Oncofusion

start / graph / ana-infant-trial-feasibility-arithmetic

Analysismoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q8-impact

What 160 infants a year actually permits: the sample-size arithmetic

also called infant trial feasibility · power calculation infant leukemia · how long would the trial take

A proper two-arm trial big enough to prove a new drug improves survival in infant leukemia would need about 300 babies, which even with every country in the world contributing takes six years to recruit and nine or ten to report. A single-arm study answering a narrower question needs about forty babies and one year.

The same thing, for a scientist

Standard log-rank sample-size arithmetic applied to realistic infant accrual. A two-arm randomised superiority trial detecting a hazard ratio of 0.66 (3-year EFS 40% to 55%) at two-sided alpha 0.05 and 80% power requires ~182 events, implying ~300 infants at a ~60% event rate, which at the Interfant consortium's historical accrual of roughly 48 KMT2A-rearranged infants per year takes ~6.3 years to accrue and ~9-10 years to report. A more modest HR of 0.75 requires ~379 events and is not feasible at all. A single-arm Simon two-stage design testing a response rate of 40% against a 20% null needs ~35-40 patients, under one year of international accrual. The arithmetic, not timidity, is why infant leukemia trials look the way they do.

Go deeper
The inputs4 deeper
InputValueSource
US infant acute leukemia cases/yr~1601
US infant KMT2A-rearranged ALL cases/yr~63-72derived 1
Interfant-06 total accrual651 infants over ~a decade2
Implied international accrual, infant ALL~65/yrderived 2
Implied international accrual, KMT2A-rearranged infant ALL (~75%)~48/yrderived 12
Baseline 3-year EFS, KMT2A-r infant ALLbelow 40%1

All accrual figures are ceilings: they assume every diagnosed infant in every participating country is eligible, consented and enrolled.

Design 1: the two-arm randomised trial everyone says they want

Log-rank sample size for a superiority trial: d = 4(z(1-alpha/2) + z(1-beta))^2 / (ln HR)^2.

Target effectln HREvents requiredPatients at 60% event rateAccrual years at 48/yrReport at +3y follow-up
EFS 40% -> 55% (HR 0.66)-0.416182~3036.3~9.3 years
EFS 40% -> 50% (HR 0.75)-0.288379~63213.2~16 years
EFS 40% -> 65% (HR 0.47)-0.75556~931.9~5 years

Read the middle row. A ten-percentage-point improvement in event-free survival is not detectable in this disease at conventional power within a working career. That is not an argument for accepting weaker evidence; it is an argument that the trial must be designed to detect a large effect, or must borrow information, or must not be randomised.

The bottom row is the encouraging one: an effect large enough to matter clinically, a twenty-five-point EFS gain, needs under a hundred infants and is genuinely feasible through a consortium in about two years of accrual.

Design 2: the single-arm study that actually gets run

A Simon-style two-stage single-arm design testing a complete remission rate of 40% against a 20% null, one-sided alpha 0.05, 80% power, needs approximately 35-40 patients, with a futility stop after the first ~15. At 48 eligible infants a year internationally, that is under a year of accrual.

The blinatumomab infant study is the real-world instance: 30 newly diagnosed infants, single-arm, compared against historical Interfant-06 controls, no protocol-defined toxic effects, and outcomes better than historical controls 3. It answered a question nobody could have answered by randomising.

Its limitation is equally real: a single-arm result against historical controls cannot separate drug effect from improvements in supportive care, patient selection and era. See trialdesign-external-and-historical-controls-infant.

Design 3: what borrowing buys, and what it costs

A meta-analytic-predictive prior built from adult KMT2A-rearranged data could contribute an effective sample size of perhaps 20-40 patients to the infant analysis, cutting required infant accrual by a third to a half 4. Applied to Design 1's feasible row, that is the difference between two years of accrual and one.

