# Lean Body Mass Task Card Task id: `lean_body_mass` Evidence status: grounded Evidence reviewed: 2026-07-29 ## Target DXA-derived total lean body mass. 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: `body_comp_total_lean_mass` - target dataset: `body_composition` - target source: None - default research stage: `00_00_visit` - valid range: 10000.0-150000.0 g - primary metric: R2 ## Why This Matters Lean body mass is relevant to sarcopenia, frailty, mobility, metabolic reserve, and aging. It is a useful body-composition target, but absolute total lean mass is heavily determined by sex, height/body size, and BMI. For PhenoBench, this is mainly a secondary body-composition probe. The claim to care about is added signal beyond crude anthropometrics. ## 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 very high demographic floor around R2 0.78. Natural comparators include plasma metabolomics, objective activity, grip strength, and frailty-system age-clock panels. A row that does not clear the floor is not informative. ## Benchmark-Track Evidence The frozen v0 tracks use the fixed nine-field V1 Nightingale amino-acid panel (`Ala`, `Gln`, `Gly`, `His`, `Ile`, `Leu`, `Val`, `Phe`, `Tyr`) plus V1 age, sex, and BMI for both V1 and V2 lean-mass targets. literature and literature support amino-acid and branched-chain-amino-acid associations with muscle mass and strength. Same-scan regional DXA fields remain excluded. Gait Fusion is not v0-benchmark-ready. Raw Newton skeleton files exist on S3, but no canonical operational source contract, checksum-bound embeddings, or accessible checkpoint was found. The available engineered-gait V1 cohort overlaps only 448 V1 and 113 V2 eligible lean-mass targets, so gait remains a future blocked candidate. ## Caveats - Absolute lean mass is strongly confounded by sex and body size. - DXA scan quality, device/calibration, hydration, and segmentation can add noise. - Total lean mass is not the same as appendicular lean mass index or sarcopenia diagnosis. - Same-DXA predictor features are likely leakage. ## Metrics Primary metric is R2. Secondary MAE is most readable in kg after converting the stored gram-scale target. Audit views should include demographic-floor delta R2, sex-stratified residuals, and support-cohort matching when comparing modality predictors. ## References - Source material: `docs/task_cards/source_material/foundation-model-surrogate-targets-v1.md` - Brown et al., 2015, sarcopenia and mortality: - Abramowitz et al., 2018, muscle mass, BMI, and mortality: - Reicher et al. (2024), phenome-wide associations of human ageing, *Nature Aging*, . Local copy in **HPP documentation**: `assets/papers/hpp/reicher_2024_phenome_wide_associations_of_human_aging/`. - Gabet et al. (2026), HPP gait foundation model, . - Moore et al. (2018), metabolomic profiles of muscle mass and strength, . ## Open Questions - Should appendicular lean mass or ALMI become the primary muscle target? - Which DXA QC fields should be enforced before leaderboard rows are trusted?