Clarification of the published survival analysis in H3F3B p.K27I-mutant diffuse midline glioma

Dear Editor,

Cheng et al. [2] recently reported the largest combined cohort of H3F3B p.K27I-mutant diffuse midline glioma (DMG) to date and provided molecular evidence that this rare tumor represents a biologically distinct subtype within H3K27-altered DMG. In reviewing their study, we attempted to reconstruct the Kaplan–Meier estimate shown for the H3F3B cohort in Cheng et al.’s Fig. 5 from the patient-level follow-up data reported in their Table 1. In doing so, we encountered an apparent discrepancy that seems difficult to reconcile and that may affect interpretation of the survival comparison.

Among the 11 H3F3B-mutant cases with follow-up in Table 1, deaths are reported at 7, 9, 18, and 40 months, whereas seven patients were alive at 2, 5, 6, 9, 10, 17, and 36 months [2]. Treating the living patients as right-censored observations and excluding the two cases without follow-up yields a standard Kaplan–Meier curve with four event-related downward steps (Fig. 1). The R code used for the Kaplan–Meier reconstruction is provided in Additional file 1. The reconstructed survival probability first reaches 0.50 at 18 months and remains at 0.50 until the final event at 40 months. Therefore, the numerical median is convention-dependent: assigning the median to the first time point at which the curve reaches 0.50 gives 18 months, whereas using the midpoint of this 0.50 plateau gives 29 months. To avoid overinterpreting this convention-dependent median, our emphasis is on the shape of the reconstructed curve: by visual inspection, the published H3F3B curve in Cheng et al.’s Fig. 5 appears to fall below 0.50 substantially earlier and to show more event-related downward steps than would be expected from four deaths.

Fig. 1.

Kaplan–Meier curve reconstructed from the patient-level follow-up times and vital status reported in Table 1 of Cheng et al. [2]. Living patients were treated as right-censored, and the two cases without follow-up were excluded. The reconstructed curve contains four event-related downward steps. The survival probability first reaches 0.50 at 18 months and remains at 0.50 until the final event at 40 months; using a midpoint convention for this 0.50 plateau gives a median overall survival of 29 months. Tick marks indicate censored observations

We may of course be missing an analytical detail not specified in the article, such as a different time origin, endpoint definition, censoring rule, or follow-up dataset. Importantly, our reconstruction was not based on digitization of a plotted curve or on aggregate survival summaries, but on the individual follow-up times and vital status entries published in Table 1 in Cheng et al. Thus, our point is that the patient-level data made available to readers in Table 1 Cheng et al. do not appear to reproduce the published Kaplan–Meier curve for the H3F3B cohort under standard right-censoring rules.

This issue matters because the article interprets H3F3B p.K27I-mutant DMG as having survival comparable to that of the spinal cord and brainstem H3K27M-mutant comparator cohorts shown in Cheng et al.‘s Fig. 5 [2]. We also note that the provenance and selection criteria of the H3K27M-mutant comparator cohorts in Cheng et al.’s Fig. 5, which appear to comprise 28 spinal cord and 57 brainstem cases according to the number-at-risk table, are not explicitly described in the article . Consequently, the basis for this survival comparison is difficult to evaluate independently. We do not argue that H3F3B p.K27I-mutant DMG has a favorable prognosis. Regardless of whether the convention-dependent median is described as 18 or 29 months, the reconstructed curve remains compatible with poor survival, and the small sample size precludes firm prognostic conclusions in either direction.

However, until the H3F3B survival curve and comparator cohorts can be reproduced from, or adequately explained by, the data and methods available to readers, the conclusion that survival is comparable to those cohorts remains difficult to assess independently. The published data also leave open the possibility of greater clinical heterogeneity within this rare subtype. In the published series, one patient was alive at 36 months and another died at 40 months [2]. Earlier reports have also included adult spinal tumors [4].

A cautious interpretation is also consistent with the available biological evidence. In peptide-based assays, K27I inhibits Polycomb Repressive Complex 2 less strongly than K27M [3]. Cheng et al. further observed complete or mosaic-like loss of H3K27me3 rather than a single uniform pattern [2]. More broadly, distinct H3K27 alterations need not have identical clinical behavior; for example, H3.1-K27M and H3.3-K27M diffuse intrinsic pontine gliomas show distinct outcomes [1]. These observations do not demonstrate a better prognosis for H3F3B p.K27I-mutant DMG, but they support caution when interpreting survival comparisons across biologically non-identical K27-altered tumors.

Our concern pertains to the reproducibility of the survival analysis from the data made available in the published article, not to the evidence that H3F3B p.K27I-mutant DMG represents a distinct subtype. We would therefore be grateful if the authors could clarify the time origin, endpoint definition, censoring rules, and dataset used to generate Fig. 5, as well as the provenance and selection criteria of the H3K27M-mutant comparator cohorts. Such clarification, or a corrected survival plot if necessary, would help readers interpret the prognostic implications of this rare tumor and would strengthen an otherwise important contribution.

Sincerely,

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

RK and HT conceived the study, extracted the published data, performed the survival reconstruction, and drafted the manuscript. HT, YY, and OA reviewed the analysis and critically revised the manuscript. All authors read and approved the final manuscript.

Funding

This work was supported by JSPS KAKENHI Grant No. 25K10905 and the Japan Agency for Medical Research and Development (AMED) under Grant Nos. 256f0137004j0001, 266f0137004j0002, and 276f0137004j0003..

Data availability

The data analyzed during the current study were extracted from Table 1 of Cheng et al. [2], which is publicly available in the published article. The R code used to reconstruct the Kaplan–Meier curve is provided in Additional file 1.

Declarations

Ethics approval and consent to participate

This Letter reports a re-analysis of anonymized patient follow-up data already published in Table 1 of Cheng et al. [2]. No new human participants were enrolled, and no identifiable participant information was accessed. Therefore, additional institutional ethics approval and consent to participate were not required for the present study.

Consent for publication

This Letter does not include any newly obtained individual participant data, images, or other identifiable personal information. The analysis is limited to data already published by Cheng et al. [2]. Therefore, additional consent for publication was not required.

Competing interests

The authors declare no competing interests.