Fever and infection: the reason most of these babies see a doctor, and the reason some of them die
also called fever · recurrent infections · neutropenia · febrile neutropenia · sepsis at presentation · induction death
Fever is present in about 53% of children with leukaemia and recurrent infections in about 49%, and it is usually fever, not anything more dramatic, that finally brings the baby to hospital. Infection is also the main thing that kills infants during the first weeks of treatment: one trial cut deaths during the first induction from about 15% to under 2% almost entirely by treating infection better.
The same thing, for a scientist
Fever at presentation has two causes that cannot be distinguished clinically: genuine infection in a functionally neutropenic child, and tumour fever from cytokine release. Fever is recorded in 53% and recurrent infections in 49% of children with leukaemia in the presenting-features meta-analysis. The greater significance is therapeutic: infants tolerate intensive induction badly, and infection - not leukaemia - was the dominant cause of early death. On COG AALL0631, de-intensifying induction and adding a supportive-care package including inpatient management until marrow recovery cut induction mortality from 4/26 (15.4%) to 2/123 (1.6%) (p = 0.009) and sterile-site infections from 57.7% to 19.5% (p = 0.0002), while complete response rates rose. All induction deaths bar one were infection-related, with Aspergillus fumigatus, Candida tropicalis, C. parapsilosis, Pseudomonas aeruginosa, Bacillus cereus and Enterococcus faecium among the organisms. This is the clearest quantified statement in infant ALL of the principle that intensity has a ceiling set by toxicity rather than by anti-leukaemic effect.
Why a leukaemic infant is infected before treatment starts
The white cell count can be 200 x 10^9/L and the child still be functionally neutropenic, because almost all of those cells are blasts with no microbicidal capacity. Add three infant-specific vulnerabilities and the picture is complete: immunological immaturity (relative hypogammaglobulinaemia after maternal IgG wanes, immature complement and neutrophil function), mucosal and skin barriers that are easily breached, and a caregiving environment full of viral exposure. Presentations are therefore often a string of 'viral illnesses' or persistent oral candidiasis before anyone checks a count.
| Feature | Frequency | Population |
|---|---|---|
| Fever | 53% | childhood leukaemia, 33 cohorts, n = 3084 1 |
| Recurrent infections | 49% | same 1 |
| Pallor | 54% | same 1 |
| Asymptomatic at diagnosis | 6% | same 1 |
Fever cannot be assumed infective. Tumour fever from cytokine release is genuine and common in high-burden leukaemia, but it is a diagnosis of exclusion made after cultures and treatment, never at the bedside - so in practice every febrile child with suspected leukaemia receives broad-spectrum antibiotics.
The AALL0631 experiment: the most important supportive-care result in infant ALL1 deeper
COG AALL0631 originally used a 3-week induction based on P9407. Early mortality was unacceptable and the protocol was amended mid-study to a less intensive 5-week Interfant-99-based induction with lower daunorubicin dosing, native L-asparaginase substituted for pegaspargase, an added 7-day prednisone prophase, omission of cyclophosphamide and filgrastim, plus enhanced supportive-care guidelines including hospitalisation through induction until marrow recovery 2.
| Cohort 1 (intensive) | Cohort 2 (de-intensified + supportive care) | p | |
|---|---|---|---|
| n | 26 | 123 | |
| Induction death | 4 (15.4%) | 2 (1.6%) | 0.009 |
| Sterile-site infection | 15 (57.7%) | 24 (19.5%) | 0.0002 |
| Complete response at end of induction intensification | 17/25 (68%) | 94/100 (94%) | 0.0012 |
The direction of the last row is the part that should be read twice. Less intensive therapy produced a higher complete-response rate, because the deaths and the dose delays caused by infection were themselves reducing the delivered anti-leukaemic dose. All induction deaths but one were infection-related; the organisms included Aspergillus fumigatus (sinus, lung), Candida tropicalis and C. parapsilosis, Pseudomonas aeruginosa, Bacillus cereus and Enterococcus faecium 2 - a fungal- and Gram-positive-heavy list typical of prolonged neutropenia with central venous access.
The same lesson from the randomised trial
Interfant-06 made the same point with a randomisation rather than an amendment: adding myeloid-style consolidation courses (ADE and MAE) to the backbone gave a 6-year disease-free survival of 39.3% against 36.8% for the lymphoid course IB, log-rank p = 0.47, and overall outcome in Interfant-06 did not improve on Interfant-99 3. Two decades of escalation bought nothing. That is the empirical basis for the entire targeted-therapy argument in this disease: the chemotherapy ceiling is set by what a baby can survive, and it has been reached.
