Oncofusion

start / graph / concept-wbc-300-threshold-is-an-administrative-choice

Explainerstrong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.Q34-mrd-monitoring

WBC 300: a stratification convenience, not a biological threshold - and the evidence says so four different ways

also called WBC >=300 x10^9/L · hyperleukocytosis cut-point · high-risk WBC criterion · white cell count threshold infant ALL

Infant leukaemia trials call a white cell count of 300,000 per microlitre the line between risk groups, but nothing in the biology changes at that number: the recognised danger threshold for a clogged circulation is a third of it, the line was never derived from a published cut-point analysis, and how many babies fall above it differs threefold between countries. It is a line drawn to split a trial into groups of workable size.

The same thing, for a scientist

The answer to the question this entry exists to settle is: **no biological justification for 300 x10^9/L has been published, a trial that tested it against a different cut-point found the different one better, and five lines of evidence say it is a stratification convenience.** (1) **A head-to-head exists and 300 loses.** COG P9407 analysed 209 infants and compared risk groupings directly: by Interfant-99 groups (KMT2A-rearranged with age <6 months and WBC >300 x10^9/L) 5-year event-free survival was 24% high risk versus 41% intermediate (P = .047), while splitting the same patients at the National Cancer Institute cut of **50 x10^9/L** gave 67% versus 29% within KMT2A-rearranged patients (P = .0002); the authors conclude that KMT2A-rearranged infants separate better by partner gene, by age above or below 90 days, or by WBC above or below 50 x10^9/L than by the Interfant risk groups [E8]. (2) The clinically defined threshold for hyperleukocytosis is **100 x10^9/L**, the level above which leukostasis carries high early morbidity and mortality [E1]; 300 is three times that. (3) The predecessor protocol Interfant-99 found *very high white blood cell count* independently adverse and states no threshold in its report [E2]; 300 appears as a protocol definition in Interfant-06 [E3][E4], not as the output of a cut-point analysis. (4) The proportion the line selects varies threefold between contemporary cohorts: 29% in Interfant-06 and 30% in JPLSG MLL-10 against 11.2% in CCCG-ALL-2015, which its own authors attribute to under-representation of high-risk infants [E5]. (5) Dichotomisation loses power, creates residual confounding and makes the effect size depend on the cut-point's position, and a data-derived optimal cut-point is seriously biased [E6]. **The contrast that settles the character of the 300 line is the 90-day age line.** COG uses age at or below 90 days to define high risk and gives an explicit biological argument for it - microarray separation of AFF1 cases at about 90 days, with B-cell maturation genes elevated in older infants and interleukin, HSP and HLA genes in younger ones - which the authors say provides biological grounds for the cut-off [E8]. A cut-point in this disease **can** be given a biological justification. Somebody did it for age. Nobody has done it for WBC 300.

Go deeper
Where the line is actually written down

The registered Interfant-06 protocol defines high risk as KMT2A rearrangement and age under 6 months (under 183 days) and WBC at or above 300 x10^9/L and/or prednisone poor response 3. Medium risk explicitly includes KMT2A-rearranged infants under 6 months with WBC below 300 and a good prednisone response 3. CCCG-ALL-2015 uses the same conjunction, stratifying to high risk infants aged <6 months and KMT2A-rearranged with WBC >=300 x10^9/L, or with induction failure 5.

Neither the Interfant-06 report 4 nor the Interfant-99 report 2 presents a cut-point derivation. Interfant-99 identified very high white blood cell count as an independent adverse factor and named no number 2. So the threshold enters the literature as a protocol definition and is thereafter cited as if it were a finding - the ordinary route by which an administrative choice acquires the appearance of biology.

What 300 would have to mark to be biological, and does not

Three candidate biological anchors, none of which is at 300, and one direct empirical test.

Leukostasis. Hyperleukocytosis is defined as a WBC above 100 x10^9/L, the level associated with high early morbidity and mortality from leukostasis-related intracranial haemorrhage and pulmonary distress 1. If 300 marked a leukostasis threshold it would coincide with that definition; it does not, and in the leukapheresis series that defines it the mean WBC in children actually treated with therapeutic leukocytapheresis was 483.2 x10^9/L overall and 547.1 in ALL 1 - clinicians intervened well above both 100 and 300, on symptoms rather than on a number.

