Oncofusion

start / graph / target-newborn-screening-fusion-transcripts

Targetstrong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.Q1b-genesis2

Newborn screening for leukaemia fusions: verdict NOT FEASIBLE TODAY, and the reason is arithmetic plus the absence of anything to offer a positive

also called population screening for KMT2A fusion · Guthrie card screening · early detection of infant leukaemia

In principle you could test every newborn's blood spot for the leukaemia fusion. In practice the disease is so rare that almost every positive result would be wrong, the test would have to be near-perfect to avoid that, and there is nothing useful to offer the families who tested positive. It fails on all three counts at once.

The same thing, for a scientist

Screening is the most emotionally compelling of the three intervention routes and the least defensible on current evidence. It fails three independent tests. **Prevalence**: KMT2A-rearranged infant leukaemia occurs in roughly 1 in 30,000-40,000 births (`epi-fusion-positive-newborns-versus-leukemia-risk`), and the prevalence of fusion-positive newborns has been measured once only, at n=65 and negative (PMID 9737688), which leaves a 95% upper bound near 4.5% and a programme still largely blind to its own positive predictive value. **Specificity arithmetic**: at a true-positive rate of ~33 per million, a test with 99.9% specificity generates about 1,000 false positives per million, giving a positive predictive value near 3%; even 99.99% specificity gives only about 25%. To reach a PPV above 75% the assay needs specificity better than 99.999%, which is a demanding target for a technology whose known failure modes include trans-spliced and read-through chimeras that produce the same transcript from an unrearranged genome [E4][E5]. **Assay reality**: RT-PCR screening of newborn samples has already been attempted at cohort scale for a different fusion and ran into RNA degradation requiring preparation within 1-2 days of birth, a constraint incompatible with dried blood spots [E2]; the one large, stringent cord-blood screen found no confirmable positives at all [E3]. **And there is nothing to offer.** No intervention is known to prevent progression in a fusion-positive newborn, and the one documented natural history of such a clone is spontaneous clearance [E1].

Go deeper
CORRECTION: the verdict survives, two of its four reasons do not

This entry rejected newborn screening on four grounds. A counter-evidence pass found that two of them fail, and the coordinator verified both sources.

1. The prevalence had been measured. This entry said it never had. Yamamoto 1998 screened 65 normal cord bloods and found no MLL/AF-4 chimeric mRNA (PMID 9737688). Small n and a weak bound, but a measurement.

2. The PPV arithmetic is an argument the screening field has already overruled in practice. Sweden's national SCID programme screens for a condition of comparable rarity — estimated incidence 1:38,500, the same order as KMT2A-rearranged infant leukaemia — at 99.94% specificity and a PPV of 4.11%, and it was adopted as national policy (PMID 34449549). A few-percent PPV is not by itself disqualifying.

3. The assay objection is an RNA objection and dissolves at the DNA level. The cited degradation constraint applies to RNA; the Swedish programme runs qPCR on DNA eluted from routine dried blood spots at a scale of 115,786 newborns in a single year. A DNA breakpoint assay is not subject to the objection this entry raised.

What actually disqualifies it, and it is sufficient on its own: there is no intervention to offer a fusion-positive, disease-free newborn, and the one documented natural history of such a clone is spontaneous clearance. That is Wilson–Jungner criterion 3, and screening which cannot change an outcome converts a rare catastrophe into a common anxiety.

This verdict has an explicit reopening condition. If a menin inhibitor were ever shown to prevent progression in a fusion-positive carrier, criterion 3 would be met and this entry would need rewriting within a year — and the DNA-breakpoint-on-dried-blood-spot assay is the technical development that would make a programme possible. The entry is kept as a conditional rejection, not a permanent one.

The Wilson-Jungner criteria, applied honestly1 deeper
CriterionStatus for KMT2A fusion screening
Important health problemYes — the outcome is catastrophic even though the disease is rare
Recognisable latent stageYes, proven by blood-spot backtracking (mechanism-in-utero-origin-neonatal-blood-spot-backtracking)
Accepted treatment for those detectedNo. There is no intervention for a fusion-positive, disease-free newborn
Suitable testNo. RNA degrades in dried spots 2; genomic assays are not validated for population scale
Test acceptable to the populationUncertain, and see the ethics section below
Natural history understoodNo. The penetrance is unmeasured
Agreed policy on whom to treatNo
Cost-effectiveCannot be assessed without penetrance

Six of eight criteria fail. This is not a close call.

