Dying in induction: the place where infant protocols actually differ
also called induction death infant ALL · treatment-related mortality infant · early death infant leukaemia · supportive care infant induction
When one American trial made its first three weeks of treatment gentler and added strict supportive-care rules, deaths during that phase fell from 15.4% to 1.6% and remission rates went up, not down. Infant protocols differ from each other far more in how they keep babies alive through induction than in which drugs they use.
The same thing, for a scientist
The single largest published swing in infant ALL outcome attributable to a protocol change is not a drug. COG AALL0631 amended induction from a 3-week P9407-based schedule to a 5-week Interfant-99-based schedule and added supportive-care guidelines including hospitalisation until marrow recovery: induction mortality fell from 4 of 26 (15.4%) to 2 of 123 (1.6%), p = 0.009; sterile-site infections fell from 57.7% to 19.5%, p = 0.0002; and complete response at end of induction intensification **rose** from 68% to 94%, p = 0.0012 [E1]. All induction deaths but one were infection-related. The comparison holds across systems: induction fatal infections were 7.5% in CCCG-ALL-2015 against 3.8% in Interfant-99 as reported by the CCCG authors, with a cumulative incidence of fatal infection over the whole of therapy of 13.5% (95% CI 7.7-19.3) [E2]. And the direction reverses when intensity rises: Interfant-21 gave full doses from 6 months of age and **reverted to Interfant-06 dosing because of a higher rate of infectious deaths in induction** [E3]. De-intensification plus supportive care improved both toxicity and efficacy in AALL0631; intensification cost lives in Interfant-21. That is an unusually clean pair of natural experiments in a rare disease.
The AALL0631 amendment, in full
| Endpoint | Cohort 1 (3-week P9407-based induction) | Cohort 2 (5-week Interfant-99-based + supportive care) | p |
|---|---|---|---|
| n | 26 | 123 | - |
| Induction mortality | 4 (15.4%) | 2 (1.6%) | 0.009 |
| Sterile-site infection | 15 (57.7%) | 24 (19.5%) | 0.0002 |
| Gram-positive infection in induction | higher | significantly lower | 0.0002 |
| Grade 3-5 non-infectious toxicity | - | no clinically significant difference | - |
| Complete response at end of induction intensification | 17/25 (68%), week 7 | 94/100 (94%), week 9 | 0.0012 |
The last row is the one that matters most and is most easily misread. Making induction less intense did not cost remissions; it gained them, because dead patients and patients whose therapy is interrupted by sepsis do not achieve remission. The two cohorts are sequential rather than randomised, so part of the effect is secular improvement in supportive care - but the assessment timepoints also differ (week 7 vs week 9), which cuts the other way and favours cohort 2 for a second, purely technical reason 1.
The Interfant-21 reversion, which is the mirror image1 deeper
Dose reduction for infants has never been based on pharmacokinetic data; it is a safety convention resting on the relatively large body surface area of an infant 3. JPLSG MLL-10 changed the convention, giving full doses from 4 months of age, and reported a 3-year EFS of 66.2% (SE 5.6) for KMT2A-rearranged infants, the best in the field 4. Interfant-21 followed suit, giving full doses from 6 months of age - and reverted to Interfant-06 dosing after a higher rate of infectious deaths in induction 3.
So the same intervention - dosing infants more like children - produced the best reported EFS in one health system and excess induction deaths in another. The most parsimonious explanation is that the dose is not the variable; the supportive care around the dose is, and MLL-10's own investigators attribute their result partly to aggressive supportive-care guidelines 3.
What this implies for a targeted-agent trial
If induction mortality is 1.6% in one system and 7.5% in another for the same disease, then a single-arm trial of a new agent run in the better system will outperform a historical control from the worse one for reasons that have nothing to do with the agent. Every comparison in this beat against an Interfant historical control - including the blinatumomab pilot's jump from 49% to 82% 2-year DFS - inherits that confound.
