How common is ZNF384-rearranged leukaemia? Published answers span 1.6% to about 20% and the spread is almost entirely denominator, not biology

FULL TEXT READ via Europe PMC fullTextXML for PMC8579251. This is the only figure in this beat's reading quoted against a CONSECUTIVE population denominator, which is why it is the headline. Limits: one country, one decade, 12 events, Wilson 95% interval roughly 1.1-3.2% on 12/643. The denominator mixes BCP-ALL with B/myeloid MPAL. The hierarchical clustering that defined the B-other group was run on 110 of the 158 cases, with the ZNF384-aberrant cases from outside that window added, so the 10/158 and 12/158 figures combine cases ascertained by slightly different routes.

• Even this cohort's own B-other figure of 6% sits at the LOW end of the 5-10% range the same paper quotes from other cohorts, which shows that between-cohort variation in the B-other denominator exceeds the entire contribution of the newly described noncanonical class (6% to 8%). Anyone comparing a new cohort's B-other share against a published one is therefore comparing definitions of B-other more than biology. This is the strongest argument for quoting only the consecutive-population figure and treating every B-other percentage as uninterpretable across studies. Zaliova M et al. 2021

The 4.1% and 2.4% figures are in the published ABSTRACT and are solid; the partner counts were extracted by a page-fetch summarisation of the PMC rendering (Europe PMC and NCBI efetch both refuse full XML for this publisher) and should be re-checked against the PDF. The cohort is trial-enrolled, not consecutive, so it is subject to enrolment selection; it is also Japanese, and the adult Japanese ZNF384 frequency is reported elsewhere as unusually high, which raises the possibility of a population effect this beat cannot assess. Note the partner counts sum to 23 against 4.1% of 291 = 12, because the paper assembled additional cases beyond the trial cohort - the two denominators must not be mixed.

• 4.1% is 2.5-fold higher than the Czech consecutive figure of 1.6%, and the difference has at least three candidate explanations that this beat cannot separate: trial enrolment selection, a Japanese population effect, and a detection difference (this study used whole-transcriptome sequencing on selected cases). Quoting 4.1% as 'the frequency of ZNF384 fusions in childhood BCP-ALL' - which is how it is usually cited - assumes all three are negligible. Zaliova M et al. 2021

FULL TEXT READ via the PMC HTML rendering of PMC6294900 (pnas.org returns 403). **THIS 6% IS NOT A ZNF384 FIGURE** - the G5 subgroup is defined by expression clustering and includes four ZNF362-fusion cases, so the ZNF384 count is 76 of 74 by the listed partners... which does not reconcile: 53 + 12 + 11 + 4 = 80 against a stated subgroup size of 74. The table also collapses all four ZNF362 cases under 'SMARCA2-ZNF362' while the body text splits them 3 SMARCA2 and 1 TAF15. Neither discrepancy is resolved here and no derived fraction should be computed from this row. The cohort is an assembled international series, not consecutive.

• The arithmetic in this table does not close - the four named partner counts sum to 80 against a stated subgroup size of 74 - and the same table contradicts the paper's own body text on the ZNF362 partner split. A frequency figure from a table that disagrees with itself should be carried as an order of magnitude and nothing more. The reason to record it at all is that 6% of an international 1,223-case series is the most-cited ZNF384 frequency after 4.1%, and a reader needs to know it is a transcriptional-subgroup size that includes a different gene. Li JF et al. 2018

The 5% and 50% are an introduction summary in a mechanism paper, i.e. a field characterisation rather than a measurement, and the 5% sits three-fold above the Czech consecutive figure. The 48% comes from Alexander et al. (PMID 30209392) and **DOES NOT RECONCILE WITH ITS OWN COHORT NUMBERS**: that paper reports 35 B/myeloid MPAL cases and names 15 ZNF384-rearranged ones, and two figure legends independently give n = 15; 15/35 is 42.9%. The denominator behind 48% is not stated anywhere in the paper. Quoted as published; do not construct a fraction from it.

