The age confound: a partner arm and an age arm are nearly the same arm
also called KMT2A partner-age collinearity · partner as a proxy for age
Which partner gene a leukemia has depends heavily on how old the patient is, and age is one of the strongest predictors of survival. So a difference between partners can easily be an age difference wearing a partner's name.
The same thing, for a scientist
KMT2A partner distribution is strongly age-dependent, and age is independently one of the most powerful prognostic variables in this disease. In 465 KMT2A-rearranged B-ALL cases spanning 1 month to 89 years, partner and age were correlated at chi-square P<0.0001: KMT2A::AFF1 was 71% of adult/elderly cases versus 48% of infant/paediatric (P=7.7E-12), KMT2A::MLLT3 occurred exclusively in infants and children, and KMT2A::MLLT1 and KMT2A::MLLT10 were 61% and 92% paediatric versus 8% and 0.6% adult (P=0.008, P=1.5E-4). Independently, a 497-patient pooled analysis of childhood 11q23 ALL found age the single most important prognostic factor (HR for EFS 1.84, 95% CI 1.38-2.47, P=0.0001 for infants versus those aged 1 year or older) and reported that partner distinctions emerged only in the older patients, with any 11q23 abnormality uniformly dismal in infants. With that collinearity, an unadjusted partner comparison in a mixed-age cohort is uninterpretable, and a single-age-band comparison cannot detect a partner effect that operates through age at all.
What would separate them
Three designs, in increasing cost:
- Adjusted model in a cross-age cohort. Done once, for MRD clearance rather than survival: partner remained independent of age and maturity (drop-one P=0.03) 1. The survival version has never been run.
- Within-age-band multi-partner comparison with adequate n. The Ponte-di-Legno cohort is the closest, and holds age above 1 year constant across 629 patients 3.
- Off-modal cells. Adults with MLLT3 and infants with a rare partner are the cells that break the collinearity. They exist but are rare, and no cohort has deliberately over-recruited them.
The specific trap for this graph
KMT2A::AFF1 dominates infants in absolute numbers and also dominates adults as a proportion 1. A naive reading - AFF1 is the infant partner - is wrong in the second direction and produces a systematic error whenever infant results are generalised to AFF1 as a fusion. AFF1's age distribution is bimodal relative to the other partners, not infant-skewed.
Age also determines what the comparator is
Infant protocols, paediatric protocols and adult protocols differ in intensity, transplant policy and treatment-related mortality. Whole-cohort EFS runs roughly 46% in Interfant-06 infants 4, 69.1% in non-infant KMT2A-r ALL children 3, and 32-42% 3-year EFS/OS in adult KMT2A-r AML. So age carries a treatment confound as well as a biological one, and the two cannot be separated from published summary data.
| confound | age |
|---|---|
| collinearity p | <0.0001 (chi-square, partner versus age group, n=305) |
| status | partially separated for MRD clearance (partner independent of age, P=0.03); never separated for survival |
| source | PMIDs 41743263, 12057554, 36256911, 31283407 |
partner by age
{
"AFF1_adult_pct": 71,
"AFF1_infant_paediatric_pct": 48,
"MLLT3": "exclusively infant/paediatric",
"MLLT1_paediatric_pct": 61,
"MLLT1_adult_pct": 8,
"MLLT10_paediatric_pct": 92,
"MLLT10_adult_pct": 0.6
}age effect size
{
"hr_efs_infant_vs_older": 1.84,
"ci": "1.38-2.47",
"p": 0.0001,
"n": 497
}separating designs
[ "adjusted model in a cross-age cohort with a survival endpoint", "within-age-band multi-partner comparison with adequate per-arm n", "deliberate over-recruitment of off-modal cells (adult MLLT3, infant rare partners)" ]
- A menin-inhibitor trial that reports response by partner without reporting age within partner will produce numbers that look partner-stratified and are age-stratified, which is worse than reporting nothing because it invites a mechanistic reading.
- If partner is largely a proxy for age, then the class abstraction survives the outcome data but for an uninteresting reason, and the real question becomes whether menin dependence varies with cell of origin.
- Does fusion partner remain an independent predictor of EFS, not just MRD clearance, in a cross-age adjusted model? That single analysis would resolve most of this beat.
- How many adults with KMT2A::MLLT3 and how many infants with KMT2A::ELL exist in assembled datasets? Those cells set the ceiling on how well the confound can ever be broken observationally.
- Is developmental maturity a mediator of the partner effect rather than a confounder, in which case adjusting for it removes real partner signal?
