Early deaths may inflate survival benefit of brain tumour therapy in young children, letter warns

A letter published in the Egyptian Journal of Neurosurgery argues that a recent study of brain tumours in children under three years old may have overstated the survival benefit of adjuvant chemotherapy and radiotherapy, because of a statistical problem known as guarantee-time bias. The letter, written by Nezihe Köker Özer, Ercan Yılmaz and Didem Tatar, responds to a study by Khalid and colleagues that described brain tumours in children younger than three years treated in a low- and middle-income setting, a population the authors note is rarely characterised in the literature.

The original study reported presentation, surgical management and outcomes for this age group, including a 30 percent in-hospital mortality rate, with one child dying before any surgical intervention could take place. Median overall survival varied sharply by tumour type: 2.4 months for medulloblastoma, the most frequent histology in the cohort, 3.2 months for atypical teratoid/rhabdoid tumour, and 0.4 months for embryonal tumour with multilayered rosettes. Data on brain tumours in children under three are scarce, and the letter writers acknowledge the value of documenting this burden in a setting where such data are seldom available.

The concern centres on how adjuvant therapy was recorded. In the cohort, adjuvant therapy was not a baseline characteristic of the children but an exposure that could only be registered in those who survived surgery and the early postoperative period. The letter notes that the original manuscript describes radiotherapy delivered after the first, second and, in one case, fourth operation. Children who died within these intervals were necessarily classified as untreated.

That classification creates the structural imbalance at the heart of guarantee-time bias. Because only survivors live long enough to receive and be recorded as receiving treatment, survival time accrues to the treated group by construction rather than by any true treatment effect. The letter stresses that the magnitude of this bias is greatest precisely where early mortality is high, as it is in this cohort, where nearly a third of children died in hospital.

The letter writers argue that the original authors’ own analysis suggests the treatment signal is fragile. In the study’s Table 6, chemotherapy carried a hazard ratio of 0.98 with a p value of 0.96, which the letter describes as no association whatsoever. The advantage attributed to radiotherapy, by contrast, rested on a Kaplan-Meier comparison for which no hazard ratio was reported. No histology-stratified treatment analysis was presented, even though the original conclusion described adjuvant therapy tailored to tumour histology.

According to the letter, this absence of stratification matters because histology governed both eligibility for adjuvant treatment and prognosis. An unstratified comparison, the writers contend, largely reflects the fact that pilocytic astrocytoma, with a median overall survival of 71.4 months, and the embryonal tumours occupy entirely different treatment pathways, rather than demonstrating a benefit of the treatments themselves.

The letter also distinguishes between two kinds of methodological concern. The original authors noted that their findings should be interpreted with caution given the retrospective design, a caveat that addresses confounding by indication. Guarantee-time bias, however, is structural rather than confounding, the letter argues, arising from how follow-up time is allocated, and it persists however carefully measured covariates are adjusted for in the analysis.

The letter writers attach clinical weight to this distinction because of the known harms of treatment in this age group. Radiotherapy in children under three years is associated with well-recognised neurocognitive, endocrine and developmental sequelae, a point the original authors themselves emphasised. A survival benefit that is partly an artefact of exposure timing, the letter warns, could shift multidisciplinary discussions towards earlier or more intensive irradiation in exactly the population in which restraint is most strongly advised.

As a remedy, the letter proposes reanalysis of the dataset. A landmark analysis restricted to children alive at a defined postdiagnosis time point, or a Cox model treating adjuvant therapy as a time-varying covariate, would show how much of the reported benefit survives correction, the writers argue. Reporting these results alongside histology-stratified estimates would, in their view, considerably strengthen the clinical implications drawn from what they describe as an otherwise valuable dataset.

The letter is published as open access, with the original article having appeared on 16 July 2026 and the letter on 22 October 2026.

Subject of Research: Guarantee-time bias in survival analyses of adjuvant therapy for brain tumours in children under three years

Article Title: Adjuvant therapy and survival in children under three years with brain tumours: accounting for guarantee-time

Article References: Köker Özer, N., Yılmaz, E., & Tatar, D. (2026). Adjuvant therapy and survival in children under three years with brain tumours: accounting for guarantee-time. Egyptian Journal of Neurosurgery, 41(1), Article 179. https://doi.org/10.1186/s41984-026-00686-9

Image Credits: AI Generated

DOI: 10.1186/s41984-026-00686-9

Keywords: guarantee-time bias, brain tumours, pediatric oncology, adjuvant chemotherapy, radiotherapy, medulloblastoma, survival analysis, Kaplan-Meier, landmark analysis, low- and middle-income countries, children under three, Egyptian Journal of Neurosurgery