Tumour burden. WBC is a poor proxy for burden in a disease with massive extramedullary and CNS involvement, and in an infant the same absolute count represents a very different burden because circulating blood volume is roughly a tenth of an older child's - the point concept-hyperleukocytosis-cytoreduction-in-a-neonate makes.

A discontinuity in outcome. This is the one that could in principle exist, and the one place it has been looked for, it was not found at 300. COG P9407 ran both groupings on the same 209 infants. The Interfant-99 grouping using WBC >300 separated high risk from intermediate risk at 24% versus 41% 5-year EFS with P = .047 - a real but marginal separation. The National Cancer Institute cut at 50 x10^9/L separated the same KMT2A-rearranged patients at 67% versus 29% with P = .0002, and separated KMT2A-germline infants at 88% versus 52% (P = .013) 8. The authors state plainly that KMT2A-rearranged infants separate better by partner gene, by age above or below 90 days, or by WBC above or below 50 x10^9/L 8.

Consistently, in the one cohort that ran a multivariable analysis restricted to KMT2A-rearranged infants, WBC >=300 was not among the factors associated with inferior EFS; age under 3 months and MRD positivity at days 19 and 49 were 5.

What follows for this graph

The threshold is real as an administrative object and unreal as a biological one. That is not a criticism of the trialists - a randomised protocol needs allocation rules applicable on day one from a blood count, and a continuous variable cannot allocate. The error is downstream: treating WBC >=300 as a disease feature rather than as a protocol rule.

Four specific consequences.

  1. A better-performing cut-point was available and is in routine use elsewhere. The National Cancer Institute risk criteria use 50 x10^9/L, and on the one cohort where both were applied it outperformed 300 by a wide margin in KMT2A-rearranged infants 8. Whatever the case for 300, it is not that nobody had an alternative.
  2. The high-risk group is a conjunction and is small. 5 of 71 KMT2A-rearranged infants in CCCG-ALL-2015 5. Interfant-06's high-risk group had a 6-year EFS of 20.9% 4, the worst outcome in the disease, estimated on a group defined by three stacked dichotomies.
  3. Cross-cohort comparison of the proportion above 300 measures enrolment, not epidemiology. 29%, 30% and 11.2% in three contemporary protocols 5.
  4. Two clinically indistinguishable infants receive different therapy. An infant at 299 x10^9/L and one at 301 x10^9/L differ in whether they are offered transplant in first remission. concept-risk-group-cut-points-are-administrative-choices-infant-all makes this argument across the WBC, age and MRD cut-points; this entry is the WBC-specific answer the graph was pointing at.

And the honest counter-position: MRD has now displaced presenting features in every cohort that measured both 57. Arguing about where to draw the WBC line may be arguing about a variable the next protocol will not stratify on at all - and P9407's own conclusion points the same way, since the variable it found most informative was the partner gene, not the count 8.

A note on how this paper was read, because the beat had logged it unreadable

PMID 36861410 returns an empty abstract from PubMed and its Europe PMC core record gives pmcid: null, isOpenAccess: N, inEPMC: N. All four routes of this project's unreadable test require a PMCID, so all four were structurally unavailable and the paper was initially logged as title-only in ana-mrd-assay-enumeration-and-bounded-rejection.

It was recovered by resolving the DOI to the publisher's own page: https://doi.org/10.3324/haematol.2022.281552 redirects to haematologica.org, which serves the full Letter to the Editor as HTML, 27,880 characters after tag stripping. Haematologica is an open-access journal that does not deposit in PMC.

The generalisable rule: the four-route PMC test is a test of PMC, not a test of readability. A journal can be fully open access and absent from PMC, in which case every PMC route returns a clean structural negative. Resolve the DOI as a fifth route before writing 'unreadable'. Tried on the other two papers this beat could not read, it failed for both - JCO returned HTTP 403 and the Lancet DOI resolved to a 13-character Elsevier redirect stub - so the fifth route is a real addition and not a universal rescue.

