad partner and four ordinary ones?
  2. Does the split survive within a single uniformly treated protocol, e.g. AAML1831, where era and protocol are held constant?
  3. Is partner risk group still prognostic once menin inhibition is part of frontline therapy, or does a class-level drug collapse the difference?
Every source behind this entry (2)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    In 1,130 children with KMT2A-rearranged AML, a fusion-partner-based high-risk group (AFDN, AFF1, MLLT10, ABI1, MLLT1; n=402) had 5-year EFS 30.3% versus 54.0% for the non-high-risk group (n=728), CIR 59.7% versus 35.2% and OS 49.2% versus 70.5% (all P<.0001), and partner group remained independent of flow-MRD in multivariable analysis (EFS HR 2.0, 95% CI 1.5-2.5, P<.0001).

    PMID 36996387DOI
    What this evidence cannot show

    Retrospective pooling across many study groups and national protocols, 2005-2016, so treatment is not uniform. The partner risk classification was imported from earlier retrospective series rather than derived or validated prospectively here. 126 patients (10.0%) with unassignable partners were excluded and differed from those retained (younger, higher WBC), so the estimate applies only to assignable partners. Flow-MRD was available for 456/1,130 patients, so the multivariable model runs on 40% of the cohort.

    Evidence that argues against this (2)
    • In adults the same comparison is null: an MD Anderson series of 172 adults with KMT2A-r AML found on multivariable analysis that translocation subtype did not predict survival, and concluded risk guidelines should treat all KMT2A-r as adverse regardless of partner. Issa GC et al. 2021
    • The HARMONY adult AML database (205 KMT2A-r patients) likewise found OS similar across translocations including t(9;11)/KMT2A::MLLT3 (P=0.756), with age, secondary AML, DNMT3A and KRAS mutations carrying the prognostic weight instead. Hernández-Sánchez A et al. 2024
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    The partner-based risk classification applied here originates in an international retrospective study of 756 children with 11q23/MLL-rearranged AML in which 5-year EFS ranged from 11%+/-5 to 92%+/-5 across translocation subgroups.

    PMID 19528532DOI
    What this evidence cannot show

    The source study is retrospective across 11 collaborative groups, and the extreme subgroups driving the 11%-92% range are small. It is the same lineage and an overlapping era, so E1 is a partly dependent re-test, not an independent replication.

    Evidence that argues against this (1)
    • The same source study explicitly could not confirm the favourable prognosis of t(9;11)/KMT2A::MLLT3 that had been reported from single-institution data, showing that partner-specific prognostic claims in this disease have not replicated reliably. Rubnitz JE et al. 2002
Connected to