The crucial asymmetry, stated precisely

For ETV6::RUNX1 the asymmetry is real in direction: low twin concordance and latencies of a decade require that many more carriers exist than cases. The concordance figure has been revised downward and this section previously used the old one. The often-quoted ~5% comes from referral series; a population-based assessment of twins in California reports 0.8% raw concordance for childhood acute leukaemia and 2.6% among estimated monozygotic pairs, attributing the higher literature figures to ascertainment bias (PMID 39815817, Haematologica 2025). A lower concordance makes the asymmetry LARGER, so the argument here is strengthened rather than weakened. The often-quoted hundred-fold figure, however, comes from one study 6 that a larger and more stringent one failed to reproduce 3. For KMT2A the asymmetry may be much smaller or absent, because twin concordance is near-total and latency is months. The implication is counterintuitive: KMT2A screening would have a much better positive predictive value than ETV6::RUNX1 screening if a carrier could be reliably detected — and would still fail, because the prevalence is fifty times lower and there is no intervention.

The ethics of telling parents

A positive result would tell a family that their well newborn carries a lesion associated with an aggressive leukaemia, with no preventive action available, an unknown probability of progression, and a monitoring plan that would itself involve repeated marrow aspirates in an infant. The false-positive families receive the same anxiety with none of the underlying risk, and at a PPV of 3% they are the overwhelming majority. Newborn screening programmes have repeatedly learned this lesson in other conditions; the specific hazard here is that the disease is rare enough, and frightening enough, that the harm falls almost entirely on families who were never at risk. The one published natural history of a detected carrier is a clone that disappeared on its own 1, which means that even a true positive might be a child who needed nothing.

What a defensible version would look like

Not population screening. A targeted, consented, research screen in a defined enriched population, with pre-specified management that is surveillance rather than treatment: the healthy monochorionic co-twins of index infant KMT2A-r cases, where the prior probability is high enough that a positive is likely real and the family is already engaged with the diagnosis. That cohort is tiny, which is exactly why it is ethically and statistically tractable, and it would produce the first prospective penetrance and clearance data in humans.

Facts and figures
modalitypopulation newborn screening
verdictNOT FEASIBLE TODAY
assumptiontrue-positive rate taken as the KMT2A-r infant leukaemia cumulative risk of ~33 per million births, i.e. assuming near-full penetrance; if penetrance is lower the true-positive rate for DISEASE falls and the PPV for disease falls with it
defensible alternativetargeted consented research screening of healthy monochorionic co-twins of index cases, with surveillance-only management
ppv arithmetic
[
  {
    "specificity_pct": 99.9,
    "false_positives_per_million": 1000,
    "true_positives_per_million": 33,
    "ppv_pct": 3.2
  },
  {
    "specificity_pct": 99.99,
    "false_positives_per_million": 100,
    "true_positives_per_million": 33,
    "ppv_pct": 24.8
  },
  {
    "specificity_pct": 99.999,
    "false_positives_per_million": 10,
    "true_positives_per_million": 33,
    "ppv_pct": 76.7
  }
]
blocking failures
[
  "no intervention to offer a positive",
  "penetrance unmeasured",
  "no validated population-scale assay",
  "RNA degrades in dried spots"
]
How this could bite us back
  • At a PPV of a few percent, the overwhelming majority of the burden of a screening programme falls on families who were never at risk.
  • A screen with no intervention to offer converts a rare catastrophe into a common anxiety, and there is no way to withdraw information once given.
  • If screening were introduced and a menin inhibitor were then used pre-emptively in screen-detected carriers, the programme would be treating mostly-healthy infants with a drug that causes differentiation syndrome, on the basis of an unmeasured penetrance.
Still open
  1. Would a genomic (DNA-level, targeted-capture) assay on a dried blood spot solve the degradation and specificity problems at once? It avoids the trans-splicing and read-through false-positive classes entirely, because those are RNA phenomena. Nobody has built or costed one for population scale, and it is the single technical development that would reopen this question.
  2. What is the actual PPV? It cannot be computed until cord-blood KMT2A prevalence is measured. Everything in this entry is conditional on that missing number.
  3. If a carrier were detected, is there any surveillance schedule that improves outcome? Earlier diagnosis of infant KMT2A-r ALL has never been shown to improve survival, and the disease is often already high-count at presentation, so even perfect detection might not help.
  4. What do parents of affected children actually want here? This beat found no published preference study, and the ethics argument above is made entirely from the clinician's side.
Every source behind this entry (6)
  1. E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    A KMT2A-fusion-positive clone detected in a healthy infant cleared spontaneously, so a true-positive screen result does not imply a need for intervention.

