Age-specific incidence and outcome by age — four CNS histologies

Direct answer

1 · Pilocytic astrocytoma — the histogram

Age-adjusted incidence per 100,000 population. The curve rises from infancy to a peak at 1–4 years, then declines monotonically through every subsequent published band. Note the under-1 rate is markedly lower than the peak — the histogram is not highest at birth.

2 · The three that cannot be charted

Shown as annotated panels rather than empty or zero bars. Zero bars would be read as zero incidence, which is false in every case — these tumours occur, they are simply not counted separately.

Can the four share one chart?

No. Three separate reasons, any one sufficient:

1. Three of the four have no incidence data at all. Plotting counts or series age-distributions on the same axes as a per-100,000 rate silently converts relative frequency into apparent incidence.
2. Band structures do not align. CBTRUS main-report strata are 0–14 / 15–39 / 40+. The outcome papers use different bands again — 0–13/14–22/23–38/≥39 for PXA, 0–19/20–39/40–59/60+ for PA, <18 vs adult for ganglioglioma, ≤9 vs >9 for DLGNT. None reconcile without interpolation.
3. Even the PA series spans three data windows.

3 · Outcome by age

Pilocytic astrocytoma — population level

Two independent SEER analyses agree on direction: outcome worsens monotonically with increasing age at diagnosis, steepest after 60.

• Tomita 2023 (PMID 36063258) — SEER 2000–2018, N=5,211, 462 PA-specific deaths; bands 0–19/20–39/40–59/60+. Observed mortality peaked in the 80–84 stratum. Age remained significant alongside tumour location and race/ethnicity.
• Yang 2022 (PMID 34875553) — SEER 1983–2016, N=4,357. Adult age independently adverse in a Fine-Gray competing-risk model containing extent of resection, location, extension and era.

Ganglioglioma — overall survival only

• Adults ≥40 vs <40: HR 4.225 (95% CI 2.540–7.026, p<0.001), adjusted for sex, site and treatment. N=703, SEER 18, 2004–2016 (Lin 2021).
• Paediatric age <1 y: significantly worse OS alongside brainstem location. N=348 (Dudley 2015, PMID 25603107) — but the infant estimate rests on roughly 12 children (3.5% of cohort).
• Anaplastic ganglioglioma: median OS 28.5 months, N=58 (Selvanathan 2011, PMID 21626070). Age was tested and was NOT an independent predictor — only resection and unifocal disease were.

Interpretation. The association is genuinely U-shaped, but evidence for age as an independent biological factor is weak. The most informative internal check is the anaplastic cohort: with grade held constant, age dropped out. That argues age is largely reading grade, resectability and location. Temporal-lobe tumours are the most resectable and had the best outcome. Radiotherapy associated with worse OS (GTR+RT HR 5.074) — confounding by indication, not treatment harm.

Pleomorphic xanthoastrocytoma

• SEER 18, 1994–2016, N=470, grade-blind: age ≥39 adjusted HR 3.78 (2.16–6.59), adjusted for GTR, size, location, RT, chemo, demographics. Bands are quartiles 0–13/14–22/23–38/≥39 (Dono 2021).
• NCDB 2004–2016, N=546 adults, WHO grade II only — the only grade-2-restricted population cohort: median OS 128.6 months; age ≥65 HR 2.20 (Khalafallah 2020). Grading is registrar-abstracted, not central review.
• Paediatric ≤18, NCDB, N=224: subtotal vs gross-total resection HR 17.44 (p<.001) — the largest single effect reported anywhere for this tumour (Scarpelli 2021).
• Grade effect: 5-year OS 80.8% (grade 2) vs 47.6% (anaplastic), p=0.0009 (Vaubel 2021).

Interpretation. The age HR persisted after adjustment for GTR, which is the strongest evidence for partial independence. But the largest cohort excluded WHO grade as a variable, because anaplastic PXA grade 3 was not introduced until 2016 — the year the study period ended. Since anaplasia becomes more frequent with age, the unadjusted age effect is substantially a grade effect. Hazard ratios are not survival rates and cannot be converted into one for charting.

Diffuse leptomeningeal glioneuronal tumour

The evidence base is four heavily overlapping publications — the same cases recur across Wiśniewski 2022, Kweh 2025 and Lee 2026. These are not four independent confirmations.

• Age >9 vs ≤9 y retained as one of four independent predictors of poor OS, alongside Ki-67 ≥7%, raised-ICP presentation and contrast-enhancing intraparenchymal disease. N=63 (Wiśniewski 2022).
• Paediatric ≤18 vs adult >18: no significant OS difference — age not prognostic at this dichotomy. N=75 (Lee 2026).
• All ages: 5-year OS 83.3% ± 8.8, 5-year PFS 15.9% ± 8.0, N=30 (Mikkelsen 2025); 5-year OS 81.8%, 5-year PFS 48.6%, N=87 IPD meta-analysis (Kweh 2025).

Interpretation. Treat age >9 as a soft risk marker, not an established prognosticator. The 9-year threshold is a data-derived optimised cutpoint from samples of n=30 and n=63, externally unvalidated. No hazard ratio with a confidence interval is published for age in any DLGNT cohort — the effect exists only as a p-value. Since KIAA1549::BRAF fusion is present in 27/28 tested (96.4%) and therefore defines the entity rather than stratifying it, age is not tracking driver genotype.

Is age independent, or a proxy?