What an acceptable new therapy has to look like
If infection is the limiting toxicity, then the value of a targeted agent is measured partly in myelosuppression avoided. A menin inhibitor works by releasing a differentiation block rather than by ablating the marrow, so in principle it can reduce leukaemic burden without the prolonged aplasia that produced the AALL0631 deaths - and the blinatumomab pilot has already shown that an immunotherapy can be added to the Interfant backbone with only 10 serious adverse events (fever 4, infection 4, hypertension 1, vomiting 1) in 30 infants and no toxic effect meeting the primary endpoint 4. The counter-consideration is that menin inhibitors carry differentiation syndrome, which presents with fever, and QTc prolongation; in a febrile neutropenic infant, distinguishing differentiation syndrome from sepsis is an unsolved clinical problem.
| sign | fever and infection |
|---|---|
| mechanism | functional neutropenia despite high total white count, plus infant immunological immaturity and barrier vulnerability; tumour fever from cytokine release is a diagnosis of exclusion |
frequencies
[
{
"finding": "fever at presentation",
"pct": 53,
"population": "childhood leukaemia 0-18 y, n=3084"
},
{
"finding": "recurrent infections",
"pct": 49,
"population": "childhood leukaemia 0-18 y, n=3084"
}
]induction mortality
[
{
"cohort": "COG AALL0631 cohort 1 (intensive 3-week induction)",
"n": 26,
"deaths": 4,
"pct": 15.4
},
{
"cohort": "COG AALL0631 cohort 2 (de-intensified 5-week induction + supportive care)",
"n": 123,
"deaths": 2,
"pct": 1.6,
"p_value": 0.009
}
]sterile site infection PCT — the numbers behind the chart
{
"cohort1": 57.7,
"cohort2": 19.5,
"p_value": 0.0002
}complete response PCT
{
"cohort1": 68,
"cohort2": 94,
"p_value": 0.0012,
"note": "assessed at week 7 vs week 9 respectively"
}organisms causing induction death
[ "Aspergillus fumigatus", "Candida tropicalis", "Candida parapsilosis", "Pseudomonas aeruginosa", "Bacillus cereus", "Enterococcus faecium" ]
supportive care measures
[ "inpatient management through induction until marrow recovery", "antifungal and antibacterial prophylaxis protocols", "central-line care", "prompt empirical broad-spectrum antibiotics for fever" ]
- The chemotherapy ceiling in infants is set by infection, not by anti-leukaemic effect, so any targeted agent that adds even modest myelosuppression to the backbone can cost more in infection deaths than it gains in leukaemia control.
- Differentiation syndrome from menin inhibition presents with fever and respiratory compromise, which is indistinguishable at the bedside from sepsis in a neutropenic infant; the two have opposite management implications for continuing the drug.
- Prolonged neutropenia selects for fungal and resistant Gram-positive organisms, as the AALL0631 death list shows, so infection risk is not linear in duration - a small extension of aplasia can cross a threshold into invasive fungal disease.
- If an effective targeted agent permits de-escalation of chemotherapy, the de-escalation itself must be tested, and a non-inferiority trial in a population of a few hundred infants a year worldwide may be impossible to power.
- How much of the survival gap between high-income and other Interfant-06 participating groups (6-year EFS 49.4% versus roughly 10-12% lower) is attributable to infection deaths rather than to leukaemia biology? If most of it, then supportive care is the highest-yield global intervention in this disease.
- Can differentiation syndrome from a menin inhibitor be distinguished from sepsis in a febrile neutropenic infant by any available biomarker, and what is the safe default management when it cannot?
- Given that de-intensification improved both mortality and complete-response rate on AALL0631, what is the minimum chemotherapy backbone that a highly active targeted agent would need to be combined with - and is there a formal de-escalation trial design acceptable in a disease with 46% event-free survival?
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
In a systematic review and meta-analysis of 33 childhood leukaemia cohort studies (n = 3084, ages 0-18), fever was present in 53% of children and recurrent infections in 49%, alongside hepatomegaly 64%, splenomegaly 61%, pallor 54% and bruising 52%; 6% were asymptomatic at diagnosis.
Clinical presentation of childhood leukaemia: a systematic review and meta-analysis · Clarke RT, Van den Bruel A, Bankhead C, et al. · Archives of Disease in Childhood 2016What this evidence cannot show
Cohort studies without control groups, so these are frequencies among diagnosed children, not predictive values - fever is extremely common in well infants and its presence carries almost no diagnostic weight in isolation. Ages 0-18 pooled with no infant stratum. 'Recurrent infections' is a retrospectively assigned label vulnerable to recall bias once a leukaemia diagnosis has been made.
