Hypothesis: the paediatric gap is set by the population ratio, not by paediatric biology
also called why children are left out · population-ratio hypothesis
The reason babies get left out of new cancer drug trials may have nothing to do with how hard babies are to treat, and everything to do with what fraction of the eligible patients they are. If so, you can predict in advance which new drugs will reach children and which will not, just by counting patients.
The same thing, for a scientist
Proposed: for a mechanism-directed oncology asset, the share of trial enrolment that is paediatric is governed primarily by the paediatric share of the biomarker-eligible population, and only secondarily by formulation difficulty, safety concern or regulatory requirement. The menin field (infants ~1.2% of the menin-addressable population, paediatric enrolment negligible, dedicated paediatric studies withdrawn, PREA requirement waived for under-1s) and the TRK field (no large adult pool, infants enrolled from phase 1, infant-predominant histologies developed) are the two poles. If the hypothesis holds, paediatric inclusion is predictable from epidemiology at the moment a target is nominated, which makes it an actionable design variable rather than a post-hoc disappointment.
The statement, and why it is not obvious2 deeper
Statement. Paediatric enrolment share in a mechanism-directed oncology programme is approximately a monotone function of the paediatric share of the biomarker-eligible population, with an additional downward step when a large adult indication for the same mechanism exists.
The competing explanations, all plausible, all testable against it:
- Paediatric biology is harder. If true, paediatric inclusion should be low regardless of population share. The TRK case contradicts this: infants were dosed from phase 1 in the hardest possible setting, solid tumours of the first year of life 2.
- Formulation is the barrier. If true, assets with an existing liquid or dispersible form should show much higher paediatric inclusion at the same population share. Testable.
- Regulation already fixes it. If true, PREA and the EU paediatric regulation should produce paediatric data. In the index case the requirement for under-1s was waived as impracticable 3, so regulation did not fix it here.
- Safety concern dominates. If true, paediatric inclusion should track the severity of the adult safety profile rather than population share. Testable.
The two poles, as the hypothesis predicts them
| Menin inhibitors | TRK inhibitors | |
|---|---|---|
| Large adult pool for the same mechanism | Yes (NPM1-mutant AML, ~6,800 US/yr) | No |
| Paediatric share of eligible population | ~1.2% infants | high |
| Infants in the pivotal programme | 13 infants <2y of 104; under-1s excluded from the label's efficacy claim 3 | infants dosed from phase 1 2 |
| Dedicated paediatric studies | three withdrawn with zero enrolment | ran |
| Second asset in class | adults-only label, 18-year protocol floor | paediatric from the start |
The hypothesis says the first row causes the rest.
If it is true, what to do about it
Three interventions follow directly, and they are different from the ones usually proposed.
- Nominate targets partly on population ratio. If two interfaces are equally druggable and one has no large adult pool, the one without the adult pool will reach children faster. That is a legitimate target-selection criterion for a paediatric lab and nobody applies it.
- Attach the infant stratum to the adult approval, not after it. A minimum infant stratum size specified at the time of adult pivotal design costs the sponsor little when the adult trial is large, and cannot be waived later on feasibility grounds because the patients are already being enrolled.
- Close the feasibility waiver. The PREA waiver for under-1s was granted precisely because the population is small and dispersed 3. The arithmetic in
ana-infant-trial-feasibility-arithmeticshows that an international consortium can accrue ~48 KMT2A-rearranged infants a year, which is not impracticable for a single-arm study. The waiver rests on a feasibility claim that is contestable with data the consortium already has.
| statement | Paediatric enrolment share in a mechanism-directed oncology programme is approximately a monotone function of the paediatric share of the biomarker-eligible population, with a further downward step when a large adult indication for the same mechanism exists. |
|---|---|
| origin | data-pattern |
| prior probability | 0.6 |
| proposed new analysis | Assemble every mechanism-directed oncology approval since 2015 (n on the order of 50-80), code each for biomarker-eligible adult and paediatric population from SEER and published biomarker frequencies, code paediatric enrolment share and paediatric study status from ClinicalTrials.gov and the label, and regress paediatric enrolment share on population ratio with response rate, formulation and adult safety profile as covariates. Entirely public data; runnable without new sample access. |
predictions
[ "Across mechanism-directed oncology assets approved since 2015, paediatric enrolment share will correlate positively with paediatric share of the biomarker-eligible population (Spearman rho > 0.5).", "Assets whose biomarker has no adult population above roughly 2,000 US cases per year will show paediatric enrolment above 20%; assets with a larger adult population will show below 10%.", "Withdrawal or non-initiation of dedicated paediatric studies will cluster in assets with the lowest paediatric population share, independent of the adult safety profile.", "Formulation availability (liquid or dispersible at first approval) will add little explanatory power once population share is in the model." ]
analyses
[ "ana-infant-trial-feasibility-arithmetic" ]
experiments
[]
falsifiers
[ "Paediatric enrolment share shows no association with paediatric population share across mechanism-directed approvals once response rate is controlled for.", "An asset with a very large adult population nonetheless enrols infants proportionately, showing that adult pool size is not the constraint.", "Formulation availability, not population share, explains most of the variance in paediatric inclusion.", "A sponsor-funded randomised frontline infant menin-inhibitor trial opens before 2029 without any change in the population ratio." ]
- If the hypothesis is right, it implies that the very success of the platform strategy predicts paediatric exclusion, and a lab arguing for platforms is arguing for a structure that systematically deprioritises its own patients unless it also argues for the corrective mechanism.