And one markup trap inside the rescue. Unescaping HTML entities before stripping tags destroys the content: &lt;50x10 9 /L became <50x10 9 /L and was then deleted as a tag, turning WBC count <50x10^9/L into WBC count with no number, in a sentence that still reads as a sentence. Strip tags first, unescape entities second.

Facts and figures
questiondoes WBC >=300 x10^9/L have a biological justification, or is it a stratification convenience?
answerstratification convenience; no published derivation, and four independent lines of evidence against a biological reading
high risk is a conjunctionKMT2A-r AND age <6 months AND (WBC >=300 OR prednisone poor response)
what would settle ita published change-point or spline analysis of WBC against event-free survival in a pooled infant KMT2A-rearranged cohort; none exists, and the nearest available evidence (COG P9407) compares two pre-chosen cut-points rather than estimating one
age direction third confirmationCOG P9407 describes the Interfant-99 high-risk group as KMT2A-rearranged with age under 6 months and WBC above 300 x10^9/L, a third independent source giving the age criterion as UNDER 6 months against the Interfant-06 abstract's 'older than'
threshold
{
  "value": 300,
  "unit": "x10^9/L",
  "used_by": [
    "Interfant-06",
    "CCCG-ALL-2015",
    "JPLSG MLL-10 (proportion reported)"
  ]
}
clinical hyperleukocytosis threshold
{
  "value": 100,
  "unit": "x10^9/L",
  "source_pmid": "29193219"
}
proportion above threshold by cohort PCT
{
  "Interfant-06": 29,
  "JPLSG MLL-10": 30,
  "CCCG-ALL-2015": 11.2,
  "source_pmid": "38634141"
}
conjunction yield cccg
{
  "kmt2a_r_patients": 71,
  "meeting_age_and_wbc": 5
}
cccg multivariable within kmt2a r
{
  "significant": [
    "age <3 months",
    "day 19 MRD positive",
    "day 49 MRD positive"
  ],
  "not_significant": [
    "WBC >=300 criterion"
  ]
}
statistical objection
{
  "source_pmid": "16217841",
  "points": [
    "loss of power",
    "residual confounding",
    "effect size depends on cut-point position",
    "data-derived optimal cut-points are seriously biased"
  ]
}
head to head cutpoint comparison
{
  "cohort": "COG P9407, 209 infants analysed (199 treatment-eligible, 157 KMT2A-rearranged)",
  "interfant99_grouping_efs_5y_pct": {
    "high_risk": 24,
    "intermediate": 41,
    "low_risk_kmt2a_germline": 69,
    "p_high_vs_intermediate": 0.047
  },
  "nci_wbc_50_cut_efs_5y_pct": {
    "all_treatment_eligible": [
      76,
      33
    ],
    "kmt2a_rearranged": [
      67,
      29
    ],
    "mllt1": [
      67,
      15
    ],
    "kmt2a_germline": [
      88,
      52
    ]
  },
  "authors_conclusion": "KMT2A-rearranged infants separate better by partner gene, by age </>90 days, or by WBC </>50e9/L than by the Interfant risk groups",
  "source_pmid": "36861410"
}
a cut point that does have a biological argument
{
  "variable": "age",
  "value": "90 days",
  "used_by": "COG",
  "argument": "microarray separation of AFF1 cases at about 90 days, with elevated B-cell maturation genes in older infants versus interleukin, HSP and HLA genes in younger ones",
  "source_pmid": "36861410"
}
alternative wbc cutpoint in routine use
{
  "value": 50,
  "unit": "x10^9/L",
  "scheme": "National Cancer Institute risk criteria"
}
How this could bite us back
  • This threshold decides who is offered allogeneic transplant in first remission before the age of one, with busulfan-based conditioning. A line with no published derivation, which lost a head-to-head against a cut-point six times lower, is allocating the most toxic intervention in paediatric oncology.
  • Because the criterion is a conjunction with age, an infant above 300 but over 6 months old is medium risk, while an infant below 300 and under 6 months with a poor steroid response is high risk. Anyone reading the criteria as a list rather than as a conjunction will misclassify both, and several secondary sources do read it as a list.
  • Moving the line to 50 x10^9/L would reclassify a large fraction of infants as high risk, because 50 is close to the median presenting count in this disease. A cut-point that discriminates better is not automatically a cut-point that should allocate therapy, and swapping one undefended number for a better-discriminating one is still choosing a number.
  • The proportion of infants above 300 varies from 11% to 30% between contemporary protocols. A trial powered on an expected high-risk fraction taken from another cohort will under- or over-enrol the arm with 20.9% event-free survival.
  • The honest reading of the one head-to-head is that partner gene beat both WBC cut-points. Spending effort re-cutting the count may be optimising the wrong variable, and the partner gene is already measured in every patient who gets a molecular diagnosis.
Still open
  1. Where did 300 x10^9/L come from? Neither Interfant-99, Interfant-06 nor the registry record gives a derivation, and this beat could not locate one. The two documents most likely to contain it are the two it could not read.
  2. Should infant protocols simply adopt the National Cancer Institute cut of 50 x10^9/L? On the one cohort where both were applied it discriminated far better within KMT2A-rearranged patients, and it is already the standard elsewhere in paediatric ALL.
  3. Is there a change point in the WBC-outcome relationship in infant KMT2A-rearranged ALL, as opposed to two pre-chosen cut-points being compared? A spline or change-point analysis on pooled Interfant cohorts would answer it and requires no new data.
  4. COG gives a published biological argument for its 90-day age cut-off from microarray separation of AFF1 cases. Could the same method be applied to WBC, or does WBC simply have no expression correlate to anchor a cut-point to?
  5. If partner gene carries adjusted hazard ratios of 2.51 for AFF1 and 3.30 for MLLT1 against KMT2A-germline while MLLT3 is indistinguishable from germline, is any WBC cut-point carrying independent stratification weight at all?
Every source behind this entry (8)
  1. E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    Hyperleukocytosis is defined as a white blood cell count above 100 x10^9/L and carries high early morbidity and mortality from leukostasis-related complications, namely intracranial haemorrhage and pulmonary distress; in a single-institution series of 19 children with leukaemia who underwent therapeutic leukocytapheresis the mean WBC was 483.2 x10^9/L overall and 547.1 x10^9/L in acute lymphoblastic leukaemia.