The numbers across systems
| Cohort | Induction fatal infection / induction mortality | Whole-therapy fatal infection |
|---|---|---|
| COG AALL0631 cohort 1 | 15.4% | - |
| COG AALL0631 cohort 2 | 1.6% | - |
| Interfant-99 | 3.8% (as reported in 2) | - |
| CCCG-ALL-2015 | 7.5% (10 of 134) | 13.5% cumulative incidence (95% CI 7.7-19.3) |
| Interfant-21, full dosing from 6 months | higher than Interfant-06; dosing reverted 3 | - |
In CCCG-ALL-2015 there were 18 deaths in remission: 8 severe pneumonia, 7 septic shock, 2 invasive fungal infection, 1 third-degree atrioventricular block 2.
| metric | induction mortality and infection |
|---|---|
| population | infants <1 year with ALL |
cohorts — the numbers behind the chart
[
{
"cohort": "COG AALL0631 cohort 1",
"n": 26,
"induction_mortality_pct": 15.4,
"sterile_site_infection_pct": 57.7,
"cr_pct": 68,
"pmid": "25407157",
"title": "Decreased induction morbidity and mortality following modification to induction therapy in infants with acute lymphoblastic leukemia enrolled on AALL0631: a report from the Children's Oncology Group.",
"authors": "Salzer WL, Jones TL, Devidas M, et al.",
"venue": "Pediatr Blood Cancer",
"year": 2015,
"title_provenance": "Title/authors/venue backfilled from PubMed esummary on 2026-09-20 because this citation stored a bare PMID with no title, which made it unscoreable by verify-citations.py. BACKFILLING IS NOT VERIFICATION: it confirms the PMID resolves to a real paper, not that this is the right paper for the claim. That still requires reading."
},
{
"cohort": "COG AALL0631 cohort 2",
"n": 123,
"induction_mortality_pct": 1.6,
"sterile_site_infection_pct": 19.5,
"cr_pct": 94,
"pmid": "25407157",
"title": "Decreased induction morbidity and mortality following modification to induction therapy in infants with acute lymphoblastic leukemia enrolled on AALL0631: a report from the Children's Oncology Group.",
"authors": "Salzer WL, Jones TL, Devidas M, et al.",
"venue": "Pediatr Blood Cancer",
"year": 2015,
"title_provenance": "Title/authors/venue backfilled from PubMed esummary on 2026-09-20 because this citation stored a bare PMID with no title, which made it unscoreable by verify-citations.py. BACKFILLING IS NOT VERIFICATION: it confirms the PMID resolves to a real paper, not that this is the right paper for the claim. That still requires reading."
},
{
"cohort": "Interfant-99",
"induction_fatal_infection_pct": 3.8,
"source": "as reported in PMID 38634141",
"pmid": "17658395",
"title": "A treatment protocol for infants younger than 1 year with acute lymphoblastic leukaemia (Interfant-99): an observational study and a multicentre randomised trial.",
"authors": "Pieters R, Schrappe M, De Lorenzo P, et al.",
"venue": "Lancet",
"year": 2007,
"title_provenance": "Title/authors/venue backfilled from PubMed esummary on 2026-09-20 because this citation stored a bare PMID with no title, which made it unscoreable by verify-citations.py. BACKFILLING IS NOT VERIFICATION: it confirms the PMID resolves to a real paper, not that this is the right paper for the claim. That still requires reading."
},
{
"cohort": "CCCG-ALL-2015",
"n": 134,
"induction_fatal_infection_pct": 7.5,
"whole_therapy_fatal_infection_cip_pct": 13.5,
"ci95": [
7.7,
19.3
],
"pmid": "38634141",
"title": "Outcome of infants with acute lymphoblastic leukemia treated with the Chinese Children's Cancer Group Acute Lymphoblastic Leukemia 2015 study protocol.",
"authors": "Leung AWK, Cai J, Wan Z, et al.",
"venue": "Haematologica",
"year": 2024,
"title_provenance": "Title/authors/venue backfilled from PubMed esummary on 2026-09-20 because this citation stored a bare PMID with no title, which made it unscoreable by verify-citations.py. BACKFILLING IS NOT VERIFICATION: it confirms the PMID resolves to a real paper, not that this is the right paper for the claim. That still requires reading."