Evidence that argues against this (1)
- This meta-analysis covers ages 0 to 18, and infants are a small minority of its 3,084 children. Infant KMT2A-rearranged leukaemia presents differently from the disease that dominates the pooled figure — higher white cell counts, more CNS and extramedullary involvement, a pro-B rather than common-ALL phenotype — so a frequency computed across the whole paediatric range is not the frequency in the patients this graph is about. The signs are also chart-recorded retrospectively, which systematically under-records soft findings and over-records the ones that prompted referral. Statistics in Medicine 2006
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
On COG AALL0631, replacing a 3-week P9407-based induction with a less intensive 5-week Interfant-99-based induction (lower daunorubicin, native L-asparaginase instead of pegaspargase, added 7-day prednisone prophase, cyclophosphamide and filgrastim omitted) plus enhanced supportive care including hospitalisation until marrow recovery reduced induction mortality from 4/26 (15.4%) to 2/123 (1.6%) (p = 0.009) and sterile-site infections from 15/26 (57.7%) to 24/123 (19.5%) (p = 0.0002), with significantly fewer Gram-positive infections (p = 0.0002) and a higher complete response rate at end of induction intensification (94/100, 94% versus 17/25, 68%; p = 0.0012); all induction deaths were infection-related except one from progressive disease, with Aspergillus fumigatus, Candida tropicalis, Candida parapsilosis, Pseudomonas aeruginosa, Bacillus cereus and Enterococcus faecium among the causative organisms.
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, et al. · Pediatric Blood & Cancer 2015What this evidence cannot show
Sequential non-randomised cohorts, and cohort 1 had only 26 patients, so the 15.4% mortality rests on 4 deaths and the confidence interval around it is wide. The regimen change and the supportive-care package were introduced together and cannot be separated. The complete-response comparison uses different assessment timepoints (week 9 versus week 7), which biases it in favour of cohort 2. Nonetheless the mortality and infection reductions are large and in the same direction.
Evidence that argues against this (1)
- Five changes were made at once — lower daunorubicin, native L-asparaginase for pegaspargase, an added prednisone prophase, cyclophosphamide and filgrastim removed, hospitalisation until count recovery — against a historical comparator. Nothing in the design attributes the reduction in induction death to any one of them, and the supportive-care change alone is a plausible whole explanation. Leukemia 2021
- E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
In Interfant-06 (n = 651), early intensification with post-induction myeloid-type courses did not improve outcome compared with the lymphoid-type course IB (6-year disease-free survival 39.3% [SE 4.0, n = 169] versus 36.8% [SE 3.9, n = 161], log-rank p = 0.47), and overall outcome for infant ALL in Interfant-06 did not improve compared with Interfant-99.
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, et al. · Journal of Clinical Oncology 2019What this evidence cannot show
A negative randomised comparison shows no overall benefit but does not exclude benefit in a subgroup - and the companion MRD analysis subsequently found that infants with high end-of-induction MRD may in fact benefit from the myeloid arm while MRD-negative infants do better with the lymphoid arm, so 'intensification does not work' is too coarse a summary.
Evidence that argues against this (1)
- The Interfant-06 MRD analysis found opposite optimal consolidation arms for high- and low-MRD infants (6-year DFS 45.9% versus 23.2% for high MRD treated myeloid versus lymphoid; 78.2% versus 45.0% for MRD-negative treated lymphoid versus myeloid), so the randomisation's null result conceals a real interaction. source 2021
- E4established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
Adding one 28-day continuous infusion of blinatumomab to Interfant-06 chemotherapy in 30 infants with newly diagnosed KMT2A-rearranged ALL produced no toxic effect meeting the primary endpoint; 10 serious adverse events were reported (fever 4, infection 4, hypertension 1, vomiting 1) and the toxicity profile was consistent with that seen in older patients.
Blinatumomab Added to Chemotherapy in Infant Lymphoblastic Leukemia · van der Sluis IM, de Lorenzo P, Kotecha RS, et al. · New England Journal of Medicine 2023What this evidence cannot show
Thirty patients is too few to characterise the rate of uncommon toxicities, and the primary endpoint was deliberately narrow (a toxic effect attributable to blinatumomab causing permanent discontinuation or death), so 'no toxic effects meeting the definition' is a weaker statement than 'well tolerated'. Half of the serious adverse events were fever or infection, which is exactly the toxicity domain that limits infant therapy. **Regraded moderate -> 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 (1)
- No toxicity endpoint met in 30 patients bounds the rate loosely: zero events in 30 is compatible with a true rate up to about 10%. Ten serious adverse events in the same 30 infants, four of them infections, is the number that should anchor the safety statement, and single-arm toxicity against no concurrent control cannot separate drug from disease. Leukemia 2021
the 53% fever figure is ascertained in a population that can report and be asked; it does not transfer to infants
the reason most of these babies reach a doctor
presenting sign: neutropenic fever, the reason most infants reach medical attention
- co-occurs with →Hepatosplenomegaly: the swollen abdomen that is often the first thing a parent notices
both follow from the same process: blast expansion displacing normal haematopoiesis and seeding organs
persistent fever to 39.8 C with respiratory infection as the presenting illness, and fever again at relapse