    PMID 29193219DOI
    What this evidence cannot show

    The 100 x10^9/L definition is stated as established background in a small retrospective apheresis series (14 ALL, 5 AML at one institution), not derived in it. The mean WBC of those who received apheresis reflects referral for the procedure, not a threshold at which leukostasis begins. Infants are included (as young as 22 days) but are not analysed separately.

    Evidence that argues against this (1)
    • The same series reports that patients presented with central nervous system symptoms in 15%, respiratory symptoms in 10% or both in 10% at a mean count of 483 x10^9/L, so most children with counts far above both 100 and 300 were asymptomatic - which argues that no single count is the right threshold for anything, including 100. Thapa N et al. 2018
  2. E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    In Interfant-99, which enrolled 482 infants aged 0-12 months, all types of MLL gene rearrangement, very high white blood cell count, age younger than 6 months and poor response to the prednisone prophase were independently associated with inferior outcomes; the report states no numerical white blood cell threshold.

    PMID 17658395DOI
    What this evidence cannot show

    Abstract only: the paper is not open access and the Europe PMC core record returns no PMCID, so routes 2 to 4 of the full-text test are unavailable. The absence of a stated threshold is an observation about the abstract; the full report may define one, and this entry does not claim otherwise. Interfant-99 stratified prospectively on prednisone response, not on WBC.

    Evidence that argues against this (1)
    • That Interfant-99 found very high WBC independently adverse is itself the justification for including a WBC criterion in the successor protocol; the objection in this entry is to the specific value of 300, not to using WBC at all. Pieters R et al. 2019
  3. E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.trial_registry

    The registered Interfant-06 protocol defines high-risk disease as MLL rearrangement together with age at diagnosis under 6 months (under 183 days) and WBC at or above 300 x10^9/L and/or prednisone poor response, and defines medium risk to include MLL-rearranged infants under 6 months with WBC below 300 x10^9/L and a prednisone good response; no derivation of the 300 x10^9/L value is given.