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
KMT2A partner and patient age are strongly collinear (chi-square P<0.0001 in 465 B-ALL cases): AFF1 71% of adult/elderly versus 48% infant/paediatric (P=7.7E-12), MLLT3 exclusively infant/paediatric, MLLT1 61% versus 8% and MLLT10 92% versus 0.6% paediatric versus adult (P=0.008, P=1.5E-4); nonetheless fusion partner remained an independent determinant of MRD clearance after adjustment for age group and developmental maturity (drop-one P=0.03).
Intra-subtype heterogeneity shapes treatment response in KMT2A-rearranged ALL across all age groups. · Hartmann AM, Bastian L, et al. · Hemasphere 2026What this evidence cannot show
B-ALL only; the adjusted result is for MRD clearance, not survival, so it does not establish that partner survives adjustment for age as a predictor of cure. Denominators vary between analyses (n=305 for the age-fusion correlation, n=321 for partner counts, n=97 for MRD clearance categories). Assembled retrospective cohort, treatment not uniform.
Evidence that argues against this (1)
- A 497-patient pooled analysis concluded that age was the most important prognostic factor and that within infants any 11q23 abnormality conferred an equally dismal outcome, which is the reading in which partner adds nothing once age is known. Pui CH et al. 2002
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
In 497 children and young adults with 11q23-rearranged ALL, age was the most important prognostic factor (infants under 1 year versus 1 year and older, HR for EFS 1.84, 95% CI 1.38-2.47, P=0.0001), with any category of 11q23 abnormality dismal in infants while t(4;11) and t(9;11) were worse than other 11q23 changes only in older patients.
Outcome of treatment in childhood acute lymphoblastic leukaemia with rearrangements of the 11q23 chromosomal region. · Pui CH, Gaynon PS, et al. · Lancet 2002What this evidence cannot show
Pooled retrospective data from 11 groups and single institutions, patients treated 1983-1995 on protocols now superseded; partner assignment largely cytogenetic. The infant-versus-older contrast is itself a subgroup comparison.
Evidence that argues against this (1)
- In children over 1 year treated 1995-2010, a 629-patient cohort found 5-year EFS spanning 41.7% to 91.2% across partner-by-lineage cells, which is a far wider partner spread than this older analysis reported for older patients and suggests the partner effect is not merely residual after age. Attarbaschi A et al. 2023
- E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Holding age constant above 1 year, 629 children with KMT2A-r ALL still show large partner-associated variation, with whole-cohort 5-year EFS 69.1% and reported cells from 41.7% to 91.2%.
Outcomes of Childhood Noninfant Acute Lymphoblastic Leukemia With 11q23/KMT2A Rearrangements in a Modern Therapy Era: A Retrospective International Study. · Attarbaschi A, Möricke A, et al. · J Clin Oncol 2023What this evidence cannot show
Age is held constant only in the coarse sense of excluding infants; within 1-18 years age still varies and was not adjusted for in the reported cells. The extreme cells have n=9 and n=34, and are partner-by-lineage rather than partner alone.
Evidence that argues against this (1)
- In infants, two cohorts that tested partner directly found no difference at all, so the partner spread does not extend across the whole paediatric age range and remains compatible with an age-dependent partner effect. Hilden JM et al. 2006
- E4established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
Whole-trial outcomes differ sharply by age band in this disease: Interfant-06 reported 6-year EFS 46.1% (SE 2.1) and OS 58.2% (SE 2.0) in infants under 1 year.
Outcome of Infants Younger Than 1 Year With Acute Lymphoblastic Leukemia Treated With the Interfant-06 Protocol: Results From an International Phase III Randomized Study. · Pieters R, De Lorenzo P, et al. · J Clin Oncol 2019What this evidence cannot show
Whole-trial figure includes KMT2A-germline infants, who do markedly better, so it is not a like-for-like comparator for a KMT2A-r cohort. Different endpoint horizon (6-year) from the 5-year figures it is being compared with.
Evidence that argues against this (1)
- JPLSG MLL-10 reported 3-year EFS of 66.2% for KMT2A-rearranged infants, roughly 20 points above the Interfant-06 whole-trial figure, showing that age band alone does not fix outcome and that protocol and cohort composition matter as much. Tomizawa D et al. 2020
quantifies the partner-age collinearity and partly separates it
an infant-only cohort cannot detect a partner effect that operates through age
holding age above 1 year constant, partner variation persists
- related to →KMT2A::AFF1 (MLL-AF4), t(4;11)(q21;q23)
AFF1 is 71% of adult versus 48% of infant/paediatric B-ALL, not an infant-only partner