# GlycA Task Card Task id: `glyca` Evidence status: grounded Evidence reviewed: 2026-08-15 ## Target Nightingale NMR GlycA concentration. Continuous-scalar regression. The default research stage is `00_00_visit`, but the Task can be configured for another available stage. Implementation facts: - target field: `GlycA` - target dataset: `nightingale` - target source: None (Nightingale's primary source is set in the loader, not on the Task) - target unit: mmol/L - default research stage: `00_00_visit` - valid range: 0.5-3.0 mmol/L; a PhenoBench plausibility filter, not a reference interval - primary metric: R2 ## Why This Matters GlycA is an NMR inflammatory biomarker reflecting glycosylated acute-phase proteins. It is more stable than hsCRP and has been associated with myocardial infarction and mortality in population cohorts. For PhenoBench, GlycA is a useful inflammation-axis target because it is measured in the same Nightingale panel as many metabolic biomarkers but is not just a standard lipid or glucose target. **What a row here does not support.** GlycA is an inflammatory-state marker measured on one NMR platform at one visit. A row is a same-visit association: it is not a diagnosis of any inflammatory condition, and it is emphatically not the myocardial-infarction or mortality prediction that the external GlycA literature reports - those are prospective endpoints in other cohorts, and citing them as motivation is not the same as claiming them here. ## Related HPP Measurements Demonstrated in HPP, and persisted in this repository rather than asserted: - **Two matched tracks are already scored on canonical test splits**, both from the 2026-07-29 deep dive in an internal review record: V1 CBC + demographics reaches test R² = 0.150 (95% CI [0.077, 0.217]) on a checksum-bound cohort of 2,760 with 8 features, and V1 DXA + demographics reaches R² = 0.120 (95% CI [0.071, 0.166]) on 6,580. TabSwift is top-ranked in both; the gradient-boosted and cross-validated linear rows sit within the intervals, so no method separation is established. - Both tracks exclude every CBC input and target history as a leakage guard, and the CBC track's native support before Task eligibility was 5,225 participants against the 2,760 that survive it - the eligibility narrowing is large enough to matter when comparing the two tracks' numbers. **Both tracks clear the demographic floor on the paired comparison**, which is the quantity that matters rather than the raw R2 above: CBC **ΔR2 = +0.074 (95% CI [+0.018, +0.127])** on n = 2,760 and DXA **ΔR2 = +0.059 ([+0.035, +0.085])** on n = 6,580. The eight-field CBC panel is therefore a real result, not a floor artefact. Proposed rather than shown: that a learned multimodal representation beats it. Nothing committed answers that, and the CBC delta is the number to beat. **These are `reviewable` rows, not `final` ones.** Measured across the committed leaderboards: all 787 rows carry `curation_state: reviewable` and none is `final`. A complete fresh v0 re-run is committed to, so every figure in this section is pre-re-run evidence that the re-run supersedes - persisted and citable today, not a frozen result. **Which curation states the Figure 1 export accepts is the export track's live question, not this card's**; read the two counts above, not a gate. ## Clinically Meaningful Variants **No categorical GlycA variant is defensible, and unlike the glycemic targets the reason is that no clinical threshold exists at all.** GlycA is a composite NMR signal from glycosylated acute-phase proteins; it has no diagnostic cutoff, no guideline-endorsed reference interval, and its published associations with myocardial infarction, stroke and mortality are reported per standard deviation or per quartile within each study cohort rather than against a fixed value. Two properties make a transferred cutoff worse here than elsewhere: - **The value is platform-defined.** GlycA is a Nightingale NMR quantity, not a standardised assay with an interconversion to another platform's inflammatory measure. A cutoff from one NMR cohort does not carry to another cohort's calibration, and it does not carry to hsCRP at all - the two are correlated markers of different things, and the card should not be read as licensing a swap. - **The distribution is shifted by transient state.** Acute infection and chronic inflammatory disease move GlycA substantially, and the default cohort applies no such exclusion, so any binning would place a state-dependent subgroup on one side of the line. Quartiles or standard-deviation bands are the honest way to report a distribution here, and they are a presentation choice, not a Task. The continuous target stays primary. ## Benchmark-Track Evidence Benchmark-Track Evidence reviewed: 2026-07-29. literature review prioritized DXA body composition as the strongest independent shared predictor panel: adiposity is consistently associated with GlycA through inflammatory state. The existing matched HPP deep dive also found useful CBC signal. These findings motivate separate DXA and CBC tracks; they do not establish causality. Same-assay Nightingale remains a ceiling only. ## Baselines And Ceilings The required floor is a no-modality-feature baseline: `NoFeaturesPredictor` contributes no modality features, and `DemographicFloorStrategy` fits age, sex, and BMI. The task code records a low demographic floor around R2 0.07, while the source material expected R2 0.10-0.20 before HPP estimation. Natural comparators include hsCRP if available, leukocyte/inflammatory blood markers, NMR engineered features, diet/microbiome features, and sleep/activity tracks. Same-panel NMR predictors need explicit leakage review. ## Caveats - GlycA is an inflammatory-state marker, not a disease diagnosis. - Acute infection, chronic inflammatory disease, medication, and batch/assay effects can dominate signal. - Nightingale features may contain near-direct correlates of GlycA. - A good GlycA score should be claimed as inflammatory biomarker signal, not MI or mortality prediction. ## Metrics Primary metric is R2. Secondary MAE and Spearman are useful. Audit views should include demographic-floor delta R2, split Ns, same-panel leakage exclusions, and sensitivity to participants with acute inflammatory outliers if available. ## References - Source material: `docs/task_cards/source_material/foundation-model-surrogate-targets-v1.md` - Riggs et al., 2022, GlycA and hsCRP associations with MI and ischemic stroke: - Gruppen et al., 2019, GlycA and mortality in PREVEND plus meta-analysis: - Kohn et al. (2025), the Human Phenotype Project metabolic abstract, *SLEEP*, . Local copy in **HPP documentation**: `assets/papers/hpp/kohn_2025_0054_the_human_phenotype_project_metabol/`. - Dullaart et al., 2015, GlycA, adiposity, and glucose-tolerance status: ## Open Questions - Is hsCRP available and mature enough to act as a clinical comparator? - Which Nightingale feature exclusions are needed to avoid measuring GlycA by proxy from the same panel? - Should inflammatory disease or acute-infection exclusions be part of the default cohort?