    PMID 19411627DOI
    What this evidence cannot show

    A single case; cannot establish how often clearance happens.

    Evidence that argues against this (1)
    • Ascertainment cuts the other way too. Greaves' review of twin leukaemia records concordant monozygotic infant pairs as the recurrent, long-recognised observation (described since 1882) and explains them by transplacental spread of one initiated clone to the co-twin, with clonotypic genomic fusion sequences confirming common clonal origin. The published literature is therefore dominated by pairs in which BOTH twins developed disease, so treating the single documented clearance case as the natural history of a screen-detected carrier is an over-read in one direction, just as citing only concordant pairs would be an over-read in the other. Neither body of case reports has a denominator, and that is the point: penetrance is unmeasured, not low. Read: abstract only. Greaves MF et al. 2003
  2. E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    Large-scale RT-PCR screening of newborn cohort samples for a leukaemia fusion requires viable cells or RNA prepared within 1-2 days of birth, a constraint incompatible with archived dried blood spots.

    RT-PCR screening for ETV6-RUNX1-positive clones in cord blood from newborns in the Danish National Birth Cohort · Olsen M, Hjalgrim H, Melbye M, Madsen HO, Schmiegelow K · Journal of Pediatric Hematology/Oncology 2012
    PMID 22217495DOI
    What this evidence cannot show

    A feasibility study for one fusion using RNA; a DNA-based genomic-breakpoint assay would not have the same degradation problem, but no such population-scale assay has been validated.

    Evidence that argues against this (1)
    • This is the strongest single counter in this entry, and it is numerical. Sweden's national newborn screening programme for SCID runs a multiplex qPCR (TREC/KREC/ACTB) on DNA eluted from routine dried blood spots. In its first year it processed 115,786 newborns, estimated SCID incidence at 1:38,500 -- the same order as the ~1 in 30,000-40,000 this entry assigns KMT2A-rearranged infant leukaemia -- achieved specificity 99.94%, referred 73 children, and reported a positive predictive value of 4.11% for SCID. It was adopted as national policy at that PPV. Two of this entry's blocking failures therefore do not survive: a nucleic-acid assay on dried blood spots IS validated at population scale (the degradation objection applies to RNA, not DNA, and a genomic-breakpoint assay is DNA), and a PPV of a few percent is NOT by itself disqualifying for a newborn screening programme. What remains disqualifying is the absence of anything to offer a positive -- SCID screening is defensible because transplant before infection changes survival. The entry's verdict survives; its stated reasoning does not, and the ranking of the three failures should be rewritten so that 'nothing to offer' carries the whole argument. Read: abstract only. Gongrich C et al. 2021
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    A stringent screen of 1,417 cord bloods with sensitive RT-PCR plus flow-sorted re-testing produced no confirmed fusion-positive samples, showing how hard it is to make such an assay specific.

    Prevalence of t(12;21)[ETV6-RUNX1]-positive cells in healthy neonates · Lausten-Thomsen U, Madsen HO, Vestergaard TR, Hjalgrim H, Nersting J, Schmiegelow K · Blood 2011
    PMID 20713965DOI
    What this evidence cannot show

    Nine of 1,417 samples were positive on first-run PCR and hybridisation before all failed confirmation, which is precisely the false-positive rate problem a screening programme would face, at scale.