For all four: partially independent, incompletely adjusted. Age survives multivariable adjustment in PA, ganglioglioma and PXA — but only against the variables registries happen to record. Three things no registry captures, all of which vary with age:

• Molecular profile. SEER, NPCR, NCDB and CBTRUS contain no molecular data. KIAA1549::BRAF fusion, BRAF V600E and CDKN2A/B deletion are all invisible. Every published age effect is unadjusted for them.
• Diagnostic misclassification in older adults. A meaningful fraction of registry-coded 9421 over age 50–60 is likely diffuse glioma or, under CNS5, high-grade astrocytoma with piloid features (HGAP) — an entity that did not exist for most of these registry years, occurs preferentially in adults, and has far worse survival. This alone can manufacture the elderly mortality signal.
• Competing mortality and treatment intensity. A mortality peak at 80–84 in a WHO grade 1 tumour strongly hints at non-tumour drivers plus lower rates of attempted GTR in comorbid patients.

4 · What the data cannot tell you

Histology coding and grouping

• PA rates are 9421 + 9425, not pure 9421 — pilomyxoid astrocytoma is folded in. Contamination is small (138 cases, 0.01/100,000) but concentrated in the youngest children, i.e. precisely in the peak bands being charted.
• Behaviour-code artefact. WHO calls 9421/1 non-malignant, but US registries have historically reported it with behaviour /3, so CBTRUS counts PA within malignant totals. Analyses restricted to malignant behaviour capture a selected, worse-prognosis subset.
• Registry cell suppression would blank several age bands in any bespoke breakout.

WHO CNS5 versus data vintage — the most consequential limitation

• CBTRUS groupings, including the 2025 report covering 2018–2022, are still built on the 2016 WHO Classification. Verbatim: “CBTRUS is using histopathology groupings according to 2016 WHO Classification.” Every rate above reflects a WHO 2016 case definition regardless of diagnosis year.
• PA: CNS5 introduced HGAP with no dedicated ICD-O-3 code, so HGAP is still absorbed into the pilocytic line. A clean CNS5-defined pilocytic astrocytoma incidence rate does not currently exist in any population registry.
• Ganglioglioma: SEER years 2004–2018 entirely predate CNS5, which carved out PLNTY, myxoid glioneuronal tumour and diffuse low-grade glioma MAPK-altered. The Chen 2024 rising trend (APC +3.31) is therefore not safely interpretable as a true rise in disease frequency.
• PXA: some tumours called grade 2 historically would be graded 3 today — biasing historical grade-2 survival downward.
• DLGNT: cases 2012–2025 were diagnosed under at least two definitions and, before 2016, under none. Must be distinguished from leptomeningeal gliomatosis, which has a worse prognosis; historical contamination inflates apparent mortality in older series.

Incidence versus relative frequency

Constantly conflated for exactly these four entities. None of the following are incidence, and none can be converted into it:

• “PA is the most common histopathology in ages 0–14”; “2.3% of primary brain tumours at 15–39”.
• “Ganglioglioma is ~1–2% of paediatric CNS tumours”; “~10–40% of temporal-lobe epilepsy specimens” — the latter carries severe referral selection.
• “PXA is <1% of astrocytomas.”
• DLGNT median age 7.5 y — describes who reached referral centres.
• All BRAF-stratified prevalence figures. There is no published population incidence of fusion-positive versus fusion-negative anything.

Only the per-100,000 values in the pilocytic astrocytoma table are incidence.

5 · Suggested next steps

1. Complete the PA histogram — fastest, highest yield

Institutional library retrieval of Neuro-Oncology 2025;27(Suppl 4):iv1–iv66 (Price M et al., CBTRUS 2018–2022, doi 10.1093/neuonc/noaf194) — specifically Table 6 for the all-ages rate and median age, plus supplementary tables for any 40+ band. Paywalled at Oxford, not in PMC.

2. Build the ganglioglioma incidence analysis that does not exist

A SEER*Stat frequency-and-rate session: morphology 9505 (behaviours /0, /1, /3 reported separately), sites C70.0–C72.9 plus C75.1 per CBTRUS convention, against SEER-22 population denominators, diagnosis years 2004–2022, age-adjusted to the 2000 US standard, 5-year bands with counts and 95% CIs. Expect small-cell suppression in several bands.

SEER*Stat is free but needs a signed data-use agreement (seer.cancer.gov/data/access.html). For full US coverage rather than SEER-only, request a custom analysis from CBTRUS directly — they hold the combined CDC NPCR + NCI SEER base and can run entity-level breakouts the published report does not contain. To our finding, this analysis has never been published.

3. PXA — accept grade 2 is unobtainable from registries

An extract on 9424/3 gives PXA incidence but cannot separate grade 2 from grade 3, because no code exists. For grade- and molecularly-resolved data, request the DKFZ/Heidelberg CNS methylation classifier reference cohort (molecularneuropathology.org, Capper/Sahm), or approach the authors of Vaubel 2021 (Mayo, Brain Pathol 31(1):20–32). Both are referral cohorts — relative frequency, never incidence.

4. DLGNT — a registry query will not work

Do not commission a SEER extract; code 9509 is shared and non-identifying, so the query cannot succeed regardless of budget. Instead contact the corresponding author of Mikkelsen 2025 (PMID 40788548), and approach the Children's Oncology Group and SIOPE rare-CNS-tumour working groups for prospective registry inclusion — a prospective consortium registry is the only mechanism that will ever produce a DLGNT denominator.