The cost is an exchangeability assumption that this graph has repeatedly documented to be doubtful: infant KMT2A-rearranged leukemia arises from a fetal cell of origin, carries a different partner distribution, has a near-silent co-mutational landscape, and behaves differently from adult disease. Borrowing across a fifty-year age gap is a strong assumption, not a technical convenience. Robust MAP formulations with a heavy-tailed mixture component exist precisely to limit the damage when the assumption fails, and any borrowing design in this disease should use one. See trialdesign-bayesian-borrowing-across-fusions.

Conclusion the lab can act on
  1. Do not propose a randomised frontline infant trial powered for a modest effect. It cannot accrue and a reviewer will know.
  2. Propose either a large-effect randomised design (target HR <= 0.5, ~93 infants, ~2 years) or a single-arm design with a pre-specified external control.
  3. Any design must be international from the outset. The US alone supplies ~40 evaluable infants a year at best; the consortium supplies ~48 and could supply more with wider participation.
  4. The binding constraint is accrual, not money. At 300 infants, even the full drug cost at list price ($240,000 per six-month course) is ~$72 million, which is within the range of a single sponsor's phase 3 budget. Nobody is declining to run this trial because the drug is expensive.
Facts and figures
scriptanalyses/infant-trial-feasibility/run.py
statusplanned
methodlog-rank event-driven sample size, d = 4(z_{1-alpha/2}+z_{1-beta})^2 / (ln HR)^2; Simon two-stage for single-arm binary endpoint
result summaryHR 0.66 needs ~182 events, ~303 infants, ~6.3 years of international accrual; HR 0.75 needs ~379 events and is infeasible; HR 0.47 needs ~56 events, ~93 infants, ~1.9 years. A single-arm 40%-vs-20% response design needs ~35-40 infants, under one year.
assumed international accrual kmt2a r infants per year48
assumed event rate0.6
alpha0.05
power0.8
scenarios
randomised superiority0.470randomised superiority: 0.47randomised superiority0.660randomised superiority: 0.66randomised superiority0.750randomised superiority: 0.75
inputs
[
  "US infant acute leukemia incidence (Brown 2013)",
  "Interfant-06 accrual (Pieters 2019)",
  "baseline 3-year EFS in KMT2A-rearranged infant ALL"
]
scenarios — the numbers behind the chart
[
  {
    "design": "randomised superiority",
    "hr": 0.66,
    "events": 182,
    "n": 303,
    "accrual_years_at_48_per_year": 6.3
  },
  {
    "design": "randomised superiority",
    "hr": 0.75,
    "events": 379,
    "n": 632,
    "accrual_years_at_48_per_year": 13.2
  },
  {
    "design": "randomised superiority",
    "hr": 0.47,
    "events": 56,
    "n": 93,
    "accrual_years_at_48_per_year": 1.9
  },
  {
    "design": "single-arm Simon two-stage, 40% vs 20%",
    "n": [
      35,
      40
    ],
    "accrual_years_at_48_per_year": 0.8
  }
]
How this could bite us back
  • Designing for a large effect because only a large effect is detectable means a genuinely useful ten-point improvement would be declared negative and abandoned. Underpowering is not a neutral choice; it kills real benefits.
  • Borrowing from adults makes the trial feasible and simultaneously imports the adult population's biology into the infant inference. If infant disease differs, borrowing produces a confidently wrong answer faster than not borrowing produces an uncertain one.
  • A single-arm design against historical controls will attribute to the drug any improvement in supportive care, diagnostics or MRD-guided stratification that occurred between the control era and the study era. In a disease where survival has been flat for decades this is less dangerous than usual, but it is not zero.
Still open
  1. What is the actual current Interfant-21 accrual rate, and how many of those infants are KMT2A-rearranged? Replacing the derived ~48/yr with a real consortium figure would firm up every number here.
  2. What effective sample size would a properly fitted MAP prior on adult KMT2A-rearranged menin-inhibitor data actually contribute to an infant analysis?
  3. Does the graph's evidence on fetal cell of origin and partner distribution formally violate exchangeability enough to make borrowing inadmissible, and can that be tested rather than assumed?
Every source behind this entry (4)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.review

    US infant acute leukemia comprises roughly 160 cases per year (~90 ALL, ~70 AML) with KMT2A rearrangement in 70-80% of infant ALL, and 3-year event-free survival in KMT2A-rearranged infant ALL is below 40%.

