🧬 Aging Wiki

thymic-health-score

Properties1
aliasesthymic health, CT thymic health score, deep-learning thymic health, thymic involution imaging biomarker

4 min read

⚠️ Auto-extracted by Claude on 2026-06-09 — not verified against the full PDFs (DOIs 10.1038/s41586-026-10242-y and 10.1038/s41586-026-10243-x; closed-access, no local full text). The model architecture, training labels, and exact effect sizes are summarized from abstracts + secondary coverage and may be approximate. gap/no-fulltext-access

Thymic health score (CT-derived)

The thymic health score is a deep-learning imaging biomarker that quantifies the functional state of the adult thymus — its size, shape, and tissue composition (functional epithelial parenchyma vs. fatty involution) — from routine chest CT. Introduced by Bernatz, Aerts, Birkbak and colleagues in a pair of 2026 Nature papers 12, it operationalizes immune aging as a single scalable readout derivable opportunistically from scans acquired for unrelated indications.

It is, in effect, an organ-level aging clock for the thymus, and the first immune-aging imaging biomarker in this wiki to carry population-scale outcome validation. It complements the wiki’s existing immune-relevant biomarkers — neutrophil-to-lymphocyte ratio, IL-6 — which index the inflammatory arm of immune aging rather than the thymic-output arm.

What it measures

Thymic involution — progressive replacement of thymic epithelial parenchyma with adipose tissue beginning at puberty — is one of the earliest and most dramatic organ-aging processes and the upstream driver of falling naive-T-cell output (see thymus, immunosenescence § Thymic involution). On CT, this manifests as a shift from soft-tissue-density parenchyma toward fat density in the anterior mediastinum. The deep-learning model scores this composition (plus size/shape) into a continuous thymic health value; higher = more preserved functional thymus.

Molecular ground truth (TRACERx validation): the score correlates with TCR repertoire diversity, T-cell-receptor excision circles (TRECs) — a direct marker of recent thymic emigrants — and immune-signaling pathway activation 2, indicating it indexes genuine thymic function, not merely anatomy.

Model architecture and training labels are not verifiable from available text; model-architecture: composite-other and training-target: morbidity are best-fit placeholders (the score is validated against mortality/morbidity and immunotherapy outcomes; whether the network was trained on age, expert ratings, or a composition metric is unconfirmed). gap/no-fulltext-access

Prognostic performance (approximate — pending full-text verification)

In ~27,600 asymptomatic adults across NLST + Framingham 1, high vs. low thymic health was associated with:

OutcomeEffect
All-cause mortality~50% lower (12-yr ~13.4% vs ~25.5%, NLST)
Cardiovascular mortality~63% lower; replicated independently in Framingham
Lung-cancer incidence~36% lower (6-yr ~3.4% vs ~5.3%)

In a pan-cancer immune-checkpoint-inhibitor cohort (n ≈ 3,476) 2, higher thymic health predicted reduced progression and mortality in NSCLC, independent of PD-L1 and TMB — establishing it as a host-side complement to tumor-intrinsic immunotherapy biomarkers.

Strengths and limitations

Strengths: free/opportunistic (uses CT already acquired); population-scale validation; an independent cohort (Framingham) replicated the cardiovascular signal; molecular validation against TRECs/TCR diversity; captures the thymic-output axis that inflammatory markers (NLR, CRP, IL-6) miss.

Limitations:

  • Single group, single method, 2026 — not yet externally replicated by independent teams. human-evidence-level: limited reflects newness, not weak internal evidence. gap/needs-replication
  • Observational — predicts outcomes but is not shown to be causal; may partly mark general systemic health. gap/no-mechanism
  • Mendelian randomization not applicable — a structural/imaging readout, not a germline-instrumentable trait (cf. AGE-Reader skin autofluorescence precedent).
  • Intervention-responsiveness only partial/inferred — thymic-regeneration interventions (the Fahy 2019 TRIIM GH+DHEA+metformin pilot; IL-7; sex-steroid blockade; FOXN1) move thymic mass on MRI, so this CT score should respond, but no trial has yet used this specific score as an endpoint. gap/dose-response-unclear
  • Requires a chest CT; not derivable from blood, unlike methylation/proteomic clocks.

Relationship to other biomarkers

  • Indexes a different axis than methylation clocks (horvath-clock-2013, grimage-2019) or proteomic clocks — those are systemic/multi-tissue; this is organ-specific (thymus).
  • Mechanistically upstream of inflammatory indices: failing thymic output → naive-T-cell decline → compensatory inflammaging → elevated il-6, NLR.

See also

Footnotes

  1. bernatz-2026-thymic-health-adults · n≈27,612 (NLST + Framingham) · observational imaging cohort · model: deep-learning on chest CT · ~50% lower all-cause mortality (high vs low thymic health) · unverified against full text gap/no-fulltext-access ↩ ↩2

  2. bernatz-2026-thymic-health-immunotherapy · n≈3,476 pan-cancer ICB cohort + TRACERx validation · observational · model: deep-learning on CT; humans · thymic health predicts NSCLC immunotherapy outcomes independent of PD-L1/TMB · unverified against full text gap/no-fulltext-access ↩ ↩2 ↩3

Graph View

Backlinks