    Evidence that argues against this (1)
    • The specificity bar this entry sets may be higher than real programmes clear. Across the five largest prospective TREC-screening cohorts in this systematic review, PPV for SCID was 0.8-11.2% (18.3-81.0% for the broader T-cell lymphopenia outcome) with 100% sensitivity for typical SCID in case series, over 3.15 million newborns screened. Screening programmes for conditions of this rarity are run, and recommended, at positive predictive values in single digits, with the false positives absorbed by a confirmatory second tier rather than treated as programme-ending. Read: abstract only. van der Spek J et al. 2015
  4. E4moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    Cancer-associated fusion transcripts can be produced in normal cells by trans-splicing without any genomic rearrangement, creating an irreducible false-positive class for RNA-based screening.

    PMID 18772439DOI
    What this evidence cannot show

    A published erratum exists (PMID 26430125) that this beat did not read; the general point is independently supported by E5 and by the read-through literature.

    Evidence that argues against this (1)
    • The trans-splicing false-positive class has been looked for at this specific junction and not found. Yamamoto et al. applied nested PCR to 65 normal cord blood samples and detected no MLL/AF-4 chimeric mRNA in any of them. What they did find, in 20 of 65 (31%), was an in-frame fusion of AF-4 (AFF1) exon 11 to exon 4 or 5 -- an aberrant transcript internal to the partner gene, present in healthy fetal blood and also in 4 of 6 t(4;11) leukaemic samples, and absent from 21 peripheral blood samples and from fetal fibroblasts. So the confounder for a KMT2A::AFF1 RNA assay is real but sits on the AFF1 side rather than in a genuine trans-spliced chimera, which changes which primer design is dangerous. Read: abstract only. Yamamoto S et al. 1998
  5. E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    In childhood B-precursor ALL, 229 of 245 RNA-seq fusion calls were normal conjoined genes and only 6.1% of those were leukaemia-exclusive, quantifying the false-positive burden of transcript-based fusion detection in this exact disease setting.

    Conjoined Genes as Common Events in Childhood Acute Lymphoblastic Leukemia · Severgnini M, D'Angio M, Bungaro S, Cazzaniga G, Cifola I, Fazio G · Cancers (Basel) 2022
    PMID 35884588DOI
    What this evidence cannot show

    Applies to untargeted RNA-seq fusion calling rather than to a targeted KMT2A assay, which would be far more specific. Cited to establish that the false-positive class is real and large in paediatric ALL transcriptomes, not to put a number on a hypothetical targeted screen.

    Evidence that argues against this (1)
    • Moving from RNA to a targeted DNA assay does not rescue a screening programme; it converts a specificity problem into a positive-predictive-value problem. DNA-based GIPFEL screening of healthy newborns found an ETV6-RUNX1 fusion in five percent of them - true rearrangements, not conjoined-gene artefacts, in children who overwhelmingly will never develop leukaemia. So the false-positive burden this claim quantifies for transcript calling is replaced, in the assay the entry proposes as more specific, by a very large number of true-positive, non-progressing clones. The screening arithmetic has to survive both. (Record and title fetched and checked; full text not open access.) Schafer D et al. 2018
  6. E6weak confidence 2 of 5: Preliminary, small, indirect, or not yet peer reviewed.paper

    The widely cited claim that leukaemia fusions are present in newborns at about a hundred times the disease rate rests on a single cord-blood study.

    Chromosome translocations and covert leukemic clones are generated during normal fetal development · Mori H, Colman SM, Xiao Z, Ford AM, Healy LE, Donaldson C, Hows JM, Navarrete C, Greaves M · Proceedings of the National Academy of Sciences of the USA 2002
    PMID 12048236DOI
    What this evidence cannot show

    Not reproduced (see E3) and did not test any KMT2A fusion.

    Evidence that argues against this (1)
    • The premise that cord-blood KMT2A fusion prevalence has never been measured is wrong, and the measurement that exists is a zero. Yamamoto et al. screened 65 normal cord bloods by nested PCR and found no MLL/AF-4 chimeric mRNA. With n=65 and no events the rule of three puts the 95% upper bound near 4.5%, which is a weak bound and nowhere near the ~1 in 30,000 birth incidence, so this does not settle penetrance -- but it does mean the correct statement is 'measured once, at small n, and negative', not 'never measured'. The PPV arithmetic in this entry's data block should be read as still conditional on a real prevalence estimate, and the open question about the missing number stands. Read: abstract only. Yamamoto S et al. 1998
Connected to