    Treatment of infant leukemias: challenge and promise · Brown P · Hematology Am Soc Hematol Educ Program 2013
    PMID 24319237DOI
    What this evidence cannot show

    2013 review, US-specific, and the survival baseline predates Interfant-06 reporting and blinatumomab. If the true baseline EFS is now higher, every sample size in this node rises.

    Evidence that argues against this (2)
    • The incidence ceiling is not the accrual rate, and registry data put the gap at an order of magnitude. Novartis's international paediatric midostaurin study NCT00866281 (relapsed/refractory paediatric leukaemia, minimum age 3 months, sites in the US, France, Italy, the Netherlands and Sweden) ran from September 2009 to September 2014 and enrolled 22 patients in total -- about 4.4 per year across six countries -- and was terminated with the registry reason 'Despite considerable efforts to boost recruitment during the final year of the study, no new patients were enrolled.' Read from the ClinicalTrials.gov API v2 record on 2026-09-18 (status TERMINATED, enrolment 22 ACTUAL). Any sample-size arithmetic that converts an incidence denominator into an accrual rate without an attrition factor will be wrong by a large multiple. Novartis Pharmaceuticals 2014
    • A KMT2A-specific instance of the same failure. St Jude's phase 2 of bortezomib and vorinostat in refractory or relapsed MLL-rearranged haematological malignancies (NCT02419755, maximum age 21, single US site) opened 2015-04-14 and closed 2016-12-31 having enrolled 12 patients, with the registry reason 'Accrual goals were no longer feasible based on restrictions imposed by the DSMB.' Read from the ClinicalTrials.gov API v2 record on 2026-09-18. This is the disease this node is about, and the realised accrual was roughly 7 per year at a single high-volume referral centre. St. Jude Children's Research Hospital 2016
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Interfant-06 accrued 651 infants with newly diagnosed acute lymphoblastic leukemia through an international randomised phase 3 consortium study.

    PMID 31283407DOI
    What this evidence cannot show

    Accrual spanned roughly a decade; the per-year rate used here is an average and actual accrual varied. The KMT2A-rearranged fraction of Interfant-06 is taken from the general infant ALL literature rather than recomputed from the paper.

    Evidence that argues against this (2)
    • This node's central input, ~48 KMT2A-rearranged infants per year of international accrual, is roughly 45% too high against the consortium's own current protocol. Interfant-21 (NCT05327894) is restricted to infants under one year with KMT2A-rearranged ALL or MPAL, runs across 24 countries, and carries a planned enrolment of 160 with a start date of 2022-12-15 and a primary completion date of 2027-09: about 33 KMT2A-rearranged infants per year, not 48. Substituting 33 multiplies every accrual-year figure in this node by ~1.45 -- the HR 0.66 design moves from 6.3 to ~9.2 years of accrual, the 'feasible' HR 0.47 design from 1.9 to ~2.8 years, and the Simon two-stage design from ~10 to ~14 months. Read from the ClinicalTrials.gov API v2 record on 2026-09-18 (status RECRUITING, enrolment 160 ESTIMATED). Princess Maxima Center for Pediatric Oncology 2026
    • The consortium's planned size has been falling, not holding steady, which matters because this node treats Interfant-06 accrual as a stable resource. Interfant-99 (NCT00015873, Dutch Childhood Oncology Group, 11 countries, RANDOMIZED phase 3) planned 350 infants with a start date of 1999-05 and a primary completion of 2006-02; Interfant-06 accrued 651; Interfant-21 plans 160 and is registered with allocation NON_RANDOMIZED and intervention model SEQUENTIAL. The world's infant-ALL consortium moved from a randomised design in 1999 to a non-randomised one in 2022 while halving planned enrolment. Both records read from ClinicalTrials.gov API v2 on 2026-09-18. Dutch Childhood Oncology Group 2009
  3. E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    A single-arm study added one course of blinatumomab to Interfant-06 chemotherapy in 30 newly diagnosed infants, met its primary safety endpoint with no protocol-defined toxic effects, and reported better outcomes than historical Interfant-06 controls at a median follow-up of 26 months.