• The ~50% MPAL figure is the one clinically important number in this entry that is NOT a denominator artefact, and it is also the one whose arithmetic cannot be reproduced from its source. Both things are true at once. A reader should take 'roughly half of paediatric B/myeloid MPAL is ZNF384-rearranged' as well supported in direction and unreliable in its second digit, and should note that MPAL diagnosis itself depends on WHO lineage-assignment criteria that changed between revisions - so the denominator is a moving definition, not a fixed population. Alexander TB et al. 2018

ABSTRACT ONLY - the Wiley full text is unreachable by every route tried, with no PMC copy and no Wayback snapshot. This is a range quoted in the introduction of a mechanism paper, with the underlying cohorts not identified in the abstract, so it cannot be traced to a denominator at all. Philadelphia-negative BCP-ALL is itself a residual category, though a much broader one than B-other, so the same inflation argument applies in weaker form.

• A 7-17% range with no named cohorts is not a frequency estimate, it is a summary of other people's summaries, and the two-fold spread inside it is larger than most of the differences this entry is trying to explain. It is recorded because it is the figure most often repeated in the mechanistic literature on this family, and a reader needs to be able to place it - against the Czech consecutive figure of 1.6-1.9%, a 7-17% claim is between four and eleven times higher and the whole difference is denominator plus population. Zaliova M et al. 2021

The 20% is a background statement in the Background section of a mechanism paper, not a measurement made in it, and the cohort behind it is not identified there. It stacks THREE narrowing conditions at once - Philadelphia-negative (a residual category), adult (where ZNF384 is commoner), and Japanese (a single population, from the group that has reported the highest frequencies) - so it is the least transferable number in this entry and the most likely to be quoted out of context. The same paper's own RNA-seq cohort was 323 Japanese ALL patients.

• This figure is roughly twelve-fold the Czech consecutive paediatric figure, and essentially all of the gap is in the three stacked conditions rather than in any disagreement about the biology. It should never be cited as 'ZNF384 fusions occur in 20% of B-ALL'. The one thing it does support, alongside the four-fold adult-versus-child difference for EP300::ZNF384 in an independent transcriptome series, is that the subtype is genuinely commoner in adults - which is an age effect and not a denominator effect, and matters because this project's population is paediatric. McClure BJ et al. 2018

FULL TEXT READ as the author-accepted PDF in the University of Tsukuba repository, the free location Unpaywall lists for this DOI. The 1.03% is explicitly a composite ESTIMATE the authors construct from two differently ascertained screens - 55 selected BCP-ALL samples without conventional genetic abnormalities, then 346 further paediatric ALL cases by RT-PCR - and is not a rate from a consecutive series. The 5.7%/1.3% age comparison is from McClure et al. (PMID 29531323), read via Europe PMC, and combines two differently selected cohorts within that study.

• The age comparison is the part of this item that carries real information, and it is also the part with the weakest design: 122 AYA/adult and 152 paediatric cases from one transcriptome study, differently recruited, with 7 and 2 events respectively. The Wilson intervals overlap considerably (roughly 2.8-11.4% versus 0.4-4.7%), so a four-fold point difference on 9 total events is suggestive rather than established. It is corroborated independently by the Ponte di Legno median-age difference between partners and by the high reported adult Japanese frequency, which is why it is carried as a real effect rather than noise. McClure BJ et al. 2018

FULL TEXT READ via Europe PMC fullTextXML for PMC8999738. n = 1 patient, in a paper about lineage switch rather than about cytogenetics, so this establishes that a cryptic presentation OCCURS for TCF3::ZNF384 and says nothing about how often. The earliest description of E2A-ZNF384 also called its t(12;19)(p13.3;p13.3) cryptic, which is consistent, but this beat did not read that 2008 paper.

• One cryptic case does not establish that TCF3::ZNF384 is usually cryptic, and the evidence for the parallel claim about EP300::ZNF384 is far stronger - six of six not detected by G-banding, five with normal karyotypes. This item is included because the B-other inflation argument needs the two commonest partners both to be cytogenetically invisible, and for TCF3 that requirement currently rests on one patient plus a 2008 title this beat did not open. It is the weakest link in this entry's argument and is marked as such. Gocho Y et al. 2015