}
]efs — the numbers behind the chart
[
{
"cohort": "Interfant-99",
"years": 4,
"efs_pct": 47,
"se": 2.6
},
{
"cohort": "Interfant-06",
"years": 6,
"efs_pct": 46.1,
"se": 2.1
},
{
"cohort": "JPLSG MLL-10 KMT2A-r",
"years": 3,
"efs_pct": 66.2,
"se": 5.6
}
]- A new agent added to induction in a system with 1.6% induction mortality will look better than the same agent added where induction mortality is 7.5%, and neither result transfers to the other system.
- De-intensification improves survival by keeping patients alive to receive therapy, so a trial that de-intensifies and adds a targeted agent cannot attribute its gain to the agent.
- Menin inhibitors and BH3 mimetics both carry differentiation-syndrome and tumour-lysis risks that land in exactly the induction window where infants are least able to tolerate an additional insult.
- Hospitalising infants until marrow recovery, which is part of what made AALL0631's numbers move, has costs in nosocomial infection, family disruption and health-system capacity that no trial endpoint captures.
- How much of the 10-12 point EFS gap between West European/North American groups and other countries in Interfant-06 is supportive-care capacity rather than disease or protocol?
- Interfant-21's dosing reversion is reported in a review; has the excess induction infectious mortality that triggered it been published with numbers, and if not, why not?
- If induction mortality is the dominant modifiable variable, what is the maximum outcome improvement available from supportive care alone, and does it exceed the improvement any current targeted agent has demonstrated?
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
In COG AALL0631, replacing a 3-week P9407-based induction with a 5-week Interfant-99-based induction plus enhanced supportive care reduced induction mortality from 4 of 26 (15.4%) to 2 of 123 (1.6%) (p = 0.009) and sterile-site infections from 57.7% to 19.5% (p = 0.0002), while complete response at end of induction intensification rose from 68% to 94% (p = 0.0012); all induction deaths but one were infection-related.
Decreased induction morbidity and mortality following modification to induction therapy in infants with acute lymphoblastic leukemia enrolled on AALL0631: a report from the Children's Oncology Group. · Salzer WL, Jones TL, Devidas M, Dreyer ZE, Gore L, Winick NJ, et al. · Pediatr Blood Cancer 2015What this evidence cannot show
Sequential non-randomised cohorts, so calendar time confounds the protocol change and secular improvement in supportive care cannot be separated from it. Cohort 1 has 26 patients, so its 15.4% has a wide interval. The complete-response comparison is at different timepoints (week 7 vs week 9), which favours cohort 2 independently of the regimen.
Evidence that argues against this (1)
- The two cohorts are separated in time, and general paediatric critical care and antifungal practice improved over the same interval, so an unknown fraction of the mortality reduction is secular rather than protocol-driven. Leung AWK et al. 2024
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
In CCCG-ALL-2015, 10 of 134 infants (7.5%) died of infection during induction against 3.8% reported for Interfant-99; sepsis occurred in 44.8%, severe pneumonia in 17.9% and invasive fungal infection in 9.0%; 18 deaths in remission occurred (8 severe pneumonia, 7 septic shock, 2 invasive fungal, 1 complete heart block), with a cumulative incidence of fatal infection of 13.5% (95% CI 7.7-19.3).
Outcome of infants with acute lymphoblastic leukemia treated with the Chinese Children's Cancer Group Acute Lymphoblastic Leukemia 2015 study protocol. · Leung AWK, Cai J, Wan Z, Qin J, Fang Y, Sun L, et al. · Haematologica 2024What this evidence cannot show
The Interfant-99 comparator (3.8%) is quoted from this paper's discussion, not measured here. Single-country cohort with documented under-enrolment of high-risk infants; 18 of 134 (13.4%) discontinued protocol treatment, 10 on parental request, which competes with death as an outcome and complicates the mortality denominator. Read in the open-access full text at PMC11290514.