    International Collaborative Treatment Protocol for Infants Under One Year With Acute Lymphoblastic or Biphenotypic Leukemia (Interfant-06) · Dutch Childhood Oncology Group (lead sponsor), with 18 named national and international collaborating groups · ClinicalTrials.gov 2006
    NCT00550992
    What this evidence cannot show

    Fetched from the raw ClinicalTrials.gov API on 2026-09-20. A registry record states the protocol rule and is not expected to contain a statistical derivation, so its silence on where 300 came from is weak evidence of absence on its own; it is used here with E2 and E4, neither of which supplies one either. The record's status is UNKNOWN and its registered enrolment estimate (445) differs from the published enrolment (651).

    Evidence that argues against this (1)
    • The published Interfant-06 report states the high-risk age criterion in the opposite direction (older than 6 months), so the registry and the publication disagree about the risk model they are both describing, and the registry is not self-evidently the correct one. Pieters R et al. 2019
  4. E4established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    Interfant-06 defined a high-risk group using KMT2A rearrangement, age relative to 6 months, WBC count at or above 300 x10^9/L and prednisone response; the 6-year event-free survival of that high-risk group was 20.9%; and across the whole 651-patient cohort KMT2A rearrangement was the strongest prognostic factor for event-free survival, followed by age, WBC count and prednisone response.

    PMID 31283407DOI
    What this evidence cannot show

    Abstract only; not open access and no PMCID. The abstract states the age criterion as older than 6 months, which contradicts the registry and the same group's MRD paper; the WBC value of 300 is consistent across all three sources and is what this entry relies on. The ordering of prognostic factors is reported without effect sizes in the abstract.

    Evidence that argues against this (1)
    • The ranking places WBC third behind KMT2A status and age, and the protocol's own subsequent MRD analysis found that MRD at end of induction carried independent prognostic weight after adjustment for risk group, so the WBC criterion is not carrying much of the stratification even within the protocol that uses it. Stutterheim J et al. 2021
  5. E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Among 134 infants in CCCG-ALL-2015, median WBC was 62.7 x10^9/L (range 1.3-777) with 11.2% at or above 300 x10^9/L, against 29% in Interfant-06 and 30% in JPLSG MLL-10, which the authors attribute to under-representation of high-risk infants in their cohort; high risk was defined as age under 6 months and KMT2A-rearranged with WBC at or above 300 x10^9/L, or induction failure; only 5 of 71 KMT2A-rearranged infants met both the age and WBC criteria; and in multivariable analysis restricted to KMT2A-rearranged patients the factors associated with inferior event-free survival were age under 3 months and MRD positivity at day 19 and day 49.

    PMID 38634141DOI
    What this evidence cannot show

    Read in full text. The authors state their cohort is unrepresentative: infants were 1.7% of the whole protocol against an expected 3-4%, 18.4% were under 6 months against 50-68% elsewhere, and KMT2A rearrangement was present in 53.0% against 74-83% elsewhere. A multivariable analysis in 71 patients with these event numbers is underpowered, so the failure of the WBC criterion to appear is weak evidence of its unimportance. Day-19 and day-49 MRD was measured by flow cytometry on a protocol with different timepoints from Interfant.

    Evidence that argues against this (1)
    • Because this cohort under-enrolled the youngest and the KMT2A-rearranged, it contains few patients above the threshold, which is exactly the circumstance in which a real threshold effect would fail to appear. The negative finding is confounded by the enrolment bias the authors themselves describe. Leung AWK et al. 2024
  6. E6established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    Dichotomising continuous predictor variables in medical research causes a considerable loss of power and residual confounding, makes the achieved simplicity costly rather than free, and using a data-derived optimal cut-point leads to serious bias; the authors conclude that dichotomisation of continuous data is unnecessary for statistical analysis and should not be applied to explanatory variables in regression models.