    Blinatumomab Added to Chemotherapy in Infant Lymphoblastic Leukemia · van der Sluis IM, de Lorenzo P, Kotecha RS, et al. · New England Journal of Medicine 2023
    PMID 37099340DOI
    What this evidence cannot show

    Thirty patients, single arm, historical comparison, short follow-up. The efficacy signal is suggestive rather than established, which is exactly the trade this node is quantifying. **Regraded strong -> established on 2026-09-20** under `concept-the-confidence-grading-rule`: this is the trial's own reported result, which is established as a fact about what the trial found.

    Evidence that argues against this (2)
    • The node's first actionable conclusion -- 'Do not propose a randomised frontline infant trial' -- is contradicted by a randomised frontline infant trial that opened while this beat was being written. NCT06317662, sponsored by the National Cancer Institute, is 'A Phase 2 Study of Blinatumomab in Combination With Chemotherapy for Infants With Newly Diagnosed Acute Lymphoblastic Leukemia With Randomization of KMT2A-Rearranged Patients to Addition of Venetoclax': maximum age 365 days, allocation RANDOMIZED, planned enrolment 153, start 2025-06-05, primary completion 2028-12-31, sites in the United States and Puerto Rico only. That is ~43 infants per year planned in North America alone, and the randomisation is specifically within the KMT2A-rearranged stratum. Read from the ClinicalTrials.gov API v2 record on 2026-09-18 (status RECRUITING). The node's arithmetic may still be right about what such a trial can detect, but its conclusion that the design cannot be proposed or accrued is empirically false as of 2025. National Cancer Institute (NCI) 2026
    • Evidence in the opposite direction to the accrual pessimism above: a broad enough international network can accrue an ultra-rare infant genotype fast. NCT06679582 (Sutro Biopharma, luveltamab tazevibulin in infants and children under 12 with CBFA2T3::GLIS2 AML, minimum age 1 day) lists 36 sites across the United States, Canada, Austria, Denmark, France, Germany, Italy, the Netherlands and Spain; it started 2024-12-04 and by its completion date of 2025-06-01 had an actual enrolment of 24, roughly four children a month, for a fusion rarer than KMT2A rearrangement. Its registry reason for stopping is the uninformative 'Terminated Early', so nothing should be inferred about why it closed. Read from ClinicalTrials.gov API v2 on 2026-09-18. The binding constraint on paediatric accrual therefore looks like network breadth rather than the disease's incidence. Sutro Biopharma et al. 2025
  4. E4established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    Historical control information can be summarised into a meta-analytic-predictive prior whose effective sample size quantifies how many patients' worth of information is being borrowed, with robust mixture formulations limiting the damage if the historical and current populations differ.

    Summarizing historical information on controls in clinical trials · Neuenschwander B, Capkun-Niggli G, Branson M, Spiegelhalter DJ · Clinical Trials 2010
    PMID 20156954DOI
    What this evidence cannot show

    A methods paper, not an application to infant leukemia. The effective sample sizes quoted in this node (20-40) are illustrative of what such a prior can contribute and are not derived from an actual fitted prior on KMT2A data.

    Evidence that argues against this (1)
    • The node treats drug cost as a non-issue at ~$240,000 per infant course, but the only published economic model of a menin inhibitor uses a much smaller number and a much shorter exposure. The Syndax-sponsored budget-impact analysis in the Journal of Managed Care & Specialty Pharmacy assumes revumenib drug acquisition of $36,867 per 28-day cycle (wholesale acquisition cost from RED BOOK, 2024 USD, US commercial payer perspective) and a median duration of treatment of 2.39 months, giving a realised per-patient drug cost on the order of $95,000 rather than $240,000-$474,000. Abstract read via PubMed and cost inputs, treatment duration, funding and the explicit exclusion of paediatric patients read from the open-access full text at PMC12205252. If the arithmetic in the node's conclusion is redone at $36,867 per cycle and 2.39 months, 300 infants cost ~$29 million, not $72 million. Abraham I et al. 2025
Connected to