Evidence that argues against this (1)
- The high treatment-abandonment rate means the mortality figures are not comparable with trials where nearly all patients complete protocol therapy; some deaths after abandonment will not be counted as treatment-related in either place. Salzer WL et al. 2015
- E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.review
Age-based dose reduction in infant protocols is a safety convention not based on pharmacokinetic data; Interfant-21 gave full doses from 6 months of age and reverted to Interfant-06 dosing because of a higher rate of infectious deaths in induction, while JPLSG MLL-10's better outcome is attributed partly to aggressive supportive care and stricter age-related dosing.
KMT2A-rearranged acute lymphoblastic leukemia in infants: current progress and challenges. · Stutterheim J, Van der Sluis IM, Vrenken KS, Pieters R, Stam RW · Haematologica 2025What this evidence cannot show
A narrative review written by the Interfant investigators about their own trials; the Interfant-21 reversion is reported here and, as far as this beat could establish, nowhere as a primary publication with numbers. The attribution of MLL-10's result to supportive care and dosing is the authors' interpretation, not a tested comparison. Read in the open-access full text at PMC12399962.
Evidence that argues against this (1)
- MLL-10 and Interfant-06 differ in country, era, supportive-care infrastructure and patient mix as well as in dosing, so attributing the EFS difference to dose is a between-trial inference of the weakest kind; the review itself notes there are no pharmacokinetic data supporting under-dosing. Tomizawa D et al. 2020
- E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
JPLSG MLL-10 reported a 3-year event-free survival of 66.2% (SE 5.6) for infants with KMT2A-rearranged ALL and 93.3% (SE 6.4) for those with germline KMT2A, with 94.4% (SE 5.4) in the intermediate-risk group and 56.6% (SE 6.8) in the high-risk group.
A risk-stratified therapy for infants with acute lymphoblastic leukemia: a report from the JPLSG MLL-10 trial. · Tomizawa D, Miyamura T, Imamura T, Watanabe T, Moriya Saito A, Ogawa A, et al. · Blood 2020What this evidence cannot show
n = 90, with 15 low-risk, 19 intermediate-risk and 56 high-risk, so the intermediate-risk 94.4% rests on 19 patients and its standard error of 5.4 understates the fragility of a figure that close to the ceiling. Single-country. Abstract read; full text paywalled.
Evidence that argues against this (1)
- Interfant-06 with 651 patients reported 6-year EFS of 46.1%, and the MLL-10 advantage has not been reproduced outside Japan; a 90-patient single-country result at this effect size is the classic profile of a finding that regresses. Pieters R et al. 2019
- E5established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
Interfant-06 enrolled 651 infants with 6-year event-free survival of 46.1% (SE 2.1) and overall survival of 58.2% (SE 2.0), and outcome did not improve compared with Interfant-99; in West European and North American groups 6-year EFS and OS were 10-12 points higher than in other countries.
Outcome of Infants Younger Than 1 Year With Acute Lymphoblastic Leukemia Treated With the Interfant-06 Protocol: Results From an International Phase III Randomized Study. · Pieters R, De Lorenzo P, Ancliffe P, Aversa LA, Brethon B, Biondi A, et al. · J Clin Oncol 2019What this evidence cannot show
The 10-12 point geographic gap is reported without adjustment for patient mix, so it mixes supportive-care capacity with case ascertainment and with what proportion of patients reach a trial centre at all. Abstract read; full text paywalled.
Evidence that argues against this (1)
- A 10-12 point outcome gap between regions on the same protocol is itself an argument that supportive-care capacity, not protocol design, is the dominant modifiable variable in infant ALL - which would make drug-development comparisons across regions largely uninterpretable. Salzer WL et al. 2015
induction mortality and infection in infant ALL
- related to →Interfant-21 (NCT05327894) — the randomised test of blinatumomab in infant KMT2A-rearranged ALL, open now
full dosing from 6 months reverted after excess infectious deaths in induction
- related to →Interfant-06 — 651 infants, and the definitive answer that myeloid-style consolidation does not help
6-year EFS 46.1%; 10-12 point regional gap on the same protocol
the infection numbers that do exist are treatment-phase, not presentation-phase
- ← related toNine survivors in ten carry a long-term complication, and it is worse the younger they were
blinatumomab era: no post-induction infection-related fatality against 4% in Interfant-06