    Dichotomizing continuous predictors in multiple regression: a bad idea. · Royston P, Altman DG, Sauerbrei W · Stat Med 2006
    PMID 16217841DOI
    What this evidence cannot show

    A methodological paper in medical statistics with a case study in primary biliary cirrhosis, not in leukaemia. It argues against dichotomisation in regression analysis; it does not address the separate problem that a randomised protocol needs a rule that allocates treatment on day one, which a continuous variable cannot do.

    Evidence that argues against this (1)
    • The objection applies to explanatory variables in regression models. A trial stratification rule is an allocation device rather than a regression covariate, so the statistical argument does not by itself show the protocol was wrong to use a cut-point - only that the resulting effect estimate depends on where the line sits. Royston P et al. 2006
  7. E7strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    In Interfant-06, MRD at end of induction and at end of consolidation predicted disease-free survival with a significant effect after adjustment for risk group and consolidation treatment, and the authors conclude MRD has independent prognostic value that will be used for improved risk stratification and treatment selection in the next Interfant protocol.

    PMID 33405950DOI
    What this evidence cannot show

    Read in full text. MRD was not mandatory in the protocol, so the 249 patients analysed are a subset of 439 eligible. Independence is shown against risk group as a whole rather than against the WBC criterion specifically, so this does not directly demonstrate that MRD replaces WBC.

    Evidence that argues against this (1)
    • In the same multivariable model, risk group also retained a significant impact on disease-free survival alongside MRD, so the presenting-feature stratification is not simply superseded by MRD in this cohort. Stutterheim J et al. 2021
  8. E8strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    In Children's Oncology Group trial P9407, among 209 infants analysed (199 treatment-eligible, 157 KMT2A-rearranged), applying Interfant-99 risk groups (high risk defined as KMT2A-rearranged with age under 6 months and WBC count above 300 x10^9/L) gave 5-year event-free survival of 24% high risk versus 41% intermediate risk versus 69% low risk (KMT2A-germline), with high risk versus intermediate P = .047; applying National Cancer Institute risk groups at a WBC cut of 50 x10^9/L gave 76% versus 33% in treatment-eligible patients overall (P < .0001), 67% versus 29% in all KMT2A-rearranged (P = .0002), 67% versus 15% in MLLT1 (P = .029) and 88% versus 52% in KMT2A-germline (P = .013); the authors conclude that KMT2A-rearranged infants separate better when analysed by partner gene, by age at or below versus above 90 days, or by WBC below versus at or above 50 x10^9/L; and they state that their microarray studies revealed AFF1 case separation at about 90 days with elevated B-cell maturation genes in older infants versus interleukin, HSP and HLA genes in younger ones, providing biological grounds for the 90-day age cut-off used by COG.

    PMID 36861410DOI
    What this evidence cannot show

    Read in full text from the publisher's open-access HTML page after all four PMC routes were structurally unavailable; PubMed returns an empty abstract for this record. It is a Letter to the Editor, so the analysis is compressed and several results sit in supplementary figures this beat did not retrieve. P9407 ran 1996-2006 with era-appropriate partner-gene assignment (karyotype, FISH, Southern blot, conventional and panhandle PCR), and 7.6% of partners were 'unknown'. The comparison of risk groupings is a retrospective application of two schemes to one cohort treated on neither, so the survival differences reflect discrimination and not the effect of being allocated by either scheme. The 50 x10^9/L comparison uses NCI criteria developed for older children, and the biological grounds offered for the 90-day age cut-off are the authors' own earlier microarray work, cited rather than reproduced here.

    Evidence that argues against this (2)
    • The Interfant grouping is a conjunction of three variables and the NCI grouping is a single variable, so comparing their separations is not a clean comparison of 300 against 50: the Interfant high-risk group is far smaller, which widens its confidence intervals and weakens its P value independently of where the WBC line sits. Leung AWK et al. 2024
    • P9407's own strongest discriminator was the partner gene, with AFF1 and MLLT1 carrying adjusted hazard ratios of 2.51 and 3.30 against KMT2A-germline while MLLT3 was indistinguishable from germline - so the paper's main message is that neither WBC cut-point is the variable that matters most, which argues against replacing 300 with 50 as much as against keeping 300. Robinson BW et al. 2023
Connected to