- Selecting targets partly by the absence of a large adult population would steer a lab away from the best-validated interfaces towards worse ones, trading druggability for access. That is a genuine harm and the trade should be made explicitly, not adopted because the access argument feels virtuous.
- Is there a threshold paediatric population share above which paediatric development reliably happens, and where is it?
- Do European PIP outcomes show the same gradient as US enrolment, which would indicate the effect is economic rather than regulatory-jurisdictional?
- Has any sponsor ever been paid, by a public or philanthropic funder, to add an infant stratum to an adult pivotal trial, and what did it cost?
- E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper
Infants with KMT2A-rearranged leukemia are roughly 1.2% of the menin-addressable US population, because the same mechanism serves NPM1-mutant AML which is roughly twice the size of the whole KMT2A-rearranged population.
The menin inhibitor revumenib in KMT2A-rearranged or NPM1-mutant leukaemia · Issa GC, Aldoss I, DiPersio J, et al. · Nature 2023What this evidence cannot show
The population shares are derived in `epi-kmt2a-rearranged-acute-leukemia-all-ages` from SEER projections combined with this paper's introductory 10% and 30% framing figures, not measured directly. The 1.2% is arithmetic, with the uncertainty of its inputs.
Evidence that argues against this (1)
- The 1.2% figure is a ratio of prevalences and the hypothesis needs it to predict developmental behaviour, which requires the further assumption that sponsors allocate paediatric effort in proportion to population size. Sponsors allocate on expected revenue, trial feasibility and regulatory obligation, and a 1.2% population that is ALSO the population with the highest unmet need and a statutory paediatric pathway may behave quite differently from its share. Nature 2023
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
TRK inhibitor development enrolled children including infants from phase 1 and developed infant-predominant, ETV6::NTRK3-driven histologies, in a setting with no large adult population for the same biomarker.
Larotrectinib for paediatric solid tumours harbouring NTRK gene fusions: phase 1 results from a multicentre, open-label, phase 1/2 study · Laetsch TW, DuBois SG, Mascarenhas L, et al. · Lancet Oncology 2018What this evidence cannot show
A single comparison pole. TRK inhibitors also had an exceptionally high response rate, which independently makes small paediatric cohorts informative, so response magnitude is confounded with population ratio in this comparison.
Evidence that argues against this (1)
- If high response rate rather than population ratio is what enabled paediatric inclusion, the hypothesis is wrong. Distinguishing them requires assets with high response rates and large adult pools, and low response rates and no adult pool. source 2018
- E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database
FDA waived the paediatric study requirement for patients under 1 year of age for revumenib as 'impossible or highly impracticable' because of the small number and geographical dispersion of patients, and the label states safety and efficacy under 1 year have not been established despite 13 infants under 2 years appearing in the pivotal cohort.
NDA 218944 Approval Letter and Full Prescribing Information, REVUFORJ (revumenib) tablets · US Food and Drug Administration · FDA Drugs@FDA 2024What this evidence cannot show
The waiver is a statutory PREA determination, not a statement that infant study is scientifically impossible; it does not prevent voluntary study. Recorded in this graph by the regulatory entries in the same beat.
Evidence that argues against this (1)
- The under-1 waiver granted as 'impossible or highly impractical' is the outcome this hypothesis predicts, and it also supplies an alternative explanation that does not need the population-ratio mechanism at all: infants under one are hard to enrol for reasons of consent, venous access, pharmacokinetics and rarity, independently of market size. The hypothesis and the mundane explanation predict the same waiver. Hematology Am Soc Hematol Educ Program 2013
- ← related toNPM1::ALK (NPM-ALK), t(2;5)(p23;q35)
Population conditioning (Savage 2008, PMID 18385450): in 1314 reviewed PTCL cases, ALK-negative ALCL formed a large share of systemic ALCL with 5-year OS 49% versus 70%, so the 90% NPM1-ALK figure is specific to the paediatric ALK-positive slice.
- proposed by →The counter-argument: the menin field is an adult field, and the infant studies were withdrawn
generalises the menin observation into a testable rule
- tests →TRK inhibitors: one drug, seventeen cancers, and a paediatric population that actually got dosed
the opposite pole of the predicted gradient
feasibility arithmetic contesting the PREA impracticability waiver
the population ratio the hypothesis is built on